Policy Research Working Paper                        10799




       Do Disasters Always Increase Intimate
                Partner Violence?
                 Evidence from the 2018 Earthquake
                       in Papua New Guinea

                                       Alyssa Leng
                                     Sharad Tandon




Poverty and Equity Global Practice
June 2024
Policy Research Working Paper 10799


  Abstract
 This paper examines how an earthquake in Papua New                                 partner violence, driven by declines among women, who
 Guinea changed people’s attitudes about and the prevalence                         are the least likely to underreport intimate partner violence.
 of intimate partner violence. Although there are several rea-                      The results highlight that the responsibilities of household
 sons why disasters can aggravate intimate partner violence,                        members and social norms can change in sufficiently tur-
 among men in disaster-affected regions, the accept-ability of                      bulent disasters, which can lead to improvements, and that
 intimate partner violence declined significantly. There was a                      measurement issues need to be better addressed to improve
 smaller and noisier decline in reported incidents of intimate                      understanding of intimate partner violence.




 This paper is a product of the Poverty and Equity Global Practice. 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 standon3@worldbank.org or alyssa.leng@anu.edu.au.




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


                                                       Produced by the Research Support Team
 Do Disasters Always Increase Intimate Partner Violence? Evidence from the 2018 Earthquake in
                                              Papua New Guinea*




                                     Alyssa Leng„ and Sharad Tandon„




JEL classification: D10; D13; O12; O56

Keywords: Gender-Based Violence; Disaster; Papua New Guinea




   * The  authors would like to thank Andrew Blackman, Stephen Howes, Rinku Murgai, Jane Sprouster, Stephen
Ndegwa, Khwima Nthara, Dhiraj Sharma, and Tara Vishwanath for comments on earlier drafts. The views expressed
here are those of the authors and may not be attributed to the World Bank.
   „ Development Policy Center, the Australian National University, 7 Liversidge St., Canberra ACT 2601, Australia,
alyssa.leng@anu.edu.au.
   „ Poverty and Equity Global Practice of the World Bank, 1818 H St. NW, Washington, D.C. 20433, USA, stan-
don3@worldbank.org
1. Introduction

       A large and growing literature illustrates that intimate partner violence (IPV) is pervasive across
the world. Worldwide estimates suggest that 26 percent of women have experienced IPV in their
lifetimes, 10 percent of women have experienced IPV in the past year, and that women in all
countries and in households of diverse backgrounds experience IPV (e.g., Devries 2013; WHO 2021;
etc.). However, the large shares of women experiencing IPV is likely underreported given substantial
social desirability biases against honestly reporting IPV experiences for both the perpetrator and
the victim (e.g., Fisher 1993; Gregson et al. 2002; Bell and Naugle 2007; etc.), and given potential
risks posed to respondents who answer honestly (e.g., Zimmerman 1995; Ellsberg and Heise 2005;
etc.).
       Although there are several interlinked causes of IPV worldwide,1 natural disasters have the
potential to increase the already high prevalence of IPV across the world (e.g., van Daalen et al.
2022). However, the empirical evidence is far from definitive regarding the impact of disasters on
IPV. Of the few settings that have been able to investigate the impact of turbulent natural disasters
on IPV, several have estimated imprecise changes that can rule out neither no effect on IPV nor
large changes (e.g., Frasier et al. 2004; Fagen et al. 2011; Diaz and Saldarriaga 2023; etc.).2 , 3
Furthermore, even evidence on the IPV impacts of changes in economic well-being, which is often a
significant component of a disaster, does not uniformly suggest that worse economic outcomes lead to
increased IPV.4 Given the lack of definitive evidence, we need to more fully understand the impacts
of different types of disasters over different time frames, along with the mechanisms by which they
change the prevalence of IPV, to ensure that disasters do not aggravate the already poor global IPV
situation.
       Corroborating the empirical ambiguity, we present a simple conceptual framework that empha-
   1
      For example, see Council of Europe (2019) for a summary of some of the cultural, legal, economic and political
factors that underlie the high rates of IPV.
    2
      Two settings find that disasters increase the prevalence of IPV, but they also illustrate that other dimensions
captured in the survey were also trending differently between disaster and disaster-affected regions (Weitzman and
Behrman 2016; Shwefer 2018).
    3
      There are two findings in Dias and Saldarriaga (2022). The article finds both that there are little estimated
changes in reported IPV following flooding disasters and that droughts increase the prevalence of reported IPV
incidents, likely through a negative income shock. The citation above references the former result.
    4
      For example, see JPAL (2022) for a summary of the conflicting evidence. Several settings illustrate instances
where worse economic outcomes lead to increased IPV (e.g., Gennetian 2003; Gibson-Davis et al. 2005; Kim et al.
2007; Gupta et al. 2013; Haushofer and Shapiro 2016; Hidrobo et al. 2016; Dias and Saldarriaga 2023; etc.). However,
several settings also cannot reject the hypothesis of no change in IPV from negative income shocks (e.g., Cools et al.
2020; Cooper et al. 2021; etc.); some settings even find that less economic resources actually reduce the potential for
conflict over resources and lead to less IPV (e.g., Macmillan et al. 1999; Heise 2011; Bulte and Lensink 2019; Cullen et
al. 2024 etc.); and some settings illustrate that whether IPV increases or decreases can depend on the characteristics
of perpetrators and victims (e.g., Hidrobo and Fernald 2013).



                                                           2
sizes that the impact of disasters does not necessarily increase the prevalence of IPV, even if it is
assumed that negative economic shocks lead to increased IPV.5 Specifically, husbands have the po-
tential to engage in IPV if consumption falls below a critical threshold. However, turbulent disasters
have the potential to decrease consumption and to change intrahousehold roles and gender norms,
which can in turn also change the critical consumption threshold and further affect the prevalence
of IPV (e.g., Panda and Agarwal 2005; Heise 2011; Doss 2013; etc.). Whether a disaster increases or
decreases the prevalence of IPV in this setting depends on the change in the consumption threshold
relative to the change in consumption.
      To more fully investigate the relationship between disasters and IPV, we analyze the impact of a
devastating earthquake and a series of aftershocks that precipitated a humanitarian and displacement
crisis in Papua New Guinea (e.g., Government of PNG et al. 2018; IOM 2018; etc.). The disaster
occurred between the third and fourth phase of a nationally representative survey that collected
extensive information on IPV. The initial phases of the survey focused on urban areas and then
progressively surveyed more rural and remote regions, and comparing the IPV changes after the third
phase in the provinces most strongly affected by the disaster to other regions allows us to estimate
the immediate impacts. Several robustness checks illustrate that the changes in IPV following
the disaster are likely not being driven by substantial differences between urban and more remote
populations across provinces or in regional differences in survey implementation.
      The data illustrate two primary empirical findings. First, there was a large decline in the accept-
ability of IPV in the disaster-affected regions in which the vast majority of the deaths occurred and
in which virtually all of the internal displacement was concentrated (e.g., Government of PNG et al.
2018; IOM 2018; etc.). The share of men who thought it was acceptable for a husband to beat his
wife declined in all five individual scenarios captured, and the total number of scenarios in which
a husband is justified to beat his wife reduced by 0.64 more than in the rest of the country. The
declines were large in magnitude, where the latter change represents a 19 percent reduction relative
to pre-disaster levels in affected regions. However, there was little change in the acceptability of IPV
among women in the same regions.
      Second, there was a large difference between changes in men’s attitudes about IPV and changes
in the prevalence of IPV reported by women. Specifically, while attitudes about IPV markedly
improved in disaster-affected regions, there was a smaller and less precisely estimated decline in
the share of women reporting to have experienced IPV in the past year in the same regions. But
  5
      As described above, there are several settings where this assumption is empirically refuted.


                                                            3
these overall changes mask significant heterogeneity, where there was no change for women who were
most likely to underreport IPV incidents and large and precisely estimated declines in IPV incidents
reported by other women. Importantly, there were improvements in IPV attitudes among men
regardless of the likelihood of their partners to underreport IPV incidents, consistent with reporting
issues driving the heterogeneity in IPV prevalence.
      We also investigate the hypothesis that these changes in IPV are the result of a potential change
in the role of women in the household. Although specific roles and responsibilities assumed by
household members are mostly unobservable, the survey utilized here captures the household member
that was responsible for specific decisions, such as who was primarily responsible for deciding how
to spend household income and whether to make large purchases. Consistent with the conceptual
framework that highlighted the possibility of turbulent disasters impacting intrahousehold roles and
social norms, we find that the share of husbands solely responsible for several important decisions
declined at the same time as the responsibility of the wife increased.
      There are three primary contributions of these empirical results. First, these results illustrate that
despite a substantial negative income shock and a rise in food insecurity in disaster-affected regions,
turbulent natural disasters can decrease the prevalence of IPV given the many other potential impacts
that disasters might have on individuals and households. Although the link between disasters and
IPV is not uniform across settings, the majority of results indicate either an increase or a null result.6
These results illustrate an important possibility in a world in which climate hazards are becoming
increasingly prevalent (e.g., Hallegatte 2014).
      Second, the large differences between changes in IPV attitudes and reported incidents illustrates
the importance of accounting for measurement issues when analyzing IPV. Although the possibility
has long since been acknowledged, most quantitative studies analyzing the IPV impacts of disasters,
income shocks, and other interventions focus only on incidents directly reported by women.7 The
few studies that address the potential mismeasurement of IPV prevalence have primarily focused on
list experiments (e.g., Bulte and Lensink 2019; Cullen 2023; etc.). However, the results here also
illustrate that triangulating between IPV incidents reported by victims and hypothetical IPV atti-
tudes of potential perpetrators, the latter of which potentially has a lower amount of underreporting
than IPV incidents (e.g., Ellsberg and Heise 2005), is also a potentially rich source of information.
  6
    See van Daalan et al. 2022 for a summary of findings across disaster settings.
  7
    See van Daalen et al. (2022) for a summary of the IPV measures analyzed in disaster setting and see Tankard
and Iyengar (2018) for a summary of IPV measures analyzed when investigating the link between economic shocks
and IPV.



                                                      4
Given the difficulty in ever fully overcoming the measurement issue in any single strategy,8 the use of
many improved measurement approaches simultaneously might best identify the level and changes
of IPV prevalence.9
       And third, these results build on work demonstrating the link between the expanded role of
women in the household and changes in IPV (e.g., Bulte and Lensink 2019; Bandiera et al. 2020;
Cullen et al. 2024; etc.). Given a simultaneous expansion of women’s roles in the household and
a decline in acceptability and prevalence of IPV estimated here, these results further illustrate the
potential importance of empowering women in the household. Better understanding the implications
of empowerment on a wider range of IPV outcomes could potentially lead to scalable interventions
that decrease the prevalence of IPV (e.g., Heisse 2011).10 However, given the empirical evidence is
mixed regarding whether IPV improves or worsens,11 there also needs to be more investigation in
how to minimize the possibility of aggravating IPV in the process of improving empowerment.
       The rest of the paper is structured as follows. Section 2 presents a conceptual framework illus-
trating the emprical ambiguity of how disasters might impact IPV; Section 3 presents background
information on IPV in PNG and the 2018 earthquake; Section 4 describes the data used in the em-
pirical analysis; Section 5 describes the empirical strategy; Section 6 reports the baseline empirical
results; Section 7 explores potential reasons for the decline in reported incidents being different from
the improvement in IPV attitudes; Section 8 explores potential mechanisms for the changes in IPV;
and Section 9 concludes.

2. Conceptual Framework

       This section illustrates how the impact of disasters on IPV is complex and empirically ambiguous
in a setting in which it is assumed that a decline in economic well-being leads to an increase in the
   8
      For example, list experiments are limited in the types of things that can be asked and the specificity with
which respondents can describe sometimes complicated viewpoints, the results of some of these approaches are highly
dependent on the wording of questions and can significantly vary from experiment to experiment, and the inability
to identify responses at the individual level limits the ability to further investigate the causes and consequences of
sensitive decisions (e.g., Tandon and Vishwanath 2022). As a consequence of some of these challenges, the estimates
from list experiments are often not very precise in the few instances that one has the ability to validate the results
(e.g., Rosenfield et al. 2015; Kramon and Weghorst 2019; etc.).
    9
      This approach is similar to other instances of using several strategies simultaneously to elicit sensitive informa-
tion from respondents, such as list experiments, endorsement experiments, randomized response design, embedding
enumerators in the community to gain trust, and utilizing more anonymous survey modalities (e.g., Glynn 2013; Blair
et al. 2013; Blair et al. 2015; Blattman et al. 2016; Tandon and Vishwanath 2022; etc.) Additionally, this approach
is also similar to triangulating between objective and subjective welfare measures to more precisely interpret levels
and changes in well-being (e.g., Stiglitz et al. 2009; OECD 2013; Krueger and Stone 2014; Tandon 2024; etc.).
   10
      A growing body of work illustrates how to best expand women’s empowerment in which the link to IPV is not
investigated (e.g., Goldin and Katz 2002; Duflo 2012; Jayachandran 2015; Dhar et al. 2022; etc.).
   11
      For example Bulte and Lensink (2019) illustrate a worsening of IPV following an intervention that improves
employment of women; and Bandiera et al. (2020) find an improvement in one IPV outcome following an intervention
that improved the productive capacity of women and other empowerment dimensions.


                                                           5
likelihood of IPV. Specifically, individuals choose a consumption bundle (Ci1 ) based on prices p
and wealth wi in the initial period; and then a disaster occurs in the second period that reduces
                    δCi1
consumption by       δS
                           < 0. In each period, a husband becomes sufficiently stressed and it becomes
possible that he might engage in IPV if the consumption bundle is below a minimum threshold:




(1)                                                             ¯i )
                                              IP Vit = I (Cit < C


where I is an indicator function equaling one if household consumption is above their minimum
       ¯i .
bundle C
                                         ¯i can be influenced by gender norms, by intrahousehold rela-
       The individual-specific threshold C
tionships and responsibilities, and all the other issues that have been illustrated to determine the
prevalence of IPV (e.g., Council of Europe 2019). The framework incorporates the two limiting cases
                                                                                               ¯i
in which husbands are always potentially willing to commit IPV regardless of economic factors (C
larger than any feasible consumption value) and in which there is no level of economic well-being
                                               ¯i = 0).
such that a husband will commit an act of IPV (C
       In this simple setting, if disasters only impact the economic well-being of a household, it is
                                                                                    δIP Vit
necessarily the case that disasters will increase the likelihood of IPV (             δS
                                                                                              > 0). Specifically,
the share of husbands that would change from being unwilling to commit IPV to those that might
would be the share in households that had pre-disaster consumption just above the consumption
threshold.12 Depending on the share of such individuals, which further depends on the size of the
consumption shock, the actual change in IPV could either be small or large.
       However, the above situation critically assumes that the threshold at which a husband might
                                                  ¯
                                       Ci
commit acts of IPV does not change ( δδS  = 0). But if the disaster is sufficiently turbulent that it
changes social norms, responsibilities of individual family members, or any of the other factors that
are incorporated in the individual-specific consumption threshold, it is possible that the economic
                                                                                         ¯i
                                                                                        δC
threshold at which a husband might commit acts of IPV also changes (e.g.,               δS
                                                                                              < 0).13 And in the
case that the consumption threshold declines significantly, the share of husbands that might commit
                                        δIP Vit
acts of IPV actually can decrease (       δS
                                                      < 0). The key comparison is the magnitude of the change
                                                                                         1                  1
  12                                                                               ¯i + δCi < C 1 < C
     Specifically, all husbands in households with consumption in the first period C                 ¯i − δCi will
                                                                                         δS     i          δS
change from being unwilling to commit an act of IPV to potentially being able to.
  13
     For example, in the case of a disaster where women assume new roles and responsibilities in the household, or
in a case where there is community trauma from many deaths and fear for safety, the threshold at which a husband
might commit IPV might possibly change.


                                                             6
in the consumption threshold relative to the magnitude of the consumption decline.14
       Thus, even in this special case of how IPV might change following the onset of a disaster, it is
still possible that IPV declines. But, as described in the Introduction, there are instances where a
change in economic well-being leads either to little change in IPV or to an increase in IPV (e.g.,
Heise 2011; Bulte and Lensink 2019; etc.), which further emphasizes the possibility that there might
be little change in IPV or even a decline following a disaster. Thus, more empirical analysis of
disaster settings is needed to better illustrate the degree to which disasters impact IPV and in what
direction.

3. Background on Intimate Partner Violence in PNG and the 2018 Earthquake

       Rates of IPV prevalence in PNG are among the highest in the world (e.g., Sardinha et al., 2022),
where 56 percent of women between the ages of 15-49 years have experienced physical violence from
the age of 15 and 63 percent of married women experienced any form of spousal violence by a current
or former partner (DHS 2019). However, the reported patterns of IPV in the country are puzzling,
suggesting that estimates of the prevalence of IPV might be imprecise. For example, reported IPV
incidence is higher for more educated and more urban women despite the fact that IPV is more
acceptable on average among both less educated and more rural men and women (Baxi et al. 2023;
and Mambon 2023).15
       A full catalog of potential causes of IPV in PNG is out of the scope of this work, but it is
important to note there is significant heterogeneity regarding gender norms and practices across the
country. There are broad expectations for males to be dominant in society at large and in the home,
while females are expected to be subservient domestically (e.g., Rooney et al. 2022), and differences
between these assumptions and actual behavior can then contribute to the occurrence of IPV (e.g.,
Eves 2018; Smith 2024). Additionally, other practices in parts of PNG, such as the payment of bride
prices, may also play a part in IPV through men perceiving a greater level of ownership or authority
over their wives (e.g., Eves 2019).
       While the PNG government has become increasingly aware and active on the issue of IPV,
  14
    The share of husbands who change from potentially committing acts of IPV to being unwilling to is increasing
                                                                  ¯i
                                                                  C
in the magnitude of the decrease in the consumption threshold ( δδS  ); and the share of husbands that change from
                                                                                                            1
                                                                                                         δCi
unwilling to potentially able to commit IPV increases in the size of the consumption decrease (i.e.,      δS ).   In the
                            δC 1
case of sufficiently small δSi or a sufficiently small number of individuals with first period consumption such that
         1                   1                                     ¯i
C¯i + δCi < C 1 < C  ¯i − δCi , coupled with a sufficiently large δC
       δS       i          δS                                     δS or a sufficiently large number of individuals with
                      ¯i − δC¯i    1    ¯     ¯i
                                             δC
initial consumption C       δS < Ci < Ci + δS , there will be a decline in the share of husbands that might potentially
engage in acts of IPV.
  15
     These patterns are corroborated by the summary statistics reported in Section 3.



                                                          7
there is still much to be done. A national strategy on IPV was published in 2016 (Government of
Papua New Guinea 2016) and a Special Parliamentary Committee on Gender-Based Violence was
established in 2020. Recommendations from this process included the establishment of a permanent
IPV parliamentary committee, the creation of national and provincial IPV secretariats to roll out
the 2016 national strategy, and an increase in funding, staffing and training for law enforcement
efforts around IPV (Government of Papua New Guinea 2022).
   But in the midst of a national crisis in IPV, there was a devastating earthquake that struck the
country early in 2018. A 7.5 magnitude earthquake struck the country on 26 February 2018, with
over 270 aftershocks registered in months after the initial event. While the earthquake’s epicenter
was in Southern Highlands, neighboring provinces including Hela, Western, and Enga were also
strongly affected (UNCT 2018; WHO 2018).
   More than 500,000 people were affected by the earthquake, with over 34,000 people displaced
and an estimated death toll of 100 (ICRC 2018; and World Health Organization 2018). The earth-
quake caused severe and widespread damage to dwellings, critical infrastructure, and farms (e.g.,
ICRC 2018). Access to clean water, food and shelter was severely constrained in areas that were
consequently considered priority areas in the period following the disaster (e.g., ICRC 2018; WHO
2018). In response to the disaster, the PNG government announced a state of emergency in March
2018 in Hela, Southern Highlands, Western and Enga provinces (UNCT 2018). K450 million was set
aside to fund disaster relief and rebuilding efforts by the PNG government, with additional support
provided by international donors including Australia and the UN (e.g., DFAT 2018; UN 2018).
   However, despite a state of emergency being declared in four provinces, the devastation in Hela
and Southern Highlands provinces was especially severe. Hela was the worst-affected region by the
Earthquake in terms of deaths and accounted for 60 percent of internal displacement, and Southern
Highlands accounted for the second largest number of deaths and virtually the rest of the internal
displacement not in Hela province (e.g., Government of Papua New Guinea et al. 2018; IOM 2018);
and the food security crisis was reported as being more severe in Hela and Southern Highlands than
in the other regions in which a state of emergency was declared (e.g., UNCT 2018). These two
provinces, in which the disaster was especially turbulent and severe, are the focus of the rest of the
analysis.

4. Data

   As described in the Introduction, the earthquake and its aftershocks struck the country in the


                                                  8
midst of the ongoing 2016-2018 Demographic and Health Survey (DHS). The survey consisted of a
household survey, and then separate surveys for eligible men and women in the household. In all
households in the survey, all women who were aged 15-49 and who were either usual residents of
the household or who spent the night before in the household were eligible to be interviewed; and
for every second household in the survey, all men who were aged 15-49 and who were either usual
residents of the household or who spent the night before in the household were eligible to complete
the men’s questionnaire. In total, 16,021 households were surveyed with a common questionnaire,
15,198 women were administered the women’s questionnaire,16 and 7,333 men were administered the
men’s questionnaire.17
       The survey is nationally and provincially representative, and can report estimates at the rural
and urban levels. The fieldwork was conducted in four waves between October 2016 and December
2018.18 However, the households interviewed were not randomized across waves. Rather, the survey
began in the most urban areas in each province, and then progressively surveyed more rural and
remote areas (DHS 2019).19
       All households to be interviewed were selected prior to the beginning of the survey in 2016, and
no households were allowed to be replaced. But data collection was difficult and fieldwork could not
be completed in 33 of the 800 census units originally selected in the sample design, reducing the
overall sample size from the initial 19,200 households selected to the 16,012 households that were
ultimately included in the survey. Reasons for the delays and for not completing work in each census
unit include difficulties in handling the terrain in the country, adverse weather, and security issues
(e.g., DHS 2019).
       Importantly, the disaster did affect the survey. The worst-affected region by many measures-
Hela- was not able to field the fourth phase of the survey following the earthquake; and the second
worst-affected region - Southern Highlands- delayed interviews in the fourth phase until seven months
after the final large aftershock while other provinces were performing interviews continuously over
the final phase. Although the disaster affected the timing of the survey in Southern Highlands, as
discussed above, the choice of households was not affected.20 But given the exclusion of Hela from
  16
     The number of women interviewed in each household varied between one and 10.
  17
     The number of men interviewed in each household varied between one and eight.
  18
     Phase one was conducted between October 1 and December 9, 2016; phase two was conducted between March and
June 2017; phase three was conducted between October and December 2017; and phase four was conducted between
April and December 2018.
  19
     The survey implementation was also corroborated by informal discussions with the National Statistical Office of
Papua New Guinea.
  20
     The survey documentation does not provide non-response rates of those initially selected to be interviewed by
phase or province. However, we investigate the possibility in the robustness section.


                                                         9
the survey following the disaster, the empirical analysis focuses on the change in IPV prevalence in
the Southern Highlands relative to the rest of the country.
      In this analysis, we first focus on the men’s survey, which includes a range of questions about the
acceptability of intimate partner violence (IPV). The question states ”in your opinion, is a husband
justified in hitting or beating his wife in the following situations,” and then presents a range of
situations. The situations are ”if she goes out without telling him,” ”if she neglects the children,”
”if she argues with him,” ”if she refuses to have sex with him,” and ”if she burns the food.” And
respondents are able to respond ”yes,” ”no,” or ”don’t know.” The question only asks about a
hypothetical scenario about a husband and wife and does not ask if the respondent would hit or
beat his wife in each situation.
      In addition to the men’s survey, we also focus on a subset of the women’s survey that admin-
istered a domestic violence module. In a random subset of the households that were selected to
be administered the men’s questionnaire, one woman who was eligible for the women’s survey was
randomly selected to respond to the added module. The module was only fielded in households in
which privacy from other household members could be secured, resulting in a sample of 4,873 women
(DHS 2019). The questions about abuse on which we focus are the questions about the incidence of
IPV, which are asked only of the 3,945 women in the sample that had either been married or had
lived with a man at one point in their lives.
      The enumerator began fielding the module with a statement that guaranteed the anonymity of
the respondent and that nobody else in the household would be informed of the responses. The
extensive module asked whether the respondent’s husband or partner engaged in three different
types of emotional abuse (e.g., humiliate or threaten, etc.) and ten different types of physical abuse
(e.g., slap or punch, etc.) directed at the respondent.21 For each type of abuse that the respondent
experienced, they were then asked if the abuse happened ”often,” ”sometimes,” or ”not in the past
12 months.” The module further asked additional relevant questions about IPV. Specifically, the
module asks whether the respondent is often afraid of her husband or partner, sometimes afraid, or
never afraid; and the module further asks the respondent if she had physically harmed her husband
or partner at times when he was not harming her.
      The respondent in the women’s survey is explicitly told that they are allowed to not answer
specific questions if they so choose. However, given the very unique circumstances of PNG and the
careful fieldwork ensuring the safety of respondents, the rates of non-response are very low. In each
 21
      An exhaustive list of types of abuse covered can be provided upon request.


                                                          10
of the 13 individual questions on types of IPV experienced, no more than eight respondents out of
the 3,945 refused to answer; and most respondents answered all questions, with only 45 respondents
out of 3,945 not answering all the 13 questions on instances of IPV on which the empirical analysis
focuses.22
       As described in the Background Section, the summary statistics illustrate a dire IPV situation
in the country.23 The prevalence of men answering that it is acceptable for a husband to beat their
wife varies between 18.1 percent and 61.6 percent of men, depending on the scenario; and the share
of men that reported it was acceptable for a husband to beat their wife in all scenarios is 10 percent.
Although the questionnaire does not ask men about IPV in which they participated, the high share
of acceptability illustrates a climate in which IPV is likely common.
       The high shares of acceptability of IPV translates into a high share of women who reported
experiencing IPV in the past year. The share of women who experienced less severe forms of physical
violence captured- pushing, slapping, or twisting an arm- was between 29 and 35 percent; and the
share of women who experienced more severe forms of physical violence captured- punching, beating
up, choking, or attacking with a knife or a gun- was between 10 and 22 percent. The share of women
experiencing sexual violence was also very high, with between 10 and 22 percent of women reporting
to have been forced to have intercourse or perform other sex acts in the past year.24 The share of
women who experienced forms of emotional IPV in the past year is also high and varies between
25 and 37 percent of women. As discussed in the background section, 63 percent of women report
having experienced at least one IPV incident over the past year, which is much larger than the 10
percent of women globally who reported experiencing IPV in the past year and the 26 percent of
women globally who reported experiencing IPV in their lifetimes.
       However, as described in the documentation of the survey implementation, Figure 2 illustrates
that the DHS progressively reached more and more rural areas over time. The share of the population
that was rural in the first three phases, which were conducted before the disaster, was 15 percentage
points lower than that in the fourth phase. Figure 2 illustrates that the survey implementation
is important for interpreting the resulting estimates, where Figures 2b and 2c illustrate that IPV
  22
     But because of the small numbers of non-response, there are slight differences in the sample size for different
dependent variables. The results report keeping all observations and treating individual non-responses as missing
observations. However, all results are qualitatively identical if the sample is restricted to the 3,900 households that
had responses recorded for each violence question.
  23
     An expanded list of summary statistics for all additional variables used in the analysis is available from the authors
upon request.
  24
     The categorization of IPV into less severe, severe, and sexual violence is one that was used in the DHS questionnaire
(DHS 2019).



                                                            11
attitudes and prevalence were significantly different between rural and urban areas.
       As one would expect given the literature on the prevalence and causes of IPV worldwide, Figure
2b illustrates that the acceptability of IPV was slightly lower in urban areas than in rural areas,
with the difference being between 6 and 9 percentage points lower in urban areas. But as discussed
in the background section and contrary to expectations, the reported prevalence of IPV was higher
in urban areas than in rural areas for nine out of the 10 types of violence experienced in the past
year. Although the confidence intervals of the estimates often overlap and only one of the differences
is statistically different at conventional significance levels when including individual-level control
variables (e.g., education, etc.), the rural-urban differences in reported IPV incidents are far different
than the clear patterns presented in the acceptability of IPV in Figure 2b.25 However, it is difficult
to interpret whether these are actual differences in the prevalence of IPV or if they might represent
differences between urban and rural areas in the willingness of victims to report the violence they
experience.

Section 5a. Empirical Strategy

       As described in the Introduction, we estimate how men’s attitudes about IPV and how the
prevalence of IPV incidents reported by women changed following the 2018 earthquake in PNG,
which immediately preceded the beginning of the fourth phase of the 2016-2018 DHS. Specifically,
we estimate the following difference-in-differences specification:




(2)        IP V Indicatoritr = τr + βP ostitr + ρAf f ectedir + γP ostitr ∗ Af f ectedir + δXi + ϵi


where P ost denotes an indicator equaling one if the individual was interviewed after the earthquake
in the fourth phase of the survey; Af f ected denotes an indicator equaling one if the household lived
in the Southern Highlands Province, which was the region that was most strongly affected by the
earthquake that was able to be included in the post-earthquake phase of the survey;26 X denotes
individual-level control variables; and t denotes month-year, r denotes the four broad regions in the
country, and i denotes individuals.27
  25
      Estimates of the rural-urban difference in the prevalence of reported IPV incidents in the past year when including
all control variables included in the baseline empirical specification are available from the authors upon request.
   26
      The other significantly affected province- Hela- was unable to conduct any interviews following the disaster. The
results are identical if Hela is entirely dropped from the sample, or if the indicator identifies either households from
Southern Highlands or Hela as being affected populations.
   27
      Treatment in this case is not staggered and there are only two treatment groups, which are the settings in
which the validity of the difference-in-difference estimator has been illustrated to be questionable (e.g., Callaway and


                                                           12
   The estimate of γ represents how much larger the share of the population responding ”yes” to the
IPV indicator increased in disaster-affected regions than in the rest of the country. Importantly, these
results are a likely underestimate of the impacts in the worst-affected regions given the omission of
the worst-affected areas from the fourth phase of the survey and the delay in interviewing households
in the region on which this analysis focuses.

Section 5b: Robustness Checks of the Identification Strategy

   As discussed in the Introduction, there are three critical assumptions for γ to identify the arguably
causal impact of the earthquake. Specifically, there cannot be substantial differences between urban
and more remote populations across provinces, the survey design and implementation needs to be
roughly identical across provinces, and the IPV change in the rest of the country need to be a good
counterfactual for the change in disaster-affected regions in absence of the earthquake. A variety of
robustness checks illustrate that these conditions are likely met and that the results are being driven
by the earthquake.
   First, we illustrate that the empirical strategy uncovers no significant differences in individual
characteristics that would be impossible to be affected by the disaster, where differences in the
sampling strategy or in how urban and more remote regions varied by province would result in
different samples where such differences would likely be evident. Tables 1 and 2 reports the changes
in each individual control variable used in the baseline specification in disaster-affected regions in
column (1) following the disaster, the change in the rest of the country in column (3), and reports
the difference between the changes in column (5). Table 1 reports the changes for the Men’s Survey
and Table 2 reports the changes for the Women’s Survey.
   Consistent with the identification strategy, the results illustrate that there were few differences
in the changes between disaster-affected regions. As expected by going from urban to rural and
remote areas over time, the estimates illustrate that households are socioeconomically worse off
following the earthquake in the fourth phase in both the disaster-affected regions and the rest of
the country. However, in the difficult-to-adjust characteristics, there is no difference in the changes
between disaster-affected regions and the rest of the country. Of the 38 variables captured in both
the Men’s and Women’s Surveys in Tables 1 and 2, the vast majority are small in magnitude and
imprecisely estimated. Only one of the differences in column (5) is statistically significant at the 5
Sant’Anna 2020; Goodman-Bacon 2021; Sun and Abraham 2021; Baker et al. 2022; Borusyak et al. forthcoming;
etc.).




                                                   13
percent level across Tables 1 and 2, and only two differences are statistically significant at the 10
percent level, which is what would be expected by chance.
       Second, further corroborating the conditions required for γ to represent the arguably causal
impacts of the earthquake, the identification strategy is able to uncover a pattern independently
reported by humanitarian agencies using other data sources at the time of the earthquake- a growing
food security crisis arising from the disaster that was worst felt in the Southern Highlands and Hela
provinces. The DHS collected the Food Insecurity Experience Scale in the household survey, which
is composed of both subjective and objective questions on food access.
       Using an identical identification strategy as described in (1) but using the household survey
information on food security as the dependent variable, Figure 3 reports the difference-in-differences
estimate of the change in food security relative to the rest of the country. As expected by the
sampling strategy that reaches more rural and remote households in later phases of the survey,
there was a worsening of food access for all households following the earthquake in the fourth phase
of the survey across the country. However, there was a substantially larger increase in the share
of households reporting poor food access in the regions worst affected by the disaster. There was
between a 6 and 15 larger percentage point increase in the share of poor food access in disaster-
affected regions for the three objective food security measures (skip meals, run out of food, and
going a whole day without food), which represents between 16 and 45 percent of the pre-disaster
share of households that reported poor food access across the entire country.28
       Third, it is possible that the disaster led to changes in survey quality or in the willingness of
individuals to respond to the IPV module in earthquake-affected regions, and this change might be
driving some of the results and not an actual change the prevalence of IPV. Even though the low
rates of non-response reported in the Data section suggest that this is likely not a significant problem,
it is possible that the few instances of non-response were concentrated in disaster-affected regions
after the earthquake. However, Figure 4 reports estimates of specification (2) using an indicator
equaling one if a woman did not respond to the IPV question and illustrates that this is not likely
to be the case. All of the estimates are small in magnitude, vary in sign, and are unable to rule out
the possibility that there is no difference in the change in non-response in disaster-affected regions
  28
     Although subjective questions on food security add information about how individuals felt about their experiences
with food insecurity, the questions are difficult to interpret given different individuals use different thresholds and
different criteria by which to report their subjective experiences, and those experiences might have little to do with
actual food consumption (e.g., Stiglitz et al. 2009; Krueger and Stone 2014; Tandon 2024; etc.). Figure 3 illustrates
that there was little difference in the change in subjective food security questions in disaster-affected regions relative
to the rest of the country, with the magnitude of the difference-in-differences being close to zero and imprecisely
estimated.


                                                           14
and the rest of the country.
       And fourth, the analysis will also decompose the average changes in IPV acceptability and
prevalence across the entire country in ways that illustrate that the changes are likely driven by the
disaster and not other potential explanations. In particular, the analysis will illustrate that provincial
variation in the share of the total population that is rural are likely not driving the results, and that
all results are qualitatively identical when restricting the sample to only individuals living in rural
areas; that the main results are driven by strong changes only in disaster-affected regions and that
there was little change in any comparison region following the disaster;29 and, to the degree that is
possible, the analysis will illustrate that there is no evidence of differential changes in the baseline
results leading up to the disaster.

Section 6. Changes in Attitudes about and Prevalence of Gender-Based Violence

       There was a large reduction in the share of men that reported it was acceptable for a husband to
beat their wife in disaster-affected regions. Figure 5a presents the difference-in-differences estimate
from the baseline empirical specification using the total number of scenarios in which a respondent
thought it was acceptable for a husband to beat their wife as the dependent variable. The first
estimate is from a sparse specification with no regional fixed effects and no control variables; the
second estimate adds regional fixed effects; and the third estimate adds both regional fixed effects
and control variables.
       In all instances, there was a substantially larger decline in the number of scenarios in which IPV
was acceptable in disaster-affected regions than in the rest of the country. In the most complete
specification that includes regional fixed effects and control variables, the number of scenarios re-
ported as acceptable declined by 0.64 more in disaster-affected regions than in the rest of the country,
which represents a 19.0 percent decrease relative to the pre-disaster average in affected regions. The
estimate was qualitatively identical across specifications that add fixed effects and control variables,
illustrating that the results are likely not being driven by unexplained factors.
       The large decline in the total number of scenarios in which it was acceptable for a husband to beat
their wife was composed of a large decline in affirmative responses to each individual scenario cap-
tured in the survey. The share of the male population that reported IPV to be acceptable declined
by between 13 and 35 percentage points more in disaster-affected regions than in the rest of the
  29
    Although the acceptability of IPV is higher in rural areas than in urban areas, summary statistics presented in
Section 3 illustrate that the magnitude of the difference is not too large. And thus, despite the progressively more
rural sample over time, we still find little difference in the acceptability of IPV over the course of the survey because
the share of the population that is rural was already very high in the initial phases of the survey (83 percent).


                                                           15
country, which represents between 25 and 59 percent of the pre-disaster shares in disaster-affected
regions. Each of these instances were precisely estimated, with four out of the seven estimates statis-
tically significant at the 1 percent level, and the remaining three out of seven estimates statistically
significant at the 5 percent level. Furthermore, the baseline results are qualitatively identical when
restricting the sample to only the 5,403 men living in rural areas in Figure 5b.
       Although there was a large decline in the acceptability of IPV by men, Figure 5c illustrates that
there was not a similar change in the acceptability reported by women. The difference-in-differences
estimates for women were small in magnitude, varied in signs between specifications, and none of
the seven measures were statistically significant at conventional significance levels. Rather, the share
of women that reported that IPV was acceptable following the disaster varied between 42 and 53
percent for each of the five scenarios, and was similar in several scenarios to that of men following
the disaster.
       Figure 6 unpacks the changes in the acceptability of IPV following the disaster across the country.
As opposed to only reporting the difference-in-differences estimate in the baseline specification,
Figure 6a reports the changes in each region separately for disaster-affected regions, for the entire
country, and for several other potential comparison regions.30 There was a large change only in
disaster-affected regions, where the there was a decline of approximately 0.86 scenarios in which
IPV was acceptable on average. Alternatively, the change in any comparison region was between
-0.17 and 0.30, with none of them being statistically significant at conventional significance levels.
       Figure 6b illustrates that the baseline difference-in-differences estimate for men is robust to using
any other comparison region in the country and that the magnitude of the estimate is qualitatively
identical across comparison regions. This also includes the estimate comparing the change in the
disaster-affected regions highlighted here to other provinces in which there was a state of emergency
declared but which did not face the same challenges (Enga and Western Provinces). This latter pat-
tern suggests that the severity of the crisis, in terms of psychological trauma and in the humanitarian
situation, is important to the change in the acceptability of IPV.
       There is little evidence that IPV was trending differently in disaster-affected regions prior to the
disaster. Data collection prior to the disaster was primarily collected in two months in disaster-
  30
    The comparison regions include Rest of PNG, which denotes the entire country included in the survey less the
Southern Highlands; Other Regions in State of Emergency, which denotes Enga and Western Provinces, where a state
of emergency was also declared, but in which the humanitarian disaster was much less severe; Least Affected, which
denotes regions aside from provinces that border the four provinces in which a state of emergency was declared; Border
Disaster, which denotes provinces that border the four provinces in which a state of emergency was declared; and
Similar MPI Profile, which denotes provinces that have the most similar multidimensional poverty rates, as measured
by the World Bank’s Multidimensional Poverty Measure, to the provinces in which a state of emergency was declared.


                                                         16
affected areas, and data collection following the disaster was also conducted in two months. The top
panel of Figure 7 re-estimates the baseline specification but estimates the difference-in-differences
in the acceptability of IPV in disaster-affected regions in the first month relative to the second in
the pre-disaster period, and then further estimates the difference-in-differences following the disaster
relative to the last pre-disaster month.31
       The magnitude of the differential change leading up to the disaster was small in magnitude and
imprecisely estimated in the top panel of Figure 7, and the differential change following the disaster
was qualitatively identical to the baseline estimate reported in Figure 5.32 Although data collection
did not occur in each phase of the survey in disaster-affected regions, Figure 7 also shows that there
is little evidence that the acceptability of IPV was changing over the course of the survey in the rest
of the country. The last panel of Figure 7 reports the change in the average number of scenarios
in which IPV was acceptable relative to the first phase, and illustrates that the magnitudes of the
changes were small relative to the change in disaster-affected regions and was imprecisely estimated.33
       Despite the large and precisely estimated decline in the acceptability of IPV among men following
the disaster, the change in the prevalence of IPV reported by women was not nearly as precisely
estimated. Figure 8a illustrates that there was a 17.4 percentage point decline in the share of women
that reported experiencing either emotional or physical IPV in the past year, which would represent
22 percent of the pre-disaster prevalence. This is composed of approximately a 17.4 percent decline
in the share of women who experienced emotional IPV and a 17.3 percent decline in the share of
women who experienced physical IPV.
       However, the estimates are not precisely estimated and none are statistically significant at con-
ventional significance levels. Figures 8b and 8c illustrate a similar pattern, where there was a decline
in every type of IPV captured. Only one of the 13 estimates was statistically significant at the 5
percent level and only two additional estimates were statistically significant at the 10 percent level.
Although more estimates are statistically significant at conventional significance levels than one
might expect by chance, the results are not nearly as strong as for the change in acceptability of
IPV.
       Combined, the large improvement in attitudes about IPV of men is corroborated by the change
  31
      The pre-disaster sample in the rest of the country is restricted only to households surveyed in the same months
as pre-disaster data collection occurred in disaster-affected regions.
   32
      In results not reported, the lack of a change in the pre-period is robust to the time period in which data collection
is restricted to in the rest of the country. For example, the results are qualitatively identical when restricting the
entire sample- disaster-affected regions and the rest of the country- to only the last two months of data collection
prior to the disaster and the first two months of data collection following the disaster.
   33
      The p-value of all the coefficients being jointly equal to zero is 0.260.


                                                            17
being restricted to only disaster-affected regions, by the robustness of the results to any comparison
region, and at least some evidence that the change in attitudes resulted in a decline in IPV. As
discussed in the Introduction, the large decline in the acceptability of IPV is novel and contrasts
with several other contexts in which the change in IPV prevalence following a disaster was unable
to be statistically discernable from no change. These results also illustrate that the severity of the
disaster is important to the baseline empirical pattern, and that other regions in a state of emergency
but less badly affected did not have a similar change.

Section 7. Why Is the Change in IPV Incidents Reported by Women Less Precise Than
the Change in IPV Attitudes of Men?

       We investigate why the change in the prevalence of IPV was less robust and less strong than the
change in the acceptability of IPV reported by men. First, one possibility that could explain why the
change in the prevalence of IPV was imprecisely estimated and potentially small is the possibility
that instances of IPV could be a low-frequency event, and that the potential for IPV decreased even
if the actual prevalence did not change. However, we illustrate that the frequency of the event is
likely not the sole cause of the difference.
       Figure 9a investigates changes in the intensity of IPV among women who experienced IPV in the
past year. As described in the Data Section, for respondents that have ever experienced each of the
13 IPV incidents, the respondent is also given the opportunity to answer whether the event occurred
either often, sometimes, within the past year, or not within the past year. But when we restrict the
sample to the 2,451 women who have experienced IPV in the past year (out of 3,945 total), and
analyze the change in the share that experience any of the IPV events often, the estimates are small
in magnitude, imprecisely estimated, and change signs between specifications.34 Thus, even among
women that experience multiple IPV incidents each year, there was little change in the severity of
the IPV.
       Second, it is possible that the time period covered by the questionnaire- one year- is too long
and might also capture events from before the disaster, which would contribute to noise in the
measure and the less precise estimates. However, this is unlikely to be the case. The surveys in
disaster-affected regions were delayed until November and December of 2018, which already is six
to seven months after the disaster and limits the time period covered that is prior to the disaster.
Additionally illustrating that this is likely not the entire explanation, the food security questions
  34
    The result is the same when estimated with the full sample of women, where the indicator for experiencing IPV
often for the excluded part of the sample is zero.


                                                       18
also have an identical reference period and are able to illustrate a more precise worsening of the food
security situation in Figure 3.
       And third, it is possible that only a subset of women who experience IPV accurately describe
those experiences to enumerators, and that the underreporting is worse than for men who are asked
about the acceptability of a hypothetical man committing acts of IPV. Not admitting to abuse
suffered in the pre-disaster period would result in no change for underreporting households even if
there was a reduction in IPV. This could result in an estimate downward biased to zero and in an
estimate that is less precisely estimated.
       Figure 9b investigates this hypothesis and estimates the baseline empirical specification separately
for women that are potentially more and less likely to report IPV experiences. Specifically, one of the
many reasons that women might not report IPV experiences is the fear of additional abuse if other
household members suspected that they might have admitted to the abuse that was occurring. It
has been shown that women who have more resources and productive capacity can result in a more
credible recourse of being able to leave an abusive household, and thus might fear this repercussion
less than women who are less able to leave the household (e.g., Agarwal 1997; Hesse 2011; etc.).
And reporting patterns in PNG are consistent with the least educated women underreporting IPV
incidents. Women who never attended school report fewer instances of IPV despite the fact that the
least educated women and men are more likely to view instances of IPV being acceptable.35
       Thus, using a school attainment indicator to proxy for productive capacity, which was fixed long
before the earthquake happened for women eligible for the women’s survey, we estimate the baseline
empirical specification separately for women that never attended school and women that attended at
least some primary school or above.36 The results illustrate a substantial difference in the experiences
of the 921 women who had never any school and the 3,024 women who had at least attended some
school. There was a 47-percentage point larger decrease in the share of women who experienced
any IPV in the past year for women who had attended at least some school than in the rest of
the country. The estimate is precise and is nearly three times the baseline estimate for the entire
sample in Figure 8a. However, there was no change in the share for women who never attended any
school relative to the rest of the country (difference-in-differences estimate of 0.1 percentage points).
And all of these patterns extend to both categories of IPV captured by the survey- emotional and
  35
     Estimates of the difference are available from the authors upon request. And this is similar to the lower instances
of IPV reported in rural areas despite IPV being more acceptable in those regions, as reported in Section 4.
  36
     This is not to say that all women who never attended school underreport IPV or that all educated women
accurately report IPV.



                                                          19
physical abuse.
       Further illustrating that these patterns were likely driven by measurement issues and not by men
in the least educated households not changing their likelihood of committing acts of IPV, Figure 9c
re-estimates the same specifications as Figure 9b but uses the IPV attitudes of men as the dependent
variable. The results are in stark contrast to the results reported in Figure 9b. The decline in IPV
attitudes was evident for both men who never attended school and for men that did, but the decline
was even larger for men who never attended any school. One factor that could contribute to the
latter pattern was the fact that the prevalence of poor IPV attitudes was worse among men with
no education in the pre-disaster period and thus had the largest share of men who would be able to
improve their attitudes about IPV.
       Table 3 more concretely estimates the difference in the baseline empirical patterns based on
whether the individual ever attended any school. Specifically, the table re-estimates the baseline
specification, but interacts the variable of interest (P ostitr ∗ Af f ectedir ) with an indicator for whether
the individual never attended any school.37 The coefficient on the triple interaction gives an estimate
of how much larger the difference-in-differences estimate was for individuals that never attended
school. A positive estimate is consistent with the decline in IPV prevalence and attitudes being
smaller for the least educated individuals, and a negative estimate is consistent with the decline
being even larger.
       The estimates in Table 3 corroborate the patterns illustrated in Figure 9. The decline in the
prevalence of IPV reported by the least educated women is less strong than that of all other women.
The estimates are positive for all three measures, strong in magnitude, and two out of the three
measures are statistically significant at the 5 percent level. Consistent with Figure 9c, the decline
in the acceptability of IPV among men is even larger for the men that have the least education. All
point estimates of the seven measures are negative, five of the seven have a strong magnitude, and
two of the seven measures are statistically significant at conventional significance levels (one at the
5 percent level and one at the 10 percent level).
       Combined, the results are consistent with there being a large drop in actual IPV prevalence
following the disaster and consistent with significant underreporting of IPV incidents in the pre-
disaster period by the women who are potentially least likely to admit to experiencing IPV. The
large decline in the IPV prevalence substantiates the large improvement of IPV attitudes of men
estimated in Section 5. However, there could also be other subsets of the population that are not
  37
       All lower-order terms, regional fixed effects, and control variables are also included in the specification.


                                                              20
fully admitting to IPV experiences and that the empirical results are capturing only a subset of the
changes that are actually occurring.

Section 8. Mechanisms for the Decline in IPV

   We also investigate a potential mechanism for the observed changes. In the Introduction, we
noted that disasters have the potential to substantially change the tasks needed to be performed
by the entire household, which has the potential to change the roles and responsibilities of each
household member. Thus, although there are reasons that negative economic shocks and increased
stress might worsen IPV attitudes and prevalence, these added roles and responsibilities that women
might support might actually improve their importance and standing in the household.
   Although most of the intrahousehold roles and responsibilities are difficult to observe in a house-
hold survey, we do find evidence of an expanded role for women inside the household. The survey
asks women who in the household is primarily responsible for four different decisions- how to spend a
husband’s income, health decisions for the respondent, large household purchases, and for decisions
about visiting other family members. For all four decisions, we re-estimate the baseline specification
using indicators equaling one if the husband alone was responsible for the decision and indicators
equaling one if the wife was alone responsible for the decision.
   Figure 10 illustrates that there was a reduction in the instances in which the husband was solely
responsible for each decision. The larger decline in the shares in disaster-affected regions varied
between 3 and 16 percentage points, which represents between 9 and 34 percent of the pre-disaster
shares in disaster-affected regions; and two of the four estimates are precisely estimated and are
statistically significant at conventional significance levels (p-values of 0.035 and 0.051). Furthermore,
the estimates for men being solely responsible for decisions are accompanied by an increase in the
share of women who are alone responsible for each decision.
   Although we are unable to list all the potential reasons for why IPV attitudes and IPV prevalence
declined following the disaster, the results suggest a key pathway- an expanded role for women in
a very tumultuous setting. More fully uncovering what these added roles and responsibilities are
inside the household in disaster settings, and comparing these added roles to other potential factors
that could aggravate IPV, could more fully explain some of these patterns and the instances in which
the prevalence of IPV might actually decline in disaster settings.

9. Conclusion

   This analysis illustrates that attitudes about IPV substantially improved following a large disaster

                                                   21
in PNG that precipitated a humanitarian disaster and a forced displacement crisis. The analysis
also demonstrates that the change in reported prevalence of IPV was not as substantial and was
less precisely estimated, and illustrates that one likely explanation is the underreporting of IPV
incidents in modules that directly ask women about the incidents that they experience. And lastly,
the analysis explores potential mechanisms that explain the decline in attitudes about IPV for the
entire population of men and the decline in reported prevalence among a subset of the population
of women.
   However, there are many avenues that the analysis is unable to explore given data limitations.
First, as discussed above, we are not fully able to identify how the roles and responsibilities of women
change in the household, and thus we cannot fully identify all the ways in which the disaster might
have changed the intrahousehold dynamics that might result in changes to IPV. More detailed survey
instruments and combining quantitative household data with some more qualitative discussions could
better identify how disasters might change household dynamics. Although more subjective in nature,
more detailed questionnaires that inquire why women think their husbands might commit acts of
intimate partner violence could help better explain these potential pathways and understand IPV
more generally.
   Second, these results are relatively short-term results and the data coverage following the disaster
limits our ability to more fully investigate the full impacts of the disaster. Whether these changes
persist more than seven months following the disaster is unable to be explored here; and we are
unable to explore whether data collected even more closely in time to the disaster or whether data
collected from regions even more strongly impacted than those analyzed here might have resulted in
similar or even stronger findings. But the continuation of a survey in the midst of such disasters are
relatively uncommon and these results add additional information relative to other settings where
data might be collected long after the disaster ends.
   And lastly, these results only investigate how disasters might change IPV. Although some of
the results might have implications for what might improve the IPV situation in the country more
generally, such as interventions that necessarily expand the roles and responsibilities for women
inside the household, the results do not explain how such a dire IPV situation evolved and how to
more fully reduce and eliminate IPV in the country.




                                                  22
References

Abiona, O., M. Koppensteiner. 2018. ”The Impact of Household Shocks on Domestic Violence:
Evidence from Tanzania.” IZA Discussion Paper No. 11992, IZA Institute of Labor Economics,
Bonn, Germany.
Agarwal, B. 1997. “Bargaining and Gender Relations: Within and Beyond the Household.” Feminist
Economics 3(1), 1-51.
Baker, A., D. Larcker, and C. Wang. 2022. ”How Much Should We Trust Staggered Difference-in-
Differences Estimates?” Journal of Financial Economics 144 (2), 370-395.
Bandiera, O., N. Buehren, R. Burgess, M. Goldstein, S. Gulesci, I. Rasul, M. Sulaiman. 2020.
”Women’s Empowerment in Action: Evidence from a Randomized Control Trial in Africa.” American
Economic Journal: Applied Economics 12 (1), 210-259.
Baxi, P., D. Naidoo, and S. Tandon. 2024. ”How Do Unfair Shares Compare? Benchmarking
Well-Being in Papua New Guinea.” Working Paper, the World Bank, Washington, D.C., USA.
Bell, K., and A. Naugle. 2007. ”Effects of Social Desirability on Students’ Self-Reporting of Partner
Abuse Perpetration and Victimization.” Violence and Victims 22 (2).
Blair, G., and K. Imai. 2012. ”Statistical Analysis of List Experiments.” Political Analysis 20,
47-77.
Blair, G., K. Imai,, and Y. Zhou. 2015. “Design and Analysis of Randomized Response Technique.”
Journal of American Statistical Association 110 (511), 1304-1319.
Blattman, C., J. Jamison, T. Koroknay-Palicz, K. Rodrigues, and M. Sheridan. 2016. ”Measuring
the Measurement Error: a Method to Qualitatively Validate Survey Data.” Journal of Development
Economics 120, 99-112.
Borusyak, K., X. Jaravel, and J. Spiess. Forthcoming. ”Revisiting Event Study Designs: Robust
and Efficient Estimation.” Review of Economic Studies, forthcoming.
Bulte, E. and R. Lensink. 2019. ”Women’s empowerment and domestic abuse: Experimental
evidence from Vietnam.” European Economic Review 115, 172-191.
Calloway, B. and P. Sant’Anna. 2020. ”Difference-in-Differences with Multiple Time Periods.”
Journal of Econometrics 225 (2), 200-230.
Cools S., M. Flato, and A. Kotsadam. 2020. ”Rainfall Shocks and Intimate Partner Violence in
Sub-Saharan Africa.” Journal of Peace Research 57 (3), 377-390 J. Peace Res., 57 (3) (2020), pp.
377-390
Cooper M., A. Sandler, S. Vitellozzi, Y. Lee, G. Seymour, B. Haile, and C. Azzari. 2021. ”Re-
Examining the Effects of Drought on Intimate-Partner Violence.” PLoS One 16e0254346.
Council of Europe. 2019. ”Gender Matters: A Manual on Addressing Gender-Based Violence
Affecting Young People.” Report, Council of Europe, Strasburg, France.
Cullen, C. 2023. “Method Matters: The Underreporting of Intimate Partner Violence.” The World
Bank Economic Review 37 (1), 49–73.
Cullen, C. A. Alik-Lagrange, M. Ngatia, and J. Vaillant. 2024. ”The Unintended Impacts of an
Intimate Partner Violence Prevention Program: Experimental Evidence from Rwanda.” Working
Paper, the World Bank, Washington, D.C., USA.
Davidson, H. 2018. ”Papua New Guinea Earthquake: Death Toll Rises as Disease Threat Grows.”


                                                 23
Report, The Guardian, London, United Kingdom. ”https : www.theguardian.comworld2018mar15papua−
new − guinea − earthquake − death − toll − rises − disease − threat − grows”
Devries, K., J. Mak, C. Garcia-Moreno, M. Petzold, J. Child, G. Falder, S. Lim, L. Bacchus, R.
Engell, L. Rosenfeld, C. Pallitto, T. Vos, N. Abrahams, and C. Watts. 2013. ”The Global Prevalence
of Intimate Partner Violence Against Women.” Science 340 (6140), 1527-1528.
DFAT. 2018. ”Highlands Earthquake Factsheet.” Report, Department of Foreign Affairs and Trade,
Government of Australia, Canberra, Australia. ”https:www.dfat.gov.ausitesdefaultfilespng-earthquake-
factsheet.pdf.”
DHS. 2019. ”2016-2018 Demographic and Health Survey Report.” Report, Demographic and Health
Survey, USAID.
Dhar, D., T. Jain, and S. Jayachandran. 2022. ”Reshaping Adolescents’ Gender Attitudes: Evidence
from a School-Based Experiment in India.” American Economic Review 112 (3), 899-927.
Diaz, J., and V. Saldarriaga. 2023. ”‘A Drop of Love? Rainfall Shocks and Spousal Abuse: Evidence
from Rural Peru.” Journal of Health Economics 89, 102739.
Doss, C. 2013. ”Intrahousehold Bargaining and Resource Allocation in Developing Countries.”
World Bank Research Observer 28 (1), 52-78.
Duflo, E. 2012. “Women Empowerment and Economic Development.” Journal of Economic Litera-
ture 50 (4), 1051–79.
Education Cannot Wait. 2019. ”After the Quake.” Report, Education Cannot Wait, New York,
USA. ”https : relief web.intreportpapua − new − guineaaf ter − quake”
Elsberg, M., and L. Heise. 2005. ”Researching Violence Against Women: A Practical Guide for Re-
searchers and Activists.” Report, World Health Organization, Geneva. ”https : iris.who.intbitstreamhandle1
1.”
Epstein, A.,E. Bendavid, D. Nash, E. Charlebois, S. Weiser. 2020. ”Drought and Intimate Partner
Violence Towards Women in 19 Countries in Sub-Saharan Africa During 2011–2018: A Population-
Based Study” PLoS Medicine 17 (3), Article e1003064.
Eves, R. 2019. ”Full Price, Full Body: Norms, Brideprice and Intimate Partner Violence in the
Highlands of Papua New Guinea.” Culture, Health & Sexuality 21 (12), 1367-1380.
Eves, R., G. Kouro S. Simiha, and I. Subalik. 2018. ”Do No Harm Research: Papua New Guinea.”
Report, The Australian National University, Canberra, Australia. ”https:dpa.bellschool.anu.edu.ausitesdefaul
03do no harm png low res.pdf.”
Fagen J., W. Sorenson, and P. Anderson. 2011. ”Why Not the University of New Orleans? Social
Disorganization and Sexual Violence Among Internally Displaced Women of Hurricane Katrina.”
Journal of Community Health 36, 721-727.
Fisher, R. 1993. “Social Desirability Bias and the Validity of Indirect Questioning.” Journal of
Consumer Research 20 (2), 303-315.
Frasier P., L. Belton, E. Hooten, M. Campbell, B. DeVellis, S. Benedict, C. Carrillo, P. Gonzalez, K.
Kelsey, and A. Meier. 2004. ”Disaster Down East: Using Participatory Action Research to Explore
Intimate Partner Violence in Eastern North Carolina.” Health Education Behavior 31, 69S-84S.
Gennetian L. 2003. ”Welfare Policies and Domestic Abuse Among Single Mothers: Experimental
Evidence from Minnesota.” Violence Against Women 9, 1171-1190.
Gibson-Davis C., K. Magnuson, L. Gennetian, G. Duncan. 2005. ”Employment and the Risk of

                                                 24
Domestic Abuse Among Low-Income Women.” Journal of Marriage and Family 67, 1149-1168.
Goldin, C., and L. Katz. 2002. “The Power of the Pill: Oral Contraceptives and Women’s Career
and Marriage Decisions.” Journal of Political Economy 110 (4): 730–70.
Glynn, A. 2013. “What Can We Learn with Statistical Truth Serum? Design and Analysis of the
List Experiment.” Public Opinion Quarterly 77, 159-172.
Goodman-Bacon, A. 2021. ”Difference-in-Differences With Variation in Treatment Timing.” Journal
of Econometrics 225 (2), 254-277.
Government of PNG. 2016. ”Papua New Guinea National Strategy to Prevent and Respond to Gen-
der Based Violence 2016-2025.” Report, Government of Papua New Guinea, Port Moresby, Papua
New Guinea. ”https:femilipng.orgwp-content uploadsNational-Strategy-to-Prevent-and-Respond-to-
IPV.pdf.”
Government of Papua New Guinea. 2022. ”Special Parliamentary Committee on Gender-Based Vi-
olence Report to Parliament.” Report, Government of Papua New Guinea, Port Moresby, Papua New
Guinea. ”https:www.unitedforequalitypng.com filesugd7f0c06 50372501b8b7406b8482e9560dbd21b4.pdf?ind
Government of PNG, OCHA, and UNCT. 2018. “Highlands Earthquake Situation Report No.
10.” Report, Port Moresby, Papua New Guinea. ”https : relief web.intreportpapua − new −
guineapapua−new −guinea−highlands−earthquake−situation−report−no−10−28−may −2018”
Gregson, S., T. Zhuwau, J. Ndlovu, and C. Nyamukapa. 2002. ”Methods to Reduce Social De-
sirability Bias in Sex Surveys in Low-Development Settings: Experience in Zimbabwe.” Sexually
Transmitted Diseases 29 (10), 568-575.
Gupta J., K. Falb, H. Lehmann, D. Kpebo, Z. Xuan, M. Hossain, and J. Annan. 2013. ”Gender
Norms and Economic Empowerment Intervention to Reduce Intimate Partner Violence Against
Women in Rural Cˆ ote d’Ivoire: A Randomized Controlled Pilot Study.” BMC International Health
and Human Rights 13(1), Article 46.
Hallegatte, S. 2014. ”Economic Resilience : Definition and Measurement.” Policy Research Work-
ing Paper 6852, the World Bank, Washington, DC. ”http : hdl.handle.net1098618341License :
CCBY 3.0IGO”
Haushofer, J., and J. Shapiro. 2016 ”The Short-Term Impact of Unconditional Cash Transfers to
the Poor: Experimental Evidence from Kenya.” Quarterly Journal of Economics 131 (4), 1973-2042.
Heise, L. 2011. ”What Works to Prevent Partner Violence? An Evidence Overview.” London:
STRIVE Research Consortium.
Hidrobo M., and L. Fernald. 2013. ”Cash Transfers and Domestic Violence.” Journal of Health
Economics 32, 304-319.
Hidrobo M., A. Peterman, and L. Heise. 2016. ”The Effect of Cash, Vouchers, and Food Transfers
on Intimate Partner Violence: Evidence from a Randomized Experiment in Northern Ecuador.”
American Economic Journal: Applied Economics 8, 284-303.
ICRC. 2018. ”Papua New Guinea: Earthquake Response Operation Update 2018 (March to May
2018).” Report, International Committee of the Red Cross, Geneva, Switzerland. ”https:reliefweb.intreportpap
new-guineapapua-new-guinea-earthquake-response-operation-update-2018-march-may-2018.”
International Organization of Migration. 2018. ”Displacement Tracking Matrix Report for Earthquake-
Affected Communities Hela, Southern Highlands and Western Provinces.” Report, International Or-
ganization of Migration, Port Moresby, Papua New Guinea. ”https : relief web.intreportpapua −
new − guineadisplacement − tracking − matrix − report − earthquake − af f ected − communities −


                                                25
hela”
Jayachandran, S. 2015. “The Roots of Gender Inequality in Developing Countries.” Annual Review
of Economics 7, 63–88.
J-PAL. 2022. ”The impacts of economic interventions on intimate partner violence.” J-PAL Policy
Insights, Abdul Latif Jameel Poverty Action Lab, Cambridge, Massachusetts.
    Abdul Latif Jameel Poverty Action Lab (J-PAL). 2022. ”The impacts of economic interventions
on intimate partner violence.” J-PAL Policy Insights. Last modified July 2022.
Kim J., C. Watts, J. Hargreaves, L. Ndhlovu, G. Phetla, L. Morison, P. Pronyk. 2007. ”Understand-
ing the Impact of a Microfinance-Based Intervention on Women’s Empowerment and the Reduction
of Intimate Partner Violence in South Africa. American Journal of Public Health 97, 1794-1802.
Kramon E, and K. Weghorst. 2019. ”(Mis) Measuring sensitive attitudes with the list experiment.”
Public Opinion Quarterly 83,236–263.
Krueger, A., and A. Stone. 2014. Measuring Subjective Well-being: Progress and Challenges.
Science 346, 42-43.
Macmillan R. and R. Gartner. 1999. ”When She Brings Home the Bacon: Labor-Force Participation
and the Risk of Spousal Violence Against Women.” Journal of Marriage and Family 61, 947-958.
Mambon, K. 2023. ”Rethinking Perceptions of Spousal Violence in PNG.” Report, Devpolicy, Devel-
opment Policy Center, Australian National University, Canberra, Australia. ”https:devpolicy.orgrethinking-
perceptions-of-spousal-violence-in-png-20230222.”
Organization for Economic Co-operation and Development. 2013. OECD guidelines on measuring
subjective well-being. Report, OECD, Paris.
Panda, P., and B. Agarwal. 2005. ”Marital Violence, Human Development and Women’s Property
Status in India.” World Development 33 (5), 823-850.
Rooney, N., M. Forsyth, J. Goa, D. Lawihin, and D. Kuir-Ayius. 2022. ”Thinking Incrementally
About Policy Interventions on Intimate Partner Violence in Papua New Guinea: Understanding
”Popcorn” and ”blanket.” Culture, Health & Sexuality DOI: 10.1080/13691058.2022.2103736.
Rosenfeld, B., K. Imai, and J. Shapiro. 2015. “An Empirical Validation Study of Popular Survey
Methodologies for Sensitive Questions.” American Journal of Political Science 60 (3), 783-802.
Sardinha, L., M. Mahue-Giroux, H. Stockl, S. Meyer, and C. Garcia-Moreno. 2022. ”Global, Re-
gional, and National Prevalence Estimates of Physical or Sexual, or Both, Intimate Partner Violence
Against Women in 2018.” The Lancet 399 (10327), 803-813.
Smith, A. 2024. ”The Paradox of Empowerment: Gender Norms and Intimate Partner Violence in
PNG.” Report, Devpolicy, Canberra, Australia. ”https:devpolicy.orgthe-paradox-of-empowerment-
gender-norms-and-intimate-partner-violence-in-png-20240222.”
Stark, L., and A. Ager. 2011. ”A Systematic Review of Prevalence Studies of Gender-Based Violence
in Complex Emergencies.” Trauma Violence Abuse 12, 127-134.
Stiglitz J, Sen A, Fitoussi J. 2009. Report of the Commission on the Measurement of Economic Per-
formance and Social Progress. (accessed January 2023) http : www.stiglitz −sen−f itoussi.f rdocumentsrappo


Sun, L. and S. Abraham. 2021. ”Estimating Dynamic Treatment Effects in Event Studies with
Heterogeneous Treatment Effects.” Journal of Econometrics 225 (2), 175-199.
Tandon, S. 2024. ”How Difficult is it to Interpret Subjective Well-being Questions During Crises?

                                                26
Evidence from the Onset of Conflict in Yemen.” Oxford Economic Papers 76 (3).
Tandon, S., and T. Vishwanath. 2022. ”Capturing Sensitive Information from Difficult-to-Reach
Populations: Evidence from a Novel Internet-Based Survey in Yemen.” World Bank Policy Research
Working Paper WPS10179, the World Bank, Washington, D.C.
Tankard, M., and R. Iyengar. 2018. ”Economic Policies and Intimate Partner Violence Prevention:
Emerging Complexities in the Literature.” Journal of Interpersonal Violence 33 (21), 3367-3387.
UNCT. 2018. “Papua New Guinea: Earthquakes- February 2018.” Report, UNCT, Port Moresby,
Papua New Guinea. ”https : relief web.intdisastereq − 2018 − 000020 − png.”
van Daalen, K., S. Kallesoe, F. Davey, S. Dada, L. Jung, L. Singh, R. Issa, C. Emilian, I. Kuhn, I.
Keygnaert, and M. Nilsson. 2022. ”Extreme Events and Gender-Based Violence: A Mixed-Methods
Systematic Review.” The Lancet Planetary Health 6 (6) E504-E523.
Weitzman, A. and J. Behrman. 2016. ”Disaster, Disruption to Family Life, and Intimate Partner
Violence: The Case of the 2010 Earthquake in Haiti.” Sociological Science 3, 167-189.
WHO. 2018. “Papua New Guinea Earthquake Situation Report No. 2.” Report, World Health Orga-
nization, Geneva, Switzerland. ”https:www.who.intdocsdefault-sourcewpro—documentsemergencypng-
earthquake20180328-png-quake-sitrep2.pdf?sfvrsn=48ace0bf 0.”
WHO. 2021. ”Violence Against Women Prevalence Estimates, 2018.” Report, World Health Orga-
nization, Geneva. ”https : www.who.intpublicationsiitem9789240022256”
Zimmerman K. ”Plates in a Basket Will Rattle: Domestic Violence in Cambodia, a Summary.”
Phnom Penh, Cambodia: Project Against Domestic Violence; 1995.




                                                27
Table 1. Changes in Difficult-to-Adjust Household Characteristics Following the Disaster- Men’s Survey

                                                                  Change Following                    Change Following
                                                              Earthquake in Disaster-           Earthquake in the Rest of
                                                                  Affected Regions                      the Country       Difference in Differences
                                                                  _______________                    ________________        ________________
                                                               Mean Standard Error                 Mean Standard Error Mean Standard Error
Indicator- Never Attended Primary                              0.108        (0.085)              0.083***       (0.030)    0.025         (0.089)
Indicator- At Least Started Primary                            0.061        (0.077)              0.102***       (0.031)    -0.041        (0.082)
Indicator- At Least Started Secondary                       -0.176***       (0.057)             -0.145***       (0.023)    -0.031        (0.060)
Indicator- At Least Started Above Secondary                    0.007        (0.032)             -0.040***       (0.012)    0.048         (0.033)
Years Schooling                                                -1.327       (0.797)             -1.638***       (0.321)    0.311         (0.848)
Indicator- Incomplete Primary (highest)                         0.041       (0.066)              0.105***       (0.030)    -0.064        (0.071)
Indicator- Completed Primary (highest)                         0.020        (0.030)                -0.002       (0.016)    0.023         (0.034)
Indicator- Incomplete Secondary (highest)                   -0.164***       (0.048)             -0.098***       (0.017)    -0.066        (0.051)
Indicator- Completed Secondary (highest)                       -0.012       (0.043)             -0.047***       (0.014)    0.035         (0.045)
Age of Head of Household                                       -2.817       (1.948)              -1.814**       (0.719)    -1.003        (2.050)
Indicator- Male-Headed Household                               0.060        (0.050)                0.025*       (0.014)    0.035         (0.051)
Indicator- Illiterate                                          0.106        (0.088)              0.100***       (0.030)    0.006         (0.092)
Indicator- Partially Literate                                  0.045        (0.057)              0.118***       (0.023)    -0.073        (0.060)
Indicator- Can Read Sentence                                 -0.206**       (0.082)             -0.224***       (0.037)    0.018         (0.089)
Indicator- Other Literacy                                      0.056        (0.040)                 0.007       (0.005)    0.049         (0.040)
Number of Children Born                                        0.060        (0.348)               0.308**       (0.124)    -0.248        (0.365)
Indicator- Rural                                               0.133        (0.089)              0.143***       (0.025)    -0.009        (0.091)
Share of Members Eligible Men's Survey                         0.040        (0.032)                -0.013       (0.018)    0.053         (0.037)
Indicator- Head of Household                                  0.163*        (0.097)              0.114***       (0.025)    0.049         (0.099)
Notes: Figures report the changes in difficult-to-adjust individual characteristics in the Men's Survey following the disaster separately for disaster-affected regions
and the rest of the country. The table further reports the difference between the first and third columns, and represents how much more did the variable increase in
disaster-affected regions than in the rest of the country. There are 281 observations from disaster-affected regions- 109 from before the disaster and 172 after;
and there are 7051 observations for the rest of the country- 5,558 from before the disaster and 1,493 after. Standard errors are clustered at the PSU level and
reported for each estimate. *** denotes statistical significance at the one percent level; ** denotes statistical significance at the 5 percent level; and * denotes
statistical significance at the 10 percent level.




                                                                                 28
Table 2. Changes in Difficult-to-Adjust Household Characteristics Following the Disaster- Women’s
Survey

                                                                  Change Following                   Change Following
                                                              Earthquake in Disaster-           Earthquake in the Rest of
                                                                  Affected Regions                      the Country       Difference in Differences
                                                                  _______________                   ________________         ________________
                                                               Mean Standard Error                 Mean Standard Error Mean Standard Error
Indicator- Never Attended Primary                            0.258***       (0.089)              0.122***      (0.043)     0.136         (0.097)
Indicator- At Least Started Primary                            -0.123       (0.100)                0.022       (0.046)     -0.145        (0.108)
Indicator- At Least Started Secondary                         -0.095*       (0.056)             -0.099***      (0.025)     0.004         (0.060)
Indicator- At Least Started Above Secondary                    -0.041       (0.036)             -0.045***      (0.017)     0.005         (0.039)
Years Schooling                                                 0.377       (2.558)             -1.674***      (0.396)     2.051         (2.550)
Indicator- Incomplete Primary (highest)                        -0.048       (0.099)                0.059       (0.038)     -0.107        (0.104)
Indicator- Completed Primary (highest)                        -0.074*       (0.042)              -0.037**      (0.017)     -0.038        (0.045)
Indicator- Incomplete Secondary (highest)                     -0.104*       (0.055)             -0.077***      (0.021)     -0.026        (0.058)
Indicator- Completed Secondary (highest)                        0.009       (0.009)               -0.021*      (0.012)    0.030**        (0.015)
Age of Head of Household                                       -0.021       (3.130)             -2.261***      (0.665)     2.241         (3.153)
Indicator- Male-Headed Household                               -0.037       (0.079)                0.029       (0.025)     -0.067        (0.081)
Indicator- Illiterate                                        0.300***       (0.081)              0.128***      (0.040)     0.172*        (0.090)
Indicator- Partially Literate                                  -0.094       (0.075)              0.051**       (0.025)    -0.144*        (0.078)
Indicator- Can Read Sentence                                 -0.210**       (0.096)             -0.179***      (0.039)     -0.030        (0.102)
Indicator- Other Literacy                                       0.004       (0.004)                0.001       (0.005)     0.002         (0.006)
Number of Children Born                                         0.265       (0.436)              0.469***      (0.171)     -0.204        (0.462)
Indicator- Rural                                                0.069       (0.055)              0.123***      (0.022)     -0.055        (0.058)
Share of Members Eligible Women's Survey                       -0.009       (0.031)             -0.019***      (0.007)     0.010         (0.031)
Indicator- Head of Household                                    0.052       (0.071)                0.003       (0.020)     0.049         (0.073)
Notes: Figures report the changes in difficult-to-adjust individual characteristics in the Women's Survey following the disaster separately for disaster-affected
regions and the rest of the country. The table further reports the difference between the first and third columns, and represents how much more did the variable
increase in disaster-affected regions than in the rest of the country. There are 170 observations from disaster-affected regions- 61 from before the disaster and
109 after; and there are 3,785 observations for the rest of the country- 2,863 from before the disaster and 922 after. Standard errors are clustered at the PSU level
and reported for each estimate. *** denotes statistical significance at the one percent level; ** denotes statistical significance at the 5 percent level; and *
denotes statistical significance at the 10 percent level.




                                                                                 29
Table 3. Differences in the Baseline Estimates by Education Level

                                    Womens' Experience with GBV                       Acceptability of Husband to Beat Wife- Men's Responses
                                     _______________________                      ________________________________________________________
                                       (1)       (2)                (3)         (4)                 (5)          (6)          (7)         (8)         (9)          (10)
                                             Experience         Experience
                                 Experience any                 any                             Go Out
                                 Any GBV Emotional              Phsyical   Total                Without          Argue                                        Justified
                                 in the Past GBV in the         GBV in the Affirmative          Telling Neglect with     Refuse Burn                          in All
VARIABLES                        Year        Past Year          Past Year Responses             Husband Children Husband Sex    Food                          Scenarios

Affected x Post x
Indicator for No
Education                          0.529**         0.486**          0.311          -1.270*        0.031       -0.015        -0.185     -0.245 -0.271           -0.342**
                                   (0.228)         (0.219)         (0.233)         (0.670)       (0.137)      (0.147)      (0.195)     (0.150) (0.187)          (0.145)

Observations                         3,912          3,909           3,910           7,203         7,203        7,203        7,203       7,203       7,203         7,203
R-squared                            0.045          0.032           0.041           0.109         0.043        0.038        0.037       0.040       0.045         0.047
Notes: The table estimates the difference in the baseline difference-in-differences estimate for key variables based on whether the respondent ever attended any school.
Specifically, each dependent variable is regressed on a term that intereacts the baseline difference-in-differences estimate with an indicator equaling one if the
respondent never attended any school, and the specifications include all lower-order terms. Columns (1)-(3) uses indicators equaling one if the women experienced any
form of GBV in the past year as the dependent variables, and columns (4)-(10) indicators equaling one if the male respondent thought it was acceptable for a husband to
beat their wife as the dependent variable. All specifications include control variables and regional fixed effects. Control variables include the number of children born in
the household, the share of household members that were eligible for the women’s survey, the years of total schooling, the age of the household head, and indicators for
the last level of schooling obtained (partially completed primary, completed primary, partially completed secondary completed secondary, or higher than secondary), for
whether the respondent was the head of the household, for the respondents literacy level (illiterate, partially literate, can read a sentence out loud), and for whether the
head of the household was male. Standard errors are clustered at the PSU level and the associated 95 percent confidence intervals for each estimate are reported.




                                                                                    30
Figure 1. Gender-Based Violence Summary Statistics

 1a. Share of Men Who Reported it is Acceptable for a Husband to Beat His Wife
  0.70
  0.60
  0.50
  0.40
  0.30
  0.20
  0.10
  0.00
           Go Out Without        Neglect Children      Argue with          Refuse Sex           Burn Food          Justified in All
           Telling Husband                              Husband                                                      Scenarios

 1b. Share of Women Who Experienced Physical Intimate Partner Violence in the Past Year
  0.70
  0.60
  0.50
  0.40
  0.30
  0.20
  0.10
  0.00
             Push         Slap      Twist Arm    Punch      Beat Up       Choke    Attack with Force     Other     Other
                                                                                    Weapon Intercourse Forced Sex Threat
                                                                                                          Act    Related to
                                                                                                                  Sex Acts

 1c. Share of Women Who Experienced Emotional Gender-Based Violence in the Past Year
  0.70
  0.60
  0.50
  0.40
  0.30
  0.20
  0.10
  0.00
           Husband Humiliates Wife- Last Year Husband Threatened Wife- Last Year             Husband Insulted Wife- Last Year
 Notes: Estimates using the 2016-2018 Demographic and Health Survey. The attitudes about gender-based violence (GBV) are obtained
 from the Men’s Questionnaire and the prevalence of physical violence experienced by women is obtained from the Women’s
 Questionnaire. Standard errors clustered at the PSU level are reported for each estimate.




                                                                 31
Figure 2. Increasing Share of the Sample that is Rural Over the Course of the Survey and Implications
for Gender-Based Violence Measurement

 2a. Rural Share Before and After Earthquake
  1.20
  1.00
  0.80
  0.60
  0.40
  0.20
  0.00
                                  Men's Survey                                              Women's Survey

                                                           Phases 1-3    Phase 4

 2b. Rural-Urban Differences in the Share of Men Who Reported it is Acceptable for a Husband to
 Beat His Wife
  0.70
  0.60
  0.50
  0.40
  0.30
  0.20
  0.10
  0.00
            Justified in All     Go Out Without     Neglect Children       Argue with           Refuse Sex          Burn Food
              Scenarios          Telling Husband                            Husband

                                                              Urban     Rural

 2c. Rural-Urban Differences in the Share of Women who Experienced Intimate Partner Violence in
 the Past Year
  0.60
  0.50
  0.40
  0.30
  0.20
  0.10
  0.00
            Punch         Slap      Twist Arm      Punch      Beat Up     Choke     Attack with Force     Other     Other
                                                                                     Weapon Intercourse Forced Sex Threat
                                                                                                           Act    Related to
                                                                                                                   Sex Acts

                                                              Urban      Rural

 Notes: Estimates using the 2016-2018 Demographic and Health Survey. The attitudes about gender-based violence (GBV) are obtained
 from the Men’s Questionnaire and the prevalence of physical violence experienced by women is obtained from the Women’s
 Questionnaire. Standard errors clustered at the PSU level are reported for each estimate.




                                                                  32
Figure 3. How Much More Food Access Worsened in Disaster-Affected Regions Relative to the Rest of
the Country

 3a. Subjective Questions on Food Security
      0.35


      0.25


      0.15


      0.05


     -0.05           Worried                Non-Healthy               Few Foods           Ate Less Than Should Go to Bed Hungry



     -0.15


     -0.25

 3b. Objective Questions on Food Security
   0.35


   0.25


   0.15


   0.05


                        Skipped Meals                              Ran Out of Food                           No Food Whole Day
   -0.05


   -0.15


   -0.25
 Notes: Figures report the difference-in-difference estimate from the baseline specification using indicators of poor food access as the
 dependent variable. Specifically, each estimate represents how much more did the share that reported poor food access (responded yes)
 increase in the worst-affected regions by the disaster relative to the change in the rest of the country. Standard errors are clustered at the
 PSU level and the associated 95 percent confidence intervals for each estimate are reported.




                                                                       33
Figure 4. Differential Non-Response for Questions on Intimate Partner Violence Experienced

 4a. Emotional Intimate Partner Violence

   0.09

   0.07

   0.05

   0.03

   0.01

  -0.01      Husband Humiliated Respondent             Husband Threatened Respondent               Husband Insulted Respondent

  -0.03

  -0.05

 4b. Physical Intimate Partner Violence
   0.10

   0.08

   0.06

   0.04

   0.02

   0.00
             Pushed       Slapped      Twisted     Punched      Beat Up       Choked      Attacked  Forced     Other     Other
  -0.02                                 Arm                                                 with Intercourse Forced Sex Threat
                                                                                          Weapon                Act    Related to
  -0.04                                                                                                                 Sex Acts

  -0.06
 Notes: The figures report the difference-in-difference estimate from the baseline specification using an indicator for non-response to the
 questions on the intimate partner violence asked to women. All specifications include control variables and regional fixed effects. Control
 variables include the number of children born in the household, the share of household members that were eligible for the women’s survey,
 the years of total schooling, the age of the household head, and indicators for the last level of schooling obtained (partially completed
 primary, completed primary, partially completed secondary completed secondary, or higher than secondary), for whether the respondent
 was the head of the household, for the respondents literacy level (illiterate, partially literate, can read a sentence out loud), and for
 whether the head of the household was male. Standard errors are clustered at the PSU level and the associated 95 percent confidence
 intervals for each estimate are reported.




                                                                    34
Figure 5. How Much Larger the Increase in the Share Who Reported it is Acceptable for a Husband to
Beat His Wife was in Disaster-Affected Regions

 5a. Men- Entire Sample
    0.20
    0.00
                Total        Total       Total             Go Out        Neglect       Argue with Refuse Sex Burn Food Justified in
   -0.20
            Affirmative- Affirmative- Affirmative         Without        Children       Husband                        All Scenarios
   -0.40     No Fixed        Fixed    Responses            Telling
             Effects or Effects, No                       Husband
   -0.60     Controls     Controls
   -0.80
   -1.00
   -1.20
 5b. Men- Only Rural Sample
     0.20
     0.00
              Total        Total       Total     Go Out                Neglect      Argue with Refuse Sex Burn Food Justified in
    -0.20
          Affirmative- Affirmative- Affirmative Without                Children      Husband                            All
    -0.40  No Fixed        Fixed    Responses Telling                                                                Scenarios
    -0.60 Effects or Effects, No                Husband
           Controls     Controls
    -0.80
    -1.00
    -1.20
 5a. Women- Entire Sample

   0.75
   0.55
   0.35
   0.15
   -0.05
                Total        Total       Total             Go Out        Neglect      Argue with Refuse Sex Burn Food Justified in
   -0.25    Affirmative- Affirmative- Affirmative         Without        Children      Husband                        All Scenarios
   -0.45     No Fixed        Fixed    Responses            Telling
             Effects or Effects, No                       Husband
   -0.65      Controls     Controls
   -0.85
 Notes: Figures report the difference-in-difference estimate from the baseline specification using indicators of respondents claiming it is
 acceptable for a husband to beat his wife in certain scenarios as the dependent variable. Specifically, each estimate represents how much
 more did the share that reported it was acceptable (responded yes) increase in the worst-affected regions by the disaster relative to the
 change in the rest of the country. All specifications include control variables and regional fixed effects. Control variables include the
 number of children born in the household, the share of household members that were eligible for the men’s survey, the years of total
 schooling, the age of the household head, and indicators for the last level of schooling obtained (partially completed primary, completed
 primary, partially completed secondary completed secondary, or higher than secondary), for whether the respondent was the head of the
 household, for the respondents literacy level (illiterate, partially literate, can read a sentence out loud), and for whether the head of the
 household was male. Standard errors are clustered at the PSU level and the associated 95 percent confidence intervals for each estimate
 are reported.




                                                                      35
Figure 6. Changes in the Acceptability of Gender-Based Violence for Men

 6a. Simple Change in Total Affirmative Responses in Disaster-Affected and Other Comparison
 Regions
   0.55

   0.35

   0.15

   -0.05
            Disaster Affected       Rest of PNG        Other Regions in      Least Affected       Border Disaster         Similar MPI
   -0.25                                                   State of                                                         Provile
                                                         Emergency
   -0.45

   -0.65

   -0.85
                                                       Comparison
   -1.05                                               Regions

   -1.25

   -1.45

 6b. Difference-in-Differences Estimate for Different Comparison Regions
    0.00
                 Rest of PNG           Other Regions in            Least Affected          Border Disaster         Similar MPI Provile
   -0.20                              State of Emergency
   -0.40

   -0.60

   -0.80

   -1.00

   -1.20

   -1.40

   -1.60

   -1.80

   -2.00
 Notes: The top panel reports the average change in the total number of scenarios in which the respondent thought it was acceptable for a
 husband to beat their wife following the disaster in disaster-affected regions and several comparison regions. The bottom panel reports the
 difference-in-difference estimate from the baseline specification using the total number of scenarios in which a respondent thought it was
 acceptable for a husband to beat their wife following the disaster using different comparison regions. Rest of PNG denotes the entire
 country included in the survey less the Southern Highlands; Other Regions in State of Emergency denotes households in Enga and Western
 Provinces, where a state of emergency was also declared, but in which the humanitarian disaster was much less severe; Least Affected
 denote regions aside from provinces that border the four provinces in which a state of emergency was declared; Border Disaster denote
 provinces that border the four provinces in which a state of emergency was declared; and Similar MPI Profile denote provinces that have
 the most similar multidimensional poverty rates, as measured by the World Bank’s Multidimensional Poverty Measure, to the provinces in
 which a state of emergency was declared. All specifications include control variables and regional fixed effects. Control variables include
 the number of children born in the household, the share of household members that were eligible for the men’s survey, the years of total
 schooling, the age of the household head, and indicators for the last level of schooling obtained (partially completed primary, completed
 primary, partially completed secondary completed secondary, or higher than secondary), for whether the respondent was the head of the
 household, for the respondents literacy level (illiterate, partially literate, can read a sentence out loud), and for whether the head of the
 household was male. Standard errors are clustered at the PSU level and the associated 95 percent confidence intervals for each estimate
 are reported.



                                                                      36
Figure 7. How Much Larger the Increase in the Acceptability of GBV was in Disaster-Affected Regions
than in the Rest of the Country Before the Disaster

 7a. Difference in Differences Estimate of Total Scenarios Acceptable in Disaster-Affected Regions
 Relative to the Rest of the Country
        0.75
        0.50
        0.25
        0.00
       -0.25                           Pre-Period                                                 Post-Period
       -0.50
       -0.75
       -1.00
       -1.25
       -1.50

 7b. Change in Total Scenarios Acceptable by Phase in the Rest of Papua New Guinea Relative to the
 First Phase
        0.75
        0.50
        0.25
        0.00
       -0.25                   Phase 2                                 Phase 3                                 Phase 4
       -0.50
       -0.75
       -1.00
       -1.25
       -1.50
 Notes: The figures report how the acceptability of gender-based violence (GBV) was changing leading up to and following the disaster. In all
 panels, the dependent variable is the total scenarios in which GBV was acceptable. The top panel reports difference-in-difference estimates
 from the baseline specification, but restricts the pre-period in the rest of the country to be the same time period as the pre-disaster data
 was collected in disaster-affected regions; and the bottom panel restricts the sample to the rest of the country and reports changes in the
 average number of scenarios in which GBV was acceptable in each phase of the survey relative to the first phase. All specifications include
 control variables and regional fixed effects. Control variables include the number of children born in the household, the share of household
 members that were eligible for the men’s survey, the years of total schooling, the age of the household head, and indicators for the last
 level of schooling obtained (partially completed primary, completed primary, partially completed secondary completed secondary, or higher
 than secondary), for whether the respondent was the head of the household, for the respondents literacy level (illiterate, partially literate,
 can read a sentence out loud), and for whether the head of the household was male. Standard errors are clustered at the PSU level and the
 associated 95 percent confidence intervals for each estimate are reported.




                                                                      37
Figure 8. How Much Larger the Increase in the Prevalence of GBV was in Disaster-Affected Regions
than in the Rest of the Country

 8a. Difference-in-Differences Estimate of Experiencing Any GBV in the Past Year
    0.20

    0.00
            Experience Any GBV in the Past Year Experience any Emotional GBV in the Experience any Phsyical GBV in the
   -0.20                                                     Past Year                          Past Year

   -0.40

   -0.60

 8b. Difference-in-Differences Estimate of Each Individual Type of Emotional GBV Captured in the
 Past Year
    0.40

    0.20

    0.00
              Husband Humiliated Respondent              Husband Threatened Respondent                Husband Insulted Respondent
   -0.20

   -0.40

   -0.60

 8c. Difference-in-Differences Estimate of Each Individual Type of Physical GBV Captured in the Past
 Year
   0.20
   0.10
   0.00
              Pushed       Slapped      Twisted      Punched       Beat Up       Choked       Attacked    Forced     Other     Other
   -0.10
                                         Arm                                                    with   Intercourse Forced Sex Threat
   -0.20                                                                                      Weapon                  Act    Related to
                                                                                                                              Sex Acts
   -0.30
   -0.40
   -0.50
   -0.60
 Notes: The figures report the difference-in-difference estimate from the baseline specification using an indicator for GBV incidents
 occurring in the past year as the dependent variable. All specifications include control variables and regional fixed effects. Control variables
 include the number of children born in the household, the share of household members that were eligible for the women’s survey, the
 years of total schooling, the age of the household head, and indicators for the last level of schooling obtained (partially completed primary,
 completed primary, partially completed secondary completed secondary, or higher than secondary), for whether the respondent was the
 head of the household, for the respondents literacy level (illiterate, partially literate, can read a sentence out loud), and for whether the
 head of the household was male. Standard errors are clustered at the PSU level and the associated 95 percent confidence intervals for each
 estimate are reported.




                                                                       38
Figure 9. Potential Reasons for the Differences Between Changes in GBV Attitudes and GBV
Prevalence

 9a. Difference-in-Differences Estimate of Experiencing GBV Often for Women that Have
 Experienced GBV in the Past Year
           0.60
           0.40
           0.20
           0.00
          -0.20 Experience GBV Often in the Past Experience Emotional GBV Often Experience Phsyical GBV Often in
                             Year                        in the Past Year                the Past Year
          -0.40
          -0.60

 9b. Difference-in-Differences Estimate of Experiencing Any GBV in the Past Year by Education Level
           0.60
           0.40
           0.20
           0.00
          -0.20    Experience Any GBV in the Past Experience any Emotional GBV in Experience any Phsyical GBV in
                               Year                        the Past Year                  the Past Year
          -0.40
          -0.60
          -0.80

                                                      Some Education          No Education

 9c. Difference-in-Differences Estimate of Acceptability of Husband Beating Their Wife by Education
 Level- Men

            0.25

                        Total          Go Out         Neglect        Argue with      Refuse Sex      Burn Food Justified in All
           -0.75     Affirmative      Without         Children        Husband                                    Scenarios
                     Responses         Telling
                                      Husband
           -1.75


           -2.75

                                                      Some Education          No Education

 Notes: The figures report the difference-in-difference estimate from the baseline specification. The top panel uses indicators equaling one
 if the women experienced any form of GBV often as the dependent variables and restricts the sample to the 2,451 women who experienced
 some form of GBV in the past year out of 3,955 total women interviewed. The second panel uses indicators equaling one if the women
 experienced GBV in the past year as the dependent variables, but reports the estimates separately for the 926 women who never attended
 any school and the 3029 women who had. And the bottom panel uses indicators equaling one if the male respondent thought it was
 acceptable for a husband to beat their wife as the dependent variable, but reports the estimates separately for the 789 men who never
 attended any school and the 6,544 men who had. All specifications include control variables and regional fixed effects. Control variables
 include the number of children born in the household, the share of household members that were eligible for the women’s survey, the
 years of total schooling, the age of the household head, and indicators for the last level of schooling obtained (partially completed primary,
 completed primary, partially completed secondary completed secondary, or higher than secondary), for whether the respondent was the
 head of the household, for the respondents literacy level (illiterate, partially literate, can read a sentence out loud), and for whether the
 head of the household was male. Standard errors are clustered at the PSU level and the associated 95 percent confidence intervals for each
 estimate are reported.


                                                                      39
Figure 10. Responsibility for Decision Making in the Household

 10a. Difference-in-Differences Estimate- Husband Responsible for Decision Alone
     0.15

     0.10

     0.05

     0.00
                 Husband's Income-          Health Decisions- Husband Large Purchases- Husband               Visit Family- Husband
    -0.05         Husband Alone                       Alone                     Alone                                Alone
    -0.10

    -0.15

    -0.20

    -0.25

    -0.30

    -0.35

 10b. Difference-in-Differences Estimate- Wife Responsible for Decision Alone
   0.40

   0.30

   0.20

   0.10

   0.00
            Husband's Income- Woman Health Decisions- Woman                  Large Purchases- Woman Visit Family- Woman Alone
   -0.10              Alone                  Alone                                    Alone


   -0.20

   -0.30
 Notes: The figures report the difference-in-difference estimate from the baseline specification. The top panel uses indicators equaling one
 if the husband alone was primarily responsible for decisions in the household; and the bottom panel uses indicators equaling one if the wife
 alone was primarily responsible for decisions in the household. The specifications use information reported by the wife in the Women’s
 Survey. All specifications include control variables and regional fixed effects. Control variables include the number of children born in the
 household, the share of household members that were eligible for the women’s survey, the years of total schooling, the age of the
 household head, and indicators for the last level of schooling obtained (partially completed primary, completed primary, partially completed
 secondary completed secondary, or higher than secondary), for whether the respondent was the head of the household, for the
 respondents literacy level (illiterate, partially literate, can read a sentence out loud), and for whether the head of the household was male.
 Standard errors are clustered at the PSU level and the associated 95 percent confidence intervals for each estimate are reported.




                                                                      40