Policy Research Working Paper 10590 The Fine Line between Nudging and Nagging Increasing Take-up Rates through Social Media Platforms Andrés Moya Sandra V. Rozo María José Urbina Development Economics A verified reproducibility package for this paper is Development Research Group available at http://reproducibility.worldbank.org, October 2023 click here for direct access. Policy Research Working Paper 10590 Abstract This study assesses if nudges in the form of informational control group. The effects are mostly driven by the treated videos sent via WhatsApp are effective in boosting take-up individuals who received the links but did not watch the rates among vulnerable populations, specifically in the videos, who are older, busier, and have less internet access context of a regularization program for Venezuelan forced relative to other treated individuals. Additionally, the study migrants in Colombia. The study randomly assigned evaluates the effectiveness of iterative WhatsApp surveys in 1,375 eligible migrants to receive one of three informa- collecting data from hard-to-reach populations. It finds that tional videos or be in a control group. The videos aimed at while iterative WhatsApp surveys had low retention rates, solving issues related to awareness, trust, and bottlenecks iterative contacts helped to reduce attrition. Furthermore, in the step-by-step registration. The main results indicate switching behaviors from nonresponse to response were that program take-up rates for individuals who received any common after iterative contact attempts. video were eight percentage points lower compared to 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 sandrarozo@worldbank.org. A verified reproducibility package for this paper is available at http:// reproducibility.worldbank.org, click here for direct access. 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 The Fine Line between Nudging and Nagging: Increasing Take-up Rates through Social Media Platforms* Maria Jos´ r † e Urbina⃝ Andr´ r ‡ es Moya⃝ Sandra V. Rozo§ Keywords: Refugees, Amnesties, Program Take-up. JEL Classification: D72, F02, F22, O15, R23 * The order in which the authors’ names appear has been randomized using the AEA Author Randomiza- tion Tool (#lzrenF4fqQRW), denoted by ⃝ r . We thank participants of the World Bank Half Baked Seminar, World Bank Human Development seminar, Economics seminar at the University of Virginia, Hilton Foun- dation seminar, and ESOC conference at Universidad del Rosario, as well as Erin Kelley, Aart Kraay, Carlos ¨ Scartascini, David McKenzie, and Berk Ozler for useful comments. We thank IPA Colombia, our partners in this study, particularly Kyle Holloway, Alejandra Rivera, Laura Vargas, Nicole Lesmes, Jacobo Morales, and Ana Mar´ ıa Rojas for their support in data collection and intervention implementation. We are also thankful to Ana Mar´ ıa Ib´ ˜ anez, who contributed in the early stages of this project, and to Kike Yra Fonton and Andr´ es Barinas, who provided excellent research support for the project. This project was approved by IPA IRB protocol 3911. It was also preregistered at the AEA RCT Registry (AEARCTR-0008672). We thank the Hilton Foundation and the RSB Funds of the World Bank for providing financial support for this project. The authors have no conflicts of interest to report. The findings, 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. † World Bank, E-mail: murbinaflorez@worldbank.org ‡ Universidad de los Andes. E-mail: a.moya@uniandes.edu.co § Development Research Group, World Bank. Corresponding author: sandrarozo@worldbank.org I INTRODUCTION Individuals from vulnerable populations often display low enrollment in public services that could enhance their welfare (Currie 2006). This is primarily because access to public programs is not automatic and applicants must satisfy stringent criteria to qualify, which imposes disproportionate costs and restrictions on disadvantaged individuals. Given this context, social media platforms like WhatsApp are a cost-effective way to disseminate in- formation about public programs and improve take-up rates through nudges. Yet, despite the widespread use of WhatsApp for this purpose, there is scant empirical evidence on its efficacy in enhancing take-up among vulnerable populations. Informational videos sent to mobile phones can reach many people at low cost, partic- ularly those who lack access to in-person outreach or distrust the government. Further- more, data collection through WhatsApp may be a productive way to elicit information from highly mobile populations. This study addresses two broad questions: (i) can in- formational videos distributed via WhatsApp increase take-up rates for public programs among vulnerable and hard-to-reach populations?, and (ii) how effective are iterative WhatsApp surveys at collecting data from these populations? This study focuses on Colombia’s Estatuto Temporal de Proteccion para Migrantes Vene- zolanos (ETPV), a 10-year regularization program for undocumented Venezuelan forced migrants (herein “migrants”). The ETPV grants legal status and benefits to migrants who arrived in Colombia before January 2021. It issues a temporary protection permit (PPT) as an identification and regularization document that grants access to numerous services including healthcare, public, and financial services as well as a work permit. Previous studies of the impacts of an earlier version of this program on migrants’ welfare docu- mented large improvements in income and consumption, but low program take-up rates by migrants (Ib´ ˜ et al. 2022). anez The registration period for the PPT began in May 2021 and ended in June 2022. Prior to the 2 program’s implementation, our research team conducted a qualitative study of the rea- sons for low take-up rates for previous regularization programs offered by the Colombian government.1 We identified three primary limitations that have long been recognized in the literature as obstacles to adquire public programs in other settings. These limitations include: (i) low awareness of the program (Chetty, Friedman and Saez 2013, Smeeding and O’Connor 2000); (ii) lack of trust in the government due to legal problems that arise during the regularization process; and (iii) lack of knowledge about the step-by-step registration process, including confusion regarding program rules and incentives (Liebman and Zeckhauser 2004) and psychological aversion to program complexity or the “hassles” involved (Bertrand, Shafir and Mullainathan 2006). Insights from these works guided the design of our experiment. We employed a randomized controlled trial design to investigate the impact of three video treatments sent through WhatsApp on the PPT take-up rates among 1,375 undoc- umented Venezuelan migrants. Participants were randomly assigned to one of the three treatment arms or a control group. Each treatment arm consisted of a video that ad- dressed one or more of the three limitations identified above. In designing these videos, the research team incorporated behavioral insights to strengthen effectiveness. Specifi- cally, we utilized the EAST methodology developed by the Behavioral Insights Team (BIT 2014, DellaVigna and Linos 2022), which emphasizes the principles of making informa- tion easy to understand, attractive, social, and timely. Video 1 aimed to increase program awareness. It provided a detailed description of the principal benefits of the PPT, including its three-step application process, simple nature, low cost (i.e., it was free), and eligibility (the program was open to any migrant who arrived before January of 2021 to Colombia). The video was narrated by an actor who resembled a Colombian official. We refer to this treatment as the awareness video. Video 2 was designed to increase trust in the program. It presented the same information as 1 The qualitative study consisted of 42 in-depth interviews with Venezuelan migrants residing in Colom- bian cities with a high density of migrants. Roughly half of the migrants were undocumented and the other half were documented. 3 Video 1 but was narrated by a Venezuelan migrant who had already registered in the program. The narrator provided a personal account of her experience, highlighting the PPT’s benefits and emphasizing its legitimacy and safety. We refer to this treatment as the trust video. Video 3, narrated by the same Venezuelan migrant as Video 2, provided more details on and support for the step-by-step registration process. It aimed to reduce confusion through clear, concise information on requirements and procedures. Video 3, which we call the step-by-step video, was also the longest one. None of the treatments offered an option for migrants to ask questions; the videos simply directed them to public offices where they could obtain more information and guidance. The first two videos (awareness and trust) were each roughly 3.5 minutes long, whereas the step-by-step video was 5:28 minutes long. We recruited undocumented Venezuelan migrants in person in the departments of Mag- antico on the Caribbean Coast of Colombia. According to the Colombian dalena and Atl´ population census of 2018, these regions have a high density of very vulnerable migrants with low rates of regularization. To ensure representativeness, we collected the sample from the largest urban center (Santa Marta) and surrounding areas. We advertised the experiment in places frequented by Venezuelan migrants and contacted local community leaders to support it by placing registration points in marginalized communities with a high concentration of undocumented migrants. The study included individuals who met the following criteria: (i) born in Venezuela, (ii) aged 18 years or older, (iii) had internet access and WhatsApp, and (iv) had not yet sched- uled an appointment to provide biometric data, a requirement for the PPT. We defined the biometric data appointment as the key point in the PPT registration process because it is the final step before the document is issued. We wanted to identify the most vulnerable individuals who might not apply for the program without external support. To do this, we collected a short, in-person, baseline survey at the registration points. We then com- 4 pared characteristics of the migrants in our sample to those of similar national surveys and found that indeed, our sample included some of the most vulnerable migrants.2 The treatment groups received their specified videos four times through WhatsApp. This transmission occurred for the first time two months after the initial recruitment and up to a total of four times, with a one-week interval between contacts. At each contact, treated participants received the designated video and a short digital survey. The control group only received the survey. Any participant who reported requesting their biometric data appointment was no longer contacted.3 We evaluate the effects of the videos on three main outcomes: intention to register, starting the registration process, and requesting the PPT (proxied by requesting or attending the biometric data appointment).4 Our ex- periment was successful in inducing random variation across groups—i.e., groups are balanced in the vast majority of variables that we collected at baseline (2 out of 56 tests of mean differences were successful). Moreover, attrition rates were not systematic and had a more random nature. Surprisingly, we find negative effects of sending the video on the three outcomes that we examine. We document that receiving a video reduced the intention to register by 12.2 pp, lessened the probability of starting the registration process by 7.7 pp, and decreased the likelihood of requesting the PPT by 8 pp. These are meaningful effects. The treatment resulted in a reduction of 15.09 percent in PPT take-up rates relative to the control group’s mean. Furthermore, the step-by-step video, which was the longest video and offered more details was the one that reduced take-up the most. When exploring for potential explanations for the negative effects of the videos on PPT 2 Details of this comparison are in Table 2, which is described in detail later. 3 For reasons of confidentiality, we were denied access to public data on actual program registration. 4 Unfortunately, IRB restrictions prevented us from collecting the ID information which complicated matching our data with public official information. When attempting to match our sample with the gov- ernment records, it was extremely common to find homonym names, which prevented us from matching the majority of our sample with the government data. As such we were not able to use the government’s information to confirm the reports from migrants on PPT registration. 5 take-up rates, we document that the results are primarily driven by the treated individ- uals who received the links but did not played the videos. These individuals account for 15.4 percent of the individuals assigned to treatment. We document that they are older, busier, and have less internet and WhatsApp access, relative to the rest of the treated group. We confirm this result by using multiple strategies such as propensity score matching, sample restrictions, and traditional heterogeneous effects analysis. All in all, the results of these exercises consistently suggest that the videos had effects close to zero for the individuals who watched the videos, but induced large and negative effects for the individuals in the treatment who did not watch the videos. Our qualitative semi-structured interviews support these conclusions suggesting that in- dividuals in the treatment group who did not watch the videos reported frustration about the multiple contacts, due to their difficulties with low technology literacy, and with their low propensity to engage in informational videos. Interestingly, these results align with other recent studies that have also suggested negative effects of nudges for populations that are not searching for the received information. For example, Damgaard and Gravert (2018) show that reminders for charity increased the intended behaviors for some, but also unintentionally increase avoidance behaviors for others who were annoyed by the messages (for instance they decided to unsubscribe from mailing lists). Kalil et al. (2023) establish that reminder text messages to parents led to a decrease in literacy skills for children— the main outcome they targeted through a literacy platform. Moreover, Costa and Kahn (2013) show that energy conservation nudges need to be targeted to be most effective. For all individuals recruited and contacted at each stage, we explored the share that opened the link and played the videos. The most successful episode of treatment was the first contact, in which more than 70 percent of participants opened and played more than half the video. Participants who were contacted more times opened the video less and 6 played it fewer times, possibly due to fatigue or familiarity with the information. We also find play rates were lower for the videos narrated by a Venezuelan migrant (treatments 2 and 3) than for the one narrated by an actor resembling a Colombian official (treatment 1). That is, even a narrator who could speak about experiences similar to theirs did not increase migrants’ interest in the information. Our study suggests that sending information videos through social platforms is an inef- fective way to increase public program take-up rates for vulnerable individuals who are hard-to-reach and that it may be even harmful for individuals who are not interested in receiving the information. We also analyzed the effectiveness of iterative WhatsApp surveys (IWS) at collecting information from vulnerable populations, particularly undocumented forced migrants. Our study revealed five key findings. First, we lost half the relevant sample when tran- sitioning from in-person interviews conducted at recruitment (1,375) to WhatsApp sur- veys post-treatment. Moreover, the attrition rate increased as more contact attempts were made. Finally, we observed a switching behavior from non-response to response in at least 20 percent of the sample. For instance, of the participants contacted four times, 13.17 percent responded twice (comprising 7.22 percent consecutive and 5.95 percent non- consecutive responses) and 13.43 percent responded three times (8.87 percent consecutive and 4.56 percent non-consecutive responses). Our findings suggest that while IWS may be useful for gathering information in some contexts, they may not be well-suited for obtaining information from hard-to-reach populations in developing countries. Other methods such as in-person interviews, focus groups, or community engagement may be more effective in these settings. Contribution to literature: Our work builds on the literature concerning public program take-up rates and their determinants. Prior studies have identified information asymme- try (Daponte, Sanders and Taylor 1999, Bartlett and Hamilton 2004, Bettinger et al. 2012, 7 Armour 2018) and the high cost of learning about program eligibility and application procedures as major obstacles to enrollment (Chetty, Friedman and Saez 2013). Misinfor- mation also contributes to low take-up rates, creating confusion about eligibility criteria and discouraging individuals from navigating complex application rules (Bhargava and Manoli 2015, Armour 2018, Finkelstein and Notowidigdo 2019). Previous studies have also identified lack of attention (Madrian and Shea 2001) and procrastination (Karlan et al. 2016) as significant barriers. Our study extends this research by examining the role of in- formation in an environment with extremely high vulnerability and government distrust, which is particularly relevant in developing countries where trust in state institutions is generally low.5 We also advance this field by assessing the efficacy of WhatsApp videos in increasing public program take-up rates among populations in jeopardy. This study also contributes to the growing body of work on the impact of informational interventions on economic decisions.6 Previous research on the effects of clear program information on public program take-up rates have yielded mixed results. While some studies have shown such interventions can increase take-up rates (Daponte, Sanders and Taylor 1999, Saez 2009, Jones 2010, Bhargava and Manoli 2015, Finkelstein and Notowidigdo 2019, Michael Hotard and Hainmueller 2019, Domurat, Menashe and Yin 2021), others have demonstrated that one-time informational interventions are insufficient (Bettinger et al. 2012, Manoli and Turner 2014, Guyton et al. 2016). Moreover, some work has found that the effect of information can be negligible or even lead to lower take-up rates, de- pending on the population (Mastrobuoni 2011, Bettinger et al. 2012, Seira, Elizondo and ¨ Laguna-Muggenburg 2017, Allcott and Greenstone 2017, Armour 2018, Hainmueller et al. 2018). Recent research suggests that online and mobile technologies may reduce informa- tion asymmetries for individuals with high technological literacy (Arteaga et al. 2022) but may have negligible effects for the general population (Bahety et al. 2021). We offer new 5 For instance, according to data from The Americas Barometer by the LAPOP Lab, Latin American countries exhibit low approval ratings for local government. 6 For a comprehensive review of the literature, see Currie (2006). 8 evidence regarding the effect of iterative information sent through WhatsApp on take-up rates of vulnerable, hard-to-reach populations. Our findings imply such methods may be ineffective and might backfire, inducing frustration. We also contribute by evaluating the effectiveness of WhatsApp videos and online sur- veys in reaching isolated or otherwise at-risk populations. Relatively new work concludes that online technologies could be a low-cost way to collect information in areas that are inaccessible due to conflict and disease as long as recipients have a certain level of tech- nological literacy (Beam 2023, Heywood, Ivey and Meuter 2022). We establish that online surveys through social media might not succeed in eliciting data from stigmatized com- munities that also have trust issues, such as undocumented forced migrants. II REGULARIZATION PROGRAMS IN COLOMBIA II.A The PEP Program The Venezuelan exodus is one of the most pressing forced migration crises today, with more than 7.1 million migrants displaced abroad as of 2023. Colombia is the principal re- cipient of this outpouring, and it has maintained a compassionate stance by offering them full mobility and several regularization programs to formalize their status there. One of the most extensive such initiatives took place in 2018, when the Colombian government gave nearly half a million irregular Venezuelan migrants the chance to regularize their documents, obtain a work permit, access safety nets (including comprehensive education and health services) and the financial sector, and validate their educational credentials. The program, known as the Permiso Especial de Permanencia (PEP), granted these benefits to migrants for a period of two years and opened the way for them to become perma- nent Colombian residents. PEP had significant impacts on migrants’ welfare including positive effects on labor income, access to public programs, bank account ownership, and health outcomes (Ib´ ˜ et al. 2022 and Urbina et al. 2023).7 Despite the program’s gen- anez 7 Previous research on PEP’s effects on hosting communities has explored its impacts on labor market outcomes and found negligible effects (Bahar, Ib´ ˜ and Rozo 2021); local crime rates, which showed anez 9 erosity, only around 60 percent of the migrants who were offered it actually registered and received benefits. Ib´ ˜ et al. (2020) found that reasons for failure to register included anez lack of awareness (both about how to register and the eligibility requirements), lack of trust in the Colombian government, and registration bottlenecks. II.B The ETPV Program Before the PEP program expired in 2020, the Colombian government opted to scale up the regularization of Venezuelan migrants through a program referred to as the Estatuto on para Migrantes Venezolanos (ETPV), which extended the eligibility Temporal de Protecci´ period for benefits. Specifically, the ETPV offered a 10-year regularization program to Venezuelan migrants who arrived in Colombia prior to January 2021 (see Figure A.1 for a timeline). The process entailed several steps, beginning with registration in an online cen- ´ sus known as the Registro Unico de Migrantes Venezolanos. Supporting documents—such as medical certificates, grade reports, labor certificates, and property rental agreements, among others—had to be uploaded, indicating proof of arrival in Colombia before Jan- edula, passport, or uary 31, 2021. Additionally, a Venezuelan ID document (such as a c´ birth certificate) was required to substantiate Venezuelan origin, along with a photo ID. Applicants next had to schedule an in-person appointment to record their biometric data. Upon completion of this appointment, a permit or visa was granted virtually, and three on Temporal (tempo- months later, the physical document known as the Permiso por Protecci´ rary protection permit) or PPT was issued. The window for registration and the biometric data appointment occurred between May 2021 and June 2022. The complete process is de- scribed in Figure A.2. The PPT is both an identification and regularization document, providing a broad range of benefits including regular legal status; a work permit; access to public health services an increase in migrants’ reports of sexual abuse and domestic violence (Ib´ ˜ anez, Rozo and Bahar 2020); firm development, which resulted in the creation of new mom-and-pop businesses (Bahar, Cowgill and Guzman 2023); and political outcomes, which experienced no observable changes in host voting behavior (Rozo, Quintana and Urbina 2023). 10 (including Covid-19 vaccinations), the pension system, education, childcare, and the fi- nancial sector; and the potential to validate Venezuelan educational certifications. More- over, the PPT allows migrants to enter and exit the country without restrictions and serves as proof of settlement in Colombia to fulfill the time requirement for obtaining a residency on Colombia, as of October visa. According to the Colombian migration agency, Migraci´ 2022, nearly 2.5 million Venezuelan migrants had completed the online RUMV census (see Figure A.3 for their geographical distribution). III THE INTERVENTION: HOW TO INCREASE TAKE-UP RATES FOR THE ETPV III.A Rationale Previous work shows results from our qualitative investigation into why Venezuelan mi- grants did not register for the PEP program.8 We identified three key barriers that im- peded participation, namely: (i) inadequate awareness of the program; (ii) lack of trust in the government due to possible legal complications during the regularization pro- cess; and (iii) confusion and insufficient information regarding registration procedures as well as reluctance related to the complex and difficult process. Our WhatsApp interven- tion targeted these barriers and tried to increase participation in the next regularization program (the PPT) by providing information through cost-effective platforms that could make this intervention scalable. 8 The study included 42 in-depth interviews with Venezuelan migrants living in cities with a high pop- ulation of Venezuelans. All interviewees had resided in Colombia since 2018, and were: (i) beneficiaries of the Special Permit of Permanence granted by the Administrative Registry of Venezuelan Migrants (PEP- RAMV) plus (ii) migrants who could have benefited from this permit but had not registered in RAMV or (iii) were unable to access PEP despite having participated in RAMV. Participants comprised generally young people (mostly women) with difficulties generating income despite being of working age, who lived in large households (25 women and 17 men took part in the study). The main finding was that the desire to obtain official documentation was a primary motivation for Venezuelan migration to Colombia. Nev- ertheless, the process was hampered chiefly due to lack of awareness of the program. Migrant networks were essential to publicize both registration in RAMV and the subsequent PEP-RAMV process. Interviews revealed that without physical or virtual contact through social networks with church and migrant group leaders, migrants would have missed the opportunity to regularize their status. There was also fear sparked by rumors that RAMV would be used by the Colombian government to deport undocumented migrants. Finally, bottlenecks in the registration process occurred as a result of: (i) lack of access to information and weak social networks; (ii) lack of money for transportation to registration points; (iii) the cost of losing part of the working day to an activity that did not generate income; and (iv) lack of incentives to register. 11 III.B Design We randomly assigned a total of 1,375 eligible individuals into four groups of equal size that consisted of three treatment arms and one control group.9 The research team inter- viewed individuals selected for the intervention in person at registration to collect basic sociodemographic characteristics. The three treatment arms involved the dissemination of a different video through WhatsApp, each addressing a specific barrier to registration for the PPT such as lack of program awareness, distrust of the government, and details regarding registration that were intended to reduce procedural difficulties. As mentioned above, we based the intervention design on the EAST methodology de- veloped by the Behavioral Insights Team (BIT 2014), which emphasizes the principles of making information easy to understand, attractive, social, and timely. We designed the videos to simplify information about benefits, eligibility criteria, and the registration pro- cess. The use of graphic design, pop-ups, and images made the videos more attractive, and a financial incentive encouraged viewership.10 To incorporate the “social” principle, the videos informed migrants that others in their community had successfully registered for the regularization program.11 We applied the “timely” rule by strategically timing the messages to reach individuals when they were most receptive, based on insights from previous research with this population. Video 1 featured a Colombian actor who portrayed an official and provided clear and concise information on program eligibility, costs, and the registration process (awareness video). Video 2, narrated by a Venezuelan forced migrant and mother of two children who had registered in the program, provided the same information as Video 1 but added anecdotal evidence about her experience with the program in order to build trust and 9 Our initial plan was to recruit 4,180 eligible Venezuelan migrants in Colombia. However, in the field, we could only identify and include 1,375 individuals who were undocumented and wanted to participate in the experiment. We revised our pre-analysis plan to reflect the new sample size and estimation strategy. 10 The value of the incentive was 10,000 Colombian pesos (COP). 11 Social norms insights have proven successful in changing individual behaviors (Allcott 2011, Hassett, Grolleau and Ibanez 2017, Donna, Roberts and Sweeney 2007). 12 empathy (trust video). Finally, Video 3, which had the same Venezuelan narrator as Video 2, offered more detail on the registration process with a step-by-step guide on how to register online (step-by-step video). All the video scripts are provided in the Appendix. Figure 1. The Intervention Panel A. Intervention Design Panel B. Geographic Location Notes: The map on the left depicts the departments where the experiment took place, and the map on the right displays the cities and the sample size of the intervention. 13 III.C Recruitment and eligibility In partnership with Innovations for Poverty (IPA) Colombia, we recruited participants antico as they are home to many highly vulnerable migrants, ac- in Magdalena and Atl´ cording to the Colombian population census of 2018. The sample was collected to be representative of both Santa Marta, one of the largest urban centers, and rural areas that included Ci´ ´ (see Panel B of Figure 1).12 ´ and Baranoa enaga, Sabana Larga, Fundacion Based on guidance from migrant organizations, public officials, and members of the com- munity, we advertised the program in areas Venezuelan migrants were known to fre- quent. We mapped these areas and subsequently contacted local community leaders there to elicit support for opening registration points in marginalized communities with many undocumented migrants. Local leaders helped us build trust by offering information about the IPA and the researchers involved in the project. We made several modifications to our data collection process to increase trust and boost response rates. Three important ones were (i) the distribution of related research on regularization programs to local lead- ers, (ii) discussion to learn how to support Venezuelan migrants more effectively, and (iii) collaborations with Venezuelan enumerators to increase trust. The study’s eligibility criteria were carefully defined to ensure that individuals were both eligible for the program and vulnerable. Specifically, we recruited Venezuelan migrants who were of legal age (18 years or older), resident in any of the selected municipalities, undocumented, and who arrived in Colombia before January 1, 2021. Additionally, par- ticipants had to have access to a phone with WhatsApp and internet. The final sample numbered 1,375 individuals who met these criteria and agreed to participate in the study. Table 1 presents descriptive statistics and provides important insights into the partici- pants. The sample was predominantly composed of young people with an average age of 33.4 years. The majority of participants were female, accounting for approximately 67 12 We did not recruit individuals in Barranquilla due to implementation costs. 14 percent of the sample. Furthermore, individuals in our sample also had low income, with an average monthly income of 250,000 COP, which represents 20 percent of the minimum wage in Colombia. The majority reported having access to the internet for at least four hours a day, which initially suggested that mobile-based interventions could be viable for this population. Notably, the descriptive statistics also seem to suggest that trust in the Colombian government was not particularly low. Yet, self-reported measures are proba- bly biased and we did not have comparable values for Colombian natives that permitted us to make any strong conclusions. Table 1. Sample Characterization Average SD Min Max N Age 33.44 11.29 18.00 75.00 1,375 Male [=1] 0.33 0.47 0.00 1.00 1,375 Ed. Level: Primary or Less [=1] 0.19 0.39 0.00 1.00 1,375 Ed. Level: General or diversified school [=1] 0.59 0.49 0.00 1.00 1,375 Number of household members 4.68 1.74 0.00 14.00 1,375 Personal Income (COP) 249,472 217,353 0.00 1,700,000 1,375 Activity spent the most time: Working [=1] 0.42 0.49 0.00 1.00 1,375 Internet Access more than 4 hours per day [=1] 0.67 0.47 0.00 1.00 1,375 Trust in Colombian Government (1-5 scale) 4.27 1.02 1.00 5.00 1,375 Personal use of WhatsApp [=1] 0.77 0.42 0.00 1.00 1,375 Facebook or Instagram account [=1] 0.53 0.50 0.00 1.00 1,375 Twitter account [=1] 0.12 0.32 0.00 1.00 1,375 E-mail account [=1] 0.17 0.38 0.00 1.00 1,375 Social desirability index (1-4 scale) 2.28 1.40 0.00 4.00 1,375 Notes: Definition dependent variables: (i) Trust in Colombian Government is the answer to the question ”Do you trust the Colombian Government?” on a five-point scale from 1-strongly disagree to 5-strongly agree; (ii) social desirability index is constructed using four questions from the Marlowe-Crowne social desirability scale (see Crowne and Marlowe 1960 for details). The four questions are: “It is sometimes hard for me to go on with my work if I am not encouraged (false corresponds to higher social desirability)”; “There have been times when I was quite jealous of the good fortune of others (false corresponds to higher social desirability)”; “I am always willing to admit when I make a mistake (true corresponds to higher social desirability)”; and “I am always courteous, even to people who are disagreeable (true is associated with higher social desirability).” Each statement gets a score of zero or one (assigned to higher social desirability answers). The total level of social desirability bias is calculated by summing the scores of all questions. Higher values indicate more social desirability bias. We compared our sample of Venezuelan migrants with those surveyed in similar recent surveys including the 2021 Labor Force Survey (Gran Encuesta Integrada de Hogares, GEIH) and the 2020 Venezuelan Refugee Panel Survey (VenRePS) of migrants without a migratory permit. The GEIH is a comprehensive survey that regularly samples house- holds in Colombia to collect data on the labor force and individual demographic char- acteristics, including those of Venezuelan migrants. The VenRePS, on the other hand, 15 is a representative sample of documented and undocumented migrants who arrived in Colombia between January 2017 and December 2018. Our analysis in Table 2 indicates that, as intended, migrants in our intervention were the most vulnerable of all according to measures of income, education, unemployment, and access to health services. This aligns with our reasons for choosing Magdalena and antico as locations where we could find migrants who might lack information on the Atl´ regularization program or might face other challenges in registering for it. Table 2. Sample Comparability Mean Difference P-value Intervention GEIH VenRePS ETPV vs. GEIH ETPV vs. VenRePS (1) (2) (3) (4) (5) Age 33.444 33.25 32.836 0.640 0.105 (11.286) (11.810) (10.882) Male [=1] 0.325 0.441 0.4 0.000 0.000 (0.469) (0.497) (0.490) Years of education 4.518 7.814 13.043 0.000 0.000 (2.051) (4.028) (2.862) Total income (Log) 0.209 0.542 0.354 0.000 0.000 (0.161) (0.268) (0.213) Health regime: Subsidized [=1] 0.149 0.195 0.016 0.001 0.000 (0.356) (0.397) (0.127) Health regime: Contributory [=1] 0.107 0.097 0.009 0.337 0.000 (0.309) (0.295) (0.095) Health regime: None [=1] 0.847 0.708 0.974 0.000 0.000 (0.360) (0.455) (0.158) Unemployed [=1] 0.226 0.083 0.304 0.000 0.000 (0.419) (0.276) (0.460) Observations 1,375 1,792 2,317 3,167 3,692 Notes: The first column presents the mean and standard deviation for the sample of this study. Column (2) shows the mean and standard deviation for the Venezuelans who responded to the Colombian Labor Force Survey of December of 2021, known as Gran Encuesta Integrada de Hogares (GEIH). Column (3) depicts the mean and standard deviation for the undocumented Venezuelans surveyed in the Venezuelan Refugee Panel Survey (VenRePS) of 2020. Columns (4) and (5) present the p-value for the mean difference between samples. IV DATA: ITERATIVE WHATSAPP SURVEYS (IWS) After the in-person registration (which included the short sociodemographic survey) and randomization, we used WhatsApp contact each participant up to a maximum of four times in order to distribute videos and/or the survey (the control group only received the survey). At each point of contact, we sent the link to the video and one hour later the sur- 16 vey. The control group only received the survey. Specifically, we only recontacted those participants who had not responded to previous attempts or who had not yet requested the biometric data appointment. We considered the PPT successfully completed once a participant had either requested or attended the biometric data appointment. It should be noted, however, that attendance at the appointment was necessary to finish the official process. We did this because the videos specifically targeted the act itself of requesting the appointment. Moreover, due to a backlog in governmental processing times, appoint- ments were often scheduled after our last point of contact. In fact, many appointments were rescheduled due to governmental delays. We conducted the surveys according to the timeline in Figure 2. Figure 2. Iterative WhatsApp Survey Collection 1 week 1 week 1 week Contact 1 Contact 2 Contact 3 Contact 4 After the Video Question 1: → Yes (Ends), No (Continue) Question 2: → Yes (Ends), No (Continue) Question 3: → Yes (Ends), No (Continue) Question 4: → Yes (Ends), No (Continue) Question 5:→ Yes (Ends), No (Ends) 1. 2. 3. Between Contacts 1 and 2 Between Contacts 2 and 3 Between Contacts 3 and 4 The entire sample is contacted People who answered yes to People who answered yes to questions 1, 2, and 3 were no questions 1, 2, and 3 were no longer contacted longer contacted The WhatsApp surveys administered in this study included a maximum of five questions 17 (but possibly fewer) that focused on different stages of the PPT registration process in reverse order. They were: (i) receiving the PPT, (ii) attending the biometric data appoint- ment, (iii) requesting the biometric data appointment, (iv) starting the RUMV registration, and (v) intending to register. If a respondent indicated they had already received the PPT, they were not contacted again and subsequent questions were not posed as completion of all previous stages could be inferred. Similarly, if respondents replied in the affirmative to question (ii), they were not contacted again and the remaining questions were not asked, as successful completion of all prior steps could be assumed. This sequential approach was followed to query stages (i) through (v) only if the respondent answered negatively to all preceding questions. The detailed procedure is illustrated in Figure 3. Figure 3. Iterative WhatsApp Survey Structure V INTERNAL VALIDITY V.A Successful randomization We examine the internal validity of our experiment in Table 3, which reports the balance test results for the baseline covariates across treatment and control groups. Our findings indicate that of the 14 covariates observed and the 56 means difference tests evaluated, only two differences were statistically significant. These results support the internal va- lidity of our experimental design and lend confidence to our estimate of the causal effect 18 of the intervention on the outcomes of interest. 19 Table 3. Successful Covariate Balance by Treatment Type Information Trust Step-by-Step Any Control P-value Video Video Video Video (1) (2) (3) (4) (5) (1)-(2) (1)-(3) (1)-(4) (1)-(5) Age 33.130 32.607 32.797 35.251 33.551 0.533 0.691 0.017 0.548 (11.151) (10.847) (10.840) (12.113) (11.335) Male [=1] 0.305 0.334 0.313 0.348 0.332 0.418 0.830 0.235 0.367 (0.461) (0.472) (0.464) (0.477) (0.471) Ed. Level: Primary or Less [=1] 0.210 0.199 0.165 0.187 0.184 0.722 0.129 0.445 0.277 (0.408) (0.400) (0.372) (0.391) (0.388) Ed. Level: General or diversified school [=1] 0.625 0.560 0.620 0.535 0.572 0.082 0.891 0.016 0.081 (0.485) (0.497) (0.486) (0.499) (0.495) Number of household members 4.593 4.613 4.751 4.770 4.711 0.880 0.225 0.193 0.273 (1.733) (1.708) (1.677) (1.822) (1.737) Personal Income (Sin*) 9.886 9.583 9.413 9.856 9.616 0.429 0.217 0.935 0.391 (4.841) (5.232) (5.234) (4.958) (5.142) Activity spent the most time: Working [=1] 0.452 0.460 0.461 0.512 0.478 0.834 0.824 0.120 0.417 (0.498) (0.499) (0.499) (0.501) (0.500) Trust in Colombian Government (SD) 0.067 0.003 -0.059 -0.012 -0.023 0.380 0.082 0.281 0.139 20 (0.912) (0.989) (0.994) (1.008) (0.997) Internet Access more than 4 hours per day [=1] 0.277 0.217 0.261 0.225 0.234 0.070 0.640 0.119 0.114 (0.448) (0.413) (0.440) (0.418) (0.424) Personal use of Whatsapp [=1] 0.795 0.754 0.774 0.740 0.756 0.191 0.493 0.084 0.133 (0.404) (0.432) (0.419) (0.439) (0.430) Facebook or Instagram account [=1] 0.516 0.537 0.557 0.523 0.539 0.585 0.284 0.843 0.457 (0.500) (0.499) (0.498) (0.500) (0.499) Twitter account [=1] 0.110 0.097 0.101 0.146 0.115 0.584 0.730 0.150 0.789 (0.313) (0.296) (0.302) (0.354) (0.319) E-mail account [=1] 0.167 0.167 0.159 0.213 0.180 1.00 0.784 0.122 0.589 (0.374) (0.374) (0.367) (0.410) (0.384) Social desirability index (SD) 0.009 0.017 0.045 -0.072 -0.003 0.917 0.631 0.302 0.848 (1.016) (0.979) (0.974) (1.030) (0.995) Observations 347 341 345 342 1,028 688 692 689 1,375 Notes: Columns (1)–(5) present the mean and standard deviation for the control, the three treatments, and any of the treatment samples. Columns (6)–(9) depict the p-value of the t-test. Definition-dependent variables: (i) Personal income was transformed using the inverse hyperbolic sine transformation (see Burbidge, Magee and Robb 1988 and MacKinnon and Magee 1990 for details). (ii) Trust in Colombian Government is the standardized answer to the question “Do you trust the Colombian Government?” on a five-point scale from 1-strongly disagree to 5-strongly agree. (iii) Social Desirability Index is constructed using four questions from the Marlowe-Crowne social desirability scale (see Crowne and Marlowe 1960 for details). V.B Attrition analysis Table 4 presents a regression of attrition rates according to sociodemographic character- istics for the full sample of individuals. In the table, attrition is defined as an indicator variable equal to one if the individual did not respond to any of the four contact attempts. The results of the exercise are reassuring because for the 14 variables that were collected at baseline, only one is statistically significant, which corresponds to having a personal WhatsApp account (compared with a shared account). Remarkably, variables such as age, gender, income, and time spent working are not correlated with attrition rates. This exercise suggests that attrition rates are not systematic and have a more random nature. To further evaluate whether attrition had a different prevalence according to treatment status, we also evaluated the correlation of attrition according to the observed covariates, their interactions with treatment status, and treatment status by itself (see Table C.2). The table illustrates this exercise for three different definitions of attrition. The results are re- assuring since of the 42 interactions tested, only two were significant and the treatment status by itself does not correlate with any definitions of attrition that we tested. Table C.1 also shows the prevalence of attrition according to multiple definitions. For the whole sample and the most straightforward definition of attrition, we observe that 22.5 percent of the sample never responded to any of the four WhatsApp contacts, and 27.6 percent responded to one of follow-ups 1–3 but had not completed the PPT registration process. The table also shows that attrition rates between the control group and individuals treated by any video are similar.13 13 For completeness, we also explored how our results would change if attrition were non-random (al- though our analysis shows this was not the case) and if we dropped individuals who were not interested in the program and never registered. This exercise amounts to using worst-case scenario imputations (as defined in Horowitz and Manski 2000, Stantcheva 2022), which here means replacing missing values for at- trited individuals with zeroes for the treatment variables. Our results are available upon request and point to similarly sized effects and direction in the outcomes of interest. 21 Table 4. Characterizing Attrition Attrition [=1 if individual never responded to IWS] Attrited Individuals (Never responded) Age -0.000 (0.001) Male [=1] 0.051* (0.026) Ed. Level: Primary or Less [=1] 0.047 (0.037) Ed. Level: General or diversified school [=1] -0.043 (0.031) Number of household members -0.003 (0.007) Personal Income (Sin*) 0.001 (0.003) Activity spent the most time: Working [=1] -0.016 (0.029) Trust in Colombian Government (SD) 0.009 (0.012) Internet Access more than 4 hour per day [=1] -0.006 (0.025) Personal use Whatsapp [=1] -0.060** (0.027) Facebook or Instagram account [=1] 0.020 (0.027) Twitter account [=1] -0.052 (0.044) E-mail account [=1] -0.050 (0.033) Social desirability index (SD) 0.014 (0.016) R-squared 0.024 Mean Dependent Variable 0.225 Observations 1,375 Notes: Attrited Individuals is an indicator [=1] for the people who did not answer the survey and could not be contacted through WhatsApp. *** significant at the 1%, ** significant at the 5%, and * significant at the 10%. To complement this analysis and learn from using IWS with hard-to-reach, at-risk pop- ulations, we also examine attrition rates by contact attempt in Figure 4. In the figure, an individual is considered attrited if they did not respond to the survey at each spe- 22 cific contact attempt. The figure highlights that nearly half of the sample was lost at the first contact attempt. Moreover, the sample size shrank as more contacts were attempted (Panel A). This observation is also confirmed when we consider the subsample of 789 individuals who were contacted four times (Panel B). 23 Figure 4. Attrition Rate by Type of Treatment Panel A: Full Sample 70.00 1400 1,375 Non-response Individuals (% Contacted Individuals) 1200 60.58 60.00 1,030 1000 925 Number of Individuals 53.73 789 800 50.11 50.49 50.00 689 600 520 497 478 400 40.00 200 30.00 0 First Second Third Fourth Contact Non-response individuals Contacted individuals Attrition rate Panel B: Contacted Four Times 100.00 1000 Non-response Individuals (% Contacted Individuals) 90.00 789 789 789 789 800 637 Number of Individuals 608 80.74 80.00 580 600 77.06 484 73.51 70.00 400 61.34 60.00 200 50.00 0 First Second Third Fourth Contact Non-response individuals Contacted individuals Attrition rate 24 In addition, nearly 40 percent of participants who were contacted four times never re- sponded to any of the IWS. We present the distribution of possible outcomes for indi- viduals contacted four times in Panel A of Table 5. Panel A delineates the 16 cases that could have occurred. In this table, a value of “0” indicates the participant was contacted but did not respond to the survey, while a value of “1” indicates that they were contacted and responded to the survey. Just under 20 percent of individuals responded to all the surveys. The table illustrates that switching behavior from non-response to response was observed in at least 20 percent of the sample of individuals contacted four times. What is more, the table reveals there were participants in all possible response scenarios, im- plying that repeated contact attempts may prove worthwhile even if someone has not previously responded. To corroborate this finding, we combined the possible number of responses for each individual (for the sample contacted four times), as detailed in Panel B of Table 5. Our analysis indicates that 39.29 percent of the sample never responded to any of the WhatsApp surveys, whereas 14.82 percent responded only once, 13.17 percent responded twice (comprising 7.22 percent consecutive and 5.95 percent non-consecutive responses), 13.43 percent responded three times (8.87 percent consecutive and 4.56 per- cent non-consecutive responses), and 19.26 percent responded to all contact attempts. 25 Table 5. Distribution of Possible Contact Combinations Individuals Contacted Four Times Panel A: Possible Contact Cases Possible Contact Number of (% of Total) Cases Individuals 0000 310 39.29 0001 22 2.79 0010 15 1.90 0011 22 2.79 0100 30 3.80 0101 24 3.04 0110 20 2.53 0111 41 5.20 1000 50 6.34 1001 15 1.90 1010 8 1.01 1011 23 2.92 1100 15 1.90 1101 13 1.65 1110 29 3.68 1111 152 19.26 Total 789 100 Panel B: Successfully Reached Combinations Successfully Reached Number of (%) Possible Combinations Individuals Never 310 39.29 One Time 117 14.82 Two Consecutive Times 57 7.22 Two Non-Consecutive Times 47 5.95 Three Consecutive Times 70 8.87 Three Non-Consecutive Times 36 4.56 Always 152 19.26 Total 789 Notes: In the table on the left “0” corresponds to the individuals who were contacted but did not answer the survey and “1” to the individuals who were contacted and completed the survey. V.C Success of video plays Figure 5 presents the proportion of individuals who opened and played the video by treatment type for all participants recruited and contacted at each stage. Our analysis reveals three notable trends. First, the contact point with the highest success rate was the initial one, when over 73.6 percent of participants opened and played more than half the video. Second, as individuals were contacted more times, their engagement with the 26 video decreased, which could be attributed to fatigue or prior knowledge of the content. Third, play rates were lower for Videos 2 and 3 with a Venezuelan migrant as narrator compared with Video 1, which featured an actor resembling a Colombian official. This implies that a narrator with personal experiences similar to those of migrants did not generate additional interest in the video content. Figure 5. Video Play Rates by Treatment Arm 90 84.1 80.5 81.2 80 76.7 75.7 75.2 75.0 73.6 73.5 73.5 72.0 72.1 67.8 67.8 70 63.7 60.6 61.7 Percentage (%) 57.2 60 50 40 30 23.5 21.3 20 14.5 12.4 12.9 9.0 10 0 First Second Third Fourth First Second Third Fourth First Second Third Fourth Awareness Video Trust Video Step-by-Step Video Played the Video Play Time (%) Notes: The percentage is calculated over the treated sample contacted in each of the treatment arms. The treated sample for the first contact corresponded to 750 individuals, for the second contact to 257 individuals, for the third contact to 176 individuals, and for the fourth contact to 105 individuals. 27 VI EMPIRICAL STRATEGY We estimate the effects of the intervention on three outcomes: intention to register for the PPT, started registration (indicated by 1 if the individual started the RUMV registration), and actual registration for the PPT program (indicated by 1 if the individual requested or attended the biometric data appointment, or received the PPT). The primary outcome information corresponds to the last WhatsApp contact with the participant. To estimate program effects, we recover the average effects of the treatment on everyone who received the videos on the outcomes, generally known as ITT estimates. This is in- clusive of individuals who received the videos but did not watch them. For this purpose, we use a standard ordinary least squares (OLS) specification for all individuals in our study, given by the following equation: Yi = α + β1 Any Videoi + εi (1) where i stands for individual, Y for the outcomes, and Any Video represent the treatment assignment. We also evaluate the effectiveness of each the three videos individually (as written in the pre-analysis plan). VII RESULTS VII.A Pooled estimates for recipients of “Any video” Table 6 presents the empirical results of the pooled regression for all treatment arms, utilizing the last recorded response of each participant in our study.14 Panel A reports results of the ITT estimates for everyone who received the video, and Panel B presents the estimates differentiated the treatment by type of video. We find negative effects of sending the video on all three outcomes that we examine. 14 Of the 1,375 individuals registered in the experiment, we excluded 245 individuals from the sample who did not respond to any WhatsApp survey. 28 Specifically, Panel A shows that receiving a video reduced the intention to register by 12.2 pp (Column 3), lessened the probability of starting the registration process by 7.7 pp (Column 2), and decreased the likelihood of requesting the PPT by 8 pp (Column 1). Interestingly, participants in the control group who registered for the regularization program after signing up for the experiment (since no one had registered beforehand) had a mean registration rate of 53.8 percent. Therefore, the treatment resulted in a reduction of 15.09 percent in PPT take-up rates relative to the control group’s mean. Panel B, reports ITT coefficients that are negative for all treatments, albeit with less preci- sion relative to the pooled estimates. However, the effects are largest and always statisti- cally significant for the intention to register and for the Step-by-Step video. This suggest that the longest video which offered more details was actually the one that reduced take- up the most. Table 6. Intervention Effects on PPT Take-up Rates (Responses from Last Contact) Request Start Registration Intention to Indicator Variables PPT Process Register (1) (2) (3) Panel A. ITT - General Effect β1 [Any Video] -0.080** -0.077** -0.122*** (0.034) (0.034) (0.030) FDR q-values [0.015] [0.015] [0.001] R-squared 0.005 0.005 0.015 Panel B. ITT - Dissaggregated Effect β1 [Awareness] -0.060 -0.057 -0.103*** (0.041) (0.041) (0.036) β2 [Trust] -0.065 -0.069* -0.098*** (0.041) (0.041) (0.036) β3 [Step-by-Step] -0.117*** -0.108** -0.168*** (0.042) (0.042) (0.037) R-squared 0.007 0.006 0.019 Mean Untreated Group 0.538 0.585 0.826 Observations (All Panels) 1,130 1,130 1,130 Notes: Dependent variables: (i) Request PPT is an indicator [=1] if the individual reported having requested the PPT, or requested or attended the biometric appointment in the last survey contact. (ii) Start Registration Process is an indicator [=1] if the individual reported starting the RUMV census in the last survey contact. (iii) Intention to Register is an indicator [=1] if the individual reported the intention to start the RUMV census in the last survey contact. The experiment had 1,375 individuals registered. This table excludes from the sample 245 individuals who did not answer any of the four WhastApp surveys. Standard errors are reported in parentheses and False Discovery Rate (FDR) q-values are reported in brackets. ∗∗∗ significant at the 1%, ∗∗ significant at the 5%, ∗ significant at the 10%. 29 Robustness exercises. The results are robust to multiple hypothesis testing as indicated by the False Discovery Rate adjusted p-values reported in brackets. To ensure the robustness of our estimates and account for potential biases from the different number of contacts and response rates, we re-estimate the effects of the videos using responses only from the first WhatsApp survey. Table B.1 presents results of this analysis. Reassuringly, the effects have the same directions and are even larger in magnitude than those in Table 6. The estimates use the same sample size as in Table 6 since the first contact maximizes the number of observations. VII.B What explains the average negative effects of the intervention? One of the most intuitive explanations behind the results is that the intervention was only effective for the individuals who actually viewed the video and might have induced discomfort for the others (on average inducing negative effects). Hence, we proceeded to examine whether watching some part of the video had differential effects following the steps outlined below. 1. Characterizing variation in take-up rates in the treatment group. We characterized if there was variation on the time that individuals assigned to treatment watch the video. This variation is illustrated in Table 7. It shows that approximately 15.40 percent of the indi- viduals assigned to treatment did not watch the video. Hence, we created a dichotomous variable for treated individuals that takes the value of zero for those who did not watch the video at all and one for those who watched at least some part of the video. This variable for our purposes corresponds to the treatment take-up. 30 Table 7. Video Play Time Percentage Distribution Play Time (%) Observations % 0% 128 15.40 1-10% 14 1.68 10%-20% 8 0.96 20%-30% 11 1.32 30%-40% 7 0.84 40%-50% 14 1.68 50%-60% 8 0.96 60%-70% 15 1.81 70%-80% 11 1.32 80%-90% 18 2.17 90%-99% 327 39.35 100% 270 32.49 Total 831 100 2. Predicting take-up rates. Who watched the video? Next, we examine if treated individu- als who did not watch the videos are statistically different from those individuals who watched at least some part of the videos in Table 8. For this purpose, we use the variables collected at baseline and a few additional variables collected at the end of the intervention for individuals who were assigned to treatment.15 We were able to identify statistically significant differences in multiple covariates between groups for the variables of age, time use, internet access, and personal vs. shared use WhatsApp. Particularly, we observe that older individuals played the videos less than younger individuals. We also observe that individuals who are busier working, looking for a job, or studying tend to play the videos less than the rest. Although the estimates are only statistically significant and negative for the individuals who are studying. Moreover, individuals with limited internet tend to play the videos less relative to those with internet half or the full day. Finally, individuals who can use WhatsApp individually play the videos more relative to those individuals who have to share it. 15 The data was gathered at the end of the intervention only for the treatment group to attempt to under- stand the surprising results of the intervention. We also attempted to collect a survey one year after the intervention took place but response rates were low and close to 20 percent. 31 Table 8. Who Played the Video? Did not Played the Video Played the Video Difference (1) (2) (1)-(2) Age 35.922 33.043 2.879*** (11.672) (10.734) Male [=1] 0.344 0.320 0.024 (0.477) (0.467) Ed. Level: Primary or Less [=1] 0.172 0.174 -0.002 (0.379) (0.379) Ed. Level: General or diversified school [=1] 0.531 0.589 -0.058 (0.501) (0.492) Number of household members 4.608 4.713 -0.105 (1.818) (1.688) Number of minors in charge 2.071 1.969 0.102 (1.507) (1.341) Personal Income (Sin*) 9.501 9.597 -0.096 (5.085) (5.113) Health regime: Subsidized healthcare [=1] 0.180 0.161 0.019 (0.385) (0.368) Health regime: None [=1] 0.836 0.832 0.004 (0.372) (0.374) Activity spent the most time: Working [=1] 0.523 0.455 0.068 (0.501) (0.498) Activity spent the most time: Looking for a job [=1] 0.227 0.213 0.013 (0.420) (0.410) Activity spent the most time: Studying [=1] 0.062 0.037 0.026 (0.243) (0.189) Activity spent the most time: Doing house chores [=1] 0.289 0.349 -0.059 (0.455) (0.477) Trust in Colombian Government (SD) -0.172 -0.015 -0.158 (1.150) (0.997) Internet Access: 1 to 4 hours [=1] 0.250 0.213 0.037 (0.435) (0.410) Internet Access: All or half of the day [=1] 0.594 0.663 -0.069 (0.493) (0.473) Individual Whatsapp [=1] 0.695 0.771 -0.076* (0.462) (0.421) Shared WhatsApp use [=1 [=1] 0.227 0.182 0.044 (0.420) (0.386) Facebook or Instagram account [=1] 0.547 0.532 0.015 (0.500) (0.499) Twitter account [=1] 0.148 0.119 0.029 (0.357) (0.325) E-mail account [=1] 0.219 0.185 0.034 (0.415) (0.389) Social desirability index (SD) -0.045 -0.025 -0.020 (1.041) (1.013) Observations 128 703 831 Notes: The first column presents the mean and standard deviation for the sample assigned to the treatment group that did not played the video. Column (2) shows the mean and standard deviation for the sample assigned to the treatment group who played less than 90% of the video. Column (3) depicts the mean and standard deviation for the sample assigned to the treatment group who played more than 90% of the video. Columns (4) and (5) present the difference between samples.∗∗∗ significant at the 1%, ∗∗ significant at the 5%, ∗ significant at the 10%. 32 3. Matching on observables. We next examine the difference in treatment effectiveness between the control and treatment groups matching the individuals in both groups based on all the variables that predict the probability of watching the videos. We do this in two separate exercises by restricting the treatment group to those who watched and did not watch the video. The results are presented in Table 9. Panel A of the table presents the results for the exercise that includes the control group and those individuals who did not play the video. Panel B presents the results for the control group and the individuals who watched the video. All in all, we observe that the videos had negative effects on the three outcomes that we examine. Yet, the effects are close to zero and mostly not statistically significant for the individuals who watched the videos (with the exception of the outcome intent to register). Yet, the effects of the videos are large, negative, and statistically significant for those individuals who did not watch the videos. This suggests that the average negative effects of the interventions are mostly driven by the individuals who received the messages but who did not watch the videos. 33 Table 9. Intervention Effects on PPT Take-up Rates (Responses from Last Contact) Propensity Score Matching Request Start Registration Intention to Indicator Variables PPT Process Register (1) (2) (3) Panel A. Propensity Score Matching - People who did not played the video β1 [AnyV ideo] -0.257*** -0.265*** -0.342*** (0.051) (0.051) (0.044) FDR q-values [0.001] [0.001] [0.001] R-squared 0.056 0.059 0.123 Mean Untreated Group 0.538 0.585 0.826 Observations 427 427 427 Panel B. Propensity Score Matching - People who played the video β1 [AnyV ideo] -0.047 -0.042 -0.083** (0.035) (0.034) (0.029) FDR q-values [0.171] [0.171] [0.016] R-squared 0.002 0.002 0.008 Mean Untreated Group 0.538 0.585 0.826 Observations 997 997 997 Notes: Dependent variables: (i) Request PPT is an indicator [=1] if the individual reported having requested the PPT, or requested or attended the biometric appointment in the last survey contact. (ii) Start Registration Process is an indicator [=1] if the individual reported starting the RUMV census in the last survey contact. (iii) Intention to Register is an indicator [=1] if the individual reported the intention to start the RUMV census in the last survey contact. The experiment had 1,375 individuals registered. This table excludes from the sample 245 individuals who did not answer any of the four WhastApp surveys. Standard errors are reported in parentheses and False Discovery Rate (FDR) q-values are reported in brackets. ∗∗∗ significant at the 1%, ∗∗ significant at the 5%, ∗ significant at the 10%. 4. Exploring for heterogeneous effects. Our results suggest that the videos had a negative effect on the PPT take-up rates (measured as intentions to register, starting the registration process or completing the process) and that those effects are driven by the individuals who were assigned to treatment but did not watch the videos. To further confirm these conclusions we carry out two additional exercises: i) we spit the treated sample between those individuals who watched and did not watch the videos and compare their outcomes with the control group (see Table 10), and ii) examine for heterogeneous effects for those individuals who played the videos through an interaction (see Table 11). Both exercises confirm that the observed negative effects of the program are coming from those treated individuals who did not watch the videos. 34 Table 10. Heterogeneous Effects on PPT Take-up Rates by Video Reproduction Indicator Variables Request PPT Start Registration Process Intention to Register Didn’t Played Played Didn’t Played Played Didn’t Played Played Video Video Video Video Video Video (1) (2) (3) (4) (5) (6) Panel A. General Effect - ITT β1 [AnyV ideo] -0.257*** -0.048 -0.265*** -0.043 -0.342*** -0.082*** (0.051) (0.035) (0.051) (0.034) (0.044) (0.029) R-squared 0.056 0.002 0.059 0.002 0.123 0.008 Panel B. Dissaggregated Effect - ITT β1 [Awareness] -0.228** -0.040 -0.240** -0.036 -0.240*** -0.087** (0.095) (0.043) (0.095) (0.042) (0.082) (0.036) β2 [T rust] -0.295*** -0.026 -0.293*** -0.032 -0.411*** -0.045 (0.081) (0.043) (0.081) (0.043) (0.070) (0.036) β3 [Step − by − Step] -0.245*** -0.082* -0.258*** -0.066 -0.343*** -0.119*** (0.070) (0.045) (0.070) (0.045) (0.060) (0.038) R-squared 0.057 0.003 0.059 0.002 0.128 0.011 Mean Untreated Group 0.538 0.538 0.585 0.585 0.826 0.826 Observations (All Panels) 427 1,002 427 1,002 427 1,002 Notes: Dependent variables: (i) Request PPT is an indicator [=1] if the individual reported having requested the PPT, or requested or attended the biometric appointment in the last survey contact. (ii) Start Registration Process is an indicator [=1] if the individual reported starting the RUMV census in the last survey contact. (iii) Intention to Register is an indicator [=1] if the individual reported the intention to start the RUMV census in the last survey contact. The experiment had 1,375 individuals registered. This table excludes from the sample 245 individuals who did not answer any of the four WhastApp surveys. Standard errors are reported in parentheses and False Discovery Rate (FDR) q-values are reported in brackets. ∗∗∗ significant at the 1%, ∗∗ significant at the 5%, ∗ significant at the 10%. Table 11. Heterogeneous Effects on PPT Take-up Rates by Video Reproduction Request Start Registration Intention to Indicator Variables PPT Process Register (1) (2) (3) Panel A. General Effect - ITT β1 [AnyV ideo]× I(Played the Video) 0.210*** 0.222*** 0.260*** (0.048) (0.047) (0.041) β2 [AnyV ideo] -0.257*** -0.265*** -0.342*** (0.052) (0.052) (0.045) Diff. Effect= β1 + β2 -0.048 -0.043 -0.082*** (0.034) (0.034) (0.030) R-squared 0.022 0.024 0.048 Mean Untreated Group 0.538 0.585 0.826 Observations (All Panels) 1,130 1,130 1,130 Notes: Dependent variables: (i) Request PPT is an indicator [=1] if the individual reported having requested the PPT, or requested or attended the biometric appointment in the last survey contact. (ii) Start Registration Process is an indicator [=1] if the individual reported starting the RUMV census in the last survey contact. (iii) Intention to Register is an indicator [=1] if the individual reported the intention to start the RUMV census in the last survey contact. The experiment had 1,375 individuals registered. This table excludes from the sample 245 individuals who did not answer any of the four WhastApp surveys. Standard errors are reported in parentheses and False Discovery Rate (FDR) q-values are reported in brackets. ∗∗∗ significant at the 1%, ∗∗ significant at the 5%, ∗ significant at the 10%. 35 VIII ADDITIONAL QUALITATIVE EVIDENCE EXPLAINING THE RESULTS In searching for insights that could inform our results, we called several participants and conducted qualitative semi-structured interviews to explore the experiences and opinions of treatment recipients. When asked about potential explanations for the negative effects of the program for individuals who did not watch the videos, respondents mentioned frustration with the multiple contacts, frustration due to technological literacy barriers (some of them did not have an email address and could not follow-up the instructions), and frustration due to internet barriers (in some cases, they could not open the video be- cause they did not have a mobile network at home or enough mobile data or Wi-Fi when the video and survey links arrived). All these results are in line with our quantitative characterization of the individuals who watched the videos. IX DISCUSSION This paper describes an experiment we conducted in Colombia to increase take-up rates of a regularization program for undocumented Venezuelan forced migrants in 2021. We recruited and screened in-person 1,375 individuals who had not yet registered for reg- ularization and randomly assigned them to three treatment arms and a control group. Each treatment offered information on registering for the regularization program but tar- geted a different issue. The first video addressed awareness; the second video, trust; and the third video aimed to increase trust and resolve administrative challenges by offering detailed information in a step-by-step description of the process. We successfully ran- domized individuals to the different groups. We document that sending informational videos had on average detrimental effects on take-up rates for everyone contacted, and that the longer videos which presented more details on the program registration had the largest negative effects on PPT take-up rates. We also find that these effects are mostly driven by the individuals who did not watch the videos who are older, busier, and have less access to internet and WhatsApp. These re- 36 sults are in line with our qualitative semi-structured interviews with participants, which yielded several explanations for the lack of effectiveness of the treatment for the indi- viduals who did not watch them. These included frustration due to the multiple contact attempts, technological literacy barriers, and limited access to reliable Wi-fi networks. We also used the experiment to examine the effectiveness of Iterative WhatsApp Surveys in collecting data for hard-to-reach populations of this type. We found that a large por- tion of the sample was lost when transitioning from in-person interviews to WhatsApp surveys. Moreover, the sample size shrank as more contacts were attempted. Here we observed the same challenges as above, with the addition of hesitation to share personal information via digital platforms. Our results suggest that information videos through digital platforms might not be an effective tool to increase take-up rates among vulnerable populations that are hard to reach, and that on the contrary, they may create frustration and even reduce take-up rates for the individuals who do not watch them. 37 References Allcott, Hunt. 2011. “Social norms and energy conservation.” Journal of Public Economics 95(9):1082–1095. Allcott, Hunt and Michael Greenstone. 2017. Measuring the Welfare Effects of Residen- tial Energy Efficiency Programs. Working Paper 23386 National Bureau of Economic Re- search. Armour, Philip. 2018. “The Role of Information in Disability Insurance Application: An Analysis of the Social Security Statement Phase-In.” American Economic Journal: Economic Policy 10(3):1–41. Arteaga, Felipe, Adam J Kapor, Christopher A Neilson and Seth D Zimmerman. 2022. “Smart Matching Platforms and Heterogeneous Beliefs in Centralized School Choice.” The Quarterly Journal of Economics 137(3):1791–1848. Bahar, Dany, Ana Mar´ ıa Ib´ ˜ and Sandra V. Rozo. 2021. “Give me your Tired and your anez Poor: Impact of a Large-scale Amnesty Program for Undocumented Refugees.” Journal of Development Economics 151:102652. Bahar, Dany, Bo Cowgill and Jorge Guzman. 2023. “Refugee Entrepreneurship.” American Economic Association Papers Proceedings. Forthcoming . Bahety, Girija, Sebastian Bauhoff, Dev Patel and James Potter. 2021. “Texts don’t nudge: An adaptive trial to prevent the spread of Covid-19 in India.” Journal of Development Economics 153:102747. Bartlett, Susan, Nancy Burstein and William Hamilton. 2004. “Food Stamp Access Study: Final Report.” United States Department of Agriculture, Economic Research Service pp. 419–423. Beam, Emily A. 2023. “Social media as a recruitment and data collection tool: Experi- 38 mental evidence on the relative effectiveness of web surveys and chatbots.” Journal of Development Economics 162:103069. Bertrand, Marianne, Eldar Shafir and Sendhil Mullainathan. 2006. “Behavioral Economics and Marketing in Aid of Decision Making Among the Poor.” Journal of Public Policy Marketing 25. Bettinger, Eric P., Bridget Terry Long, Philip Oreopoulos and Lisa Sanbonmatsu. 2012. “The Role of Application Assistance and Information in College Decisions: Re- sults from the HR Block FAFSA Experiment.” The Quarterly Journal of Economics 127(3):1205–1242. Bhargava, Saurabh and Dayanand Manoli. 2015. “Psychological Frictions and the Incom- plete Take-Up of Social Benefits: Evidence from an IRS Field Experiment.” American Economic Review 105(11):3489–3529. BIT. 2014. “EAST: Four Simple Ways to Apply Behavioural Insights.”. Burbidge, John B, Lonnie Magee and A Leslie Robb. 1988. “Alternative transformations to handle extreme values of the dependent variable.” Journal of the American Statistical Association 83(401):123–127. Chetty, Raj, John N. Friedman and Emmanuel Saez. 2013. “Using Differences in Knowledge across Neighborhoods to Uncover the Impacts of the EITC on Earnings.” American Economic Review 103(7):2683–2721. Costa, Dora L and Matthew E Kahn. 2013. “Energy conservation “nudges” and envi- ronmentalist ideology: Evidence from a randomized residential electricity field experi- ment.” Journal of the European Economic Association 11(3):680–702. Crowne, Douglas and David Marlowe. 1960. “A New Scale of Social Desirability Inde- pendent of Psychopathology.” Journal of Consulting Psychology 24:349–54. 39 Currie, Janet. 2006. The Take Up of Social Benefits. Technical report. Damgaard, Mette Trier and Christina Gravert. 2018. “The hidden costs of nudging: Exper- imental evidence from reminders in fundraising.” Journal of Public Economics 157:15– 26. Daponte, Beth Osborne, Seth Sanders and Lowell Taylor. 1999. “Why Do Low-Income Households not Use Food Stamps? Evidence from an Experiment.” Journal of Human Resources 34(3):612–628. DellaVigna, Stefano and Elizabeth Linos. 2022. “RCTs to Scale: Comprehensive Evidence From Two Nudge Units.” Econometrica 90(1):81–116. Domurat, Richard, Isaac Menashe and Wesley Yin. 2021. “The Role of Behavioral Fric- tions in Health Insurance Marketplace Enrollment and Risk: Evidence from a Field Experiment.” American Economic Review 111(5):1549–74. Donna, Bobek, Robin W. Roberts and John T. Sweeney. 2007. “The Social Norms of Tax Compliance: Evidence from Australia, Singapore, and the United States.” Journal of Business Ethics 74(1):49–64. Finkelstein, Amy and Matthew J Notowidigdo. 2019. “Take-Up and Targeting: Experi- mental Evidence from SNAP.” The Quarterly Journal of Economics 134(3):1505–1556. Guyton, John, Dayanand S Manoli, Brenda Schafer and Michael Sebastiani. 2016. Re- minders Recidivism: Evidence from Tax Filing EITC Participation among Low-Income Nonfilers. Working Paper 21904 National Bureau of Economic Research. Hainmueller, Jens, Duncan Lawrence, Justin Gest, Michael Hotard, Rey Koslowski and David Laitin. 2018. “A randomized controlled design reveals barriers to citizen- ship for low-income immigrants.” Proceedings of the National Academy of Sciences 115:201714254. 40 Hassett, Katherine, Gilles Grolleau and Lisette Ibanez. 2017. “Social Norms and Pro- environmental Behavior: A Review of the Evidence.” Ecological Economics 140:1–13. Heywood, Emma, Beatrice Ivey and Sacha Meuter. 2022. “Reaching hard-to-reach com- munities: Using WhatsApp to give conflict-affected audiences a voice.” International Journal of Social Research Methodology pp. 1–15. Horowitz, Joel L and Charles F Manski. 2000. “Nonparametric analysis of random- ized experiments with missing covariate and outcome data.” Journal of the American Statistical Association 95(449):77–84. Ib´ ˜ anez, ıa, Sandra V. Rozo, Andr´ Ana Mar´ ıa Adelaida Ortega and es Moya, Mar´ ıguez Chatruc. 2020. “Pre-Analysis Plan: Life out of the Shadows: Im- Marisol Rodr´ pacts of Amnesty Programs on Migrant’s Life.”. Ib´ ˜ anez, ıa, Andres Moya, Mar´ Ana Mar´ ıa Jose ıa Adelaida Ortega, Sandra V Rozo and Mar´ Urbina. 2022. “Life out of the Shadows.” World Bank. Working Paper . Ib´ ˜ anez, Ana, Sandra Rozo and Dany Bahar. 2020. “Empowering Migrants: Impacts of a Migrant’s Amnesty on Crime Reports.” IZA DP No. 13889. Working Paper . Jones, Damon. 2010. “Information, Preferences, and Public Benefit Participation: Exper- imental Evidence from the Advance EITC and 401(k) Savings.” American Economic Journal: Applied Economics 2(2):147–63. Kalil, Ariel, Haoxuan Liu, Susan Mayer, Derek Rury and Rohen Shah. 2023. “Nudging or Nagging? Conflicting Effects of Behavioral Tools.” Friedman Institute for Economics Working Paper . Karlan, Dean, Margaret McConnell, Sendhil Mullainathan and Jonathan Zinman. 2016. “Getting to the Top of Mind: How Reminders Increase Saving.” Management Science 62(12):3393–3411. 41 Liebman, Jeffrey B. and Richard J. Zeckhauser. 2004. “Schmeduling.” Working Paper . MacKinnon, James G and Lonnie Magee. 1990. “Transforming the dependent variable in regression models.” International Economic Review pp. 315–339. Madrian, Brigitte C. and Dennis F. Shea. 2001. “The Power of Suggestion: Inertia in 401(k) Participation and Savings Behavior.” The Quarterly Journal of Economics 116(4):1149– 1187. Manoli, Dayanand S. and Nicholas Turner. 2014. Nudges and Learning: Evidence from Informational Interventions for Low-Income Taxpayers. NBER Working Papers 20718 National Bureau of Economic Research, Inc. Mastrobuoni, Giovanni. 2011. “The role of information for retirement behavior: Evidence based on the stepwise introduction of the Social Security Statement.” Journal of Public Economics 95(7):913–925. Michael Hotard, Duncan Lawrence, David D. Laitin and Jens Hainmueller. 2019. “A low- cost information nudge increases citizenship application rates among low-income im- migrants.” Nature Human Behaviour 647(3). Rozo, Sandra V., Alejandra Quintana and Maria J. Urbina. 2023. The Electoral Conse- quences of Easing the Integration of Forced Migrants : Evidence from a Southern Coun- try. Policy Research Working Paper Series 10342 World Bank. Saez, Emmanuel. 2009. “Details Matter: The Impact of Presentation and Information on the Take-Up of Financial Incentives for Retirement Saving.” American Economic Journal: Economic Policy 1(1):204–28. ¨ Seira, Enrique, Alan Elizondo and Eduardo Laguna-Muggenburg. 2017. “Are Information Disclosures Effective? Evidence from the Credit Card Market.” American Economic Journal: Economic Policy 9(1):277–307. 42 Smeeding, Phillips and O’Connor. 2000. “The EITC: Expectation, Knowledge, Use and Economic and Social Mobility.” National Tax Journal 53(4.2):1187–1209. Stantcheva, Stefanie. 2022. “How to run surveys: A guide to creating your own iden- tifying variation and revealing the invisible.” National Bureau of Economic Research. Working Paper . Urbina, Mar´ ıa, Sandra Rozo, Andres Moya and Ana Ib´ ˜ anez. 2023. “Least Protected, Most Affected: Impacts of Migration Regularization Programs on Pandemic Resilience.” American Economic Association Papers & Proceedings. Forthcoming . 43 Contents A Details on the ETPV 45 A.A Treatment 1 Script: Information Video . . . . . . . . . . . . . . . . . . . . . 47 A.B Treatment 2 Script: Information Video Leveraging In-group Trust . . . . . 49 A.C Treatment 3 Script: Registration Process Video Leveraging In-group Trust 51 B Additional Analysis 54 C More on Attrition 55 44 A Details on the ETPV Figure A.1. ETPV Registry and Program Rollout February 8, 2021 Residency permit July 25, 2018 ( Estatuto de Protección Temporal para Residency permit Migrantes Venezolanos – ETPV) for (Permiso Especial de Venezuelans in Colombia who crossed Permanencia – PEP) the border prior to January 31, 2021 for irregular is announced Venezuelans is announced May 5, 2021 September October 13, June 28, RUMV 2021 2021 PPT 2022 registry Biometric document RUMV August 2, 2018 December 21, 2018 starts registry starts delivery starts registry ends PEP program PEP program starts ends 2018 2019 2020 2021 2022 Around 281,307 Around 1,963,864 Around people received the 2,351,351 attended the PEP document registered in the biometric appointment RUMV Census Around 1,492,295 PPT documents were delivered Figure A.2. ETPV Application Process 45 Figure A.3. Geographical Distribution of Venezuelans Registered in the RUMV Census Venezuelans with RUMV 14 - 5,500 5,501 - 18,500 18,501 - 30,000 30,001 - 100,797 100,798 - 432,251 46 APPENDIX B. Treatment Scripts A.A Treatment 1 Script: Information Video [A Colombian actor resembling an official provides the information.] Good morning, I am going to tell you what the Temporary Statute for Venezuelan Mi- grants is, better known as ETPV. The ETPV is a measure created for the regularization of Venezuelans for 10 years in Colombia. It will allow you to apply for the Temporary Protection Permit, known as PPT, which will give you access to the following benefits: • Get vaccinated against Covid-19 • Full access to health services for you and your family • Access to government subsidies through SISBEN • Access to any job with an employment contract in Colombia • Apply for a resident visa to be permanently legal in Colombia • Validate professional degrees • Open a bank account and apply for credits • Enter and leave the country without restriction • Access to the retirement system You are eligible to apply to the PPT and it’s free. In addition, 1,434,975 Venezuelans have already registered. I am going to explain how to apply, everything is done online and you just have to follow the following three steps: 1. Enter the page https://www.migracioncolombia.gov.co/visibles to reg- ister in the Unique Registry of Venezuelan Migrants, more known as RUMV 47 2. After registering for the RUMV, you schedule the appointment for the collection of biometric data on the page: https://agendamigracoletp.emtelco.co/#/. You must confirm the appointment in your email and attend the biometric data collection in person on the assigned date 3. You will receive the PPT virtually and three months later they will deliver it to you physically. I will tell you what you need to register in the RUMV: 1. Computer with internet 2. Active email 3. Have the following three documents scanned: • Identity Document: the passport, the Venezuelan ID or the Special Permit of Permanence are valid. • Photography with a white background. Remember that you can take it from your cell phone. • “Prueba Sumaria”: this is a document that proves that you arrived in Colom- bia before January 31, 2021. It could be a certificate of medical attention, the certificate of your child’s grades, the certification of your work, or any similar document that certifies that you were in Colombia before the stipulated date. Remember that all persons of legal age in your household must register separately. How- ever, when you make the RUMV registration, you will have the option of adding the minors in your charge, the system will schedule the appointment for taking biometric data for children between seven and 18 years old. Children under seven do not need an appointment because they have access to benefits with your PPT. I WILL SUMMARIZE THE STEPS FOR YOU: • REGISTER IN THE RUMV • APPOINTMENT FOR THE BIOMETRIC DATA • OBTAINING THE PPT DON’T FORGET TO SCAN: • YOUR PHOTOGRAPH • YOUR IDENTITY DOCUMENT • YOUR “PRUEBA SUMARIA” SAVE YOUR EMAIL AND PASSWORD, YOU WILL RECEIVE YOUR DOCUMENT THERE Do you need more information? Enter the website of https://www.migracioncolombia. gov.co/visibles 48 A.B Treatment 2 Script: Information Video Leveraging In-group Trust [A Venezuelan woman with children provides the information; the goal is for the vulnerable migrant to identify with the person providing the message.] Good morning, my name is Mar´ alez, I am a Venezuelan immigrant, I arrived in ıa Gonz´ Colombia irregularly with my children in July 2020, and I am going to tell you what is the Temporary Statute for Venezuelan Migrants, better known as ETPV. The ETPV is a measure created for the regularization of Venezuelans for 10 years in Colombia. It will allow you to apply for the Temporary Protection Permit, known as PPT, which will give you access to the following benefits: • Get vaccinated against Covid-19 • Full access to health services for you and your family • Access to government subsidies through SISBEN • Access to any job with an employment contract in Colombia • Apply for a resident visa to be permanently legal in Colombia • Validate professional degrees • Open a bank account and apply for credits • Enter and leave the country without restriction • Access to the retirement system You are eligible to apply to the PPT and it’s free. In addition, 1,434,975 Venezuelans have already registered. I am going to explain how to apply, everything is done online and you just have to follow the following three steps: 1. Enter the page https://www.migracioncolombia.gov.co/visibles to reg- ister in the Unique Registry of Venezuelan Migrants, better known as RUMV 2. After registering for the RUMV, you schedule the appointment for the collection of biometric data on the page https://agendamigracoletp.emtelco.co/#/. You must confirm the appointment in your email and attend the biometric data collection in person on the assigned date 3. You will receive the PPT virtually and three months later they will deliver it to you physically. I will tell you what you need to register in the RUMV: 1. Computer with internet 2. Active email 3. Have the following 3 documents scanned: 49 • Identity Document: the passport, the Venezuelan ID or the Special Permit of Permanence are valid. • Photography with a white background. Remember that you can take it from your cell phone. • “Prueba Sumaria”: this is a document that proves that you arrived in Colom- bia before January 31, 2021. It could be a certificate of medical attention, the certificate of your child’s grades, the certification of your work, or any similar document that certifies that you were in Colombia before the stipulated date. Remember that all persons of legal age in your household must register separately. How- ever, when you make the RUMV registration, you will have the option of adding the minors in your charge, the system will schedule the appointment for taking biometric data for children between seven and 18 years old. Children under seven do not need an appointment because they have access to benefits with your PPT. I WILL SUMMARIZE THE STEPS FOR YOU: • REGISTRATION IN THE RUMV • APPOINTMENT FOR THE BIOMETRIC DATA • OBTAINING THE PPT DON’T FORGET TO SCAN: • YOUR PHOTOGRAPH • YOUR IDENTITY DOCUMENT • YOUR “PRUEBA SUMARIA” SAVE YOUR EMAIL AND PASSWORD, YOU WILL RECEIVE YOUR DOCUMENT THERE Do you need more information? Enter the website at https://www.migracioncolombia. gov.co/visibles 50 A.C Treatment 3 Script: Registration Process Video Leveraging In-group Trust [A Venezuelan woman with children provides the information; the goal is for the vulnerable migrant to identify with the person providing the message.] Good morning, my name is Mar´ ıa Gonz´ alez, I am a Venezuelan immigrant, I arrived in Colombia irregularly with my children in July 2020, and I will explain to you step-by-step how I applied for the Temporary Protection Permit, better known as the PPT. 51 Figure B.1. Registration Process Video Step-by-Step 1. I entered the Migración Colombia page: https://www.migracioncolombia.gov.co/ 2. Click on the button "MAKE THE REGISTRATION IN THE RUMV HERE" 3. Click on the button "MAKE THE REGISTRATION IN THE RUMV HERE" 4. I entered my account and username, if you do not have it, follow the following procedure, select the option REGISTER 6. After filling in the information and selecting to register, you will receive an account activation email to your email, select the activation link and with this you will have your registration done. 7. With your registration done, you will have an active username and password. The username corresponds to the email and password you used to register. Now, you must enter them in the window of the home page, it will appear just as soon as you finish the registration. 5. In the register option, fill in the corresponding information. Remember that you must have an active email and you must have the number of one of the following types of documents: Passport, PEP, identity card, or birth certificate. ■ Upload a document type photograph on a white background ■ Upload a photo of the identification document ■ Upload the “prueba sumaria”. Remember that this is a document that proves that you arrived in Colombia before January 31, 2021. It could be a certificate of medical attention, the certificate of your child’s grades, the certification of your work by the employer, or any similar document that certifies that you were in Colombia before the stipulated date. This document is only for people who were irregularly in Colombia before January 31, 2021. If not, you should not upload it After uploading all the documents, select UPLOAD DOCUMENTS 13. When uploading the documents, you will get an ad with the indication to read and accept the terms, select the I agree box and then the button FINISH 14. Once the registration is finished, a window will appear with the announcement that will refer you to the completion of the Characterization Survey: Select go to the survey 8. Once you have entered the username and password, a window will appear for you to review your information and verify that this is the same as the one you entered when registering. 15. When you go to the survey, you will have to fill out the following information (show list on screen): 9. Review the information, select next and enter the data that will be part of your resumé. In the Operation Type box, ■ Questions of recognition and permanence select the only option that appears, the rest you can easily fill out. ■ Questions about your documentation, ethnicity and identity ■ Questions about your family group ■ Questions about living condition ■ Questions about occupation and study ■ Questions about social security ■ Questions about health ■ Question about reasons for migration ■ Questions about perception of integration ■ Questions about vulnerability ad 16. Once the survey is finished, you will be ready to schedule your face-to-face appointment for biometric registration. 10. Select next and enter your requested address and contact information. 11. Select next and register your family group information. Here you can add minors in your charge. If you are only going to make your registration, you do not need to fill out this information. Remember that all persons of legal age must register, it is not a registration per household, but per individual person. 12. Select the following and attach the required documents: 52 Figure B.2. Registration Process Video Step-by-Step 17. After completing the survey, you will be directed to the window that allows you to schedule your appointment. There you must select the button, schedule appointment 18. You will get an informative notice, after reading it, click accept You must bear in mind that in the consecutive field of the RUMV and in the consecutive field of the socioeconomic survey, you must enter the document number generated in the pre-registration. Which I point out to you in the following image. 19. Fill in the data for the appointment scheduling, you must fill out the following information: ■ City: The city in which you are going to carry out the procedure ■ Headquarters: The closest office to the place where you live ■ Type of Procedure: Temporary Protection Status - Biometric ■ Sub-procedure: Biometric Registration ■ Date available to attend the appointment The system will show you the dates and times available to carry out your procedure, select the date and venue that is of interest to you, by clicking the green button to the left of the appointment. 21. In the same window, you must upload the certificate of your registration. 22. Click I'm not a robot and follow the safety instructions. Finally, click on register and with this your appointment will be assigned. 20. When you select the button, a window will appear for you to fill in your personal data. 23. Your appointment will be correctly assigned with a number of file 24. Finally, you must check your mail both in received and in unwanted messages, or in spam to see if an email with the confirmation has arrived of the appointment. There you must confirm whether or not you will attend the appointment. 53 B Additional Analysis Table B.1. Intervention Effects on PPT Take-up Rates (Responses from First Contact) Request Start Registration Intention to Indicator Variables PPT Process Register (1) (2) (3) Panel A. ITT - General Effect β1 [Any Video] -0.094*** -0.107*** -0.178*** (0.027) (0.030) (0.032) FDR q-values [0.001] [0.001] [0.001] R-squared 0.010 0.011 0.027 Panel B. ITT - Dissaggregated Effect β1 [Awareness] -0.069** -0.095** -0.151*** (0.033) (0.037) (0.039) β2 [Trust] -0.080** -0.061 -0.139*** (0.033) (0.037) (0.039) β3 [Step-by-Step] -0.135*** -0.169*** -0.248*** (0.034) (0.038) (0.039) R-squared 0.014 0.018 0.035 Mean Untreated Group 0.274 0.365 0.786 Observations (All Panels) 1,130 1,130 1,130 Notes: Dependent variables: (i) Request PPT is an indicator [=1] if the individual reported having requested the PPT, or requested or attended the biometric appointment in the last survey contact. (ii) Start Registration Process is an indicator [=1] if the individual reported starting the RUMV census in the last survey contact. (iii) Intention to Register is an indicator [=1] if the individual reported the intention to start the RUMV census in the last survey contact. The experiment had 1,375 individuals registered. This table excludes from the sample 245 individuals who did not answer any of the four WhastApp surveys. Standard errors are reported in parentheses and False Discovery Rate (FDR) q-values are reported in brackets. ∗∗∗ significant at the 1%, ∗∗ significant at the 5%, ∗ significant at the 10%. 54 C More on Attrition Table C.1. Distribution of Attrition by Treatment Status and Different Definitions of Attrition Awareness Trust Step-by-step All Sample Control Any Video Video Video Video (1) (2) (3) (4) (5) (6) Attrited Four Contacts 0.225 0.167 0.223 0.232 0.281 0.245 Lost-to-Follow-Up 0.276 0.285 0.296 0.287 0.237 0.273 Lost-to-Follow-Up and Attrited Four Contacts 0.501 0.452 0.519 0.519 0.518 0.518 Observations 1,375 347 341 345 342 1,028 Notes: Attrited Four Contacts is an indicator [=1] for the people who did not answer the survey. Lost-to-Follow-Up is a dichotomous variable if the individual responded to at least one of the follow-ups 1 through 3 and responded that he/she had not completed the registration process for the PPT or requested the biometric appointment. 55 Table C.2. Characterizing Attrition Attrited Four Lost-to-Follow-Up or Lost-to-Follow-Up Contacts Attrited Four Contacts (1) (2) (3) Age -0.002 0.001 -0.001 (0.002) (0.002) (0.003) Male [=1] 0.090* 0.032 0.122* (0.054) (0.058) (0.064) Ed. Level: Primary or Less [=1] 0.074 -0.162** -0.087 (0.076) (0.082) (0.090) Ed. Level: General or diversified school [=1] -0.076 -0.021 -0.097 (0.065) (0.070) (0.077) Number of household members -0.004 -0.019 -0.023 (0.013) (0.014) (0.016) Personal Income (Sin*) -0.004 0.002 -0.002 (0.005) (0.006) (0.006) Activity spent the most time: Working [=1] 0.003 -0.019 -0.016 (0.057) (0.061) (0.068) Internet Access more than 4 hour per day [=1] -0.005 -0.015 -0.020 (0.048) (0.052) (0.058) Trust in Colombian Government (SD) 0.001 -0.005 -0.003 (0.026) (0.028) (0.031) Personal use Whatsapp [=1] -0.116** -0.023 -0.139** (0.058) (0.062) (0.069) Facebook or Instagram account [=1] 0.039 -0.010 0.029 (0.054) (0.058) (0.065) Twitter account [=1] -0.060 0.046 -0.014 (0.088) (0.095) (0.105) E-mail account [=1] -0.095 0.096 0.002 (0.067) (0.072) (0.080) Social desirability index (SD) 0.007 0.087*** 0.094** (0.031) (0.033) (0.037) Age x 1[Any Video Treatment] 0.002 -0.003 -0.001 (0.002) (0.003) (0.003) Male [=1] x 1[Any Video Treatment] -0.054 0.012 -0.042 (0.061) (0.066) (0.073) Ed. Level: Primary or Less [=1] x 1[Any Video Treatment] -0.022 0.163* 0.141 (0.087) (0.094) (0.104) Ed. Level: General or diversified school [=1] x 1[Any Video Treatment] 0.059 0.016 0.075 (0.074) (0.080) (0.088) Number of household members x 1[Any Video Treatment] -0.000 0.021 0.021 (0.015) (0.016) (0.018) Personal Income (Sin*) x 1[Any Video Treatment] 0.007 -0.006 0.001 (0.006) (0.007) (0.007) Activity spent the most time: Working [=1] x 1[Any Video Treatment] -0.033 0.010 -0.023 (0.066) (0.071) (0.079) Internet Access more than 4 hour per day [=1] x 1[Any Video Treatment] -0.001 0.012 0.011 (0.056) (0.061) (0.067) Trust in Colombian Government (SD) x 1[Any Video Treatment] 0.010 0.008 0.018 (0.029) (0.031) (0.035) Personal use WhatsApp [=1] x 1[Any Video Treatment] 0.077 0.031 0.108 (0.066) (0.071) (0.078) Facebook or Instagram account [=1] x 1[Any Video Treatment] -0.032 0.052 0.020 (0.062) (0.067) (0.074) Twitter account [=1] x 1[Any Video Treatment] 0.014 -0.032 -0.018 (0.102) (0.110) (0.121) E-mail account [=1] x 1[Any Video Treatment] 0.057 -0.132 -0.075 (0.077) (0.083) (0.092) Social desirability index (SD) x 1[Any Video Treatment] 0.008 -0.068* -0.060 (0.036) (0.039) (0.043) Any Video Treatment [=1] -0.105 -0.043 -0.149 (0.160) (0.172) (0.191) R-squared 0.035 0.024 0.044 Observations 1,375 1,375 1,375 Mean Dependent Variable 0.225 0.276 0.502 Notes: Attrited Four Contacts is an indicator [=1] for the people who did not answer the survey. Lost-to-Follow-Up is a dichotomous variable if the individual responded to at least one of the follow-ups 1 through 3 and responded that he/she had not completed the registration process for the PPT or requested the biometric appointment. 56