Policy Research Working Paper                              10746




Do Patients Value High-Quality Medical Care?
        Experimental Evidence from Malaria Diagnosis
                       and Treatment

                             Carolina Lopez
                             Anja Sautmann
                             Simone Schaner




Development Economics                         A verified reproducibility package for this paper is
Development Research Group                    available at http://reproducibility.worldbank.org,

April 2024                                    click here for direct access.
Policy Research Working Paper 10746


  Abstract
  Can better information on the value of diagnostic tests                                  care they received, driven by those whose prior beliefs did
  improve adoption and help patients recognize higher qual-                                not match their malaria status. The patient information
  ity of care? In a randomized experiment in public clinics in                             intervention reduced malaria testing and did not improve
  Mali, providers and patients received tailored information                               treatment outcomes or patient satisfaction. These findings
  about the importance of rapid diagnostic tests (RDT) for                                 are consistent with highly persistent patient beliefs and
  malaria. The provider training increased reliance on RDTs                                distrust of the promoted diagnostic technology, which
  and improved the match between a patient’s malaria status                                translate into low demand and limit patients’ ability to
  and treatment with antimalarials by 15–30 percent. None-                                 recognize improved quality of care.
  theless, patients were significantly less satisfied with the




 This paper is a product of the Development Research Group, Development Economics. It is part of a larger effort by the
 World Bank to provide open access to its research and make a contribution to development policy discussions around the
 world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be
 contacted at carolina_lopez@worldbank.org, asautmann@worldbank.org, and schaner@usc.edu. A verified reproducibility
 package for this paper is available at http://reproducibility.worldbank.org, click here for direct access.




                                                                                         RESEA
                                                                                    CY
                                                                               LI                R
                                                                                                 CH
                                                                          PO




                                                                    TRANSPARENT
                                                                                ANALYSIS
                                                                                                     S
                                                                           W




                                                                                                 R




                                                                               R                     E
                                                                            O




                                                                                    KI           P
                                                                                         NG PA




         The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development
         issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the
         names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those
         of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and
         its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.


                                                       Produced by the Research Support Team
         Do Patients Value High-Quality Medical Care?
       Experimental Evidence from Malaria Diagnosis and
                          Treatment
              Carolina Lopez               Anja Sautmann                 Simone Schaner∗




         Originally published in the Policy Research Working Paper Series on April 2024.
         This version is updated on November 2024.
         To obtain the originally published version, please email prwp@worldbank.org.




        JEL Codes: I11, I12, O15 – Keywords: malaria treatment, demand for medication,
        patient satisfaction, patients vs. provider beliefs, Mali


   ∗
    We thank the staff of IPA Mali and especially Alassane Koulibaly for expert field coordination. This
project would not have been possible without the advice and support of Dr. Seydou Doumbia and Dr.
Issaka Sagara (Malaria Research and Training Center, University of Bamako) as well as Dr. Seydou
Fomba (Programme National de Lutte contre le Paludisme (PNLP)). We thank the Ministry of Health
and the participating community clinics for allowing us to conduct this research. The study was funded
by ESRC/DFID Development Frontiers Award ES/N00583X/1. The research protocol was approved by the
IRB of La Faculté de Médecine de Pharmacie et d’Odontostomatologie de Bamako (FMPOS) (protocol No.
2016/47/CE/FMPOS) and registered in the AEA RCT registry (Sautmann and Schaner, 2016). The find-
ings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not
necessarily represent the views of the World Bank and its affiliated organizations, or those of the Executive
Directors of the World Bank or the governments they represent. All errors are our own. Lopez: World Bank,
carolina_lopez@worldbank.org, Sautmann: World Bank and BREAD, asautmann@worldbank.org, Schaner:
University of Southern California, BREAD, and NBER, schaner@usc.edu.
1       Introduction
Health systems in many low-income countries struggle to provide quality care to citizens,
resulting in lost life years, human suffering, and reduced economic productivity (Das and
Hammer, 2014; Bariş et al., 2022). A fundamental challenge is the high rate of misdiagnosis
and misallocation of medical treatment. Drugs like antimalarials and antibiotics are lifesaving
for patients in need, and too many cases remain untreated (e.g. Macarayan et al., 2020); but
at the same time, health workers distribute an alarming share of medications to patients that
do not match a medical need (see for example Busfield, 2015; Brownlee et al., 2017). This
wastes resources, negatively affects patient health due to adverse drug interactions and side
effects, and can promote emergence of drug-resistant pathogens, which put future patients’
lives at risk (World Health Organization, 2022).
    In this paper, we evaluate the effects of patient- and provider-side information treatments
promoting diagnostic testing on treatment outcomes and patient satisfaction. Diagnostic
testing has the potential to reduce misallocation by providing credible information on a
patient’s cause of illness and reducing frictions that arise because patients cannot assess
the quality of the care they receive (Dulleck and Kerschbamer, 2006). By creating common
knowledge between provider and patient about whether a treatment is needed, diagnostic
testing may leave less room for information asymmetries, lack of trust, and mis-aligned
incentives that can lead providers to prescribe wrong or unnecessary drugs. Reducing this
form of misallocation may in turn improve quality of care by accelerating learning and
adoption of effective treatments (Adhvaryu, 2014).
    However, the slow adoption of accurate diagnostic tools remains a pervasive challenge
in health care (Baker, 2001; Fleming et al., 2021). Many efforts in low- and middle-income
countries have focused on education and training interventions for health care providers.
While systematic reviews indicate that provider-targeted training can improve patient out-
comes, the overall quality of evidence remains low, and there are still relatively few studies
that directly address adherence to test results (Rowe et al., 2018, 2021). Moreover, most such
studies do not consider patient responses. But health workers do not operate in a vacuum:
care outcomes are a product of interactions between providers and patients. If patients do
not value diagnostic testing, providers may have limited incentives to routinely verify their
diagnoses with medical tests.1 The empirical evidence on this channel is limited, in part
because identification requires downstream data on provider behavior, care outcomes, and
patient satisfaction.

    1
     For a review documenting the inconsistent linkages between patient satisfaction and quality of care, see
Farley et al. (2014).


                                                     1
    Our paper tackles this gap in the context of malaria, where misallocation is a major
problem. In our study setting in urban Mali, around 60 percent of malaria-negative patients
at public health clinics received an antimalarial prescription.2 Market forces do not appear to
reward better malaria care – neither price nor patient satisfaction are positively correlated
with better malaria treatment outcomes. Moreover, even though rapid diagnostic tests
(RDTs) for malaria are accurate, easy to administer, and available for free in the public sector,
adoption and adherence are poor: in our sample, over one-third of patients prescribed an
antimalarial were not tested, and around two-thirds of RDT-tested malaria negative patients
were prescribed an antimalarial.
    Our analysis uses data from a randomized controlled trial we conducted with 58 public
health clinics in the greater Bamako region. We collaborated with Mali’s National Malaria
Control Program to design two cross-randomized interventions that provided information
about RDTs to health workers and patients, respectively. Half of the clinics were randomly
selected for a training intervention with providers, delivered as an extension to a basic
training on malaria testing and policy that all clinics received. The extended training was
designed to increase the use of RDTs as a high-quality and low-cost diagnostic technology
and to improve adherence to test results. It emphasized the low error rate of rapid detection
tests in both laboratory and field settings and the dangers of overusing antimalarials. The
patient-side intervention was randomized at the clinic×day level and was delivered as a video,
shown on a tablet, on the importance of malaria testing and how to read an RDT result,
recorded in the local language. The objective was to align patients’ understanding of the
use RDTs with recommended medical practice and to empower them to resist unnecessary
treatment.
    We collected detailed data from health workers and patients to connect quality of care
with beliefs, knowledge, and patient satisfaction. We elicited provider beliefs about the
reliability of malaria RDTs before and after the training intervention. Using a detailed
patient survey at the clinic, we captured pre-consultation beliefs about malaria status as
well as malaria testing and treatment outcomes. We also conducted a post-consultation
survey visit at patients’ homes. At this visit we administered a malaria RDT to observe
their true malaria status and measured their ex-ante beliefs about the outcome of this test.
Finally, we recorded patient satisfaction with the provider’s testing and treatment decisions.
    We find that under the status quo, providers incorrectly believed that RDTs had limited
ability to detect mild malaria cases and relied too strongly on more time- and skill-intensive
(and therefore less accurate in practice) microscopy-based tests, sometimes carrying out

   2
    Other studies in Sub-Saharan Africa find similarly high rates of overprescription (see for example Rey-
burn et al., 2004; Hamer et al., 2007; Bisoffi et al., 2009; Ansah et al., 2010).


                                                    2
multiple tests. From qualitative observation, this often reflects cases where a provider first
conducts an RDT and, if the RDT is negative, seeks confirmation via microscopy. In the
control group, we estimate that only 41-44 percent of patients received treatment recom-
mendations for malaria that matched their underlying illness status (i.e. an antimalarial
prescription if malaria positive, no prescription if negative).
    The provider information intervention was successful in shifting beliefs about the accuracy
of RDTs and essentially closed the perceived effectiveness gap between RDTs and microscopy-
based tests. As a result, use of RDTs as the sole tool for malaria diagnosis increased by 12
percentage points (55 percent), while use of microscopy alone declined by 9 percentage points
(37 percent, not significant), and double-testing declined by 5 percentage points (68 percent).
These changes improved the allocation of malaria treatment by 7-13 percentage points (15-30
percent).3
    Despite significant improvements in the quality of care in the form of better allocation
of antimalarials, the provider information treatment decreased patient satisfaction by 0.09
standard deviation units, with point estimates indicating displeasure with both testing and
medications dispensed at the clinic. What is driving this effect? We find no evidence that
the training affected other aspects of care, including use of antibiotics and other medications.
Several pieces of evidence indicate that patients do not value RDT testing. First, the patient
information treatment – with the main message to “take an RDT and only treat if positive,”
in line with national malaria policies – did not increase malaria testing or adherence to test
results. Instead, use of malaria tests in the treatment group is 6 percentage points lower
than in the control (1% significance level), with no change in the allocation of antimalarials.
Second, the negative effect of the provider information treatment on patient satisfaction is
entirely driven by patients who were ex-ante misinformed about their malaria status and
therefore did not receive the treatment they expected. This is despite the fact that patients
in this treatment arm are better able to predict the outcome of the home malaria RDT.
    These findings suggest that, even though 49 percent of patients came to the clinic with
incorrect beliefs about their malaria status, there was no unrealized demand for more in-
formation or for basing treatment on diagnostic results. Unwelcome prescription outcomes
after more frequent RDT testing seem to have driven decreased patient satisfaction, per-
haps because RDT results at the clinic did not significantly shift patients’ beliefs about
their underlying malaria status. In other words, adoption of the technology by providers did
not succeed in creating common knowledge and reducing information asymmetries. Instead,

   3
     These effects are not driven by simply reminding providers of best-practice treatment guidelines. Before
the study, providers at all clinics received a short training emphasizing that they should only dispense
antimalarials to patients with a positive test.


                                                     3
the results suggest a highly persistent lack of trust in the technology on the patient side.
As a result, patients’ and providers’ preferences over the allocation of malaria drugs were
paradoxically less aligned as RDT adoption increased.
    Our findings build on the large literature on the drivers of health care quality, which
distinguishes shortfalls in provider knowledge and provider effort.4 This literature highlights
that knowledge deficits coexist with substantial gaps between what providers know to do,
and what they actually do during patient consultations (see de Walque et al., 2022, for
a summary of the framework and evidence).5 Our paper contributes to this literature by
showing that despite the know-do gap, a relatively low-cost information intervention can
improve providers’ beliefs in the value of a diagnostic technology and induce both behavior
change and better care. However, we also show that patients do not reward this improvement
in quality, and we identify persistent misaligned beliefs about the underlying cause of illness
as a mechanism: a focused one-on-one information intervention had no effect on patients’
trust in the technology.
    We also contribute to the ongoing debate regarding the role of regulation and the private
sector in health care markets, which dates back to Arrow (1963) and Friedman and Friedman
(1962):6 in settings where patients do not reward or even punish marginal improvements in
quality of care, competitive pressure may have perverse effects on market outcomes. Our
results suggest that absent improvements in patient perception, closing provider knowledge
gaps may in the longer run only contribute to a larger effort gap, without actually improving
care. Indeed, Banerjee et al. (2020) argue that providers may rationally and intentionally
misallocate treatment in settings where patients have low confidence in providers’ ability to
deliver quality care, and instead make inferences regarding provider quality based on prior
beliefs and the outcome of the consultation.
    The rest of the paper is structured as follows: in section 2 we provide background on
health care and malaria treatment in Mali and present evidence of status quo health provider
beliefs and knowledge as well as current treatment practices. Then, section 3 describes the
experimental design, section 4 presents our empirical results, and section 5 concludes.




   4
      Together with the lack in capacity, these deficits are sometimes termed the “three gaps” (Ibnat et al.,
2019).
    5
      Research studying mechanisms driving this gap have focused on financial incentives (Iizuka, 2012; Currie
et al., 2014; Das et al., 2016; Lagarde and Blaauw, 2022), demand-side pressure from patients Kravitz et al.
(2005); Currie et al. (2014); Lopez et al. (2022), and issues related to low motivation and effort among
providers (Banerjee et al., 2008; Das et al., 2008; Das and Hammer, 2014).
    6
      For more recent work, see, e.g. Chandra et al. (2016) and Skinner (2011) for higher-income country
settings and Das et al. (2016) and Banerjee et al. (2020) for lower-income country settings.


                                                     4
2       Background
Health Care and Status Quo Malaria Treatment in Mali. Mali’s public health care
sector is organized after the model of community-funded public health care laid out in the
Bamako Initiative, launched in 1987.7 For most patients in the public sector, the first points
of contact are community health clinics or centres de santé communautaires (CSComs),
which are run by local health associations and charge moderate fees to cover costs. Almost all
of these community clinics also have a pharmacy that supplies basic medications.8 CSComs
are an essential pillar of primary care. In our study area around Bamako city, a typical
CSCom has 1-2 physicians and around 10 other staff who can administer malaria treatment.
    Mali is among the 10 countries with the most malaria cases in Africa (World Health Or-
ganization et al., 2021), but in urban areas such as Bamako, the incidence is comparatively
low: in 2018, only 2.9 percent of children 6-59 months old tested positive for malaria, com-
pared to 18.9 percent countrywide (Koenker et al., 2020). Most malaria infections present as
“simple” or “uncomplicated” malaria, characterized by flu-like symptoms such as fever, chills,
body aches, and fatigue. However, if left untreated, simple malaria cases can progress to a
more severe stage characterized by life-threatening complications, with symptoms ranging
from convulsions to coma and organ failure (Trampuz et al., 2003). Prompt treatment of
simple malaria is therefore a cornerstone of good care.
    Mali’s national malaria policy states that all suspected malaria cases must first be tested
via microscopy or RDT, with antimalarials only given to those who test positive (Ministère de
la Santé, 2013).9 This conforms with World Health Organization (WHO) recommendations
formulated to counteract the rise of drug-resistant malaria parasites, which have rendered
several classes of antimalarials ineffective across much of Sub-Saharan Africa and Asia (World
Health Organization, 2014a; Arrow et al., 2004). Over and above the risk of fostering drug-
resistant malaria strains, overtreatment with antimalarials wastes resources and may cause
medical harm by delaying the patient from receiving the correct treatment and potentially
triggering side effects.
    Public health facilities are expected to offer free RDTs to all patients. The RDT brand

    7
      In early 2019, the Malian Ministry of Health announced plans to expand health care access and move
towards universal health care by removing user fees and expanding the network of community health workers
by 2022, but these reforms have been stalled by political instability and the COVID-19 crisis (Adepoju, 2019).
    8
      The local health association collects revenues from the sale of medications and other user fees (including
consultation fees) to support the operations of the CSCom. For instance, in our sample, 87 percent of the
clinics use pharmacy sales to fund staff salaries, but provider remuneration does not directly depend on
revenue generation.
    9
      Mali’s testing requirement for malaria was first introduced in its five-year strategic plan for 2007-2011.
Previously, the primary approach to malaria control included presumptive treatment of any fever cases (Koné
et al., 2015).


                                                       5
adopted by the Malian government, SD Bioline, shows zero false positives in WHO tests, a
detection rate of over 90% for low parasite loads, and an almost 100 percent detection rate
for high parasite loads, which accompany severe malaria cases (World Health Organization,
2014b). Despite the wide availability of RDTs, many clinics have blood test laboratories and
use microscopes to perform blood smear tests on site. In our study area, RDTs were free
of charge 70 percent of the time, while microscopy tests were free less than 3 percent of the
time.
    The relative performance of RDTs and microscopy differ depending on the setting. While
microscopy tests are typically considered a “gold standard” malaria test (Lee et al., 2002;
Trampuz et al., 2003), they require high-quality equipment and skilled technicians who
visually identify malaria parasites on blood slides. The advantages of microscopy include
the ability to assess parasite load and differentiate between types of malaria parasites. A
primary disadvantage is variable performance in resource-constrained settings, which can
result in misdiagnosis of malaria, including false positives (Wongsrichanalai et al., 2007;
Berzosa et al., 2018). By comparison, RDTs are much simpler to administer (apart from a
finger prick, procedures are similar to those required for a COVID-19 rapid test) and can be
performed by trained laypeople..10 RDTs test for antigens that remain in the bloodstream
for some time after a malaria infection has cleared (Abba et al., 2011). We take advantage
of this property in this study because it allows us to test for malaria even after treatment
was potentially started.
Health Provider Beliefs in the Study Sample. To describe status quo provider at-
titudes, Appendix Table A1 reports average beliefs about malaria prevalence, confidence in
RDTs, and confidence in microscopy tests among health providers in the group of clinics that
did not receive training on RDT accuracy.11 Panel A shows that providers think malaria
prevalence is very high, afflicting 28 percent of the general population and 44 percent of
patients at their clinic. Alongside, 51 percent of the workers report feeling pressure from
patients to prescribe unnecessary medications, with 68 percent of those workers saying that
most of the pressure they receive is to prescribe antimalarials.
    Panel B summarizes confidence in RDTs among providers. Ninety-eight percent of
providers use RDTs to diagnose malaria, but they believe that only 56 percent of patients
with simple malaria will test positive. In line with this concern, 47 percent of providers
report they would give an antimalarial to a sick relative despite a negative RDT. Providers
  10
      A limitation of some RDT brands is that they can detect only p. falciparum malaria. P. falciparum is
the most dangerous malaria parasite and accounts for more than 85 percent of infections in Mali (Cissoko
et al., 2022).
   11
      These data come from a post-intervention survey that we conducted with health providers at the end
of the study (see a detailed description in subsection 3.3).


                                                    6
put significantly more faith in RDTs to diagnose severe malaria, reporting that 89 out of
100 such patients will test positive. These results coincide with our qualitative observation
that doctors believe RDTs become more effective as the parasite load in the blood increases.
While this belief is founded in fact, RDTs perform very well when presented with parasite
loads typical of simple malaria (World Health Organization, 2014b).
   Finally, Panel C shows that 70 percent of providers use microscopy tests to detect malaria.
They trust this method more; on average, they think 84 and 95 out of 100 patients with
simple and severe malaria respectively will have a positive microscopy result. Consequently,
only 33 percent of providers would treat a relative with a negative microscopy test for malaria.
Malaria Testing and Treatment in the Study Sample. Health workers’ beliefs indi-
cate widespread skepticism of tests – especially RDTs – coupled with inflated beliefs regarding
malaria prevalence. To understand how these beliefs translate into patient care, we use data
from surveys conducted with patients visiting our study clinics. We focus on two sets of out-
comes: patients’ reports of whether they received a malaria test at the clinic and whether
they were prescribed an antimalarial, and patients’ “true” malaria status, which we measure
via follow-up RDTs conducted by our research team at patients’ homes, 1-2 days after they
visited the clinic (see subsection 3.3 for more detail). To capture status quo outcomes, we
limit attention to patients visiting clinics that were not trained on RDT accuracy.
    Figure 1 studies how the match between the underlying cause of illness (malaria positivity
per our home follow-up tests) and treatment (antimalarial prescriptions) varies with in-
clinic testing. Panel A shows that compared to those who received RDTs at the clinic,
patients tested via microscopy are less likely to be malaria positive yet more likely to receive
antimalarials. As a result, the darkest bar in Panel B shows that the match between malaria
status and malaria treatment is worst in the microscopy group and best in the group of
patients who received an RDT only. Yet even in this group, the match remains under 50
percent. The lighter bars in Figure 1, Panel B report the match rate for malaria negative
patients (middle gray bar) and malaria positive patients (light gray bar). Here we see that
the poorer performance of microscopy patients is largely driven by overtreatment – roughly
a third of malaria negative patients in the RDT-only group do not receive an antimalarial
prescription, while just 17 percent of malaria negative patients in the microscopy group
receive the same. These patterns suggest two issues are at play: first, providers may ignore
test results when writing prescriptions, either due to patient pressure (Lopez et al., 2022)
or due to their own misgivings regarding the test results – especially in the case of RDTs.
Second, lab technician errors (unintended or due to poor infrastructure/supplies) may lead
to false positives (poor specificity) and, therefore, overtreatment among malaria negative
patients tested via microscopy.

                                               7
3        Experimental Design
3.1      Sample
Our experiment was carried out in a set of public community clinics (CSComs) in the greater
Bamako area. We obtained an administrative list of all clinics in the area, dropping those
that had closed or were more than 15 kilometers away from Bamako city. We also excluded 21
clinics that were working with a local NGO to offer subsidized care to mothers and children.
This left us with a sample of 60 clinics, which we divided into three geographical clusters.
Within each cluster, we formed matched pairs based on the average number of patients per
day. The matched pairs serve as strata for the provider-centered RDT training, described in
more detail below.12
Pre-Study Training. Before the study started, four health workers at each participating
clinic were invited to attend a refresher training that covered Mali’s official malaria diagnosis
and treatment guidelines. Malian doctors conducted all training via our partnership with
the University of Bamako, the Malaria Research and Training Centre, and Mali’s National
Malaria Control Program. Key points covered included: (i) all suspected malaria cases
should receive a diagnostic test; (ii) microscopy is the “gold standard” test and should be
used when available; (iii) RDTs should be used when microscopy is not available or cannot
deliver a result within two hours; (iv) symptoms and recommended treatment for simple and
severe malaria; (v) procedures for microscopy and rapid diagnostic tests for malaria; (vi) a
hands-on training on how to administer an RDT.
    At the end of the training, participants were informed of upcoming study activities –
namely that research staff would be visiting the clinic during a future two-week period, and
that on some days, a separate set of study staff would offer vouchers for free malaria treatment
and/or information for patients on RDTs (see next sub-section on the study interventions).
Health workers were not told what days these interventions would be offered.
    Our analysis sample drops two clinics in one stratum where one of the clinics had to be
replaced after the training period ended, leaving a final sample of 58 clinics in 29 strata.13

3.2      Interventions
Our full experimental design included four cross-randomized treatments. In this paper, we
focus on the effects of two information treatments on the quality of malaria care and patient

    12
      We could not collect census data for one clinic; we assigned the median caseload to this clinic before the
randomization. This and all other random assignment was conducted in-office using Stata’s random number
generator.
   13
      The clinic in question was replaced because we learned it was hosting other interventions unrelated to
our study.


                                                       8
satisfaction. Below, we describe each treatment arm.
Extended Training Treatment. We randomly assigned one clinic in each matched pair
to receive more intensive pre-study RDT training. This “extended training” was designed to
increase providers’ trust in the diagnostic accuracy of RDTs and adherence to test results.
Another aim was to empower providers with evidence in case patients pushed to get unneces-
sary medication. The treatment was administered as part of the pre-study training described
above, following the core “basic training” which every clinic received. The content covered
the high sensitivity (and specificity) of rapid diagnostic tests available at clinics, both in
WHO quality assurance testing and in field studies conducted in Mali.14 The training also
reviewed the dangers of unnecessary treatment with antimalarials, from drug interactions to
missed diagnoses to parasite resistance. The objective was to improve provider reliance on
RDT results and to heighten awareness of the problem of overtreatment.
Patient Information Treatment. The “patient information” intervention was randomized
within clinic and consisted of a seven-minute video shown to patients individually while
they were waiting to consult a health provider. The video depicted dialogue between a
doctor and a mother visiting a clinic with her child to explain the symptoms of malaria, the
use of an RDT, the importance of testing for malaria before treating, and the risks (and
unnecessary costs) associated with treatment for severe malaria in cases where the patient
is not at high risk. The main message was that antimalarials should only be prescribed
to individuals with positive tests. The video also demonstrated how to interpret RDT
results. The video’s objective was to teach patients about Mali’s malaria diagnosis and
treatment policies while empowering them to ask questions about treatments they do not
understand. The empowerment component was designed to help patients in case a health
provider prescribes unnecessary medication.
    We have records from surveyors whether the video was watched as well as administra-
tive notes on the length of time that was watched. 89 percent of patients for whom we
have administrative data watched the full video. Combining this with the surveyor records,
we estimate that 79.5% of patients watched at least some of the video (Appendix Table
A2). While this one-on-one intervention took some time, we see no statistically significant
difference in self-reported waiting time prior to seeing the health care provider among pa-
tients in the patient information arm versus the control, suggesting intervention officers were
successful at administering the treatment during normal wait times.15
Voucher Treatments. The patient information treatment was cross-cut with two other
  14
      The training specifically addressed the performance of SD Bioline tests, the brand purchased and dis-
tributed by the government for free distribution in the public health system (Djimde et al., 2016).
   15
      The average waiting time before consultation is 17.328 minutes in the control group and 17.994 minutes
in the patient information group.


                                                     9
treatments randomized within clinic, described and analyzed in detail in Lopez et al. (2022).
In both, patients were eligible to receive a free course of antimalarials as part of simple
malaria treatment. This was operationalized via vouchers which could – with provider pre-
scription and certification – be redeemed for free antimalarials at the clinic pharmacy. The
voucher reduced the cost of treatment, leaving both the revenue to the clinic and the cost of
other types of treatment to the patient fixed. We provide a summary of the main results in
Appendix C.
    Trained “intervention officers” oversaw the voucher distribution and showed the patient
information videos to patients. Intervention officers did not perform any survey activities
and were stationed in a separate part of the clinic from the data collection staff. Patients
did not need to participate in any survey or other study activities to be eligible to receive
the voucher or information intervention.

3.3       Data Collection
Study field activities were conducted in November and December 2016, covering the end of
the rainy season and, therefore, the period of highest malaria risk. Our primary analysis
uses data from the following sources.
Clinic Census. When building the sampling frame, we conducted a short survey of can-
didate clinics, which captured information on total reported caseloads, malaria caseloads,
clinic testing, and pharmacy capabilities. This information was used to conduct the ran-
domization.16
Health Workers: Pre- and Post-Training Tests. As part of the training for health
workers, we administered a pre-training and post-training test to all health care providers
in attendance. The test included questions on providers’ knowledge of topics covered in
the basic training (e.g., recommended malaria treatments, symptoms of severe malaria).
The post-training test additionally included topics covered only in the “extended training”
treatment (e.g., sensitivity and specificity of RDTs).
Patients: In-Clinic Survey. The geography-based clinic cohorts rotated through two
weeks of data collection and experimental intervention during the study period. Within each
cohort, we randomly assigned each clinic to one of the 20 intervention schedules depicted in
Figure 2. Intervention delivery and in-clinic survey of patients occurred on 6 days during a
two-week study period. Each clinic received one control day (no within-clinic interventions),
one day with patient information only, two days with voucher interventions only, and two
days with both patient information and vouchers. The clinic staff was not informed of
the intervention schedule in advance; instead, intervention officers communicated the day’s

  16
       One clinic in the sample was not surveyed during the census and is missing information.


                                                     10
interventions at the beginning of each day.
    Enumerators attempted to interview all acutely ill patients with malaria-like symptoms
at clinic intake on these observation days. We classified a patient as “acutely ill” if they were
feeling sick (neither preventive care nor follow-up visit for earlier treatment) and exhibited
any of the following symptoms: fever, chills, excessive sweating, nausea, vomiting, diarrhea,
poor appetite, headache, cough, weakness, fatigue, or reduced consciousness. We interviewed
the patient (or a caretaker, in the case of children or the very sick) both before and after
meeting with a provider. The survey covered demographic characteristics, ex-ante percep-
tions about their condition, symptoms, any prior treatment and/or diagnosis, medications
prescribed and purchased, and blood tests taken at the clinic. Finally, the survey collected
data on the price of consultation and treatment.
Patients: Home Follow-Up Survey. In order to collect independent data on patients’
underlying malaria status and satisfaction, we conducted a follow-up survey of patients at
their homes. This survey targeted a random subset of 1,669 patients who participated in the
in-clinic survey. In addition to conducting an RDT to learn the true malaria status of the
patient and asking a series of questions designed to measure satisfaction with care received
at the clinic, the survey also collected data on any additional medications or tests taken after
the clinic visit. Appendix Table A3 shows that the probability of taking the home survey
and home-based RDT are uncorrelated with the extended training and patient information
treatments. It also shows that 87 percent of patients selected for the home survey were
successfully interviewed, and 64 percent took a home-based RDT.
Health Workers: Endline Survey. We selected up to three care providers per clinic for
a post-intervention endline survey. We interviewed the clinic director and randomly selected
one other doctor and one other care provider for an interview, including nurses, health
technicians, and midwives (subject to staffing). The survey included questions on providers
beliefs about the performance of RDTs and microscopy tests, perceived patient knowledge,
patient requests for medications, and personal preferences regarding malaria diagnosis and
treatment.
Timeline. The health worker trainings were conducted on November 2-4, 2016. In-clinic
data collection ran from November 14 to December 30, 2016, with the provider endline fol-
lowing shortly thereafter, from December 10, 2016, to January 6, 2017. Thus, our experiment
can only capture the short-run (0-2 month) effect of the extended provider training.

3.4    Predicting Malaria Risk
One of our primary research aims is to understand how improving information about RDTs
impacts the quality of care at clinics. While home-based RDTs give a very precise signal


                                              11
of a patient’s underlying malaria status, test results are not available for those who did not
participate in the home survey or did not consent to take the RDT. To address this, we
follow Cohen et al. (2015) and Lopez et al. (2022) and use the home-tested sample to predict
malaria positivity using patient symptoms and demographics. Specifically, we use probit
regression to estimate the following:

                                      E [posict | xict ] = Φ xict λ                                         (1)

where posict is a dummy variable identifying those who test RDT positive and xict includes
dummy variables for symptoms listed in Appendix Table A4: days since onset of illness,
patient age, a dummy equal to one if the patient is under age 5, the interaction between age
and the under 5 dummy, patient gender, and patient pregnancy status.17 Appendix Table
A5 reports results.18 When analyzing the allocation of malaria treatment for the full sample
                                          ˆ ict as a measure of each patient’s malaria risk,
of patients, we use the predicted value pos
including those who were not tested at home.

3.5     Randomization Verification
Table 1 verifies that clinic characteristics and provider knowledge are balanced across the
extended and basic training groups. Column 1 of the table reports the average value in the
control group, and column 2 reports the difference between the two groups, conditional on
randomization strata (clinic pair) fixed effects. Throughout the paper we cluster standard
errors at the clinic-pair level (de Chaisemartin and Ramirez-Cuellar, 2024).
    Overall, the extended training randomization was balanced, with no treatment-control
differences significant at conventional levels. Point estimates do, however, suggest that
providers in the extended training arm had greater faith in RDTs at pre-test than their
peers in the basic training arm. Extended-training clinics were also 11 percentage points less
likely to have a lab capable of performing microscopy tests. These differences could bias us
toward finding more use of RDTs in the extended training arm – to address this risk while
tying our hands in terms of covariate selection, we use the post-double-lasso procedure by
Belloni et al. (2014) to select covariates throughout our main analysis.19

   17
      We also control for demographic characteristics that may correlate with malaria risk, including the
survey respondent’s ethnicity, ability to speak French, literacy in French, education, and a dummy variable
indicating cases where the patient and respondent are different people. Our results are similar if we omit
these demographic characteristics from the specification.
   18
      This regression specification matches that in Lopez et al. (2022), but for this paper, we only include
observations from the 58 clinics included in our analysis sample.
   19
      We include clinic averages of all covariates listed in Table 1 and, for patient-level regressions, charac-
teristics considered for double lasso selection in Lopez et al. (2022): all individual-level characteristics in
Table A4 (excluding predicted malaria risk and characteristics only measured in the home survey) as well as


                                                      12
    Appendix Table A4 uses data from the in-clinic survey to assess whether patient volumes
and patient characteristics are balanced across the patient information and extended training
treatment groups. Each row represents a single regression, where we regress the outcome
of interest on an indicator for clinic-days where the patient information intervention was
in place, and an indicator for clinics selected for extended training. All regressions include
strata and survey date fixed effects.
    On average, enumerators interviewed 6 eligible acutely ill patients per observation day,
with no significant differences across treatment arms. The average patient suspects malaria
59 percent of the time, presents with 3.4 symptoms, is 17 years old, and is malaria positive
(per RDTs administered during the home survey) 22 percent of the time. Panel B shows that
most patient characteristics are balanced across treatment, though patients in the patient
information arm are less likely to report fever, slightly older, and less likely to be pregnant;
those in the extended training arm are more likely to present with nausea/vomiting/diarrhea
and had been sick for slightly longer. Importantly, malaria positivity – both measured via
RDTs and predicted based on symptoms – is balanced across treatment arms.
    Panel C studies balance in terms of survey respondent and household characteristics.
Those randomly selected for patient information are less likely to report Bambara as their
ethnic group and more likely to be literate (and consequently less likely to have a primary
school education or less); those visiting clinics selected for the extended training are less
likely to be literate, have less education, and belong to slightly bigger households. Given our
design, we cannot rule out the possibility that some of these differences are due to selection.
To address concerns regarding balance, our main results use double lasso to select covariates
as described in footnote 19. As a robustness check, Appendix B reports results without
additional controls. Overall, our findings are very similar – we therefore believe it is unlikely
that balance issues affect our main conclusions.

4     Results
4.1     Empirical Approach
We begin our analysis by documenting the impact of the extended training treatment on
health providers’ beliefs about the accuracy of RDT and microscopy tests. Next, we study
how this treatment – and the patient information intervention – impacted malaria testing
at the clinic, the allocation of malaria treatment, and patient satisfaction.

the square of patient age and illness duration. We also include pairwise interactions of the aforementioned
characteristics. Missing values for all characteristics are dummied out and recoded to the mean prior to
forming interactions. Missing dummies are also included in the potential covariate set and set of pairwise
interactions.


                                                    13
   We use the following regression to study the impacts of our treatments on patient out-
comes:
                    yict = β0 + βE ETc + βP P Ict + xitc α + γc + δt + εic            (2)

where yict is the outcome of interest for patient i visiting clinic c on date t, ETc is a dummy
variable indicating clinic c was selected for the extended training intervention, P Ict identifies
clinic-days with the patient information intervention, xitc is a vector of individual and clinic-
level covariates selected by double lasso, γc are strata (clinic-pair) fixed effects and δt are
date fixed effects. When studying effects of the extended training intervention on provider
outcomes, for which we have only one post-treatment observation, we omit the patient
information dummy and the date fixed effects. In all cases, we cluster standard errors
at the clinic-pair (strata) level (following de Chaisemartin and Ramirez-Cuellar, 2024).

4.2    Effects of Extended Training on Provider Trust in RDTs
Table 2 evaluates the impact of the extended training intervention on provider beliefs about
the diagnostic capability of RDTs and microscopy tests.
    Panel A uses the knowledge test carried out with participants right after the training to
assess short-term effects. As discussed in section 2, the averages from the basic-training group
indicate that trust in the sensitivity of RDTs is low at the outset: participants think that an
RDT will detect simple malaria only in 60 percent of cases, compared to 85 percent of cases
for a microscopy test. The extended training leads to large, significant growth in confidence
in RDTs: providers who received extended training estimate that 80 percent of patients with
simple malaria will test positive with an RDT, a 20 percentage point increase (column 2).
We also see a 7.1 percentage point increase in providers’ estimate of the sensitivity of an RDT
for severe malaria patients. Column 4 shows no significant change in the estimated share
of malaria negative patients who would test RDT negative. Column 1 reports treatment
effects on an index that averages standardized versions of the outcomes in columns 2-4,
following Kling et al. (2007). We estimate a 0.31 standard deviation unit increase in this
index, significant at the 1 percent level.
    Panel B uses data from the provider endline survey, carried out after in-clinic data col-
lection, to assess whether shifts in beliefs persisted over a longer term. To make consistent
comparisons vis-a-vis Panel A, we limit the sample to providers who attended the train-
ing. Control group means and treatment effects for RDT-related beliefs are similar to those
in Panel A, suggesting that increased trust in RDTs was durable over the course of the
experiment.
    The extended training did not provide information on microscopy tests. In line with
this, we find no systematic positive effects on beliefs about microscopy. Inspection of point

                                               14
estimates shows that the extended training roughly equalized beliefs about the performance
of the two diagnostic testing technologies.

4.3     Impacts on Testing and Treatment Allocation
We now ask whether our treatments induced changes in how malaria tests are used and who
received malaria treatment. We begin by examining the impacts of the extended training
for providers.
Effects of the Extended Training. Table 3 shows treatment effects on providers’ use
of malaria diagnostics. In the basic training arm, 58 percent of patients received a malaria
test. While the extended training had no significant effect on overall rates of malaria testing,
we do find evidence of significant reallocation from microscopy testing to RDT testing only
– patients are 11.6 percentage points more likely to report receiving just an RDT test and
4.8 percentage points less likely to report receiving multiple malaria tests (from 7 percent in
the basic training arm). While not significant, point estimates also indicate a 9 percentage
point decline in use of microscopy alone.
     Next, we consider the impact of extended training on the allocation of malaria treatment.
Quality of care is higher if (i) a higher proportion of patients with malaria receive antimalar-
ials and (ii) a lower proportion of patients without malaria do not receive antimalarials; that
is, if the “match” between treatment and illness improves. To measure this, we need to link
receipt of an antimalarial prescription to patients’ underlying malaria status. As discussed
in section 3.4, measuring malaria status is straightforward for the home-tested subsample
because we have access to actual RDT results. For the full sample, we use predicted malaria
       ˆ ict based on symptoms and demographics.
risk pos
     We explore treatment effects on misallocation graphically in Figure 3. Panel A shows
results of local linear regressions where the outcome is whether the patient received an
antimalarial prescription and the running variable is the patient’s predicted malaria risk.
Confidence intervals are based on bootstrapped standard errors clustered at the clinic-pair
level.20 The graph shows that prescription rates were lower in the extended training group
up to the 75th percentile of predicted risk; at the highest levels of predicted risk, prescription
rates in the extended training group exceeded those in the basic training group. Panels B
and C compare patterns among patients with and without an RDT test at the clinic. While
we interpret these results with caution, since malaria testing decisions are endogenous to
treatment, the graphs indicate that allocative differences are more pronounced among RDT-
tested patients, consistent with the interpretation that the intervention increased adherence

  20
     To address the fact that predicted malaria risk is a generated regressor, we re-calculate predicted risk
on each bootstrap replication.


                                                     15
to RDT results.
    Appendix Table A6 formally tests whether the slope of the malaria risk-malaria pre-
scription line differs for patients in the basic versus extended training groups.21 We find
that higher malaria risk is indeed associated with a higher probability of receiving a malaria
prescription in the intervention groups, consistent with improved allocation in the extended
training group. Moreover, differences persist when we limit the sample to patients who took
a home-based RDT and use RDT-verified malaria status as a measure of malaria risk.
    To better assess the welfare consequences of these changes in testing and prescription
behavior, we directly examine the effect of the extended training on the match between
underlying malaria status and prescribed treatment. We construct a measure of this match
as
                                 M                  M
                          mi = πi  × AMi + (1 − πi    ) × (1 − AMi ).

The variable πi  M
                   denotes patient i’s malaria status, either measured by their actual status
in the home-testing subsample (posict ∈ {0, 1}), or by their predicted risk (posˆ ict ∈ [0, 1]).
AMi is a dummy variable indicating that the patient was prescribed an antimalarial. Table
4 presents results – first for all patients using predicted risk (column 1), then for home-
tested patients using predicted risk (column 2) and actual malaria status (column 3). We
find significant improvements in quality of care that are meaningful in magnitude: whereas
just 41-44 percent of patients in the basic training arm received the appropriate malaria
treatment per our measure, the extended training increased this by 7-13 percentage points,
an up to 30 percent improvement in antimalarial allocation. Notably, this improvement was
realized without significantly changing the number of other medications prescribed, use of
antibiotics, or time spent at the clinic (Appendix Table A7).
    Taken together, the results are consistent with the hypothesis that the extended training
increased providers’ trust in RDTs, which in turn improved the quality of malaria care.
Given that extended training largely shifted the composition of tests, rather than the overall
testing rate, it stands to reason that this effect is driven by RDTs performing better in our
setting than microscopy. As discussed in section 2, patients who received microscopy testing
at baseline were less likely to be malaria positive in a second home-based RDT test, but more
likely to receive antimalarials than patients tested with RDTs. Thus, RDT-tested patients
have better treatment outcomes – largely due to a reduction in the use of antimalarials
among those who test malaria negative (see Figure 1).
  21
      To do this, we study treatment effects on antimalarial prescriptions, modifying the regression specifi-
cation given by equation 2 to include a measure of malaria risk (either home RDT result, predicted malaria
risk, or a dummy variable identifying patients with above median predicted risk) and the interaction be-
tween malaria risk and extended training. If extended training improved the allocation of antimalarials, the
interaction term should be positive and statistically significant.


                                                    16
Effects of the Patient Information Intervention. Returning to Table 3, we see that
the effects of patient information on testing and treatment outcomes are notably different.
Counter to the aim of the intervention, we find a 6 percentage point decrease in the incidence
of malaria testing, significant at the one percent level. This result is mainly driven by the 4
percentage point reduction in the share of patients tested with an RDT only (significant at
the 5 percent level). Table 4 shows that the patient information treatment had no discernible
effect on treatment allocation, despite the reduction in malaria testing.
    A possible mechanism for this result is that (at least some) patients distrusted RDTs
or arrived at the clinic with a strong ex-ante preference for a specific treatment. Reduced
testing without a change in treatment allocation could have arisen because marginal patients
who “opted out” of malaria testing were those who had very strong preferences regarding
treatment and, therefore, took steps to ensure providers dispensed their preferred prescription
regardless of the diagnostic outcome. This hypothesis is consistent with evidence from our
companion paper, which shows that patients can successfully exert pressure on providers
to prescribe antimalarials, worsening the match between treatment and underlying illness
(Lopez et al., 2022). In the next section, we first discuss the effect of our interventions on
patient knowledge and satisfaction and then explore this idea further.

4.4    Impacts on Patient Knowledge and Satisfaction
Figure 4 uses data from the home survey to assess whether the extended training improved
patients’ knowledge of their malaria testing outcome. Before carrying out the home RDT
test, enumerators asked patients about the likely outcome of their test (on a scale from
1, definitely negative, to 5, definitely positive). The figure shows average patient scores
conditional on the patient’s actual home RDT result for both the basic and extended training
groups. Patients treated at extended-training clinics were better able to predict their test
result, with those who tested negative reporting lower scores (relative to their peers in the
basic training group) and those who tested positive reporting higher scores. For a formal
test, we run a difference-in-difference regression paralleling that in column 5 of Table A6.
The coefficient on the interaction term is 0.516, with a p-value of 0.097.
    Since patients at extended-training clinics got higher-quality care and became better
informed about their malaria status, it is natural to ask whether they recognize and reward
this quality of care improvement. The home survey included a series of questions to measure
patients’ satisfaction with testing at the clinic, medicines prescribed at the clinic, and overall
care. Specifically, patients were asked about the extent to which they agreed with a series
of statements, such as “the doctor/nurse should have done additional medical testing before
prescribing treatment”, “the doctor/nurse based the treatment decision strongly on the result


                                               17
of my medical tests”, and “the doctor should have given me a different or additional treatment
or drug”. We aggregate these into three standardized indices following Kling et al. (2007):
an overall satisfaction index, which includes measures of satisfaction with tests, medications
received, and overall care, as well as two sub-indices that focus on satisfaction with tests and
medications separately.22
    Table 5 reports treatment effects. Surprisingly, the extended training had significant
negative effects on patient satisfaction: column 1 shows that patients in the extended training
arm are 0.092 standard deviation units less satisfied (significant at the 5 percent level),
with declines in satisfaction with both tests (column 2, significant at the 1 percent level)
and medications (column 3, not significant, but similar in magnitude).23 Figure 5 graphs
CDFs of the satisfaction indices by provider training arm; the distribution of satisfaction
in the basic training arm first-order stochastically dominates that in the extended training
arm. Figure A1 uses local linear regression to explore how treatment effects on the three
patient satisfaction indices vary with predicted malaria risk. The graphs show that patient
satisfaction is relatively constant across the risk distribution, with satisfaction for patients
visiting extended-training clinics consistently lower than satisfaction among patients visiting
basic training clinics.
    By contrast, Table 5 shows that the patient information treatment had no effect on
patient satisfaction. In other words, the additional information about the proper malaria
testing protocols did not change their views about the care they received at the clinic despite
the overall high rates of treatment misallocation.
Mechanisms for the Decline in Patient Satisfaction. The extended training im-
proved quality of care and increased patients’ ability to predict their home RDT results.
Why did patient satisfaction nonetheless decline? One plausible mechanism is that patients
themselves are skeptical of malaria RDTs and dislike providers relying on them. Since the
extended training increased RDT (only) testing rates, it may be that learning the result of
the clinic RDT successfully shifted patients’ beliefs about the outcome of the home RDT,
but nonetheless had limited effects on their subjective belief about their own malaria status.
   The patient information video explained that RDTs are accurate and that Mali’s official
malaria policy is to “only treat confirmed malaria positive cases.” One might therefore
expect that the patient information intervention helped ameliorate any negative effects of
the extended training on patient satisfaction – at least to the extent patient information
  22
      See Appendix C for a complete list of questions included in the patient satisfaction indices and details
on how we constructed index components.
   23
      Appendix Table A8 reports treatment effects on individual index components. The extended training’s
treatment effects on individual components are generally modest and insignificant but consistently point to
less satisfaction.


                                                     18
increases trust in RDTs. Appendix Table A9 tests this hypothesis by formally studying
the interaction between the two interventions on satisfaction indices. However, we find no
evidence that patient information significantly moderated the negative effect of extended
training – interaction terms are generally small in magnitude, vary in sign across the testing
and medication sub-indices, and are never significantly different from zero. Instead, we see
that patients in the patient information group reject the RDT significantly more often. This
suggests that patients understood the message of the video, but it had no effect on their
trust in the promoted technology.
    A related hypothesis is that patients visiting the clinic have strong priors about their
illness and resulting preferences regarding treatment, and react negatively when providers
override their beliefs to follow test results. This is consistent with our finding that pa-
tient information reduced RDT testing, as well as evidence that patients exert pressure on
providers to prescribe antimalarials (Lopez et al., 2022). To test this hypothesis, we examine
whether negative satisfaction effects are concentrated among patients whose priors are incor-
rect. Here, we exploit the fact that we elicited patients’ prior beliefs regarding the cause of
their illness before they consulted with a provider. We construct a measure of “prior match”
given by
                                    M                M
                             ˆ i = πi
                             m        × p0 + (1 − πi   ) × (1 − p0 ),

where p0 is a dummy variable equal to one if the patient suspected malaria pre-consultation.
As above, πi M
                                                               ˆ ict or actual malaria status as
               is either the patient’s predicted malaria risk pos
measured by the home test posict .
    Table 6 studies how the effect of the extended training varies with patients’ priors.24
Column 1 uses the prior match measure constructed with predicted malaria positivity to
examine patterns in the full sample, column 2 repeats the exercise for the subset of patients
with valid home RDT, and finally, column 3 uses the match between stated prior and actual
malaria status per the home RDT. In all cases, we can interpret the coefficient on the
extended training dummy (row 1) as the effect of the extended training on patient satisfaction
for patients whose priors were completely misaligned with their underlying malaria status.
The coefficient on the interaction term (row 2) tests whether the effect of extended training
differed for better-informed patients, and the sum of the two rows provides the estimated
effect of extended training for patients whose priors are exactly correct.
    In all three columns, we see that the extended training had a negative effect on the
satisfaction of the ex-ante most misinformed patients (between -0.16 and -0.13 standard
deviation units in magnitude, significant at the 5 percent level in columns 1 and 2 and at

  24
       The regression equation is yict = β0 + βE ETc + βEM ETc × m          ˆ ict + βP P Ict + xict α + γc + δt + εic .
                                                                 ˆ ict + βm m


                                                          19
the 10 percent level in column 3). On the other hand, the interaction terms are positive,
roughly equal in magnitude to the main effects, and significantly different from zero in
columns 1 and 2. In other words, we find that patient priors moderate the effect of the
extended training, with no ill-effects on satisfaction for patients whose priors were ex-ante
correct. This suggests that the extended training reduced satisfaction by increasing the rate
of perceived mis-diagnosis among patients who were e ante incorrect about the cause of their
illness.
    To further explore this channel, we create an outcome variable that indicates whether
the patient got the medications they wanted (or expected) based on their prior beliefs about
their malaria status. This match between patient prior and treatment is given by p0 × AMi +
(1 − p0 ) × (1 − AMi ), where, as before, p0 is a dummy equal to one if the patient suspects
malaria pre-consultation and AMi is a dummy variable indicating that the patient was
prescribed an antimalarial. Following the structure of Table 6, we ask whether the extended
training impacted the likelihood of receiving expected/desired treatment and whether this
varies with patient priors, in Appendix Table A10. The first row in the table reports the
effect of extended training for patients with incorrect beliefs about their true malaria status.
We see that the extended training reduced the likelihood of these patients receiving their
desired/expected treatment by 12-16 percentage points (significant at the 10 percent level
in column 2 and at the 5 percent level in column 3). The sum of the first and second rows
shows that the extended training had the opposite effect for patients with correct priors
(significant at the 10 percent level in column 3). While the analysis is underpowered, the
point estimates therefore suggest that the extended training reduced patient satisfaction
among those individuals who were less likely to receive their desired/expected treatment.
Overall, we interpret this as evidence that patient beliefs about their true malaria status were
affected very little by test results or provider recommendations, and the extended training
increased the disagreement between their priors and the consultation outcome, leading to a
negative effect on patient satisfaction.
    An alternative possibility is that the extended training negatively affected other aspects
of caregiving – providers may, for example, use fewer soft skills and/or engage in less patient
counseling when they are more motivated to use test results to inform treatment. Our
measured outcomes that best capture this channel are patients’ assessment of whether the
provider clearly explained their diagnosis and/or test results (columns 1 and 2 in Appendix
Table A8) and time spent in consultation (column 5 of Appendix Table A7). There are no
significant impacts on these outcomes. It is also unclear why a change in soft skills alone
would differentially impact the satisfaction of patients with correct vs. incorrect priors about



                                              20
their illness.25 Given these patterns, we conclude that a soft-skills channel alone is unlikely
to account for the impacts we observe.

5        Conclusion
Consistent with research in other low-income settings, we document significant gaps in the
quality of malaria care at Malian medical clinics. These gaps are especially striking given that
easy-to-administer, low-cost rapid diagnostic tests (RDTs) for malaria are readily available.
    We present descriptive and experimental evidence that providers’ beliefs about the quality
of RDTs are part of the problem. Before the intervention, health workers were too pessimistic
regarding the tests’ accuracy. We evaluate the impact of an information intervention that
provided information to health providers about RDTs’ accuracy using data from quality
assurance testing and field research in a Malian setting. In response, providers positively
updated their beliefs about RDT accuracy.
    Importantly, beliefs translated into diagnostic and treatment outcomes. First, we find
changes in testing behavior that indicate providers were more comfortable relying on RDTs
to establish a malaria diagnosis. Second, we find evidence that this led to improved quality
of care – post-consultation, the match between underlying malaria status and prescribed
treatment improved significantly. We estimate that the share of patients receiving the “right”
malaria treatment (an antimalarial if malaria positive, no antimalarial if malaria negative)
increased by 7-13 percentage points at treated clinics, relative to the 41-44 percent rate of
correct treatment in the comparison group.
    Despite these benefits, our results also highlight important risks of health information
interventions, associated with the incorrect (and persistent) beliefs of patients. First, even
though the extended provider training improved patient knowledge, demonstrated by their
increased ability to predict the outcome of their home malaria RDT, the training reduced
patient satisfaction by 0.09 standard deviation units, with evidence that this effect was con-
centrated among patients who had incorrect pre-consultation beliefs about their underlying
malaria status. Second, an intensive, one-on-one patient-centered information intervention
designed to increase demand for correct diagnosis of malaria (via malaria tests) had few
effects and, if anything, backfired: patient information reduced use of malaria tests by 6
percentage points (11 percent) and had no impact on prescription outcomes. Both these
findings suggest that patients may have hard-to-move priors about their cause of illness
and/or persistently distrust diagnostic tests. While our study cannot speak to long-term
effects, there is a risk that dissatisfied patients could reduce their demand for clinic services,

    25
     Poorer soft skills could, however, exacerbate the dissatisfaction patients felt when not prescribed what
they wanted/expected.


                                                     21
or lead providers concerned about keeping patients satisfied to revert to their old behaviors.
    In summary, our results demonstrate that simple provider-centered information interven-
tions have the potential to effectively improve the quality of care at clinics. Still, patients
may not recognize or reward these improvements, especially when they run counter to their
preferences over treatment. While this points to the need to address patient preferences
alongside provider practices, the unexpected result of our patient information intervention
highlights that the success of such approaches is by no means assured. To the extent that
patients have strong priors, they may need more powerful interventions; a better under-
standing of how best to shift patient beliefs and create demand for high-quality care is an
important area for future research.




                                             22
References
Abba, K., J. J. Deeks, P. L. Olliaro, C.-M. Naing, S. M. Jackson, Y. Takwoingi, S. Donegan,
  and P. Garner (2011). Rapid diagnostic tests for diagnosing uncomplicated P. falciparum
  malaria in endemic countries. Cochrane database of systematic reviews (7).
Adepoju, P. (2019). Mali announces far-reaching health reform. The Lancet 393 (10177),
  1192.
Adhvaryu, A. R. (2014). Learning, misallocation, and technology adoption: Evidence from
  new malaria therapy in Tanzania. Review of Economic Studies 81 (4), 1331–1365.
Ansah, E. K., S. Narh-Bana, M. Epokor, S. Akanpigbiam, A. A. Quartey, J. Gyapong, and
  C. J. Whitty (2010). Rapid testing for malaria in settings where microscopy is available and
  peripheral clinics where only presumptive treatment is available: A randomised controlled
  trial in Ghana. BMJ 340, c930.
Arrow, K., C. Panosian, and H. Gelband (Eds.) (2004). Saving Lives, Buying Time: Eco-
  nomics of Malaria Drugs in an Age of Resistance. Washington, DC: Institute of Medicine:
  National Academies Press.
Arrow, K. J. (1963). Uncertainty and the welfare economics of medical care. The American
  Economic Review 53 (5), 941–973.
Baker, A. (2001). Crossing the quality chasm: a new health system for the 21st century,
  Volume 323. British Medical Journal Publishing Group.
Banerjee, A., J. Das, J. Hammer, R. Hussam, and A. Mohpal (2020). The market for
  healthcare in low income countries. Mimeo.
Banerjee, A. V., E. Duflo, and R. Glennerster (2008). Putting a band-aid on a corpse:
  Incentives for nurses in the Indian public health care system. Journal of the European
  Economic Association 6 (2-3), 487–500.
Bariş, E., R. Silverman, H. Wang, and F. Zhao (2022). Walking the talk: reimagining primary
  health care after COVID-19. World Bank Publications.
Belloni, A., V. Chernozhukov, and C. Hansen (2014, May). High-dimensional methods and
  inference on structural and treatment effects. Journal of Economic Perspectives 28 (2),
  29–50.
Berzosa, P., A. de Lucio, M. Romay-Barja, Z. Herrador, V. González, L. García,
  A. Fernández-Martínez, M. Santana-Morales, P. Ncogo, B. Valladares, et al. (2018).
  Comparison of three diagnostic methods (microscopy, RDT, and PCR) for the detection
  of malaria parasites in representative samples from Equatorial Guinea. Malaria Jour-
  nal 17 (1), 1–12.
Bisoffi, Z., B. S. Sirima, A. Angheben, C. Lodesani, F. Gobbi, H. Tinto, and J. Van den


                                             23
  Ende (2009). Rapid malaria diagnostic tests vs. clinical management of malaria in rural
  Burkina Faso: safety and effect on clinical decisions. A randomized trial. Tropical Medicine
  & International Health 14 (5), 491–498.
Brownlee, S., K. Chalkidou, J. Doust, A. G. Elshaug, P. Glasziou, I. Heath, S. Nagpal,
  V. Saini, D. Srivastava, K. Chalmers, et al. (2017). Evidence for overuse of medical
  services around the world. The Lancet 390 (10090), 156–168.
Busfield, J. (2015). Assessing the overuse of medicines. Social science & medicine 131,
  199–206.
Chandra, A., A. Finkelstein, A. Sacarny, and C. Syverson (2016). Health care exception-
  alism? Performance and allocation in the US health care sector. American Economic
  Review 106 (8), 2110–44.
Cissoko, M., M. Magassa, V. Sanogo, A. Ouologuem, L. Sangaré, M. Diarra, C. S. Bationo,
  M. Dolo, M. D. Bah, S. Doumbia, et al. (2022). Stratification at the health district level
  for targeting malaria control interventions in Mali. Scientific reports 12 (1), 1–17.
Cohen, J., P. Dupas, and S. Schaner (2015). Price subsidies, diagnostic tests, and targeting
  of malaria treatment: Evidence from a randomized controlled trial. American Economic
  Review 105 (2), 609–45.
Currie, J., W. Lin, and J. Meng (2014). Addressing Antibiotic Abuse in China: An Experi-
  mental Audit Study. Journal of Development Economics 110, 39–51.
Das, J. and J. Hammer (2014). Quality of primary care in low-income countries: Facts and
  economics. 6, 525–553.
Das, J., J. Hammer, and K. Leonard (2008). The Quality of Medical Advice in Low-Income
  Countries. The Journal of Economic Perspectives 22 (2), pp. 93–114.
Das, J., A. Holla, A. Mohpal, and K. Muralidharan (2016). Quality and Accountability
  in Health Care Delivery: Audit-Study Evidence from Primary Care in India. American
  Economic Review 106 (12), 3765–99.
de Chaisemartin, C. and J. Ramirez-Cuellar (2024). At What Level Should one Cluster
  Standard Errors in Paired Experiments, and in Stratified Experiments with Small Strata?
  American Econonmic Journal: Applied Economics 16 (1), 193–212.
de Walque, D., E. Kandpal, A. Wagstaff, J. Friedman, M. Piatti-Fünfkirchen, A. Sautmann,
  G. Shapira, and E. Van de Poel (2022). Improving Effective Coverage in Health: Do
  Financial Incentives Work? World Bank Publications.
Djimde, A. A., H. Maiga, I. Sagara, K. Traore, H. Niangaly, A. Bamadio, A. Grivoyannis,
  M. Tekete, O. B. Traore, A. Kodio, et al. (2016). Field Assessment of SD Bioline Malaria
  Species Antigen Detection by Rapid Diagnostic Tests in Mali. Journal of Parasitology and
  Vector Biology 8 (1), 1–9.


                                             24
Dulleck, U. and R. Kerschbamer (2006). On doctors, mechanics, and computer specialists:
   The economics of credence goods. Journal of Economic literature 44 (1), 5–42.
Farley, H., E. R. Enguidanos, C. M. Coletti, L. Honigman, A. Mazzeo, T. B. Pinson, K. Reed,
   and J. L. Wiler (2014). Patient satisfaction surveys and quality of care: An information
   paper. Annals of Emergency Medicine 64 (4), 351–357.
Fleming, K. A., S. Horton, M. L. Wilson, R. Atun, K. DeStigter, J. Flanigan, S. Sayed,
   P. Adam, B. Aguilar, S. Andronikou, et al. (2021). The lancet commission on diagnostics:
   transforming access to diagnostics. The Lancet 398 (10315), 1997–2050.
Friedman, M. and R. D. Friedman (1962). Capitalism and Freedom. Phoenix Book: Busi-
   ness/Economics. University of Chicago Press.
Hamer, D. H., M. Ndhlovu, D. Zurovac, M. Fox, K. Yeboah-Antwi, P. Chanda, N. Sip-
   ilinyambe, J. L. Simon, and R. W. Snow (2007). Improved diagnostic testing and malaria
   treatment practices in Zambia. JAMA 297 (20), 2227–2231.
Ibnat, F., K. L. Leonard, L. Bawo, and R. L. Mohammed-Roberts (2019). The three-gap
   model of health worker performance. World Bank Policy Research Working Paper (8782).
Iizuka, T. (2012, May). Physician Agency and Adoption of Generic Pharmaceuticals. Amer-
   ican Economic Review 102 (6), 2826–58.
Kling, J. R., J. B. Liebman, and L. F. Katz (2007). Experimental analysis of neighborhood
   effects. Econometrica 75 (1), 83–119.
Koenker, H., M. K. Coulibaly, and I. Bouare (2020). Trends in and factors associated with
   malaria prevention in Mali: Further analysis of the Mali Demographic and Health Surveys
   and Malaria Indicator Surveys 2006-2018. Demographic and Health Surveys.
Koné, D., D. Coulibaly, O. Doumbo, I.-S. Fall, E. Kibuchi, B. Mitto, G. Sang, D. Kyalo,
   R. W. Snow, and A. M. Noor (2015). An epidemiological profile of malaria in Mali.
   Programme National de Lutte contre le Paludisme (PNLP).
Kravitz, R. L., R. M. Epstein, M. D. Feldman, C. E. Franz, R. Azari, M. S. Wilkes, L. Hinton,
   and P. Franks (2005). Influence of patients’ requests for direct-to-consumer advertised
   antidepressants: A randomized controlled trial. JAMA 293 (16), 1995–2002.
Lagarde, M. and D. Blaauw (2022). Overtreatment and benevolent provider moral hazard:
   evidence from South African doctors. Journal of Development Economics .
Lee, S., U. Kara, E. Koay, M. Lee, S. Lam, and D. Teo (2002). New strategies for the
   diagnosis and screening of malaria. 76 (S1), 291–293.
Lopez, C., A. Sautmann, and S. Schaner (2022). Does patient demand contribute to the
   overuse of prescription drugs? American Economic Journal: Applied Economics 14 (1),
   225–60.
Macarayan, E., I. Papanicolas, and A. Jha (2020). The quality of malaria care in 25 low-


                                             25
  income and middle-income countries. BMJ global health 5 (2), e002023.
Ministère de la Santé (2013, August). Politique Nationale De Lutte Contre Le Paludisme.
  Republique du Mali.
Reyburn, H., R. Mbatia, C. Drakeley, I. Carneiro, E. Mwakasungula, O. Mwerinde,
  K. Saganda, J. Shao, A. Kitua, R. Olomi, B. Greenwood, and C. Whitty (2004). Overdiag-
  nosis of malaria in patients with severe febrile illness in Tanzania: A prospective study.
  British Medical Journal 329 (7476), 1212.
Rowe, A. K., S. Y. Rowe, D. H. Peters, K. A. Holloway, J. Chalker, and D. Ross-Degnan
  (2018). Effectiveness of strategies to improve health-care provider practices in low-income
  and middle-income countries: a systematic review. The Lancet Global Health 6 (11), e1163–
  e1175.
Rowe, A. K., S. Y. Rowe, D. H. Peters, K. A. Holloway, and D. Ross-Degnan (2021). The
  effectiveness of training strategies to improve healthcare provider practices in low-income
  and middle-income countries. BMJ Global Health 6 (1), e003229.
Sautmann, A. and S. Schaner (2016). Incentives for Accurate Diagnosis: Improving
  Health Care Quality in Mali. Technical report, AEA RCT Registry. October 24.
  https://doi.org/10.1257/rct.1707-1.0.
Skinner, J. (2011). Causes and consequences of regional variations in health care. In Handbook
  of Health Economics, Volume 2, pp. 45–93. Elsevier.
Trampuz, A., M. Jereb, I. Muzlovic, and R. M. Prabhu (2003). Clinical Review: Severe
  Malaria. Critical Care 7 (4), 315.
Wongsrichanalai, C., M. J. Barcus, S. Muth, A. Sutamihardja, and W. H. Wernsdorfer
  (2007). A review of malaria diagnostic tools: microscopy and rapid diagnostic test (RDT).
  Defining and Defeating the Intolerable Burden of Malaria III: Progress and Perspectives:
  Supplement to Volume 77 (6) of American Journal of Tropical Medicine and Hygiene .
World Health Organization (2014a). Emergence and spread of Artemisinin resistance calls
  for intensified efforts to withdraw oral Artemisinin-based monotherapy from the market.
  Technical report, World Health Organization.
World Health Organization (2014b). Malaria rapid diagnostic test performance: summary
  results of WHO product testing of malaria RDTs: Round 1-5 (2008-2013). World Health
  Organization.
World Health Organization (2022). World Malaria Report 2022.
World Health Organization et al. (2021). World Malaria Report 2021.




                                             26
Tables and Figures

    Table 1: Randomization Verification: Clinic Census and Health Worker Pre-Training Survey

                                                                            (1)       (2)          (3)
                                                                           Basic   Extended
                                                                          Training Provider
                                                                           Mean    Training         N
     A. Provider-Reported Clinic Characteristics
      Average Patient Load                                                 30.371       -1.192     57
                                                                          [22.172]     (1.207)
      Clinic Has Microscopy                                                 0.897       -0.111     57
                                                                           [0.310]     (0.094)
      Fraction Patients Tested with RDT                                     0.648        0.065     57
                                                                           [0.221]     (0.057)
      Fraction Patients Tested with Microscopy                              0.473       -0.134     57
                                                                           [0.306]     (0.083)
      Fraction Tested Malaria Positive                                      0.543        0.062     57
                                                                           [0.196]     (0.048)
     B. Provider Attendance
      Number of Staff Attended per Clinic                                    3.862       0.000      58
                                                                           [0.441]     (0.158)
      Pre-Tests Submitted per Clinic                                        2.793       -0.061     58
                                                                           [1.424]     (0.558)
      Post tests submitted per clinic                                       3.897       -0.172     58
                                                                           [0.900]     (0.293)
     C. Provider Pre-Test Knowledge
      Knows Malian Malaria Policy                                           0.519       -0.045     160
                                                                           [0.503]     (0.095)
      Correct Meds for Simple Malaria Treatment                             0.568       -0.038     160
                                                                           [0.498]     (0.111)
      RDT Detection Rate w. Low Parasite Load 90% or Higher                 0.111        0.082     160
                                                                           [0.316]     (0.052)
      RDT Detection Rate w. High Parasite Load 90% or Higher                0.333        0.195     160
                                                                           [0.474]     (0.128)
      Correct Symptoms Severe Malaria                                       0.506       -0.024     160
                                                                           [0.503]     (0.108)
      Correct Time Interval to Read RDT                                     0.136        0.026     160
                                                                           [0.345]     (0.092)
      Number of Correct Answers                                             2.173        0.197     160
                                                                           [1.473]     (0.384)
     Notes : Panel A uses clinics census information (one observation missing). Panels B and C use
    the health worker pre-training survey. Robust standard errors clustered at the clinic-pair level in
    parentheses, standard deviation in brackets. All regressions include strata fixed effects. *, **, and ***
    indicate significance at the 10, 5, and 1 percent levels respectively.




                                                     27
                 Table 2: Effect of Extended Provider Training on Learning: Provider Post-Training and Clinic Endline Surveys

                                     (1)          (2)          (3)               (4)            (5)             (6)         (7)              (8)
                                                  Beliefs About RDTs                                    Beliefs About Microscopy Tests
                                                                                                        Share      Share
                                              Share RDT      Share RDT                                Microscopy Microscopy   Share
                                   RDT         Positive       Positive      Share RDT      Microscopy  Positive:  Positive: Microscopy
                                   Beliefs      Simple         Severe        Negative       Beliefs     Simple     Severe    Negative:
                                   Index       Malaria        Malaria       No Malaria       Index     Malaria    Malaria   No Malaria
      A. Data from Post-Training Survey
       Extended Training      0.312∗∗∗          0.200∗∗∗       0.071∗∗∗        0.046           0.027          -0.002         -0.005         0.037
                               (0.071)          (0.038)        (0.019)        (0.041)         (0.116)        (0.035)        (0.028)        (0.048)

       Mean (Basic Training)        0.000        0.599          0.888          0.863           0.000          0.846          0.921          0.829
       N                             205          198            196            176             201            197            195            173
28




      B. Data from Provider Follow-Up Survey
       Extended Training      0.406∗∗∗    0.158∗∗              0.072∗∗         0.038          0.277∗          0.055          0.024          0.065
                               (0.129)   (0.071)               (0.028)        (0.025)         (0.159)        (0.050)        (0.016)        (0.047)

       Mean (Basic Training)        0.000        0.560          0.889          0.938           0.000          0.840          0.954          0.868
       N                              85           85            85             85               85             85             85             85
      Notes : Robust standard errors clustered at the clinic-pair level in parentheses. All regressions include strata fixed effects. Additional covariates
     are selected using double lasso, from the set of all characteristics listed in Table 1 as well as their pairwise interactions. Missing variables for
     all characteristics are dummied out and recoded to the mean prior to forming interactions. We include these missing dummies in the potential
     covariate set and the set of pairwise interactions. Sample in both panels limited to providers who attended the training. Index components are
     standardized relative to the basic provider training mean. In cases where some (but not all) index components for a respondent are missing,
     missing components are imputed to the treatment group mean prior to standardization. *, **, and *** indicate significance at the 10, 5, and 1
     percent levels respectively.
Table 3: Effect of Extended Provider Training and Patient Information on Malaria Testing at
the Clinic: Patient Survey

                                                 (1)           (2)             (3)             (4)
                                                 Any         RDT Only       Microscopy       Multiple
                                                Malaria                       Only            Tests
                                                 Test
 Extended Training                               -0.0257       0.116∗∗         -0.0919       -0.0475∗∗
                                                (0.0547)      (0.0486)        (0.0729)        (0.0198)
 Patient Information                           -0.0613∗∗∗     -0.0432∗∗        -0.0171        0.00632
                                                (0.0177)      (0.0189)        (0.0245)       (0.00837)

 Mean (Basic Training, No Pat. Info.)             0.58           0.21           0.25           0.07
 N                                                1973           1973           1973           1973
 Notes : Robust standard errors clustered at the clinic-pair level in parentheses. All regressions include
strata and survey date fixed effects. Additional covariates are selected using double lasso, from the
set of all covariates listed in Table 1 and all individual-level characteristics in Table A4 (excluding
predicted malaria risk and characteristics only measured at the home survey) as well as the square
of patient age and illness duration. We also include pairwise interactions of the aforementioned
characteristics. Missing values for all characteristics are dummied out and recoded to the mean
prior to forming interactions. Missing dummies are also included in the potential covariate set and
set of pairwise interactions. *, **, and *** indicate significance at the 10, 5, and 1 percent levels
respectively.




                                                  29
Table 4: Effect of Extended Provider Training and Patient Information on the Allocation
of Malaria Treatment

                                                  (1)             (2)           (3)
                                             Full Sample      Has Valid Home-Based RDT
                                               Expected         Expected       Actual Match
                                                Match            Match
 Extended Training                             0.0664∗∗          0.0671∗∗         0.131∗∗∗
                                               (0.0336)          (0.0292)         (0.0427)
 Patient Infomation                            0.00191            0.0124          -0.00943
                                               (0.0171)          (0.0169)         (0.0339)

 Mean (Basic Training, No Pat. Info.)              0.44            0.41             0.43
 N                                                 1971            1093             1093
 Notes : Robust standard errors clustered at the level in parentheses. All regressions include
strata and survey date fixed effects. Additional covariates are selected using double lasso, from
the set of all covariates listed in Table 1 and all individual-level characteristics in Table A4
(excluding predicted malaria risk and characteristics only measured at the home survey) as
well as the square of patient age and illness duration. We also include pairwise interactions of
the aforementioned characteristics. Missing values for all characteristics are dummied out and
recoded to the mean prior to forming interactions. Missing dummies are also included in the
potential covariate set and set of pairwise interactions. *, **, and *** indicate significance at
the 10, 5, and 1 percent levels respectively.




                                              30
Table 5: Effect of Extended Provider Training and Patient Information on Pa-
tient Satisfaction

                                                 (1)            (2)            (3)
                                                Overall       Testing       Medications
                                                Index        Sub-Index      Sub-Index
 Extended Training                              -0.0921∗∗    -0.0994∗∗∗       -0.0806
                                                (0.0448)      (0.0357)       (0.0904)
 Patient Information                             0.0326        0.0386        0.00795
                                                (0.0237)      (0.0316)       (0.0328)

 Mean (Basic Training, No Pat. Info.)             0.00          0.00           0.00
 N                                                1429          1419           1418
 Notes : Robust standard errors clustered at the clinic-pair level in parentheses. All re-
gressions include strata and survey date fixed effects. Additional covariates are selected
using double lasso, from the set of all covariates listed in Table 1 and all individual-
level characteristics in Table A4 (excluding predicted malaria risk and characteristics
only measured at the home survey) as well as the square of patient age and illness
duration. We also include pairwise interactions of the aforementioned characteristics.
Missing values for all characteristics are dummied out and recoded to the mean prior
to forming interactions. Missing dummies are also included in the potential covariate
set and set of pairwise interactions. Index components are standardized relative to
the basic provider training mean. In cases where some (but not all) index components
for a respondent are missing, missing components are imputed to the treatment group
mean prior to standardization. *, **, and *** indicate significance at the 10, 5, and 1
percent levels respectively.




                                           31
Table 6: Effect of Extended Provider Training on Patient Satisfaction by Accuracy of
Pre-Consultation Malaria Beliefs

                                                      (1)              (2)           (3)
                                                 Full Sample       Has Valid Home-Based RDT
                                                    Overall           Overall           Overall
                                                  Satisfaction      Satisfaction      Satisfaction
                                                     Index             Index             Index
 Extended Training                                 -0.154∗∗           -0.161∗∗           -0.129∗
                                                   (0.0695)           (0.0791)          (0.0688)
 Extended Training × Prior-Risk Match               0.128∗             0.162∗∗           0.0874
                                                   (0.0659)           (0.0766)          (0.0587)
 Match: Malaria Prior and Malaria Risk             -0.114∗∗            -0.0935           -0.0481
                                                   (0.0475)           (0.0578)          (0.0453)

 P-value: ET + ET × Prior Match = 0                   0.532            0.985             0.326

 Prior Match Measure                              Expected          Expected          Actual Prior
                                                 Prior Match       Prior Match          Match
                                                                                     (Home RDT)

 Mean (Basic Training, No Pat. Info.)                 0.00              -0.02             -0.02
 N                                                    1429              1092              1092
 Notes : Robust standard errors clustered at the clinic-pair level in parentheses. Standard errors for
regressions that include a derivative of predicted malaria risk in the set of independent variables
are bootstrapped using 500 replications to account for the fact that predicted malaria risk is
a generated regressor. All regressions include strata and survey date fixed effects. Additional
covariates are selected using double lasso, from the set of all covariates listed in Table 1 and all
individual-level characteristics in Table A4 (excluding predicted malaria risk and characteristics
only measured at the home survey) as well as the square of patient age and illness duration.
We also include pairwise interactions of the aforementioned characteristics. Missing values for all
characteristics are dummied out and recoded to the mean prior to forming interactions. Missing
dummies are also included in the potential covariate set and set of pairwise interactions. *, **,
and *** indicate significance at the 10, 5, and 1 percent levels respectively.




                                                 32
Figure 1: Malaria Incident, Antimalarial Prescriptions, and Match Between Underlying
Illness and Treatment


                              A. Malaria Incidence and Antimalarial Prescriptions by In-Clinic Testing Status

                          1

                         .8
     Share of Patients




                         .6

                         .4

                         .2

                          0
                              Not tested                               RDT Only                          Microscopy

                                                Malaria Positive (Home Test)                    Antimalarial Prescribed




                                     B. Match between Malaria Test Result and Prescribed Treatment

                         1

                         .8
     Share of Patients




                         .6

                         .4

                         .2

                         0
                              Not tested                               RDT Only                          Microscopy

                                 All Patients                       Malaria Negative Patients                 Malaria Positive Patients




Notes: Sample limited to the subset of patients who consented to take an RDT during the home follow-up
survey in the Basic Training group. Panel A graphs malaria positivity (measured in home RDT) and receipt
of antimalarial prescriptions (recorded during the clinic survey), by type of malaria test conducted at the
clinic (recorded during the clinic survey). Overall positivity rate is 23.12% in the control group. Panel
B graphs the match between true malaria status (measured in home RDT) and receipt of antimalarial
prescription by type of malaria test at the clinic. The share of patients prescribed an antimalarial is 69.36%
in the control group.




                                                                         33
                                              Figure 2: Within-CSCOM Randomization Design


                                             WEEK 1                                                           WEEK 2
       CSCOM
       Number     Mon       Tues     Weds      Thurs      Fri       Sat        Sun     Mon     Tues   Weds     Thurs    Fri     Sat    Sun
         1         C         --       PV         --       DV         --         --      PI      --    PI-PV      --    PI-DV     --     --
         2        DV         --        C         --       PV         --         --    PI-DV     --      PI       --    PI-PV     --     --
         3        PV         --       DV         --        C         --         --    PI-PV     --    PI-DV      --      PI      --     --
         4         C         --       DV         --       PV         --         --      PI      --    PI-DV      --    PI-PV     --     --
         5        DV         --       PV         --        C         --         --    PI-DV     --    PI-PV      --      PI      --     --
         6         PI        --      PI-PV       --      PI-DV       --         --      C       --     PV        --     DV       --     --
         7       PI-DV       --        PI        --      PI-PV       --         --     DV       --      C        --     PV       --     --
         8       PI-PV       --      PI-DV       --        PI        --         --     PV       --     DV        --      C       --     --
         9         PI        --      PI-DV       --      PI-PV       --         --      C       --     DV        --     PV       --     --
         10      PI-DV       --      PI-PV       --        PI        --         --     DV       --     PV        --      C       --     --
         11        --        C         --       PV         --       DV          --      --      PI      --     PI-PV     --    PI-DV    --
         12        --       DV         --        C         --       PV          --      --    PI-DV     --       PI      --    PI-PV    --
34




         13        --       PV         --       DV         --        C          --      --    PI-PV     --     PI-DV     --      PI     --
         14        --        C         --       DV         --       PV          --      --      PI      --     PI-DV     --    PI-PV    --
         15        --       DV         --       PV         --        C          --      --    PI-DV     --     PI-PV     --      PI     --
         16        --        PI        --      PI-PV       --      PI-DV        --      --      C       --      PV       --     DV      --
         17        --      PI-DV       --        PI        --      PI-PV        --      --     DV       --       C       --     PV      --
         18        --      PI-PV       --      PI-DV       --        PI         --      --     PV       --      DV       --      C      --
         19        --        PI        --      PI-DV       --      PI-PV        --      --      C       --      DV       --     PV      --
         20        --      PI-DV       --      PI-PV       --        PI         --      --     DV       --      PV       --      C      --

                                        LEGEND
          --    No data collection or interventions at CSCOM
          C     Data collection at CSCOM, no interventions
         DV     Doctor vouchers and data collection at CSCOM
         PV     Patient vouchers and data collection at CSCOM
          PI    Patient information and data collection at CSCOM
        PI-DV   Patient information, doctor vouchers, and data collection at CSCOM
        PI-PV   Patient information, patient vouchers, and data collection at CSCOM


     Notes: The interventions listed above ran between November 14-December 30 2016 in three two-week blocks, with 20 CSCOMs active in
     each two-week block.
                                                 Figure 3: Effect of Extended Provider Training on Allocation of Malaria Treatment at the Clinic

                                                        A. All Patients                                                         B. Patients with RDT at Clinic                                                      C. Patients without RDT at Clinic



                                        .8                                                                                                                                                                 .8
       Share Patients with Malaria Rx




                                                                                      Share Patients with Malaria Rx




                                                                                                                                                                          Share Patients with Malaria Rx
                                                                                                                       .8
                                        .6                                                                                                                                                                 .6

                                                                                                                       .6

                                        .4                                                                                                                                                                 .4
                                                                                                                       .4


                                        .2                                                                                                                                                                 .2
                                                                                                                       .2



                                        0                                                                              0                                                                                   0

                                             0         .2               .4     .6                                           0         .2               .4            .6                                         0           .2               .4         .6
35




                                                      Predicted Malaria Risk                                                         Predicted Malaria Risk                                                                Predicted Malaria Risk


                                                                                    Basic Provider Training                                             Extended Provider Training

     Notes: Outcome in all panels is whether a patient was prescribed an antimalarial. Results from local linear regressions. Graphs omit
     results for top and bottom 2.5 percent of malaria risk distribution to avoid influence of outliers. Vertical dashed lines indicate 25th, 50th,
     and 75th percentiles of predicted malaria risk respectively.
Figure 4: Effect of Extended Training on Perceived Malaria Risk, by Home RDT Result


                                                              Basic Training                                        Extended Training
                                        4
   Perceived Malaria Risk (Out of 5)




                                        3




                                        2




                                        1




                                        0

                                                                            RDT Negative                          RDT Positive
                                       Coefficient and p-value on Provider Training x RDT result difference-in-difference is 0.516 (p=0.097).


 Notes: Data from ex-ante patient predictions of their RDT test results at home survey. Patients
 were asked to score risk on a scale of 1 to 5, with 1 being definitely malaria negative, 2 being malaria
 negative more likely, 3 being positive/negative equally likely, 4 being malaria positive more likely,
 and 5 being definitely malaria positive. The graph shows average patient scores by provider training
 arm and actual home RDT result.




                                                                                           36
                                                                                Figure 5: Effect of Extended Provider Training on Patient Satisfaction

                                                                     Overall Index                                                                             Testing Sub-Index                                                                         Medications Sub-Index
                                1                                                                                             1                                                                                             1



                                .8                                                                                            .8                                                                                            .8
       Cumulative probability




                                                                                                     Cumulative probability




                                                                                                                                                                                                   Cumulative probability
                                .6                                                                                            .6                                                                                            .6



                                .4                                                                                            .4                                                                                            .4



                                .2                                                                                            .2                                                                                            .2



                                0                                                                                             0                                                                                             0
                                     -2             -1.5             -1             -.5     0   .5                                 -2            -1.5             -1             -.5      0   .5                                       -2            -1.5             -1            -.5   0   .5
                                     KS p-value for equality of distributions is p=0.000.                                          KS p-value for equality of distributions is p=0.000.                                          KS p-value for equality of distributions is p=0.011.
37




                                                                                                                                                              Basic Training
                                                                                                                                                              Extended Training

     Notes: Patient satisfaction data from home follow-up of the patient survey.
ONLINE APPENDIX
A   Additional Results

       Table A1: Provider Beliefs in the Control Group: Health Worker Post-Intervention Survey

                                                                                     (1)    (2)       (3)
                                                                                      N    Mean       SD
     Panel A. Malaria Prevalence and Pressure
      Malaria Prevalence: General Population                                          43    0.277    0.226
      Malaria Prevalence: CSCom Patients                                              43    0.443    0.200
      Feels Pressure from Patients to Prescribe Unnecessary Medication                43    0.512    0.506
       Feels Pressure: Antimalarials                                                  22    0.682    0.477
     Panel B. Confidence in RDT
      Uses RDT to Diagnose Malaria                                                    43    0.977 0.152
      Positive out of 100: Simple Malaria                                             43   55.953 32.969
      Patients with Positive RDT out of 100 with Severe Malaria                       43   88.884 16.449
      Would Treat Relative with Negative RDT                                          43    0.465 0.505
     Panel C. Confidence in Microscopy Test
      Uses Microscopy to Diagnose Malaria                                             43    0.698 0.465
      Patients with Positive Microscopy Test out of 100 with Simple Malaria           43   83.977 21.381
      Patients with Positive Microscopy Test out of 100 with Severe Malaria           43   95.442 8.525
      Would Treat Relative with Negative Microscopy Test                              43    0.326 0.474
     Notes: Results from post-intervention health worker survey. Sample limited to clinics that received the
    basic provider training. Malaria prevalence refers to prevalence estimated by the health worker. A health
    worker is coded as feeling pressure to prescribe if s/he answers yes to the question: Do you ever feel
    pressure from patients to prescribe certain medicines when you think they are not necessary? Providers
    answering yes were then asked to specify which medications. Antimalarial also includes quinine.




                                                      A0
                       Table A2: Patient Information Implementation

                                                                   (1)          (2)
                                                               Patient Info
                                                                  Mean           N
                Watched any video (constructed)                    0.795       1002
                                                                  [0.404]
                Watched any video: patient survey                  0.722       1002
                                                                  [0.448]
                Watched any video: admin records                   0.978        686
                                                                  [0.146]
                Watched full video: admin records                  0.892        686
                                                                  [0.310]
                Share of video watched: admin records              0.943        686
                                                                  [0.192]
              Notes: Standard deviations in brackets. Information on whether the
             patient video was watched from two sources. Administrative records
             were kept by intervention officers and contain information whether the
             video was partially watched. Patient survey records were collected by
             surveyors based on observed participation in the intervention. The con-
             structed variable is coded as 1 if at least one record lists the patient as
             receiving the intervention and the other record agrees or is missing for
             this patient (but 0 if neither record lists the patient as yes or one record
             list the patient as no ).

                        Table A3: Selection into Sample by Treatment

                                                (1)         (2)                 (3)        (4)
                                                  Full Sample               Selected: Home Survey
                                               Took          Took             Took          Took
                                              Home           Home-           Home           Home-
                                              Survey         Based           Survey         Based
                                                             RDT                            RDT
 Extended Training                            -0.0260        0.0106         -0.00535         0.0253
                                             (0.0278)       (0.0333)        (0.0205)        (0.0358)
 Patient Information                          -0.0204       -0.00419        0.000465         0.0141
                                             (0.0186)       (0.0170)        (0.0162)        (0.0192)

 Mean (Basic Training, No Pat Info)            0.74           0.54            0.87           0.64
 N                                             1973           1973            1669           1669
 Notes : Robust standard errors clustered at the clinic-pair level in parentheses. All regressions
include survey date fixed effects. *, **, and *** indicate significance at the 10, 5, and 1 percent
levels respectively.


                                                 A1
            Table A4: Demographic Characteristics and Randomization Verification

                                             (1)      (2)          (3)              (4)         (5)
                                                     Regression Coefficients        P-Values
                                          Control Patient         Extended       Joint Test
                                           Mean    Info           Training       PI=ET=0         N
 A. Sample Frame (Clinic × Day-Level Observations)

  Number Eligible Logged Patients           5.89      0.071         0.357           0.775       283
                                           [3.23]    (0.261)        (0.57)
 B. Patient Characteristics

  Number of symptoms                         3.42     0.065         0.165           0.355       1973
                                            [1.57]    (0.06)       (0.188)
  Fever                                     0.828    -.035∗∗         -.007          0.072   ∗
                                                                                                1973
                                           [0.378]   (0.015)       (0.028)
  Chills or Excessive Sweating              0.236     0.001         0.053           0.383       1973
                                           [0.425]    (0.02)       (0.045)
  Nausea, Vomiting, or Diarrhea             0.458     0.002        0.082∗∗          0.087   ∗
                                                                                                1973
                                           [0.499]   (0.019)       (0.035)
  Poor Appetite                             0.483     0.006         0.021           0.868       1973
                                             [0.5]   (0.022)       (0.046)
  Headache                                  0.583     0.037         0.001           0.446       1973
                                           [0.494]   (0.029)       (0.036)
  Cough                                     0.376     0.015         0.005           0.758       1973
                                           [0.485]   (0.022)       (0.028)
  Weakness/Fatigue                          0.458     0.039          0.01           0.248       1973
                                           [0.499]   (0.023)       (0.067)
  Duration of Illness in Days                4.09     0.178        0.513∗∗         0.037   ∗∗
                                                                                                1973
                                            [4.38]   (0.182)       (0.192)
  Age                                        16.8     1.62∗∗         -.175          0.133       1973
                                            [15.5]   (0.784)       (0.804)
  Under 5 Years Old                         0.308    -.037∗∗        0.016           0.103       1973
                                           [0.462]   (0.018)       (0.024)
  Male                                      0.429     0.003          -.023          0.618       1973
                                           [0.495]   (0.022)       (0.023)
  Pregnant (Females Only)                     0.1    -.033∗∗        0.031          0.015∗∗      1101
                                           [0.301]   (0.012)        (0.02)
  Positive RDT (Home Test)+                 0.219      -.016         -.021          0.767       1093
                                           [0.414]   (0.032)       (0.039)
  Predicted Malaria Risk                    0.205      -.003         0.01            0.55       1973
                                           [0.153]   (0.006)       (0.012)
 C. Respondent and Household Characteristics

  Suspects Malaria                         0.587       -.021         -.014          0.765       1973
                                          [0.493]    (0.032)       (0.056)
  Patient Answered Clinic Survey           0.464      0.039         0.004           0.249       1973
                                          [0.499]    (0.023)       (0.026)
  Male                                     0.312       -.003         -.035          0.474       1973
                                          [0.464]    (0.024)       (0.029)
  Bambara                                    0.4      -.038∗        0.026           0.093∗      1971
                                           [0.49]    (0.021)       (0.025)
  Speaks French                            0.532      0.037          -.045          0.313       1973
                                          [0.499]     (0.03)       (0.035)
  Literate (in French)                     0.287     0.048∗∗       -.092∗∗          0.04∗∗
                                                                                                1973
                                          [0.453]    (0.022)       (0.043)
  Primary School or Less                   0.433      -.044∗       0.055∗           0.075   ∗
                                                                                                1973
                                          [0.496]    (0.023)        (0.03)
  Household Size   +
                                            9.95      0.491         1.42∗            0.13       1430
                                           [8.18]    (0.517)       (0.752)
  Share HH Under 15     +
                                           0.417        -.01        0.002           0.611       1427
                                            [0.2]    (0.011)       (0.018)
  Share HH Working+                         0.27       -.002         -.021          0.573       1427
                                          [0.191]    (0.012)       (0.019)
  Household Income Per Capita+             22963        963         -2774           0.356       1374
                                          [24675]    (1410)        (1896)
  Rental Value Home+                       60970       3122          6335           0.248       1408
                                          [76888]    (4045)        (5820)
  Mosquito Nets Per Capita      +
                                           0.488      0.002          -.041          0.423       1424
                                          [0.349]    (0.019)       (0.032)
 Notes: Robust standard errors clustered at the clinic-pair level in parentheses. All regressions include
strata and survey date fixed effects. + indicates that variable was recorded in the home survey only.
Variables measured in CFA and duration of illness top-coded at the 99th percentile. CFA610 ≈ USD1.
*, **, and *** indicate significance at the 10, 5, and 1 percent levels respectively.




                                                   A2
    Table A5: Predicting RDT Positivity With Observables

                                                        (1)
                                                     RDT Positive
 X
 Fever                                                    0.373∗∗
                                                          (0.168)
 Chills or Excessive Sweating                             0.189∗
                                                          (0.108)
 Nausea, Vomiting, or Diarrhea                           0.381∗∗∗
                                                         (0.0960)
 Reduced Appetite                                        0.00691
                                                          (0.102)
 Headache                                                 0.240∗
                                                          (0.126)
 Cough                                                   -0.159∗∗
                                                         (0.0783)
 Weakness, Fatigue, or Reduced Consciousness               0.139
                                                         (0.0976)
 Duration of Illness in Days                            -0.0182∗∗
                                                        (0.00908)
 Age Patient                                             -0.00339
                                                        (0.00562)
 Patient Under 5 Years Old                              -1.378∗∗∗
                                                          (0.216)
 Under 5 × Age                                            0.226∗∗
                                                         (0.0969)
 Patient is Male                                          1.051∗∗
                                                          (0.414)
 Patient is Pregnant                                      -0.348∗
                                                          (0.205)
 Ethnic group: Bambara                                     0.124
                                                         (0.0854)
 Respondent Speaks French                                  -0.223
                                                          (0.138)
 Respondent is Literate in French                       -0.510∗∗∗
                                                          (0.144)
 Respondent Has Primary Education or Less                  -0.140
                                                          (0.120)
 Patient Answered Clinic Survey                          -0.367∗∗
                                                          (0.174)
 Pseudo R-Squared                                          0.141
 N                                                          1093
 Notes : Standard errors clustered at the clinic level in parentheses.
Respondent refers to individual who answered clinic survey. ***, **,
and * indicate significance at the 1, 5, and 10 percent significance
levels respectively.



                                 A3
       Table A6: Effect of Extended Provider Training on the Allocation of Malaria Treatment

                                                 (1)         (2)            (3)         (4)       (5)
                                                   Full Sample              Has Valid Home-Based RDT
                                             Prescribed Prescribed Prescribed Prescribed Prescribed
                                               Anti-      Anti-      Anti-      Anti-      Anti-
                                              malarial   malarial   malarial   malarial   malarial
 Extended Training                             -0.124∗        -0.141∗      -0.0911        -0.114       -0.0896
                                              (0.0718)       (0.0769)     (0.0713)      (0.0748)      (0.0570)
 Extended Training × Malaria Risk               0.107∗        0.293∗∗       0.0980       0.282∗        0.189∗∗
                                              (0.0574)        (0.139)     (0.0642)       (0.165)      (0.0747)
 Malaria Risk Measure                         0.150∗∗∗       0.622∗∗∗     0.157∗∗∗      0.635∗∗∗       0.103∗∗
                                              (0.0407)        (0.104)     (0.0504)       (0.124)      (0.0470)

 ET + ET × Risk = 0                             0.803          0.161        0.918         0.200         0.158

 Malaria Risk Indicator                        Above         Predicted    Above        Predicted       RDT
                                               Median         Malaria     Median        Malaria       Positive
                                               Malaria         Risk       Malaria        Risk
                                                Risk                       Risk

 Mean (Basic Training, No Pat. Info.)           0.67           0.67         0.71          0.71          0.71
 N                                              1971           1971         1093          1093          1093
 Notes : Robust standard errors clustered at the clinic-pair level in parentheses. Standard errors for regressions
that include a derivative of predicted malaria risk in the set of independent variables are bootstrapped using 500
replications to account for the fact that predicted malaria risk is a generated regressor. All regressions include
strata and survey date fixed effects. Additional covariates are selected using double lasso, from the set of all
covariates listed in Table 1 and all individual-level characteristics in Table A4 (excluding predicted malaria
risk and characteristics only measured at the home survey) as well as the square of patient age and illness
duration. We also include pairwise interactions of the aforementioned characteristics. Missing values for all
characteristics are dummied out and recoded to the mean prior to forming interactions. Missing dummies are
also included in the potential covariate set and set of pairwise interactions. *, **, and *** indicate significance
at the 10, 5, and 1 percent levels respectively.




                                                        A4
                      Table A7: Impacts of Extended Provider Training and Patient Information on Other Aspects of Care

                                                         (1)                 (2)                  (3)                  (4)                   (5)
                                                                        Clinic Survey                                         Home Survey
                                                    Number of            Number of            Prescribed         Time Waiting           Time
                                                    Medications         Medications           Antibiotic           at Clinic        Consulting with
                                                    Prescribed           Prescribed                                                    Provider
                                                                         Excluding
                                                                        Antimalarial
      Extended Training                                -0.0698              -0.0853             -0.0200               -0.707              -1.081
                                                       (0.165)              (0.123)            (0.0474)              (1.471)             (0.822)
      Patient Infomation                                0.101                0.106             -0.0410∗               0.479               -0.363
                                                      (0.0728)             (0.0688)            (0.0209)              (0.983)             (0.627)

      Mean (Basic Training, No Pat. Info.)              3.78                 3.00                0.64                 18.25                 13.42
      N                                                 1971                 1971                1971                 1427                  1427
      Notes : Robust standard errors clustered at the clinic-pair level in parentheses. All regressions include strata and survey date fixed effects.
A5




     Additional covariates are selected using double lasso, from the set of all covariates listed in Table 1 and all individual-level characteristics in
     Table A4 (excluding predicted malaria risk and characteristics only measured at the home survey) as well as the square of patient age and illness
     duration. We also include pairwise interactions of the aforementioned characteristics. Missing values for all characteristics are dummied out
     and recoded to the mean prior to forming interactions. Missing dummies are also included in the potential covariate set and set of pairwise
     interactions. *, **, and *** indicate significance at the 10, 5, and 1 percent levels respectively.
                                   Table A8: Effect of Extended Provider Training and Patient Information on Satisfaction Index Components

                                                        (1)               (2)                  (3)               (4)                 (5)                (6)                (7)
                                                      Agrees:           Agrees:          Agrees: Based      Agrees: Should      Agrees: Should     Agrees: Should     Agrees: Should
                                                      Clearly           Clearly          Treatment on        Have Done           Have Done          Have Given         Have Given
                                                     Explained       Explained Test       Test Results       More Tests          Fewer Tests           Differ-             Fewer
                                                     Diagnosis          Results                                                                    ent/Additional      Medications
                                                                                                                                                    Medications
      Extended Training                                -0.0147            -0.0250             -0.0255            0.0182             0.0221              0.0264             0.0238
                                                      (0.0307)           (0.0386)            (0.0239)           (0.0413)           (0.0376)            (0.0337)           (0.0371)
      Patient Information                              0.0337            0.00573              -0.0231            -0.0198            -0.0304            0.00359            -0.00736
                                                      (0.0242)           (0.0293)            (0.0237)           (0.0314)           (0.0192)            (0.0253)           (0.0157)

      Mean (Basic Training, No Pat. Info.)              0.70               0.71                0.91               0.34                0.18               0.18               0.16
      N                                                 1424               850                 845                1407                839                1407               1413
      Notes : Robust standard errors clustered at the clinic-pair level in parentheses. All regressions include strata and survey date fixed effects. Additional covariates are selected
     using double lasso, from the set of all covariates listed in Table 1 and all individual-level characteristics in Table A4 (excluding predicted malaria risk and characteristics only
     measured at the home survey) as well as the square of patient age and illness duration. We also include pairwise interactions of the aforementioned characteristics. Missing
     values for all characteristics are dummied out and recoded to the mean prior to forming interactions. Missing dummies are also included in the potential covariate set and set
     of pairwise interactions. Outcomes indicating dissatifaction (cols 4-7) are multipled by negative 1 when creating overall satisfaction index. *, **, and *** indicate significance
     at the 10, 5, and 1 percent levels respectively.
A6
Table A9: Effect of Extended Provider Training and Patient Information on Patient
Satisfaction

                                                       (1)           (2)            (3)
                                                    Overall        Testing       Medications
                                                    Satisfac-     Sub-Index      Sub-Index
                                                      tion
                                                     Index
 Extended Training                                   -0.0879        -0.0847         -0.123
                                                    (0.0590)       (0.0559)       (0.0989)
 Patient Information                                 0.0366         0.0523         -0.0314
                                                    (0.0326)       (0.0389)       (0.0552)
 Extended Training × Patient Information            -0.00849        -0.0293         0.0838
                                                    (0.0679)       (0.0731)       (0.0912)

 P-Value: ET + ET × PI=0                             0.071∗         0.013∗∗         0.707

 Mean (Basic Training, No Pat. Info.)                 0.00           0.00           0.00
 N                                                    1429           1419           1418
 Notes : Robust standard errors clustered at the clinic-pair level in parentheses. All regres-
sions include strata and survey date fixed effects. Additional covariates are selected using
double lasso, from the set of all covariates listed in Table 1 and all individual-level charac-
teristics in Table A4 (excluding predicted malaria risk and characteristics only measured at
the home survey) as well as the square of patient age and illness duration. We also include
pairwise interactions of the aforementioned characteristics. Missing values for all charac-
teristics are dummied out and recoded to the mean prior to forming interactions. Missing
dummies are also included in the potential covariate set and set of pairwise interactions.
Index components are standardized relative to the basic provider training mean. In cases
where some (but not all) index components for a respondent are missing, missing compo-
nents are imputed to the treatment group mean prior to standardization. *, **, and ***
indicate significance at the 10, 5, and 1 percent levels respectively.




                                             A7
Table A10: Effect of Extended Provider Training on Match Between Patient Prior and
Prescription by Accuracy of Pre-Consultation Malaria Beliefs

                                                      (1)              (2)           (3)
                                                 Full Sample       Has Valid Home-Based RDT
                                                Prescription       Prescription      Prescription
                                                 Matches            Matches           Matches
                                                Malaria Prior      Malaria Prior     Malaria Prior
 Extended Training                                   -0.123            -0.156∗          -0.129∗∗
                                                   (0.0841)           (0.0880)          (0.0590)
 Extended Training × Prior-Risk Match                0.222              0.249            0.183∗
                                                    (0.163)            (0.171)          (0.0942)
 Match: Malaria Prior and Malaria Risk             0.615∗∗∗           0.526∗∗∗          0.0902∗
                                                   (0.0836)           (0.0808)          (0.0533)

 P-value: ET + ET × Prior Match = 0                   0.245            0.318             0.304

 Prior Match Measure                              Expected          Expected          Actual Prior
                                                 Prior Match       Prior Match          Match
                                                                                     (Home RDT)

 Mean (Basic Training, No Pat. Info.)                 0.63              0.66              0.66
 N                                                    1971              1093              1093
 Notes : Robust standard errors clustered at the clinic-pair level in parentheses. Standard errors for
regressions that include a derivative of predicted malaria risk in the set of independent variables
are bootstrapped using 500 replications to account for the fact that predicted malaria risk is
a generated regressor. All regressions include strata and survey date fixed effects. Additional
covariates are selected using double lasso, from the set of all covariates listed in Table 1 and all
individual-level characteristics in Table A4 (excluding predicted malaria risk and characteristics
only measured at the home survey) as well as the square of patient age and illness duration.
We also include pairwise interactions of the aforementioned characteristics. Missing values for all
characteristics are dummied out and recoded to the mean prior to forming interactions. Missing
dummies are also included in the potential covariate set and set of pairwise interactions. Outcome
is equal to one for patients who, pre-consultation, believe they have malaria and are prescribed
an antimalarial or for patients who believe they do not have malaria and are not prescribed an
antimalarial. *, **, and *** indicate significance at the 10, 5, and 1 percent levels respectively.




                                                 A8
                                                Figure A1: Effect of Extended Provider Training on Patient Satisfaction

                                      A. Aggregated Index                                            B. Testing Sub-Index                                               C. Medication Sub-Index
                            .4                                                             .4
                                                                                                                                                              .4



                                                                                                                                                              .2
                            .2                                                             .2
       Satisfaction Index




                                                                      Satisfaction Index




                                                                                                                                         Satisfaction Index
                                                                                                                                                               0

                             0                                                              0

                                                                                                                                                              -.2


                            -.2                                                            -.2                                                                -.4




                            -.4                                                            -.4

                                  0     .2              .4     .6                                0     .2              .4           .6                              0       .2              .4     .6
A9




                                      Predicted Malaria Risk                                         Predicted Malaria Risk                                               Predicted Malaria Risk


                                                                    Basic Provider Training                            Extended Provider Training

     Notes: Patient satisfaction data from home follow-up of the patient survey. Results from local linear regressions. Graphs omit results
     for top and bottom 2.5 percent of malaria risk distribution to avoid influence of outliers. Vertical dashed lines indicate 25th, 50th, and
     75th percentiles of predicted malaria risk respectively. Shaded areas give 90% confidence intervals, based on bootstrapped standard errors
     clustered at the clinic-pair level. To account for the fact that predicted malaria risk is a generated regressor, we re-calculate predicted risk
     for each bootstrap replication.
     B     Main Results Without Double-Lasso Selected Controls

     Table B1: Effect of Extended Provider Training on Learning: Provider Post-Training and Clinic Endline Surveys - No Additional
     Controls

                                    (1)          (2)          (3)             (4)            (5)             (6)         (7)            (8)
                                                 Beliefs About RDTs                                  Beliefs About Microscopy Tests
                                                                                                     Share      Share
                                            Share RDT      Share RDT                               Microscopy Microscopy   Share
                                  RDT        Positive       Positive      Share RDT     Microscopy  Positive:  Positive: Microscopy
                                  Beliefs     Simple         Severe        Negative      Beliefs     Simple     Severe    Negative:
                                  Index      Malaria        Malaria       No Malaria      Index     Malaria    Malaria   No Malaria
      A. Data from Post-Training Survey
       Extended Training      0.347∗∗∗        0.221∗∗∗       0.071∗∗∗        0.016          0.027          -0.002        -0.005        0.037
                               (0.075)        (0.046)        (0.019)        (0.045)        (0.118)        (0.036)       (0.029)       (0.049)
B0




       Mean (Basic Training)       0.000        0.599         0.888          0.863          0.000          0.846        0.921          0.829
       N                            205          198           196            176            201            197          195            173

      B. Data from Provider Follow-Up Survey
       Extended Training      0.380∗∗∗    0.186∗∗            0.056∗          0.039          0.188          0.038         0.008         0.074
                               (0.132)   (0.084)             (0.033)        (0.026)        (0.166)        (0.046)       (0.022)       (0.048)
      B. Provider Attendance
       Mean (Basic Training)       0.000        0.560         0.889          0.938          0.000          0.840        0.954          0.868
       N                             85           85           85             85              85             85           85             85
      Notes : Robust standard errors clustered at the clinic-pair level in parentheses. All regressions include strata fixed effects. Sample in both
     panels limited to providers who attended the training. Index components are standardized relative to the basic provider training mean. In cases
     where some (but not all) index components for a respondent are missing, missing components are imputed to the treatment group mean prior to
     standardization. *, **, and *** indicate significance at the 10, 5, and 1 percent levels respectively.
Table B2: Effect of Extended Provider Training and Patient Information on Malaria Testing
at the Clinic: Patient Survey – No Additional Controls

                                                 (1)            (2)            (3)              (4)
                                                 Any          RDT Only      Microscopy        Multiple
                                                Malaria                       Only             Tests
                                                 Test
 Extended Training                               -0.0428       0.0525           -0.0359       -0.0667∗∗
                                                (0.0550)      (0.0565)         (0.0688)        (0.0250)
 Patient Information                           -0.0699∗∗∗     -0.0470∗∗         -0.0184        0.00677
                                                (0.0187)      (0.0202)         (0.0251)       (0.00868)

 Mean (Basic Training, No Pat. Info.)             0.58           0.21            0.25             0.07
 N                                                1973           1973            1973             1973
 Notes : Robust standard errors clustered at the clinic-pair level in parentheses. All regressions include
strata and survey date fixed effects. *, **, and *** indicate significance at the 10, 5, and 1 percent
levels respectively.


  Table B3: Effect of Extended Provider Training and Patient Information on the Alloca-
  tion of Malaria Treatment – No Additional Controls

                                                      (1)              (2)           (3)
                                                 Full Sample       Has Valid Home-Based RDT
                                                   Expected          Expected        Actual Match
                                                    Match             Match
   Extended Training                                0.0506             0.0377             0.0888∗
                                                   (0.0329)           (0.0336)            (0.0478)
   Patient Infomation                              0.00134             0.0124              -0.0110
                                                   (0.0177)           (0.0182)            (0.0341)

   Mean (Basic Training, No Pat. Info.)                0.44             0.41               0.43
   N                                                   1971             1093               1093
   Notes : Robust standard errors clustered at the level in parentheses. All regressions include
  strata and survey date fixed effects. *, **, and *** indicate significance at the 10, 5, and 1
  percent levels respectively.




                                                  B1
Table B4: Effect of Extended Provider Training and Patient Information on
Patient Satisfaction – No Additional Controls

                                                (1)            (2)           (3)
                                               Overall       Testing      Medications
                                               Index        Sub-Index     Sub-Index
 Extended Training                             -0.0985∗∗     -0.107∗∗∗       -0.0928
                                               (0.0395)      (0.0321)       (0.0793)
 Patient Information                            0.0326        0.0386        0.00797
                                               (0.0243)      (0.0324)       (0.0335)

 Mean (Basic Training, No Pat. Info.)            0.00          0.00           0.00
 N                                               1429          1419           1418
 Notes : Robust standard errors clustered at the clinic-pair level in parentheses. All
regressions include strata and survey date fixed effects. Index components are stan-
dardized relative to the basic provider training mean. In cases where some (but not all)
index components for a respondent are missing, missing components are imputed to
the treatment group mean prior to standardization. *, **, and *** indicate significance
at the 10, 5, and 1 percent levels respectively.




                                          B2
Table B5: Effect of Extended Provider Training on Patient Satisfaction by Accuracy of
Pre-Consultation Malaria Beliefs - No Additional Controls

                                                      (1)              (2)           (3)
                                                 Full Sample       Has Valid Home-Based RDT
                                                    Overall           Overall           Overall
                                                  Satisfaction      Satisfaction      Satisfaction
                                                     Index             Index             Index
 Extended Training                                 -0.156∗∗           -0.170∗∗          -0.138∗∗
                                                   (0.0634)           (0.0730)          (0.0622)
 Extended Training × Prior-Risk Match               0.127∗             0.164∗∗           0.0885
                                                   (0.0652)           (0.0759)          (0.0579)
 Match: Malaria Prior and Malaria Risk             -0.116∗∗            -0.0934           -0.0480
                                                   (0.0467)           (0.0580)          (0.0455)

 P-value: ET + ET × Prior Match = 0                   0.420            0.883             0.189

 Prior Match Measure                              Expected          Expected          Actual Prior
                                                 Prior Match       Prior Match          Match
                                                                                     (Home RDT)

 Mean (Basic Training, No Pat. Info.)                 0.00              -0.02             -0.02
 N                                                    1429              1092              1092
 Notes : Robust standard errors clustered at the clinic-pair level in parentheses. Standard errors for
regressions that include a derivative of predicted malaria risk in the set of independent variables
are bootstrapped using 500 replications to account for the fact that predicted malaria risk is a
generated regressor. All regressions include strata and survey date fixed effects. *, **, and ***
indicate significance at the 10, 5, and 1 percent levels respectively.




                                                 B3
Table B6: Effect of Extended Provider Training on Match Between Patient Prior and
Prescription by Accuracy of Pre-Consultation Malaria Beliefs - No Additional Controls

                                                      (1)              (2)           (3)
                                                 Full Sample       Has Valid Home-Based RDT
                                                Prescription       Prescription      Prescription
                                                 Matches            Matches           Matches
                                                Malaria Prior      Malaria Prior     Malaria Prior
 Extended Training                                  -0.0768            -0.120            -0.113∗
                                                   (0.0887)          (0.0941)           (0.0660)
 Extended Training × Prior-Risk Match                0.211             0.264             0.237∗∗
                                                    (0.162)           (0.167)           (0.0942)
 Match: Malaria Prior and Malaria Risk             -0.282∗∗∗         -0.352∗∗∗           -0.139∗
                                                    (0.107)           (0.117)           (0.0728)

 P-value: ET + ET × Prior Match = 0                   0.126            0.133            0.033∗∗

 Prior Match Measure                              Expected          Expected          Actual Prior
                                                 Prior Match       Prior Match          Match
                                                                                     (Home RDT)

 Mean (Basic Training, No Pat. Info.)                 0.63              0.66              0.66
 N                                                    1971              1093              1093
 Notes : Robust standard errors clustered at the clinic-pair level in parentheses. Standard errors for
regressions that include a derivative of predicted malaria risk in the set of independent variables
are bootstrapped using 500 replications to account for the fact that predicted malaria risk is a
generated regressor. All regressions include strata and survey date fixed effects. Outcome is
equal to one for patients who, pre-consultation, believe they have malaria and are prescribed
an antimalarial or for patients who believe they do not have malaria and are not prescribed an
antimalarial. *, **, and *** indicate significance at the 10, 5, and 1 percent levels respectively.




                                                 B4
C     Additional Information
C.1    Patient and Doctor Vouchers’ Results
In Lopez et al. (2022), we study whether overprescription of antimalarials is driven by pa-
tients or health providers. We develop a theoretical model of doctor-patient interactions
where the two sides may have different preferences about the treatment for an illness. We
determine the conditions under which patient-driven demand or doctor-driven (induced)
demand for antimalarials can occur. To test the predictions of the model, we conducted
an experiment that offered a discount for simple malaria treatment through vouchers for a
free course of ACT tablets. This subsidy made simple malaria treatment more appealing
compared to both no treatment and treatment for severe malaria. The experiment varied
whether patients were given the discount voucher directly (PV) or whether providers re-
ceived the vouchers to distribute at their discretion (DV), allowing them to either inform
the patients about the voucher to enhance the appeal of simple treatment or not.
    We find that patients are nine percentage points (35 percent) more likely to redeem a
voucher on PV days compared to DV days. Patients visiting the clinic on PV days were 5.2
percentage points more likely to leave with a malaria prescription and 14 percentage points
more likely to purchase an antimalarial treatment compared to Control days. In contrast,
DV days did not affect prescription rates and increased purchases by 8.5 percentage points.
Thus, providers shared information selectively and gave vouchers more often to patients they
were already prescribing to, rather than issuing new prescriptions. The increased demand on
PV days was driven by patients with the fewest malaria symptoms. These findings strongly
support the patient demand channel. If doctors were promoting overuse, they would inform
low-risk patients about the vouchers to boost sales; however, the data show that doctors tend
to avoid prescribing to patients who are least likely to have malaria and use the voucher to
reduce costs for malaria-positive patients. We rule out that providers in DV simply forgot
about the vouchers or were increasing revenue, or that PV patients were stockpiling drugs.
    We also assess the match between malaria treatment and patientsâ malaria status (from
the home-based malaria RDTs) by creating a measure that indicates a “correct positive”
(a malaria-positive patient purchased an antimalarial) and a “correct negative” (a malaria-
negative patient did not purchase treatment). In the Control group, we estimate that 57
percent of patients received the correct malaria treatment. The PV condition reduced this
match by 8 percentage points, compared to only 4 percentage points in the DV condition.




                                             C0
C.2    Patient Information Video
Below is a full transcript of the patient information video to show the points we wanted
to convey to the patients before the consultation with the doctor. These were: the correct
diagnosis of malaria by conducting a test, the explanation of how an RDT test works, which
is the correct treatment in case of a positive test, to encourage patients to ask questions,
and to explain the consequences of taking unnecessary medication.
Transcript of the video [Original version in French]
[Voice-off] (rain images, mosquitoes) Malaria is a common disease during the rainy season
     in Mali. It is caused by a parasite that enters the blood through the bite of a mosquito.
     Fortunately, malaria tests can detect the disease, and artemisinin-based combination
     therapies (ACT) can completely cure the disease.
[Office of the doctor, a mother with her 7-year-old child]
Mother : Hello, doctor.
Doctor : Hello Madam. Take a seat. What brings you here today?
Mother : Ali was very ill for two days. He has a very hot body and does not want to eat. He
     sweats a lot. I think he could have malaria.
[Silent images: close-up of the sweaty child; view from afar: the mother putting her hand on
     the forehead of the child and looking worried]
[Voice-off] High fever that comes at regular intervals, chills and excessive sweating, vomiting
     and nausea are typical symptoms of malaria. Some patients also have diarrhea.
Doctor (standing): His illness could be malaria, but we need to do a blood test to be sure.
     Indeed, many different diseases have symptoms similar to malaria, for example typhoid
     fever or pneumonia. It is important to find the true cause of the disease so Ali can get
     proper treatment.
[Scene: The doctor lays out the test materials on the table.]
Doctor (to child) : For this rapid detection test, I am going to prick your finger and take a
     small amount of blood.
[Scene: Showing the needle and pipette.]
Doctor (in voice off, close up of test) : Your blood and this liquid (show solvent) go into
     the test cassette and will flow to the other end. Look at the two marks C and Pf. A
     red line should appear where the C is. This shows that blood is flowing correctly in the
     test cassette and that the test is going well. The mark with the sign “Pf” is the most
     important. There is a line at this mark which will turn red if the passing blood has been
     contaminated with the malaria parasite. You must wait 15 minutes before reading the
     test result, because all the blood must flow through the test cassette first. If after 15


                                             C1
    minutes there is only the red line at the “C” symbol, then we will continue reading the
    test. If after 30 minutes there is only one line, we can conclude that Ali does not have
    malaria. Only when there is a red line on the Pf before the end of 30 minutes, that we
    can conclude that the patient needs treatment for malaria.
Child to mother, moaning : My head really hurts.
Mother : A little patience my son. The doctor will soon prescribe you a medicine against
    malaria.
Doctor : I understand that you are worried about Ali. But we wouldn’t want to treat him
    for malaria if he has something else, like typhoid fever. Malaria drugs would be of no
    use, and you would be spending money on treatment he doesn’t need. Also, Ali’s real
    illness will get worse during this time. I want him to have the right treatment, so he will
    be better as soon as possible. I will start the test now and check the time.
[Voice-off] Video showing the image of the policy document and the relevant text of the
    policy, first without, then with highlights on the relevant passages: It is really important
    not to start treatment without confirming with a test that the patient has malaria – this
    is the international standard for malaria care, and it is also Mali’s national policy. RDTs
    are provided free of charge by the government and its partners.
[Voice-off] Video of a doctor drawing blood and a lab technician putting a slide under a
    microscope and looking through it: If your CSCom has a laboratory, it can do a thick-
    drop or thin-smear test. This method takes longer than an RDT, at least an hour, but
    it can tell how many parasites there are in the blood. No matter what test is done, you
    only need malaria medicine if the test confirms that you have the disease.
[Scene: Doctor and mother again, with the child]
Doctor, looking at his watch : The 15 minutes are up and the test is ready. (Close-up test):
    Blood has flowed through the test and the line marked âCâ shows that the test is working.
[Scene: Doctor showing test to mother, then close up of test again]
Doctor : Look, there are two lines on this test. This confirms that Ali needs malaria treat-
    ment. I’m going to prescribe you an ACT pack. Ali must take the ACT for three days.
    Even if his health improves, be sure to give him all the tablets at the agreed times.
    Otherwise, he could have a relapse!
Mother : My neighbor’s child had malaria too. He was given an injection (here consider
    injection and infusion; in Bambara “serum and shot”) which had an effect very quickly.
    Will Ali also receive an injection (“serum and shot”)?
Doctor : I’m glad you asked that question. It is a good thing that patients ask questions
    during consultations. Itâs true, malaria is sometimes treated with an injection (“serum
    and shot”). But injections (“serum and shot”) pose more risks than tablets, and they


                                              C2
       are more expensive. An injection (“serum and shot”) is reserved for what we call severe
       malaria. Severe malaria has symptoms such as coma, difficulty breathing, or several
       convulsions over a short period. A sign in small children is that they are unable to
       nurse. A patient suffering from severe malaria urgently needs to go to the hospital.
       Severe malaria most often occurs in children under five and pregnant women.
   Ali does not show signs of severe malaria, and he can swallow the pills I prescribed. This
       means that the injection (“serum and shot”) is not necessary.
   Mother : Oh, I see.
   Doctor : Start taking the ACTs right away. Ali will be much better soon. If his symptoms
       don’t change, come back to me, so we can make sure he’s on the road to recovery.
   Mother : Thank you, doctor. Goodbye!
   Doctor : Goodbye. And better health.
   [Scene taken from afar: Mother buying tablets at the pharmacy, the child swallowing his
       first tablet.]
   [Voice-off] Remember: Before taking any medicine for malaria, always have your blood tested
       to confirm malaria. Rapid detection tests are free at the CSCom. The recommended
       treatment for uncomplicated malaria is a drug called an ACT, taken for three days. In
       some cases, patients have severe malaria. Severe malaria has specific symptoms that a
       doctor can diagnose and requires hospitalization at the health center. Injections (“serum
       and shot”) are for the treatment of severe malaria only or for patients who cannot swallow
       tablets.
[END]




                                                 C3
                     Figure C1: Patient Information Video Outtakes




Notes: The doctor is about to explain to the mother that a positive test result is needed to treat
malaria.


                           Figure C2: Patient Information Video




Notes: The video included subtitles in case of a noisy environment or sound difficulties.




                                               C4
C.3     Construction of Satisfaction Indices
As described in Section 4, we followed the procedure in (Kling et al., 2007) to construct three
standardized indices of patient satisfaction: a sub-index that focuses on satisfaction with
tests, a sub-index focused on satisfaction with medications, and an overall satisfaction index
that includes both sub-indices and the question about care. Below, we list the individual
questions asked in our home survey, and then we explain how the indices were constructed
step by step.
Questions to Measure Satisfaction with Testing

   • How strongly do you agree or disagree with the statement “The doctor/nurse clearly
     explained the medical tests and their results to me?”

   • How strongly do you agree or disagree with the statement “The doctor/nurse based
     the treatment decision strongly on the result of my medical tests?” OR How strongly
     do you agree or disagree with the statement “The doctor/nurse based the treatment
     decision strongly on the result of [patient name]’s medical tests.”

   • How strongly do you agree or disagree with the statement “The doctor/nurse should
     have done additional medical testing before prescribing treatment”?

   • How strongly do you agree or disagree with the statement “The doctor/nurse should
     have carried out fewer tests”?

Questions to Measure Satisfaction with Medications

   • Think of the treatment you got at the clinic or the drugs the doctor prescribed. How
     strongly do you agree or disagree with the statement “The doctor should have given me
     a different or additional treatment or drug”? OR Think of the treatment the patient
     got at the clinic or the drugs the doctor prescribed. How strongly do you agree or
     disagree with the statement “The doctor should have given [patient name] a different
     or additional treatment or drug”?

   • Think of the treatment you got at the clinic or the drugs the doctor prescribed. How
     strongly do you agree or disagree with the statement “The doctor should have pre-
     scribed fewer drugs”? OR Think of the treatment [patient name] got at the clinic
     or the drugs the doctor prescribed. How strongly do you agree or disagree with the
     statement “The doctor should have prescribed fewer drugs”?




                                              C5
Overall Care

   • How strongly do you agree or disagree with the statement “The doctor/nurse clearly
     explained to me what illness I have”? OR How strongly do you agree or disagree with
     the statement, “The doctor/nurse clearly explained to me what illness [patient name]
     has”?

   To facilitate the understanding of these questions, the surveyors showed the following
image:




Construction of Indices We computed dummy variables to indicate agreement (strongly
agree and somewhat agree) for those statements with a positive implication for quality of care
(e.g. “the doctor explained the medical tests and their results"), and negative of agreement if
the question had a negative implication for quality of care (e.g. “the doctor should have done
additional medical testing before prescribing treatment”). We imputed the mean of these
dummy variables within each treatment group in case of missing index components (only if at
least some components of each index were non-missing values). We then standardized each
component relative to the group that received neither the Extended Training nor the Patient
Information interventions. Finally, we computed the average of each index’s standardized
components.




                                              C6