Policy Research Working Paper                        8458




       Quantifying the Impacts of Capturing
         Territory from the Government
            in the Republic of Yemen
                                     Sharad Tandon




Poverty and Equity Global Practice
May 2018
Policy Research Working Paper 8458


  Abstract
 This article estimates how households were aﬀected by the                           credit, increased their reliance on self-employment to deal
 mostly non-violent capture of Yemen’s capital in 2014 using                         with a decline in the economic climate, and reduced both
 a survey conducted as the capital was captured. Although                            the quantity and quality of foods consumed. Furthermore,
 socioeconomically advantaged households were initially                              women were aﬀected by the capture more so than men.
 better able to cope with the shock than other households,                           These results demonstrate that the capture of territory with-
 the capture resulted in a robust decline in expenditure for                         out wide-spread violence can result in a nearly immediate
 the entire population within three months. Struggling                               decline in standards of living and further illustrate the man-
 households turned to a number of coping strategies to                               ners in which households were able to cope with the shock.
 meet their basic needs- they increasingly made purchases on




  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/research. The author may be contacted at
  standon3@worldbank.org.




          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
  Quantifying the Impacts of Capturing Territory from the Government in the
                              Republic of Yemen




        Originally published in the Policy Research Working Paper Series on May 2018. This
        version is updated on December 2021.
        To obtain the originally published version, please email prwp@worldbank.org.




Keywords: Conﬂict, Poverty, Gender, Yemen
JEL classiﬁcation: D12, E24, J16, O12, O53




   ∗
     The World Bank, 1818 H St. NW, Washington, D.C. 20420, USA, standon3@worldbank.org, +1-202-458-
4043. I would like to thank Benu Bidani, Anna D‘Souza, Ruth Hill, Sailesh Tiwari, and Tara Vishwanath for
comments on earlier drafts. I would also like to thank audience members at the World Bank. All remaining
errors are attributable to the author. The views expressed here are those of the author and may not be
attributed to the World Bank.
1. Introduction
    Fragile and conﬂict settings are becoming increasingly important to poverty reduction,
where the share of the world’s poor living in fragile settings is expected to increase from 17
percent in 2017 to approximately 60 percent by 2030.1 Although there are numerous articles
investigating the consequences of violent conﬂict on economic outcomes, there is less evidence
regarding the non-violent capture of territory.2 Furthermore, there is less evidence regarding
how household behavior changes as these tumultuous events unfold- both the types of coping
strategies used and their timing. A better understanding of both the degree to which standards
of living might be aﬀected by such uncertainty, and the manners in which households might
cope with these events, can help identify interventions that might support households in these
increasingly important settings.
    This article addresses these issues by investigating the consequences of the unexpected
and mostly non-violent capture of Yemen‘s capital, Sana‘a, by Houthi rebels in September of
2014. This setting is unique, where the capture was very quick and accomplished with little
violence.3 The rebels placed the president under house arrest and immediately took control
of key government agencies- including those responsible for defense and ﬁnance. However,
because there was little violence, and because the rebels were not initially concerned with
many of the less prominent government agencies, the Central Statistical Organization was able
to complete an ongoing multi-themed household survey as planned for the remaining three
months of the year. This setting allows one to analyze the degree to which households were
aﬀected by the capture, the ways in which households coped with the shock, and the diﬀerent
impacts the capture might have had on diﬀerent household members.
    The capture of the city potentially aﬀected households in a number of ways. First, the
capture and resulting uncertainty potentially had an adverse impact on the economic climate
and peoples’ ability to earn a living. Second, although there was not widespread violence, the
change in the rule of law did result in a perceived decline in safety that might have made it
more diﬃcult for individuals to leave their households. For example, there were reports of
targeted violence by Houthi forces4 , and there was an increase in terrorist events.5 And third,
the capture of the city could have led to a disruption of supply chains and increases in the
prices of goods and services.
    Although the household survey continued following the capture of the capital, the identi-
ﬁcation strategy requires the sampling of households to not signiﬁcantly change following the
capture. The sampling design was such that respondents were randomized over space and time,
   1
      See (accessed March 2018): http://www.worldbank.org/en/topic/fragilityconﬂictviolence/overview.
   2
      For a survey of the economic eﬀects of conﬂict, see Blattman and Miguel (2010). Alternativley, a separate
literature has investigated the eﬀects of political instability, where the instability could be caused by reasons
aside from conﬂict (e.g., Alesina et al. 1996; etc.).
    3
      The city was captured after a 4-day siege that was mostly centered around a military base in the northern
part of the city, and abruptly ended when the internationally-recognized government ceded control of the city.
See (accessed July 2017): http://www.bbc.com/news/world-29380668.
    4
      See (accessed July 2017): https://www.hrw.org/news/2017/01/12/yemen-no-accountability-war-crimes.
    5
      See (accessed July 2017): https://www.theguardian.com/world/2014/oct/09/yemen-suicide-attack-kills-
20.


                                                       2
and there should be no diﬀerences in the households surveyed in the months before and the
months after the capture of the capital if the survey was conducted as planned (World Bank
2017). However, enumerators might have both consciously and unconsciously avoided more
dangerous areas, which might have aﬀected the representativeness of the sample.
    Despite this potential challenge, there is little evidence that sampling was signiﬁcantly
aﬀected. The sampling was reported to have been carried out as planned- all enumeration
areas and initial households to interview were selected approximately a year before the city
was captured, none of the enumeration areas were replaced, and there were no reports of
diﬃculties in completing the survey in the ﬁnal quarter. Also, there is little diﬀerence in
relatively ﬁxed household characteristics between households surveyed prior to the capture to
those surveyed after, corroborating the reports of the sampling not being signiﬁcantly aﬀected.
    Given this setting, this article focuses on the evolution of household expenditure following
the capture, as well as the ability of households to cope with the shock. Following the capture
of Sana’a, one cannot reject the hypothesis that average expenditure did not initially change
in the ﬁrst month. However, by the third month following the capture, expenditure had
signiﬁcantly dropped in the capital relative to the rest of the country.
    Importantly, the timing of the change in expenditure varied based on pre-existing household
characteristics. Households where the head had not completed primary school immediately de-
creased their expenditure, and this decline remained for the rest of the time period under anal-
ysis. Whereas, households where the head ﬁnished primary school were able to maintain their
expenditure in October, but decreased their expenditure by December. Although the house-
hold survey did not collect information on savings, this evolution of expenditure is consistent
with better-oﬀ households being able to utilize savings to maintain expenditure, but other
households having to sacriﬁce expenditure and immediately utilize other coping strategies.
    Additionally, this article analyzes a number of diﬀerent types of coping strategies that
households employed during this time. First, less educated households, who were not able to
maintain their expenditure immediately following the capture of the city, signiﬁcantly increased
their purchases using credit from local stores. The increase in these types of arrangements likely
helped these households at least initially maintain their food consumption following the shock.
    Second, there was a large change in the nature of employment in the city that began
immediately after the capture of the city. There is some evidence of a decline in household
wage income, but little evidence of a change in the overall number of households that reported
working in the past week. Rather, there was a large increase in the share of households that
turned to self-employment following the capture, and the share turning to self-employment
continued to increase throughout the occupation. The results are consistent with households
turning to self-employment as a way to cope with the poor economic climate, as the earnings of
self-owned enterprises declined and the vast majority did not earn enough to reach the poverty
line following the capture.
    Third, there was a signiﬁcant worsening in both the quantity and quality of food that was
consumed during this time period. Initially, there were not large changes in micro and macro-


                                                3
nutrient consumption following the capture. However, by the third month, the magnitude of
the changes increased and also were more precisely estimated. Furthermore, the decline in
calorie consumption suggests that households were signiﬁcantly struggling by the third month
of the occupation, as a number of settings have demonstrated that diet quality is much more
responsive to shocks than the quantity of calories (e.g., D’Souza and Joliﬀe 2012; D’Souza and
Joliﬀe 2014; etc.).
    Lastly, in addition to the decline in expenditure and the identiﬁcation of a number of key
coping strategies used by households, there is evidence that women fared particularly poorly
in this setting. The capture of the city had adverse consequences on women‘s decision-making
reported in the survey, where women had less control over measures of spending, and this
control was transferred to primarily the male household head. The decline in control over
food expenditure was especially pronounced, which was also one of the areas of spending
where women in Sana‘a had the most control prior to the capture of the city. This decline
in decision-making is consistent with women being less able to leave the household than men
due to safety concerns, where the change was immediate and did not get worse as household
expenditure signiﬁcantly declined in the later months of the occupation.
    There are two primary contributions of these results. First, this article presents estimates
of the economic eﬀects of the capture of territory without wide-spread violence. Although
previous settings have demonstrated that the severity of violence is correlated with economic
activity (e.g., Abadie and Gardeazaba 2003; Guidolin and La Ferrara 2007; Besley and Mueller
2012; Klapper et al. 2013; Singh 2013, Amodio and Di Maio 2017, etc.), there is less evidence
of how types of fragility that are largely non-violent might have aﬀected household living
standards. Furthermore, this setting allows a thorough investigation of the type of coping
strategies used and their timing. Despite the signiﬁcant reliance on coping strategies beginning
immediately after the occupation, the estimates suggest that expenditure signiﬁcantly dropped
for a majority of individuals within a relatively short time period.
    Lastly, the decline in women‘s decision-making demonstrates an important way in which
fragility might impact individuals within the household diﬀerently. Many news organizations
and humanitarian agencies report that, due to economic necessity, women become more in-
volved in decision making and economic activity during conﬂict, but then revert back to the
pre-conﬂict baseline once the conﬂict ends. For example, news reports in Yemen following the
escalation of conﬂict in 2015 suggest that women are being forced to take a more active role in
household decisions and income generation as the humanitarian situation became more dire.6
However, to my knowledge there has been little corroboration of these results.7 These results,
along with the fact that there was little change in the small labor force participation of women
during the occupation, contrast with the narrative of more responsibilities for women during
   6
     For an example from Yemen following the escalation in conﬂict in 2015,                               see
http://www.aljazeera.com/news/2015/12/yemen-widows-weakest-victim-war-151215061011411.html.             How-
ever, these reports do not suggest that women are necessarily treated better, but rather that their
responsibilities have increased.
   7
     Mkutu (2008) demonstrates that an increase in widow-led households necessarily led to more participation
of women in Uganda in a pastoral setting.


                                                     4
conﬂict.
    This article is structured as follows. Section 2 describes the Yemen context and various
manners in which the capture of territory from the government might aﬀect households in
non-violent settings; Section 3 describes the data; Section 4 describes the empirical strategy;
Section 5 presents estimates of the change in expenditure and the change in coping strategies
following the capture of the city; Section 6 presents estimates of the change in intra-household
decision making; and Section 7 concludes.

2. The Capture of Sana‘a and the Potential Eﬀects on Households

    The causes of the conﬂict in Yemen are complex and involve a number of factors. The
Houthi rebels, originating from the northern governorate of Sa’dah, had repeatedly fought
against the national government beginning in the mid 2000‘s. The Houthis’ movement began
as a theological movement that protested the national government. However, following this
protest, the government used the military to try and arrest their leader, which led to a a
number of wars during which the group made few gains, and was repeatedly beaten back by
the military.8
    However, following the Arab Spring in 2011, the government led by Ali Abdullah Saleh was
forced to hand over power to his vice president and a transitional unity government. Using
this political vacuum and the unpopularity of the transition government, the Houthis began
making military gains against the internationally-recognized government after its formation in
2012 in the hopes of having more representation in the transitional government. These military
gains eventually led to the quick capture of the capital Sana‘a in September of 2014, and then
eventually an attempt to capture the southern part of the country.9 , 10
    The capture of Sana’a resulted in a number of shocks that could have aﬀected households.
First, the political uncertainty likely signiﬁcantly impacted economic activity and investment.
Although the evacuation of foreigners accelerated at the beginning of 201511 , it is possible that
the climate immediately following the capture of Sana’a could have caused many international
   8
     See (accessed July 2017): http://www.aljazeera.com/news/middleeast/2014/08/yemen-houthis-hadi-
protests-201482132719818986.html.
   9
     See (accessed July 2017): http://www.bbc.com/news/world-middle-east-29319423.
  10
     Some have posited that one of the potential reasons that the capture of the city was accomplished
so quickly and with little violence is that it is possible that the military gains were aided by the for-
mer president, Ali Abdullah Saleh, who might have directed loyal military forces to stand down in the
face of the Houthi aggression (see (accessed July 2017): http://www.reuters.com/article/us-yemen-security-
saleh-idUSKBN0MM1MV20150326). Although it seems unlikely that the former president who waged many
wars against the Houthis would then form an alliance with them, some have speculated that they each
had a set of common rivals and could develop a mutually beneﬁcial arrangement (see (accessed July 2017):
https://intpolicydigest.org/2017/05/11/houthi-saleh-alliance-convenience/). This potential alliance is further
corroborated by the targeting of enemies of both the Houthis and the former president by Houthi forces af-
ter the capture of Sana‘a (see (accessed July 2017): http://america.aljazeera.com/articles/2014/9/25/houthi-
yemen-takeover.html), the relative ease with which the Houthis were making military gains against a gov-
ernment that were elusive before Saleh was forced out of oﬃce, reports of the Yemen army ﬁghting
along the Houthi rebels (see (accessed July 2017): http://www.reuters.com/article/us-yemen-security-saleh-
idUSKBN0MM1MV20150326), and a formal announcement of an alliance after the conﬂict escalated in March of
2015 (see (accessed July 2017): http://www.aljazeera.com/news/2015/05/cloneofcloneofcloneofstrikes-yemen-
saada-breach–150510143647004.html).
  11
     See (accessed August 2018) https://www.bbc.com/news/world-middle-east-32162364.


                                                      5
businesses and foreign governments from removing some of its personnel from the country.
The country already suﬀered from signiﬁcant underemployment even before the capture of the
capital (e.g., World Bank 2017), and one of the goals of this article is to investigate the degree
to which the political uncertainty could have exacerbated these conditions.
    Second, news reports suggest that safety was negatively impacted12 , which likely aﬀected
household members‘ ability to leave their households and seek employment opportunities. The
decline in safety reported by news agencies is further corroborated by the ongoing household
survey that was being conducted as the capital was captured. The survey included a subjective
welfare module, which separately asked the head of the household and their spouse whether
they were satisﬁed with security. Following the capture of the capital, there was a signiﬁcant
decline in the share of households where either a household head or spouse reported to be very
satisﬁed with security from 11.6 percent of households to 1 percent, and this decline was not
observed in the rest of the country. Importantly, this decline in safety might have had a worse
impact on women than men, where not a single woman interviewed following the capture of
the city (out of 235 total) reported to be very satisﬁed with security, whereas women were able
to report being satisﬁed with other aspects of their life.
    And third, the supply chains providing goods and services to the capital city were likely
disrupted. Prior to the conﬂict, Yemen relied on international sources for the vast majority of
its consumption, where approximately 90 percent of its food was imported from abroad (e.g.,
World Bank 2017). Imports to the capital city primarily came through a port in Al Hudaydah,
and were then transported by truck to the capital city.13 Given the diﬃculty and uncertainty
of transporting food and other goods across new checkpoints where safety and access were not
guaranteed likely made it more diﬃcult for goods to reach the capital city.
    There is evidence consistent with this being a signiﬁcant problem in Sana‘a following the
capture of the city. Utilizing average prices collected at markets in the city by the World Food
Programme (WFP) each month for 14 food items and commodities, there was a signiﬁcant
increase in the price of a number of goods following the capture of the city. In particular,
ﬁgure 1 demonstrates that fresh food items and fuel, all of which are imported on a regular
basis from abroad, had large price increases in the months following the capture of the city
that were not evident over the same time period in the two years prior to the capture of the
city.
    Thus, at exactly the time that prices of necessary goods and services were potentially rising,
households potentially had a decline in household income. A vast literature has investigated the
potential impacts of adverse income shocks across a number of dimensions. For example, it has
been shown that adverse income shocks adversely aﬀect the quality of food consumption but
not overall calorie consumption (e.g., Block et al. 2004), can reduce school attendance amongst
school-aged children (e.g., Edmonds et al. 2009; etc.), and potentially force households to sell a
  12
     See (accessed July 2017): https://www.hrw.org/news/2016/11/17/yemen-abusive-detention-rife-under-
houthis.
  13
     See    FAO’s    summary    of    food    imports   in  Yemen    at   (accessed    July     2017):
http://www.fao.org/giews/countrybrief/country.jsp?code=YEM.



                                                  6
limited share of productive assets and reduce possible future income generation (e.g., Fafchamps
et al. 1998). Alternatively, a number of contexts have demonstrated that household welfare
signiﬁcantly declines in the face of an adverse food price shock, where households reduce diet
quality (e.g., Brinkman et al. 2009; D‘Souza and Jolliﬀe 2012; D‘Souza and Joliﬀe 2014).
    Based on these other contexts, households in Sana‘a might also be forced to adopt a num-
ber of these coping strategies. However, the coping strategies in this circumstance might be
substantially diﬀerent. For example, households might ﬁnd it diﬃcult to sell productive assets
given the overall poor condition of economic activity; or, for example, it might be diﬃcult to
further sacriﬁce diet quality given the widespread food insecurity even before the capture of
the city (e.g., WFP 2012).
    Importantly, these shocks- both the potential tightening of household budgets and the
decline in safety- might have signiﬁcantly diﬀerent eﬀects on women in the household than men.
For example, increasing the constraints on households tends to lead to an excess burden being
placed on vulnerable members of the household in a variety of settings (e.g., Rose 1999; Khanna
et al. 2003; Miguel 2005; D’Souza and Tandon forthcoming; etc.), which could place an excess
burden on women in Yemeni households given the country is typically ranked as having one of
the highest (if not the highest) gender gaps in the world (e.g., World Economic Forum 2017).
Furthermore, the potential increase in prices might similarly force households to substitute
more home-produced goods and services for market-produced goods and services, which would
most likely place a higher time burden on household women (e.g., World Development Report
2012). Additionally, the decline in safety might make it more diﬃcult for women to leave the
household and lead productive lives than men. In particular, women’s movement and freedom
are reported to be signiﬁcantly curtailed in the country due to safety and family concerns (e.g.,
World Bank 2012), suggesting that they might even be less free to leave the household and
make decisions following the introduction of conﬂict.
    However, it is important to note that the eﬀects of these shocks on women are empirically
ambiguous. The decline in household resources might suﬃciently burden households such
that women are force to take up income generating possibilities and other responsibilities
that they were unable to do so before the conﬂict. As discussed in the Introduction, many
news organizations and humanitarian agencies report that women are forced to take on more
responsibilities, both inside and outside the household, in response to conﬂict.

3. Data

    The 2014 Household Budget Survey (HBS), conducted by the Central Statistical Organi-
zation of Yemen with support from the World Bank, was in progress as the Houthi rebels
captured the city of Sana‘a. The survey‘s primary purpose was to update the oﬃcial poverty
rate in Yemen, which had not been updated since 2005. Given the relative lack of violence
in the capture of the capital city, the survey was administered before, during, and after the
capture of the city. The ﬁeld work was conducted between January and December of 2014,
and the total sample size was of 9391 households.


                                               7
    However, given the sampling design, the survey is able to be used to identify how house-
holds were aﬀected by the capture. First, the sampling design was such that respondents
were randomized over space and time, and there should be no diﬀerences in the households
surveyed in the months before and the months after the capture of the capital if the survey
was conducted as planned (World Bank 2017).14 Second, the survey is representative at the
governorate/capital city level.
    The 2014 HBS included a complete set of modules, including standard information on
household consumption, employment, education, health, etc. that allow a thorough investiga-
tion of expenditure and household coping strategies. Appendix 1 presents summary statistics
of expenditure and coping strategies, and demonstrate that households in Sana‘a were far
better oﬀ than the rest of the country. Poverty in Sana’a was approximately one-quarter of
the rate in the rest of the country, and all other measures were similarly better in the capital
city. Additionally, women were signiﬁcantly more empowered in the capital city than in the
rest of the country, but subjective welfare was approximately equal across all regions of the
country. However, it is important to place these characteristics in historical context- poverty
has signiﬁcantly increased in Yemen since 2005, in part driven by the political uncertainty of
2012 and also the sharp increase in food prices beginning in 2011 (e.g., WFP 2012; Dang and
Ianchovichina 2016; World Bank 2017).
    Another advantage of the 2014 HBS is that there was a nearly identical 2005 HBS that allow
one to conduct additional robustness checks. One can investigate whether changes following
the capture of Sana’a were potentially due to seasonal factors, or due to a country-wide shock
as opposed to the capture of Sana‘a. In particular, the 2005 data allow an investigation of
the diﬀerence-in-diﬀerence estimate of expenditure in Sana‘a following the capture of the city.
Additionally, the data further allow us to investigate the triple diﬀerence as well, where the
diﬀerence-in-diﬀerence estimate in the city of Sana‘a is compared to the diﬀerence-in-diﬀerence
in the rest of the country. However, the women‘s decision-making module was not implemented
in the 2005 survey, and speciﬁcations analyzing those measures can only analyze the diﬀerence
in Sana‘a following the capture of the city in 2014, and compare that diﬀerence to the change
in the rest of the country.
    Although the household survey continued following the capture of the capital, the identi-
ﬁcation strategy requires the sampling of households to not signiﬁcantly change following the
capture. Importantly, given the lack of violence, the survey was reported to have been carried
out as planned. All enumeration areas and initial households to interview were selected ap-
proximately a year before the city was captured, none of the enumeration areas were replaced,
and there were no reports of diﬃculties in completing the survey in the ﬁnal quarter.
    Despite these reports of little trouble, it is always possible that enumerators consciously
  14
    Roughly equal numbers of households were surveyed in each month, where 78 households were surveyed in
October, 80 households were surveyed in November, and 77 households were surveyed in December. A total
of 235 households were surveyed after the capture of Sana’a was complete and 269 households were surveyed
in the 3 non-Ramadan months leading up to the capture. However, the responses to the women’s decision-
making module were missing for one of the households in the pre-period, leaving one less observation in those
speciﬁcations than the others. However, all results are robust to dropping that household from all speciﬁcations.


                                                       8
or unconsciously avoided more dangerous areas when replacing households. However, inves-
tigating the possibility, there does not appear to be any evidence that households surveyed
before the capture were signiﬁcantly diﬀerent in observable characteristics that are unable to
adjust to the conﬂict. Table 1 presents observable characteristics of households surveyed in
the third and fourth quarter of 2014. Households surveyed in the third quarter all began their
surveys prior to the capture, and households surveyed in the fourth quarter all began their
surveys after the city had been captured. Consistent with the capture having little eﬀect on
the sampling, column (3) demonstrates that there is little diﬀerence between diﬃcult-to-adjust
characteristics following the capture of Sana’a. Additionally, column (4) presents estimates of
a regression of a fourth quarter indicator on each of the variables analyzed in the table (re-
stricting the sample to households surveyed in the third and fourth quarter), and demonstrates
that one cannot reject the hypothesis that the all variables are jointly uncorrelated with the
quarter indicator at standard signiﬁcance levels (p-value of 0.366).15
     While it has been documented that there are no average diﬀerences between the households
sampled before and after the capture, there still might be diﬀerences in speciﬁc months. Figures
2a-2c address this concern by reporting conﬁdence intervals for the means of a wide variety of
household characteristics are reported by month, as well as the average pre-capture means of
each of the variables.
     Figures 2a-2c demonstrate that the 95 percent conﬁdence interval of each monthly estimate
of diﬃcult-to-adjust control variable contains the pre-capture mean of the variable in the vast
majority of instances. Of all the 95 percent conﬁdence intervals reported (180), the pre-capture
mean is contained in 96.7 percent of them (174). Importantly, of those conﬁdence intervals
that do not contain the pre-capture mean, the distance from the bound of the interval to
the pre-capture mean is very small; there is no evidence that any particular variables had a
large proportion of monthly conﬁdence intervals that did not contain the pre-capture mean
(the maximum for any variable is 2 of the 12 monthly conﬁdence intervals not containing the
pre-capture mean); there is no evidence that the post-capture months of October-December
were contributing the majority of the instances where the pre-capture means were outside
of the conﬁdence intervals (1 of the 7 total instances were from those months); and there is
little evidence that the conﬁdence intervals were uniformly larger following the capture of the
capital.
     However, there is an important data issue to highlight that might aﬀect estimating the
impact of capturing Sana‘a. Ramadan occurred primarily in July in 2014, and this might bias
expenditure during the pre-period given that consumption patterns are signiﬁcantly diﬀerent
during Ramadan (e.g., Joliﬀe and Serajuddin 2012, World Bank 2017). Thus, in speciﬁcations
that use only the 2014 survey, all households surveyed during Ramadan are excluded from
the sample, and the baseline speciﬁcation uses the three non-Ramadan months prior to the
capture of the capital as the pre-period to compare against the three post-capture months
  15
    The results are similar when comparing changes between the third and fourth quarter in the rest of Yemen
as well (see Appendix 2). The results are also similar when comparing the summary statistics of quarter 1 and
2 to quarter 4 (see Appendix 3).


                                                     9
of the survey. However, in speciﬁcations that compare the 2014 survey to the 2005 survey,
households surveyed during July (Ramadan in 2014) or October (Ramadan in 2005) in either
survey were excluded. However, it is important to note that the estimates are qualitatively
identical when all interviewed households are included.

4. Empirical Strategy

    Based on this setting, this article utilizes a diﬀerence-in-diﬀerence strategy to analyze the
eﬀects that the capture of Sana’a had on household expenditure and coping strategies. Figure 3
illustrates the strategy by plotting average annual household expenditure and a number of key
coping strategies by month of the survey. Expenditure in Sana’a was maintained in October
of 2014 on average, but by November and December, there was a sharp decline in expenditure
such that there was little diﬀerence in expenditure between the city of Sana’a and the rest
of Yemen despite the poverty rate being approximately four times less in the city of Sana’a
(Appendix 1). Importantly, over the same time period, there was little change in expenditure
in the rest of Yemen, and this pattern is similar for a number of other coping strategies that
are observable in the survey.
    Formalizing this empirical strategy, this article estimates the following speciﬁcation:



(1)            Outcomeir = Σ3
                            j =1 [βj P ost M onth jir ∗ Capitalir ] + αYir + γXir + υir


where i refers to household, r refers to region, P ost M onth j are indicators equaling one
if the household was interviewed during the j’th month following the occupation16 , Capital
denotes an indicator equal to one if the household is from the city of Sana‘a, Y includes all
lower-order terms (each of P ost M onthj and Capital), and X is a vector of individual control
variables.17 The baseline estimates restrict the households in speciﬁcation (1) to households
surveyed in the three non-Ramadan months prior to the capture and the three months following
the capture of the city that the survey continued. Standard errors are clustered at the PSU
level, and observations are weighted to make the estimates representative of the population.
The coeﬃcients of interest are βj , which represent how much larger the percentage point change
was in Sana’a (relative to the three months before the capture) than in the rest of Yemen for
a variety of outcomes following the capture of the city.18
  16
      Speciﬁcally, P ost M onth 1, P ost M onth 2, and P ost M onth 3 respectively are indicators equaling one if
the household was surveyed in October, November, and December.
   17
      Control variables include household size, an indicator for whether the household head is married, the
household head ever attended school, an indicator if the household head was a smoker, an indicator if the
household head chews Qat, the years of schooling of the household head, an indicator whether the household
has a child, an indicator if the household’s dwelling is made of cut stone, an indicator equaling one if the walls if
the household’s dwelling were made of other material, an indicator equaling one if the household’s roof that was
made of reinforced concrete, an indicator equaling one if the household’s roof was made of wood and concrete,
an indicator equaling one if the household’s roof was made of mud, an indicator if the household’s ﬂoor was
made of concrete, an indicator equaling one if the household’s ﬂoor was made of tile, an indicator equaling one
if the household had a ﬂush toilet, and an indicator if the household was rural.
   18
      Although the baseline speciﬁcation is restricted to balance the number of pre and post-months and to use
the pre-months that are closest in time to the capture of the city, Appendix 4 demonstrates that the results


                                                         10
   However, in addition to estimating speciﬁcation (1), this article also presents estimates of
the simple diﬀerence in Sana’a following the capture of Sana’a to demonstrate the results are
being driven by changes in Sana’a as opposed to changes in the rest of Yemen:



(2)                        Outcomei = Σ3
                                       j =1 [βj P ost M onth ji ] + γXi +          i


where all variables are the same as above.19 Speciﬁcation (2) is restricted to households from
the city of Sana’a, and the coeﬃcients of interest are βj , which represent the percentage point
change in a variety of outcomes following the capture of the city relative to the three months
before the capture.
    As an additional robustness check, this article also estimates speciﬁcations that allow the
outcome to vary by month to better demonstrate when the changes in expenditure and coping
strategies only occurred:



(3)               Outcomeir = Σ5
                               j =1 [βj M onth jir ∗ Capitalir ] + αYir + γXir + υir


where M onth j are indicators equaling one if the household was interviewed during the j’th
month following the ﬁrst month used in the base period20 , and all other variables are the
same as above.21 The coeﬃcients of interest are βj , which represent how much larger the
percentage point change was in Sana’a for a variety of outcomes month-by-month, including
the two months leading up to the survey and the three months after. In order to attribute
the decline in expenditure to the capture of the city, it is important to verify that expenditure
was not trending diﬀerently leading up the capture. In such a scenario, it is possible that the
expenditure declines could be uncovering the response to a shock that actually led to rebels
capturing the city, as opposed to uncovering the response to the capture.

5a. Decline in Expenditure Following the Capture of the Capital

   Table 2 presents estimates of how annualized expenditure changed following the capture.
Estimates of speciﬁcation (1) are presented in columns (1)-(2). Column (1) reports a sparse
speciﬁcation including only lower-order terms and not the control variables; and column (2)
adds control variables to absorb unobserved heterogeneity and increase the precision of the
are qualitatively identical regardless of which months are used to construct the pre-period.
  19
     Control variables are identical, but the rural indicator is omitted given that all households in Sana’a are
urban.
  20
     Speciﬁcally, M onth 1 - M onth 5 are month indicators for August-December respectively.
  21
     The appendix also investigates speciﬁcations that can better account for seasonality. Speciﬁcally, Appendix
4 estimates a diﬀerence-in-diﬀerence speciﬁcation comparing the change in Sana’a in 2014 to the change in
Sana’a in 2005; and Appendix 5 also estimates a triple diﬀerence speciﬁcation that compares the diﬀerence-
in-diﬀerence in Sana’a (between 2005 and 2014) to the diﬀerence-in-diﬀerence for the rest of Yemen. All
results from those speciﬁcations are qualitatively identical to the baseline speciﬁcation. However, given the
long time period between the 2005 and 2014 surveys and the numerous events that happened between- political
transition and food price shock- this article focuses on estimates from speciﬁcation (1) that compare the change
in household outcomes in the city of Sana’a to the rest of Yemen using only data from the 2014 survey.


                                                      11
estimates.
    In both speciﬁcations, expenditure in the three months following the capture of the city is
less than in the three months leading up to the capture. However, the change for households
surveyed in October and November are lower in magnitude and less precisely estimated than
the change for households surveyed in December. In the most complete speciﬁcation reported
in column (2), households surveyed in December had an expenditure that was 35.9 percentage
points lower than in the three months leading up to the capture of the city. However, despite the
increase in magnitude and precision as the occupation progresses, one can reject the hypothesis
that the three post-conﬂict coeﬃcients jointly equal zero at standard levels of signiﬁcance (p-
value of 0.039).
    It is important to verify that the decline in expenditure is being driven by a reduction in
expenditure in Sana’a, as opposed to a change in consumption in the rest of Yemen. Estimates
of the single diﬀerence (speciﬁcation (2)) in Sana’a are presented in column (3) of table 2. The
estimates are nearly identical to the diﬀerence-in-diﬀerence comparing the change in Sana’a to
the rest of the country, which suggests that the diﬀerence is being driven by a large change in
expenditure in Sana’a and little change in expenditure in the rest of Yemen. The similarity of
the coeﬃcients in columns (2) and (3) are also consistent with Figure 3, which shows a similar
pattern. Thus, the baseline estimates do not appear to be capturing a shock that is common
to the entire country.
    Additionally, it is important to verify that these changes in expenditure are being driven
by the capture of Sana‘a, and not uncovering a pre-existing change in expenditure. Column
(4) of table 2 reports estimates of a speciﬁcation similar to speciﬁcation (3), but restricts the
sample only to households surveyed before the capture; and column (5) reports estimates of
the month-by-month diﬀerence in expenditure (speciﬁcation 3).
    The estimates in columns (4) and (5) are consistent with expenditure only changing in
the months after the capture of Sana’a, and the change in expenditure being largest and
most precisely estimated by December. In both columns (4) and (5), the magnitude of the
estimates on the pre-capture months- August and September- are smaller than the change
found in November and December, and one cannot reject the hypothesis that both coeﬃcients
are jointly equal to zero at conventional signiﬁcance levels (p-values of 0.779 and 0.767 in
columns (4) and (5) respectively). Furthermore, the coeﬃcient estimates of the post-capture
months (October, November, and December) are nearly identical in magnitude to the most
complete baseline speciﬁcation in column (2), and the change in expenditure only becomes
statistically diﬀerent from zero by December (ﬁgure 4).
    Table 2 further decomposes expenditure into expenditure on food and expenditure on non-
food items in columns (6) and (7) respectively. The estimates demonstrate that the changes
in each type of expenditure are qualitatively identical to the baseline estimates in column (2).
Importantly, expenditure on food was collected in a two-week diary, and all expenditure on
food in households surveyed in post-capture months was collected after the occupation began.
Alternatively, non-food expenditure had a number of diﬀerent recall periods- two weeks, one


                                               12
month, three months, and one year depending on the type of expense. Given the similarities
between the change in food expenditure and the total household expenditure, it does not seem
likely that the longer reference period of some consumption items is aﬀecting the timing of the
expenditure declines being uncovered in the baseline pattern.
    Lastly, table 2 investigates potential heterogeneity in the drop in expenditure. Speciﬁcally,
column (8) re-estimates the baseline speciﬁcation, but restricts the sample to households where
the head ﬁnished primary school; and column (9) re-estimates the baseline speciﬁcation, but
restricts the sample to households where the head did not ﬁnish primary school. The estimates
in columns (8) and (9) demonstrate that the timing of expenditure drops varies between the
two types of households. Less-educated households observe an immediate drop in expenditure
in October, while one cannot reject the hypothesis at conventional signiﬁcance levels that more
educated households did not reduce their expenditure in October.22
    Thus, although there is an average drop in expenditure for all households by December,
there is signiﬁcant heterogeneity in when this decline in expenditure occurred. The results are
consistent with better-oﬀ households being better able to cope with the occupation initially
than less better-oﬀ households. Although the survey did not capture savings, one potential
explanation for this pattern was that better-oﬀ households might have been able to rely more
on savings than other households.

5b. Brief Discussion of Interpreting the Expenditure Changes

    The above expenditure patterns are consistent with the declines being caused by the capture
of Sana’a. It is important to note that these results are robust to two important concerns.
First, one might be worried about the possibility that the poor economic conditions might
have led to the capture of the city by the rebels, and the above patterns are capturing this
pre-existing trend (e.g., Miguel et al. 2004). However, the Houthi capture of Yemen‘s capital
city is a setting in which this concern is minimized. First, expenditure was not trending in the
months prior to the capture of Sana’a. Only after the capital was captured did expenditure
begin to change. Additionally, the capture of the city appears to be more facilitated by
powerful politicians rather than the will of either the rebels or the population in response to
economic conditions. Although the Houthis took advantage of the anti-government protests in
September23 , their success seems more related to the fact that the majority of the army loyal
to the former president did not resist the assault.24 This is corroborated by the fact that the
Houthis fought and lost numerous wars with the government over the prior decade, but only
once the army did not resist their advances were they able to achieve more success.25
    Second, as is likely consistent with other conﬂict settings, the time immediately prior to
  22
     See Appendix 6 for estimates of a speciﬁcation that interacts the diﬀerence-in-diﬀerence coeﬃcients with
an indicator that the household did not ﬁnish primary school.
  23
     See Al Jazeera (Accessed July 2017): http://america.aljazeera.com/articles/2014/9/25/houthi-yemen-
takeover.html.
  24
     See     (accessed     July     2017):           http://www.reuters.com/article/us-yemen-security-saleh-
idUSKBN0MM1MV20150326)
  25
     See (accessed July 2017): http://www.aljazeera.com/news/middleeast/2014/08/yemen-houthis-hadi-
protests-201482132719818986.html.


                                                     13
the conﬂict was turbulent. In particular, the government temporarily removed a large fuel
subsidy for the month of August that not only supported direct consumers of fuel, but also
supported the transport of goods across the country and the use of fuel as an intermediate
input of production. However, this temporary removal of the fuel subsidy did not appear
to be driving the results. First, there was little-to-no immediate impact on expenditure for
that month, where these outcomes only changed following the capture of Sana’a. Second, the
subsidy removal was targeted at the entire country and not just Sana’a, whereas all the results
analyzed here show a pattern of a clear change in Sana’a coupled with little change in the rest
of the country.26

5c. Coping Strategies

    Given the unique opportunity to exploit a number of modules in the survey, this article
further analyzes how households potentially coped with the capture of Sana’a. First, Table 3
re-estimates speciﬁcation (3), but utilizes a number of indicators on the number, value, and
source of loans made to the household.
    First, the survey breaks up loans by source- where one of the most common sources of
loans in the country was for households to make purchases on credit from a trader. Column
(1) of table 3 estimates the change in the loans made from traders, and column (2) estimates
the change in loans made from other sources (primarily relatives, friends, and banks). Column
(1) demonstrates that there was an increase in the share of loans made by traders in all three
months. However, the estimates are imprecise, and one cannot reject the hypothesis that there
was no change in loans from traders over the entire post-conﬂict period (p-value of 0.245).
Column (2) demonstrates very little change in loans from other sources as well.
    However, when breaking up the pattern by education of the household head in columns
(3)-(8) of table 3, a clear pattern emerges. Columns (3)-(5) respectively demonstrate that for
households where the head did not ﬁnish primary school, there was a large increase in the
number of loans from traders, there was a higher likelihood of a household having at least a
single loan from a trader, and there was a larger total amount borrowed from traders. One can
reject the hypothesis that there was no change in these three variables describing loans from
traders (number of loans, likelihood of having a loan, and the value of loans) over the entire
post-capture period at the one-percent signiﬁcance level (p-values of 0.005, 0.009, and 0.008
respectively). Additionally, one can see in columns (3)-(5) of table 3 that the increase in loans
from traders began immediately following the capture, which is also a time when these types
of households were struggling and had lower expenditures (table 2, column 9).
    However, columns (6)-(8) of table 3 demonstrate that there was little change in borrowing
from traders for households where the head ﬁnished primary school. The magnitude of the
coeﬃcients are lower, and they are more imprecisely estimated. Although there are likely
many coping strategies that are not observable using this household survey, the ones that are
observable suggest there is heterogeneity in the types of coping strategies used by diﬀerent
  26
     Appendix 7 further demonstrates that the change in expenditure is similar when using the matching
estimator proposed by Abadie et al. (2004).


                                                 14
types of households throughout the occupation.27
     Alternatively, table 4 further investigates coping strategies involving employment. Column
(1) re-estimates speciﬁcation (3) but uses total income from wage activities in the past week as
the dependent variable and restricts the sample to households that reported any wage activity,
and columns (2)-(6) estimate the same speciﬁcation but use measures of the amount worked by
household members and utilize the entire sample. Column (1) demonstrates that there is some
evidence of a wage decline following the occupation. Although the magnitude of the estimates
suggest that the potential wage declines were large, the estimates are not precise, and the joint
signiﬁcance of the post-occupation eﬀects are only bordering statistical signiﬁcance (p-value of
0.133).
     Furthermore, there was little evidence of individuals working less during this time period
as expenditure signiﬁcantly declined. Column (2) demonstrates that there appears to be very
little change in the likelihood that at least one member of the household worked in the past
week- the post-capture coeﬃcients vary in sign, and are close to zero. Column (3) demonstrates
that there was actually an increase in household heads reporting to having worked in the past
week in December; and column (4) demonstrates that there was a decline in the share of
household heads where non-heads reported working. However, the estimates in columns (4)
and (5) are not precisely estimated, and it is diﬃcult to claim that there was a signiﬁcant
reduction in the likelihood of working for any household member. Furthermore, columns (5)
and (6) are qualitatively identical to the estimates in columns (4) and (5), demonstrating that
there was little change to the total number of hours worked.
     The lack of strong employment and wage eﬀects as expenditure signiﬁcantly dropped is
puzzling. Furthermore, although estimates on the likelihood of heads working and the number
of hours worked are imprecise, the sign and magnitude of the coeﬃcient by December in column
(3) is such that one can rule out the possibility of a large decline in in their labor supply as
expenditure dropped the most.
     However, one can observe large changes to the nature of employment. Namely, there was
a large increase in the share of households reporting owning their own enterprise. Column (7)
demonstrates that there was a 17 percentage point increase in the share of households reporting
owning their own enterprise in October, an 11.9 percentage point increase in November, and
a 29.1 percentage point increase in December. One can reject the hypothesis that the three
post-conﬂict coeﬃcients are jointly zero at the one-percent signiﬁcance level. Furthermore, one
can see the large increase in Figure 3, as the share of households reporting their own enterprise
nearly doubled to over 50 percent of the population in Sana’a between October and December.
     At the same time that there was a signiﬁcant increase in the share of individuals who
were residing in households reporting their own enterprise, there is evidence that suggests that
households were turning to self-employment due to a less favorable economic climate rather
than households taking proﬁtable opportunities. First, column (8) of table 4 re-estimates spec-
  27
   See Appendix 6 for estimates of a speciﬁcation that allows the eﬀects of the occupation to vary based on
whether the household head ﬁnished primary school.



                                                    15
iﬁcation (3), but restricts the sample to households that report owning their own enterprise.
The estimates in column (8) demonstrate that the income of self-owned enterprises that re-
ported any income actually declined by December, suggesting that enterprises became even
less proﬁtable as more of them were started. Second, of the self-owned enterprises reported
following the occupation of Sana’a, approximately 87 percent of them reported income that
would not allow the households to reach the national poverty line (162,528 annual Yemeni
Riyals per person).
    In addition to investigating loans and employment, this section further investigates changes
to food consumption following the capture of Sana’a. Table 5 reports estimates from speci-
ﬁcation (3), but uses the share of total expenditure devoted to food and consumption of all
micro and macro-nutrients for which estimated average requirements are reported (Institute of
Medicine of the National Academies 2006). There is little change to the share of expenditure
devoted to food (row 1)- the estimate is positive, but one cannot reject the hypothesis that
there was no change in the share following the capture. However, there is a signiﬁcant drop in
the consumption of nearly all micro and macro nutrients by December (14 out of 18 categories
are statistically diﬀerent from zero at the 10 percent level in December). Interestingly, there
is a decline in calories consumed (row 2) as well as the quality of those calories (rows 3-19).
These estimates would be consistent with households signiﬁcantly struggling during the shock
by December, as calorie consumption has been uncorrelated with a number of large price and
income shocks across a variety of settings (e.g., Brinkman et al. 2009; D’Souza and Joliﬀe
2012; D’Souza and Joliﬀe 2014; etc.).
    Additionally, the timing of the changes in food consumption in table 5 are similar to those
reported in table 2- the magnitude of the estimates are largest in December, and the precision of
the estimates similarly increases as the occupation progresses. Thus, the evidence is consistent
with households trying to maintain their food consumption (through loans and unobservable
means), but at some point being unable to maintain their consumption.

6. Women’s Decision-Making

    One of the advantages of this setting is the thoroughness of the 2014 HBS survey, which
allows one to investigate the changes to a number of intra-household dynamics. As discussed
above, the 2014 HBS also included a very thorough investigation of women‘s decision-making.
Yemen had large inequities in opportunities within the household based on gender prior to the
capture of Sana‘a (e.g., Krishnan et al. 2016), and it is important to understand whether those
with fewer opportunities and resources within the household were forced to bear a signiﬁcantly
larger burden of this shock, as has been shown in other contexts (e.g., Rose 1999; Khanna et
al. 2003; Miguel 2005; D’Souza and Tandon forthcoming; etc.).
    In particular, the survey inquired about the control women had over a number of diﬀerent
types of assets and spending in the household. However, this module was included only in
the 2014 survey, and thus one cannot investigate the diﬀerence-in-diﬀerence comparing the
2014 change to the 2005 change. Thus, this article analyzes the simple diﬀerence in women‘s


                                               16
decision-making in Sana‘a and compares this diﬀerence to the rest of Yemen.
    Speciﬁcally, this section re-estimates speciﬁcations (1) and (2), but uses an indicator equal-
ing one if the spouse or the head of women-headed households made the decision to make
a number of diﬀerent types of purchases alone. Table 6 presents estimates of the women‘s
decision-making speciﬁcations. Each cell represents an estimate of either the change in decision-
making following the capture of Sana‘a or the diﬀerence-in-diﬀerence in Sana‘a relative to the
rest of Yemen. Column (1) restricts the sample to households from the city of Sana’a; column
(2) restricts the sample to households from the rest of Yemen; column (3) reports estimates of
the diﬀerence-in-diﬀerence following the capture of Sana‘a; and column (4) reports p-values of
a test of the diﬀerence-in-diﬀerence estimate in column (3) being diﬀerent from zero, adjusted
for the fact that the estimation is using a number of dependent variables measuring women’s
decision-making.28 Each row utilizes a diﬀerent measure of women‘s decision-making as the
dependent variable.
    Table 6 demonstrates that there was a signiﬁcant reduction in women‘s decision-making
in one of the three types of spending that women actually had a signiﬁcant role in making
decisions (See Appendix 1). Speciﬁcally, row (1) of table 6 demonstrates there was a large
reduction in the share of women that made decisions about food purchases in Sana’a, that did
not occur in the rest of Yemen. The most complete speciﬁcation in column (3) demonstrates
that there was a 16.0 percentage point decline in the share of women who made decisions about
food purchases.
    This change represents over three-quarters of the women who were making decisions regard-
ing food purchases prior to the capture, and the share of women who controlled food purchases
in the city of Sana’a approached and even dropped below levels seen in the rest of Yemen.
This fact is even more striking given the relative lack of power women had in food purchase
decisions in the rest of Yemen.
    The other two areas where women had signiﬁcant control over purchases were clothing for
women and children. The estimates suggest that there was a decline in the control over those
purchases as well, but the estimates are not as precisely estimated. However, one potential
reason for the lower precision might have been the frequency with which those purchases have
to be made. Food has to be purchased at relatively frequent intervals, whereas if women were
not able to leave the household during a tumultuous period, the purchase of women’s and
children clothing could be delayed. Thus, households might not have yet grappled with who
would make purchases in other dimensions that required less frequent purchasing.
    Similarly, the estimates in table 6 demonstrate that the estimated change in control over ex-
penditures was negative for nearly all dimensions (elderly care is the only exception). Although
the estimates are not as precisely estimated as the change in control over food spending, nearly
one-third of the diﬀerence-in-diﬀerence estimates in column (3) are statistically signiﬁcant at
conventional levels.
    Table 7 investigates this change in women’s decision making more thoroughly. First,
 28
      The p-values use the correction used in Anderson (2008).



                                                      17
columns (1)-(2) investigate who might actually be taking over for women in the control of
food purchases. Column (1) demonstrates that there is a signiﬁcant increase in men having
control over the food purchasing decision. Tellingly, the magnitude of the coeﬃcient is roughly
equal to the magnitude of the estimate in the ﬁrst row and third column of table 4. On the
other hand, column (2) demonstrates that there appears to be very little change in other adult
members taking over food purchases in the household. Thus, the change in women’s decision
making in the household appears to be transferred to the male head.
    Alternatively, one might be concerned exactly what the question about decision-making is
capturing. Women’s control over spending decisions and decision-making in general could be
diﬃcult to capture in simple survey instrument, and the responses might not align well with
actual behaviors. However, column (3) presents evidence demonstrating that there did appear
to be a change in behavior that was consistent with the change in responses. Speciﬁcally, the
women’s decision-making module was supposed to be administered to the most senior woman
(whether head, spouse, or otherwise), but in the case that either the spouse was not there or
the head would not allow the spouse to answer, other household members could respond to
the module.
    Following the capture of the city, not only was there a substantial reported transfer of
decision making authority from women to men, there was also a signiﬁcant increase in the
share of women who responded to the module. This is broadly consistent with women being
more likely to be in the household and less able to move about freely following the capture of
the city. Following the capture of the city, the increase in the share of women who answered
the module was 11 percentage points larger in Sana’a than in the rest of Yemen.29 Thus, it
does not appear that the change in decision making reported in table 6 is only describing a
change in peoples’ attitudes with little change in behavior.
    Additionally, there are signiﬁcant diﬀerences between the change in women’s decision mak-
ing and the change in expenditure described in Section 5. Speciﬁcally, column (4) estimates the
changes in decision making but allows the eﬀect to diﬀer by month of the occupation. Unlike
the estimates presented in table 2, there is an immediate drop in women’s decision making
following the capture, and very few households had women continue to control food purchases.
This decline is relatively consistent throughout the entire occupation period analyzed, where
one cannot reject the hypothesis that the change relative to the months prior to the occupation
was identical in each month (p-value 0.332).
    Furthermore, there was little diﬀerence in the change based on women’s bargaining power
inside the household. Using assets owned exclusively by the spouse as a proxy of bargain-
ing power, columns (5) and (6) re-estimate speciﬁcation (1) but restricts the sample to the
approximately 45 percent of the households where the spouse owns gold and the rest of house-
holds where spouse does not own gold. The estimates are qualitatively identical, with both
types of households demonstrating similar declines in the share of women who controlled food
  29
   The baseline estimates described in table 6 are qualitatively identical if the sample is restricted to only
households in which the women answered the module, and also qualitatively identical if women answered the
module not in the presence of a man.


                                                     18
purchases.30
    Similarly, the socio-economic status of households also did not seem to matter to the re-
sponse. Columns (7) and (8) re-estimate speciﬁcation (1), but restrict the sample to those
in which the head ﬁnished and did not ﬁnish primary school respectively. In contrast to the
expenditure changes estimated in table 2, there was little diﬀerence in the change in women’s
decision making between the two types of households.
    All of these results suggest that the decline in women’s decision-making in households in
Sana’a following the capture of the city was nearly universal, and did not depend on either the
status of the household or the status of women in the household. Given the two potential causes
of the change in women’s decision making- a worsening of household resources and a decline
in safety- these patterns are more consistent with the decline in women’s decision making
being driven by a decline in safety. The change was immediate and did not seem to change as
households’ budgets got signiﬁcantly worse in November and December. Furthermore, there
was minimal amounts of variation based household characteristics and the standing of the
spouse in the household. All of these factors similarly point to the initial decline in safety as
opposed to the evolving nature of household budgets.
    Alternatively, table 8 investigates how women’s employment changed as the conﬂict pro-
gressed. As described in the Introduction, a number of news agencies and humanitarian agen-
cies report that women actually have to increase their employment and income generation in
times of conﬂict due to increased need. However, in every potential speciﬁcation investigated
so far, there is no evidence that women actually increased employment. In the baseline speci-
ﬁcation in column (3), the estimate is actually negative (-0.002), and the conﬁdence interval is
relatively small. Thus, the estimate precludes a large increase in employment as the city was
captured, and certainly demonstrates that the potential increase in employment at the bounds
of the conﬁdence interval was signiﬁcantly less than the share of women who lost their ability
to make decisions about food purchases in tables 6 and 7. Furthermore, the estimates broken
up by month in column (4) further demonstrate a pattern similar to that reported in column
(3).

7. Conclusion

    This article demonstrates that in response to a relatively non-violent capture of a city,
expenditure signiﬁcantly declined in a matter of months, households engaged in a number of
coping strategies beginning immediately after the capture, and there were changes to intra-
household dynamics. These ﬁndings are evident even though this setting lacked a number
of aggravating factors that are generally present in longer and more violent conﬂicts, such as
worse disruptions to supply chains (e.g., port blockades) and worse shocks to economic activity
and safety.
    Interestingly, the response of Yemeni households to the fragility analyzed here appears
to be strong when compared to the response to other shocks in Yemen, and shocks in other
 30
      Gold was the only asset that spouses owned in non-negligible numbers.



                                                     19
settings. Although Yemen is an understudied setting, the country has experienced both sharp
and protracted shocks over the past decade that have increased poverty and food insecurity
in the country. These shocks include a long-term issue with joblessness, fallout from the Arab
Srping, and a signiﬁcant increase in food prices in 2008 and 2011. However, the change in
well-being due to these other shocks over the long term appear to be less pronounced than the
short-term shocks analyzed in the current setting (e.g., WFP 2012; Dang and Ianchovichina
2016; World Bank 2017). Furthermore, comparing the capture analyzed here to shocks in other
settings- there is a more precisely estimated decline in calorie consumption, which is relatively
insensitive to shocks, in a matter of three months than was observed in response to the World
Food Price Crisis in Afghanistan and the Global Financial Crisis (e.g., Brinkman et al. 2009;
D’Souza and Joliﬀe 2012; and D’Souza and Joliﬀe 2014).
    However, there are a number of questions unanswered by the current context. First, it is
diﬃcult to infer whether the subset of the population that initially maintained their consump-
tion following the capture were able to do so through coping strategies that were no longer
viable by December, or whether things progressively got worse over the course of the occupa-
tion. Section 5b demonstrates that there were coping strategies that households engaged in
immediately following the capture, and thus some of consumption of households was probably
maintained in this manner. However, the fact that there was a decline in calorie consumption,
and the fact the magnitude of the expenditure changes increased so much by December, it is
diﬃcult to argue that things did not get worse as the occupation progressed. Furthermore, it is
important to note that the survey ended one month before the Houthis signiﬁcantly escalated
began marching on Aden in the south of Yemen.
    Second, these are all short-term changes, and it is unclear what might happen in the
long term. For example, the coping strategies uncovered in this setting, such as loans from
traders, might eventually become unavailable and the level of deprivation might signiﬁcantly
worsen; more sustained conﬂict might lead to more deprivation and cause women to be more
rather than less engaged in household decision making out of necessity; and it is unclear how
households might adapt after the cessation of conﬂict to avoid long-term adverse impacts of
coping strategies (e.g., the speed with which households can improve diets and acquire basic
health services following the conﬂict, etc.).
    Third, as is the case in all conﬂicts, the times surrounding the empirical investigation in
this setting were tumultuous. There were protests against the government, there was a quickly
reversed removal of fuel subsidies prior to the capture of Sana‘a, and there was lingering political
uncertainty from the political transition that occurred in 2012. Thus, it is possible that the
results described here might be context-speciﬁc, and it is possible that other countries might
not experience similar shocks from the capture of territory from the internationally recognized
government. However, the robustness of the empirical ﬁndings suggest that it is unlikely that
these other factors are driving the results.
    And lastly, it is important to emphasize that it is diﬃcult to generalize the results from
this setting to present-day Yemen. As noted before, the conﬂict is more violent, there are


                                                20
substantial impediments to importing food, fuel, and medical equipment that were not present
during the capture of Sana‘a, and the escalation of the conﬂict has been going on for more
than three years. Thus, it is likely that the challenges currently presented to households
might be substantially diﬀerent from those described in Sana‘a in 2014, which could result in
a diﬀerent set of coping strategies (e.g., internal displacement) or place a diﬀerent burden on
disadvantaged household members (e.g. women and girls).




                                              21
References

Abadie, A., and J. Gardeazabal. 2003. “The Economic Costs of Conﬂict: A Case Study of the
Basque Country.” American Economic Review 93 (1), 113-132.

Abadie, A., D. Drukker, J. Herr, and G. Imbens. 2004. “Implementing Matching Estimators
for Average Treatment Eﬀects in Stata.” The Stata Journal 4 (3), 290-311.

Alesina, A., S. Ozler, N. Roubini, and P. Swagel. 1996. “Political Instability and Economic
Growth.” Journal of Economic Growth 1 (2), 189-211.

Anderson, M. 2008. “Multiple Inference and Gender Diﬀerences in the Eﬀects of Early Inter-
vention: A Reevaluation of the Abecedarian, Perry Preschool, and Early Training Projects.”
Journal of the American Statistical Association 103 (484), 1481-1495.

Besley, T., and H. Mueller. 2012. “Estimating the Peace Dividend: The Impact of Violence
on House Prices in Northern Ireland.” American Economic Review 102 (2), 810-833.

Blattman, C., and E. Miguel. 2010. “Civil War.” Journal of Economic Literature 48 (1), 3-57.

Block, S., L. Kiess, P. Webb, S. Kosen, R. Moench-Pfanner, M. Bloem, and C. Timmer. 2004.
“Macro Shocks and Micro Outcomes: Child Nutrition During Indonesia’s Crisis.” Economics
and Human Biology 2 (1), 21-44.

Brinkman, H., S. de Pee, I. Sanogo, L. Subran, and M. Bloem. 2009. “High Food Prices and
the Global Financial Crisis Have Reduced Access to Nutritious Food and Worsened Nutritional
Status and Health.” Journal of Nutrition 140 (1), 153-161.

Dang, H., and E. Ianchovichina. 2016. “Welfare Dynamics with Synthetic Panels: the Case of
the Arab World in Transition.” Policy Research Working Paper 7595, World Bank, Washing-
ton, D.C.

D‘Souza, A., and D. Joliﬀe. 2012. ”Rising Food Prices and Coping Strategies: Household-level
Evidence from Afghanistan.” Journal of Development Studies 48 (2), 282-299.

D‘Souza, A., and D. Joliﬀe. 2014. ”Food Insecurity in Vulnerable Populations: Coping with
Food Price Shocks in Afghanistan.” American Journal of Agricultural Economics 96 (3), 790-
812.

D‘Souza, A., and S. Tandon. Forthcoming. “Intra-Household Nutritional Inequities in Rural
Bangladesh.” Economic Development and Cultural Change.

Edmonds, E., N. Pavcnik, and P. Topalova. 2010. ”Trade Adjustment and Human Capital
Investments: Evidence from Indian Tariﬀ Reform.” American Economic Journal: Applied
Economics 2 (3), 42-75.

Fafchamps, M., C. Udry, and K. Czukas. 1998. “Drought and Saving in West Africa: are
Livestock a Buﬀer?” Journal of Development Economics 55, 273-305.

                                            22
Guidolin, M., and E. La Ferrara. 2007. “Diamonds are Forever, Wars are Not: Is Conﬂict Bad
for Private Firms?” American Economic Review 97 (5), 1978-1993.

Institute of Medicine of the National Academies. 2006. “Dietary Reference Intakes: the
Essential Guide to Nutrient Requirements,” J. Otten, J. Pitzi Hellwig, and L. Meyers, eds.
National Academies Press: Washington, D.C.

Jollife, D., and U. Serajuddin. 2012. “Estimating Poverty with Panel Data, Comparably: An
Example from Jordan.” Policy Research Working Paper 7373, World Bank, Washington, D.C.


Khanna, R., A. Kumar, J. Vaghela, and J. Puliyel. 2003. “Community Based Retrospective
Study of Sex in Infant Mortality in India.” British Medical Journal 327: 126-129.

Klapper, L., C. Richmond, and T. Tran. 2013. “Civil Conﬂict and Firm Performance: Evidence
from Cote D’Ivoire.” Working Paper, the World Bank, Washington, D.C.

Krishnan, N., G. Lara Ibara, A. Narayan, S. Tiwari, and T. Vishwanath. 2016. “Uneven Odds,
Unequal Outcomes.” Working Paper, World Bank, Washington, D.C.

Miguel, E. 2005. “Poverty and Witch Killing.” Review of Economic Studies 72 (4): 1153-1172.

Miguel, E., S. Satyanath, and E. Sergenti. 2004. “Economic Shocks and Civil Conﬂict: An
Instrumental Variables Approach.” Journal of Political Economy 112 (4), 725-753.

Mkutu, K. 2008. “Uganda: Pastoral Conﬂict and Gender Relations.” Review of African Po-
litical Economy 35 (116), 237-254.

Singh, P., and J. Pedersen. 2013. “Copying with Conﬂict: Palestinian Communities Two
Years into the Intifada.” FAFO Report 408, Oslo, Norway.

Rose, E. 1999. “Consumption Smoothing andExcess Femaile Mortality in Rural India.” Review
of Economics and Statistics 81 (1): 41-49.

World Bank. 2012. “World Development Report 2012: Gender Equality and Development.”
Report, World Bank, Washington, D.C.

World Bank. 2017. “Free Falling: Living Standards During Times of Conﬂict in Yemen.”
Working Paper, World Bank, Washington, D.C.

World Economic Forum. 2017. “Global Gender Gap Report 2017.” Report, World Economic
Forum, Geneva.

World Food Programme. 2012. “The State of Food Security and Nutrition in Yemen, 2012.”
Comprehensive Food Security Survey, WFP, Rome, Italy.




                                            23
Table 1. Differences between Control Variables in the Third and Fourth Quarters




                                                                                                                      Difference                     Dependent
                                                                Sampled in                Sampled in              (Quarter 4 - Quarter              Variable: Q4
                                                               Third Quarter            Fourth Quarter                     3)                         Indicator
          Household Size                                           6.897                     6.439                       -0.458                         -0.007
                                                                  [0.301]                   [0.238]                     [0.381]                        [0.007]
          Household Head Married                                   0.899                     0.878                       -0.021                         -0.087
                                                                  [0.021]                   [0.033]                     [0.039]                        [0.104]
          Head Attended School                                     0.696                     0.748                       0.052                          -0.083
                                                                  [0.051]                   [0.043]                     [0.066]                        [0.110]
          Head a Smoker                                            0.384                     0.378                       -0.006                         -0.013
                                                                  [0.051]                   [0.041]                     [0.065]                        [0.062]
          Head Chews Qat                                           0.829                     0.825                       -0.004                          0.012
                                                                  [0.031]                   [0.036]                     [0.047]                        [0.081]
          Years of Schooling- Head                                 7.936                     9.184                       1.248                         0.012*
                                                                  [0.630]                   [0.612]                     [0.873]                        [0.006]
          Household Has a Child                                    0.805                     0.856                       0.051                          0.053
                                                                  [0.027]                   [0.032]                     [0.042]                        [0.085]
          Walls Made of Cut Stone                                  0.306                     0.227                       -0.079                          0.045
                                                                  [0.060]                   [0.046]                     [0.075]                        [0.121]
          Walls Made of Other Stone                                0.509                     0.645                      0.136*                         0.160*
                                                                  [0.062]                   [0.048]                     [0.077]                        [0.096]
          Roof Made of Reinforced
                                                                     0.562                     0.540                        -0.023
          Concrete                                                                                                                                        0.014
                                                                   [0.075]                    [0.058]                      [0.094]                      [0.100]
          Roof Made of Wood and Concrete                            0.268                      0.283                        0.015                           -
                                                                   [0.064]                    [0.056]                      [0.085]                          -
          Roof Made of Mud                                          0.170                      0.177                        0.007                         0.087
                                                                   [0.049]                    [0.055]                      [0.073]                      [0.139]
          Floor Made of Concrete                                    0.232                      0.297                        0.065                       0.279*
                                                                   [0.046]                    [0.051]                      [0.068]                      [0.164]
          Floor Made of Tile                                        0.627                      0.623                        -0.004                        0.242
                                                                   [0.066]                    [0.058]                      [0.087]                      [0.158]
          Household has a Flush Toilet                              0.309                      0.225                        -0.084                       -0.103
                                                                   [0.052]                    [0.045]                      [0.068]                      [0.094]
          P-value of test of all coefficients
                                                                        -                         -                             -                       0.3660
          jointly equaling zero
          Notes: This table reports the average characteristics of surveyed households in the third and fourth quarter in the city of Sana'a in 2014 respectively in
          columns (1) and (2), and further reports that simple difference in column (3). Column (4) reports the estimates of a regression of a quarter 4 indicator on
          all control variables for a sample of households surveyed in the third and fourth quarters, and further reports a joint test of significance. The survey was
          designed such that there should be no differences between households between months of the survey, and thus if the sampling was done according to
          plan, the two estimates should be similar. There were 266 households surveyed in quarter 3 and 231 households surveyed in quarter 4. Estimates are
          weighted so as to be representative of the population, and standard errors are clustered at the PSU level. The statistical significance of the difference is
          reported for column (3). *** denotes statistical significance at the 1 percent level; ** denotes significance at the five percent level; and * denotes
          significance at the 10 percent level.




                                                                                       24
Table 2. Expenditure Changes in Response to the Capture
                      Dependent Variable: ln(Expenditure)




                                                                                                                             Restrict
                       Baseline Specification:                                                                              Sample to         Restrict
                      How Much Larger Effects                                                                                 More           Sample to
                       were in Sana'a Relative        Single         Including Pre-          Food           Non-food        Educated       Less Educated
                        to the Rest of Yemen        Difference      Conflict Trends       Expenditure      Expenditure        Heads            Heads
                           _____________            ________        ____________          _________        __________      __________       __________
                          (1)            (2)            (3)         (4)         (5)           (6)              (7)             (8)               (9)
    Post-Capture
      Months
 October Indicator
                                                                      -
 x Sana'a Indicator      -0.154        -0.041         0.016                   -0.083         -0.012           -0.092          0.160           -0.361**
                        [0.202]        [0.195]        [0.162]                 [0.233]        [0.253]         [0.232]         [0.282]          [0.171]
 November
 Indicator x Sana'a
 Indicator               -0.219        -0.177         -0.093         -        -0.220         -0.140           -0.234          -0.220          -0.396**
                        [0.179]        [0.156]        [0.128]                 [0.202]        [0.223]         [0.202]         [0.200]          [0.166]
 December
 Indicator x Sana'a
 Indicator             -0.399***      -0.359***     -0.301***        -       -0.402**       -0.410**        -0.407**        -0.364**         -0.554***
                         [0.153]        [0.128]       [0.099]                 [0.182]        [0.192]         [0.183]         [0.167]           [0.175]
    Post-Capture
      Months
 August Indicator           -
 x Sana'a Indicator                       -              -                                                                       -                -
                                                                   0.019      -0.001            -                -
                                                                  [0.209]     [0.210]
 September
 Indicator x Sana'a         -             -              -                                                                       -                -
 Indicator                                                        -0.115      -0.129            -                -
                                                                  [0.207]     [0.209]
 Control Variables         N              Y              Y           Y           Y              Y               Y               Y                 Y
 P-value of Test of
 all Post-Capture
                         0.076*        0.039**      0.0142**         -         0.106        0.004***         0.012**         0.098*           0.012**
 Coeff. Equaling
 Zero
 P-value- Test of
 Pre-Trend Coeff.           -             -              -         0.779       0.767            -                -               -                -
 Equaling Zero
 Observations           4317              4317           504        2,189       4,317           4,317           4,317           2,681             1,636
 Notes: This table estimates the change in expenditure following the capture of the Yemeni capital by rebels. Columns (1) and (2) estimate the
 baseline specification comparing the change in Sana'a to the rest of Yemen; column (3) estimates the simple difference in Sana'a alone; columns (4)
 and (5) estimate the change in expenditure in Sana'a relative to the rest of the country leading up to the capture; columns (6) and (7) analyze food
 and non-food expenditure respectively; and columns (8) and (9) estimate the effects separately for households where the head did and did not
 finish primary school respectively. All specifications exclude households surveyed during the month of Ramadan- July. All specifications include all
 lower-order terms and control variables. Control variables include all variables reported in Table 1 and an urban indicator. Standard errors
 clustered at the PSU level are reported in parentheses, and all estimates are weighted so as to be representative of the entire population. ***
 denotes statistical significance at the 1 percent level; ** denotes significance at the five percent level; and * denotes significance at the 10 percent
 level.




                                                                                  25
Table 3. Changes in Loans



                                            Dependent Variable:

                                                                                                                                                          Indicator
                                             Number           Number of           Number                                                   Number          for any
                                             of Loans         Loans from          of Loans         Indicator for         ln(Amount         of Loans         Loan         ln(Amount
                                               from              Other              from          any Loan from           Owed to            from          from a         Owed to
                                             Traders            Sources           Traders            a Trader             Traders)         Traders         Trader         Traders)
                                                (1)               (2)                (3)                (4)                  (5)              (6)            (7)             (8)
      Post-Capture Months
 October Indicator x Sana'a
 Indicator                                    0.166*             -0.036          0.304***            0.272***             2.849***           0.057          0.083           1.085
                                              [0.097]           [0.164]           [0.100]             [0.091]              [0.924]          [0.115]        [0.107]         [1.204]
 November Indicator x Sana'a
 Indicator                                     0.038             -0.053            0.237*              0.237*              2.628*           -0.004          0.010           0.041
                                              [0.083]           [0.129]           [0.134]             [0.127]              [1.431]          [0.084]        [0.073]         [0.833]
 December Indicator x Sana'a
 Indicator                                     0.286             0.066             0.777               0.108                1.293            0.177          0.152           1.624
                                              [0.221]           [0.138]           [0.626]             [0.156]              [1.716]          [0.129]        [0.115]         [1.133]
      Post-Capture Months
 August Indicator x Sana'a
 Indicator                                     -0.027            0.001             -0.075              -0.050              -0.480           -0.019          -0.012          -0.221
                                              [0.054]           [0.167]           [0.047]             [0.036]              [0.389]          [0.069]        [0.048]         [0.546]
 September Indicator x Sana'a
 Indicator                                     0.040             -0.003            0.099               0.070                0.631           -0.023          -0.005          -0.133
                                              [0.065]           [0.192]           [0.122]             [0.064]              [0.711]          [0.060]        [0.050]         [0.559]
 Restrict Households to Those
 Where Household Head Did                         N                 N                 Y                   Y                    Y               N               N               N
 Not Finish Primary School
 Restrict Households to Those
 Where Household Head                             N                 N                 N                   N                    N                Y              Y               Y
 Finished Primary School


 P-value of Test of all Post-                  0.245             0.876           0.005***            0.009***             0.008***           0.511          0.534           0.432
 Capture Coeff. Equaling Zero
 Observations                                  4,317             4,317             1,636               1,636                1,636            2,681          2,681           2,681
 Notes: This table estimates the change in the use of loans following the capture of the Yemeni capital by rebels. Columns (1)-(2) utilize the entire sample; columns (3)-(5)
 utilizes only households where the head has not finished primary school; and columns (6)-(8) utilizes only households where the head has finished primary school. All
 specifications exclude households surveyed during the month of Ramadan (July). All specifications include all lower-order terms and control variables. Control variables include
 all variables reported in Table 1 and an urban indicator. Standard errors clustered at the PSU level are reported in parentheses; all estimates are weighted so as to be
 representative of the entire population. *** denotes statistical significance at the 1 percent level; ** denotes significance at the five percent level; and * denotes significance at
 the 10 percent level.




                                                                                            26
Table 4. Changes in Employment



                                                 Dependent Variable:

                                                                 Indicator
                                                                if anybody
                                                                   in the
                                                                Household         Indicator       Indicator                         Hours
                                                                Worked in           if the         if non-        Hours the           non-
                                                 ln(wages         the Past       Household          Heads         Household         Heads          Indicator
                                                    from           Week             Head           Worked           Head            Worked            for            ln(Total
                                                  work in        (Wage or        Worked in          in the        Worked in          in the         Owning         Profit from
                                                  the past      non-Wage           the Past          Past          the Past           Past            an               Own
                                                   week)           Work)            Week            Week            Week             Week         Enterprise       Enterprises)
                                                     (1)             (2)              (3)             (4)            (5)               (6)            (7)               (8)
       Post-Capture Months
 October Ind. x Sana'a Ind.
                                                   -0.532          -0.012           -0.008          -0.074           3.402           0.139          0.170*             0.479

                                                  [0.368]         [0.057]          [0.083]          [0.105]         [4.849]        [16.347]         [0.092]           [0.650]
 November Ind. x Sana'a Ind.
                                                 -1.004**          -0.033           -0.012           0.006           0.315           6.774           0.119             0.378

                                                  [0.438]         [0.085]          [0.101]          [0.117]         [5.073]        [12.815]         [0.086]           [0.471]
 December Ind. x Sana'a Ind.
                                                   -0.359          0.034            0.110          -0.172*           6.261          -12.031        0.291***          -1.143*

                                                  [0.339]         [0.044]          [0.088]          [0.103]         [4.832]        [13.951]         [0.075]           [0.610]
       Post-Capture Months
 August Ind. x Sana'a Ind.                         -0.278          -0.046           -0.080          -0.041           5.240          -14.618          0.028            -0.720
                                                  [0.350]         [0.056]          [0.086]          [0.087]         [9.017]        [10.824]         [0.073]           [0.535]
 September Ind. x Sana'a Ind.                      -0.127          -0.038           -0.036           0.032           2.501          -11.692          0.022            -0.075
                                                  [0.428]         [0.054]          [0.079]          [0.089]         [6.747]        [13.064]         [0.072]           [0.554]
 Restrict to Households where
 Head Did Not Finish Primary                          N               N                Y               Y                Y               N               N                N
 School
 Restrict to Households where
 Head Finished Primary School                         N               N                N               N               N                Y               Y                 Y



 P-value of Test of all Post-Capture
                                                   0.133           0.698            0.496            0.306           0.532           0.494         0.002***            0.131
 Coeff. Equaling Zero
 Observations                                      2,458           4,317            4,317            4,317           4,317           4,317           4,317              644
 Notes: This table estimates the change in employment and wages following the capture of the Yemeni capital by rebels. Columns (1) utilizes the sample of households that
 report any wage income in the past week; columns (2)-(7) utilize the entire sample; and column (8) utilizes the sample of households that report income from a self-owned
 enterprise. All specifications exclude households surveyed during the month of Ramadan (July). All specifications include all lower-order terms and control variables.
 Control variables include all variables reported in Table 1 and an urban indicator. Standard errors clustered at the PSU level are reported in parentheses; all estimates are
 weighted so as to be representative of the entire population. *** denotes statistical significance at the 1 percent level; ** denotes significance at the five percent level; and
 * denotes significance at the 10 percent level.




                                                                                            27
Table 5. Changes in Nutrient Consumption



                                              Post-Capture Months                                                                 Pre-Capture Months
                                _____________________________________________                                              ______________________________
                         October              November            December                                            August                September
                         Indicator x          Indicator x         Indicator x                                         Indicator x           Indicator x
                         Sana'a               Sana'a              Sana'a                                              Sana'a                Sana'a
                         Indicator            Indicator           Indicator                                           Indicator             Indicator
 Share of
 Expenditure
                             0.006          [0.006]         0.010          [0.007]         0.002          [0.007]         0.001          [0.005]          0.008          [0.012]
 Devoted to
 Food
 ln(Calories)                0.018          [0.224]         -0.146         [0.245]        -0.292*         [0.156]         -0.053         [0.194]         -0.067          [0.221]

 ln(Choline)                -0.095          [0.142]         -0.106         [0.155]       -0.265**         [0.122]         -0.011         [0.144]         -0.061          [0.143]

 ln(Protein)                -0.011          [0.135]         -0.145         [0.132]       -0.271**         [0.128]         0.011          [0.126]         -0.107          [0.148]

 ln(Vitamin A)              -0.007          [0.169]         -0.035         [0.153]         -0.153         [0.156]         0.060          [0.102]          0.028          [0.123]
 ln(Vitamin C)               0.085          [0.197]         -0.122         [0.259]         -0.192         [0.182]         -0.041         [0.158]         -0.019          [0.189]

 ln(Vitamin E)              -0.097          [0.263]         -0.350         [0.281]         -0.425         [0.283]         -0.159         [0.236]          0.036          [0.265]

 ln(Thiamin)                 0.023          [0.181]         -0.297         [0.199]        -0.324*         [0.169]         0.036          [0.171]         -0.148          [0.199]

 ln(Riboflavin)              0.013          [0.155]         -0.169         [0.158]       -0.292**         [0.137]         -0.004         [0.140]         -0.086          [0.159]

 ln(Niacin)                 -0.001          [0.164]         -0.252         [0.173]        -0.313*         [0.160]         0.067          [0.164]         -0.115          [0.175]

 ln(Vitamin B6)             -0.099          [0.215]         -0.222         [0.192]       -0.370**         [0.173]         0.062          [0.178]         -0.100          [0.176]

 ln(Folate)                  0.009          [0.150]         -0.231         [0.167]        -0.251*         [0.148]         0.028          [0.148]         -0.106          [0.176]

 ln(Vitamin B12)            -0.217          [0.220]         -0.251         [0.208]         -0.221         [0.213]         -0.088         [0.147]         -0.099          [0.180]

 ln(Copper)                 -0.004          [0.205]         -0.182         [0.211]        -0.286*         [0.162]         0.072          [0.172]         -0.075          [0.195]
 ln(Iron)                    0.044          [0.209]         -0.197         [0.219]        -0.314*         [0.166]         0.049          [0.175]         -0.109          [0.207]

 ln(Magnesium)              -0.052          [0.249]         -0.269         [0.242]       -0.412**         [0.197]         0.046          [0.212]         -0.130          [0.217]

 ln(Phosphorous)            -0.037          [0.197]         -0.213         [0.192]       -0.359**         [0.172]         0.062          [0.183]         -0.093          [0.180]

 ln(Selenium)               -0.015          [0.200]         -0.226         [0.182]       -0.386**         [0.174]         0.079          [0.186]         -0.125          [0.183]

 ln(Zinc)                    0.007          [0.228]         -0.145         [0.239]       -0.349**         [0.175]         0.056          [0.202]         -0.079          [0.215]
 Notes: This table estimates the change in consumption of micro and macro-nutrients following the capture of the Yemeni capital by rebels. Each row reports estimates
 from a separate regression analyzing different nutrients. All specifications include the entire sample, and exclude households surveyed during the month of Ramadan
 (July). All specifications include all lower-order terms and control variables. Control variables include all variables reported in Table 1 and an urban indicator. Standard
 errors clustered at the PSU level are reported in parentheses; all estimates are weighted so as to be representative of the entire population. *** denotes statistical
 significance at the 1 percent level; ** denotes significance at the five percent level; and * denotes significance at the 10 percent level.




                                                                                           28
Table 6. Control Over Household Expenditure by Women




                                                                                   Coefficient on Conflict      Coefficient Yielding How much larger the        P-Value Adjusted
                                            Coefficient on Conflict Indicator     Indicator for Households       change in Sana'a was than the Rest of            for Multiple
                                             for Households in Sana'a Only          in the Rest of Yemen                          Yemen                            Inference
                                                      __________                        ___________                             _________                       ______________
 Dependent Variable:                                       (1)                               (2)                                    (3)                                (4)
 Women Responsible for Purchase of
 Food                                                  -0.134***                            0.021                              -0.160***                             .058*
                                                         [0.041]                           [0.021]                               [0.056]
 Women Responsible for Purchase of
 Heads Clothes                                           -0.015                             0.007                                -0.025                               .448
                                                         [0.019]                           [0.015]                               [0.025]
 Women Responsible for Purchase of
 Womens' Clothes                                         -0.060                             0.008                                -0.070                               .513
                                                         [0.083]                           [0.034]                               [0.089]
 Women Responsible for Purchase of
 Childrens' Clothes                                      -0.043                             0.005                                -0.048                               .448
                                                         [0.044]                           [0.022]                               [0.051]
 Women Responsible for Purchase of
 Medicine for Spouse                                    -0.040*                           0.014**                               -0.051**                              .169
                                                         [0.023]                           [0.007]                               [0.025]
 Women Responsible for Purchase of
 Medicine for Other Adult Women                          -0.019                            -0.000                                -0.022                               .439
                                                         [0.019]                           [0.006]                               [0.020]
 Women Responsible for Purchase of
 Medicine for Children                                  -0.040**                            0.002                               -0.042**                              .169
                                                         [0.018]                           [0.007]                               [0.020]
 Women Responsible for Expenditure
 on Son's Marriage                                       -0.018                             0.001                                -0.019*                              .225
                                                         [0.011]                           [0.003]                               [0.011]
 Women Responsible for Expenditure
 on Daughter's Marriage                                  -0.010                             0.005                                -0.019                               .225
                                                         [0.007]                           [0.004]                               [0.011]
 Women Responsible for Expenditure
 on Boys' Education                                      -0.012                            -0.000                                -0.017                               .409
                                                         [0.008]                           [0.009]                               [0.013]
 Women Responsible for Expenditure
 on Girls' Education                                     -0.019                             0.000                                -0.025                               .225
                                                         [0.012]                           [0.008]                               [0.015]
 Women Responsible for Expenditure
 on Elderly Care                                          0.003                            -0.007                                 0.014                                .73
                                                         [0.027]                           [0.023]                               [0.040]
 Women Responsible for Taking on
 Debt                                                    -0.010                            -0.002                                -0.010                               .676
                                                         [0.019]                           [0.007]                               [0.019]
 Observations                                              503                              3,789                                 4,292


 Notes: This table estimates how women's decision-making ability changed following the capture of Sana'a. Each cell reports coefficients from a separate regression.
 Column (1) reports the simple difference in the share of households where women were responsible for the decision described in each row; column (2) reports the same
 specification but restricts the sample to the rest of Yemen; and column (3) presents estimates by how much larger the change was in the city of Sana'a than in the rest of
 Yemen. Column (4) reports p-values for the coefficient of interest adjusted for multiple inference as described in Anderson (2008). The difference-in-difference
 specification in column (3) includes all lower-order terms, and all specifications include control variables. Control variables include all variables reported in Table 1 and an
 urban indicator. Standard errors clustered at the PSU level are reported in parentheses, and estimates are weighted so as to be representative of the entire population.
 *** denotes statistical significance at the 1 percent level; ** denotes significance at the five percent level; and * denotes significance at the 10 percent level.



                                                                                            29
Table 7. A Closer Look at Decision-Making in the Household



                                                     Dependent Variable:
                                                     Indicator if      Indicator        Indicator
                                                       Head is         if Neither       if Spouse
                                                     Involved in         Spouse         Answered
                                                      the Food         nor Head         Decision-
                                                      Purchase             are           Making
                                                       Decision         Involved         Module               Indicator if Woman Responsible for Food Purchases
                                                      ________          _______          _______                ________________________________________
                                                         (1)               (2)              (3)              (4)           (5)       (6)          (7)        (8)
 Post-Conflict Indicator*Sana'a
 Indicator                                             0.155**            0.005         0.110***               -          -0.180*        -0.136***        -0.131**        -0.181**
                                                        [0.069]          [0.042]          [0.042]              -           [0.093]         [0.052]         [0.059]         [0.071]
 October Ind. x Sana'a Ind.                                 -                -                 -           -0.139              -               -               -               -
                                                            -                -                 -           [0.092]             -               -               -               -
 November Ind. x Sana'a Ind.                                -                -                 -         -0.178**              -               -               -               -
                                                            -                -                 -           [0.087]             -               -               -               -
 December Ind. x Sana'a Ind.                                -                -                 -           -0.077              -               -               -               -
                                                            -                -                 -           [0.089]             -               -               -               -
          Pre-Capture Months
  _____________________________
 August Ind. x Sana'a Ind.                                                                                 0.030
                                                                                                           [0.122]
 September Ind. x Sana'a Ind.                                                                              0.063
                                                                                                           [0.105]
 Restrict Sample to Those Where
                                                           N                N                N                N               Y               N               N                N
 Spouse Owned Gold
 Restrict Sample to Those Where
                                                           N                N                N                N               N               Y               N                N
 Spouse Did Not Own Gold
 Restrict Sample to those where Head
                                                           N                N                N                N               N               N                Y               N
 Did Complete Primary School
 Restrict Sample to those where Head
                                                           N                N                N                N               N                N              N                Y
 Did Not Complete Primary School
 P-value of Test of Coefficients on
 October and December Higher-Order                          -                -                 -           0.332               -               -               -               -
 Terms Being Equal
 Observations                                            4292             4292             4292             4292            1000            3292            2668            1624
 Notes: This table investigates a number of different specifications better describing the change in women's decision making following the capture of Sana'a. Each coefficient
 represents how much larger the change in the dependent variable was in Sana'a than the rest of Yemen in 2014. Column (1) utilizes an indicator equaling one if the household
 head was involved in the food purchase decision; column (2) utilizes an indicator equaling one if neither the head nor the spouse is involved in the decision to purchase food;
 column (3) utilizes an indicator equaling one if the spouse (or head if the household head was a women) responded to the decision-making module; and columns (4)-(9) utilize an
 indicator if the spouse (or head if the household head was a woman) was in charge of the decision to purchase food. Columns (1)-(4) include the whole sample, while columns (5)-
 (8) restrict households based on measures of women's bargaining power and socioeconomic status. All specifications exclude households surveyed during the month of Ramadan
 (July). All specifications include all lower-order terms and control variables. Control variables include all variables reported in Table 1 and an urban indicator. Standard errors
 clustered at the PSU level are reported in parentheses; all estimates are weighted so as to be representative of the entire population. *** denotes statistical significance at the 1
 percent level; ** denotes significance at the five percent level; and * denotes significance at the 10 percent level.




                                                                                          30
Table 8. Changes in Women’s Employment in Response to the Capture



                           Dependent Variable: Indicator Equaling One if Spouse is Employed
                                                                                                                                       Restrict Sample           Restrict Sample
                                          Restrict                                                Restrict           Restrict           to Women in               to Women in
                           Restrict        Sample                                                Sample to          Sample to            Households                Households
                            Sample         to Rest        Difference-          Estimate           Women              Women               where Head              where Head Did
                              to              of         in-Difference        Change by          who Own           who Do Not            Completed                Not Complete
                            Sana'a         Yemen         Specification          Month              Gold             Own Gold           Primary School            Primary School
                           _______        _______        ___________         __________         __________         __________          ____________              _____________
                              (1)            (2)               (3)                (4)               (5)                (6)                    (7)                       (8)
 Post-Conflict
 Indicator                   0.006          0.009             0.008                 -               0.007              0.010                 0.006                    0.012**
                            [0.025]        [0.006]           [0.006]                -              [0.012]            [0.006]               [0.008]                   [0.006]
 Post-Conflict
 Indicator*Sana'a
 Indicator                      -              -             -0.002                 -               0.032             -0.017                 -0.003                    0.008
                                -              -             [0.025]                -              [0.048]            [0.027]               [0.033]                   [0.020]
 Oct Ind. x Sana'a
 Ind.                           -              -                 -               -0.045                -                  -                     -                         -
                                -              -                 -              [0.036]                -                  -                     -                         -
 Nov Ind. x Sana'a
 Ind.                           -              -                 -               0.065                 -                  -                     -                         -
                                -              -                 -              [0.054]                -                  -                     -                         -
 Dec Ind. x Sana'a
 Ind.                           -              -                 -               -0.039                -                  -                     -                         -
                                -              -                 -              [0.031]                -                  -                     -                         -
    Pre-Capture
      Months
 Aug Ind. x Sana'a
 Ind.                           -              -                 -               -0.020                -                  -                     -                         -
                                -              -                 -              [0.035]                -                  -                     -                         -
 Sep. Ind. x Sana'a
 Ind.                           -              -                 -               0.005                 -                  -                     -                         -
                                -              -                 -              [0.034]                -                  -                     -                         -
 Observations                  504          3,813             4,317              4,317              1,000              3,317                 2,681                     1,636
 Notes: This table investigates a number of different specifications describing the change in women's employment following the capture of Sana'a. Columns (1) and (2) estimate a
 simple change in employment following the capture of Sana'a in the city of Sana'a and the rest of Yemen respectively. Columns (3)-(8) estimate how much larger the change in
 women's employment was in Sana'a than the rest of Yemen in 2014. Columns (1)-(4) utilize the entire sample, and columns (5)-(8) restrict households based on measures of
 women's bargaining power and socioeconomic status. All specifications exclude households surveyed during the month of Ramadan (July). All specifications include all lower-
 order terms and control variables. Control variables include all variables reported in Table 1 and an urban indicator. Standard errors clustered at the PSU level are reported in
 parentheses; all estimates are weighted so as to be representative of the entire population. *** denotes statistical significance at the 1 percent level; ** denotes significance at
 the five percent level; and * denotes significance at the 10 percent level.




                                                                                          31
Figure 1. Changes in Commodity Prices in Sana’a Following the Capture




                                                     Percentage Change in Food Prices


                       0.4                                                                                    Fresh Food Items and Fuel
                      0.35                                                                                    Non-Perishable Food Items'

                       0.3

                      0.25

                       0.2

                      0.15

                       0.1

                      0.05

                         0

                     -0.05

                      -0.1
                                          Percentage Change in 2014                             Percentage Change in 2012-2013

                  Notes: These figures show average commodity prices in Sana’a markets in 2014, and the average for the two years prior to the
                  capture of the capital. These figures were obtained from the monthly database maintained by the World Food Programme.




                                                                                32
Figure 2. Monthly Confidence Intervals of Control Variables

 Figure 2a- Summary Statistics by Month
 1) Household Size                                              2) Household Head Married
  9                                                                    1
                                                                 0.95
  8
                                                                  0.9
  7                                                              0.85
                                                                  0.8
  6
                                                                 0.75
  5                                                               0.7
           Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec                  Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
                                                                                               Lower Bound- 95% CI
               Lower Bound- 95% CI        Upper Bound- 95% CI
                                                                                               Upper Bound- 95% CI
               Pre-Conflict Mean                                                               Pre-Conflict Mean

 3) Household Head Never Attended School                        4) Household Head a Smoker
      1.2                                                        0.7
       1                                                         0.6

      0.8                                                        0.5
                                                                 0.4
      0.6
                                                                 0.3
      0.4
                                                                 0.2
      0.2                                                        0.1
       0                                                           0
             Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec               Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
                                     Lower Bound- 95% CI                                          Lower Bound- 95% CI
                                     Upper Bound- 95% CI                                          Upper Bound- 95% CI
                                     Pre-Conflict Mean                                            Pre-Conflict Mean

 5) Household Head Chews Qat                                    6) Years Schooling
   1.1                                                           13

       1                                                         11

   0.9                                                            9

                                                                  7
   0.8
                                                                  5
   0.7
                                                                  3
   0.6                                                                     Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
             Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
                              Lower Bound- 95% CI                              Lower Bound- 95% CI       Upper Bound- 95% CI
                              Upper Bound- 95% CI                              Pre-Conflict Mean
                              Pre-Conflict Mean
 Source: 2014 Household Budget Survey, Author’s Calculation




                                                                             33
Figure 2b- Summary Statistics by Month
7) Household has a Child                                       8) Household has a Flush Toilet
    1                                                            0.6
  0.9
                                                                 0.4
  0.8
  0.7
                                                                 0.2
  0.6
  0.5                                                                 0
           Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec                 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
                                                                                               Lower Bound- 95% CI
              Lower Bound- 95% CI        Upper Bound- 95% CI                                   Upper Bound- 95% CI
              Pre-Conflict Mean                                                                Pre-Conflict Mean

9) Walls Composed of “Cut Stone”                               10) Walls Composed of Other Material
   0.8                                                                1

                                                                 0.8
   0.6
                                                                 0.6
   0.4
                                                                 0.4
   0.2                                                           0.2
                                                                           Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
       0
                                                                                             Lower Bound- 95% CI
           Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
                                                                                             Upper Bound- 95% CI
                               Lower Bound- 95% CI                                           Pre-Conflict Mean
                               Upper Bound- 95% CI
                               Pre-Conflict Mean

11) Roof Composed of Reinforced Concrete                       12) Roof Composed of Wood and Concrete
   1                                                            0.8
 0.8
                                                                0.6
 0.6
 0.4                                                            0.4
 0.2
                                                                0.2
   0
           Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec       0
                                                                          Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
                                  Lower Bound- 95% CI                                        Lower Bound- 95% CI
                                  Upper Bound- 95% CI                                        Upper Bound- 95% CI
                                  Pre-Conflict Mean                                          Pre-Conflict Mean
Source: 2014 Household Budget Survey, Author’s Calculation




                                                                           34
Figure 2c- Summary Statistics by Month
13) Roof Composed of Mud                                         14) Floor Composed of Concrete
  0.6                                                             0.6


  0.4                                                             0.4


  0.2                                                             0.2


                                                                    0
    0
                                                                        Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
         Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
                                                                                               Lower Bound- 95% CI
                                 Lower Bound- 95% CI                                           Upper Bound- 95% CI
                                 Upper Bound- 95% CI                                           Pre-Conflict Mean

                                             15) Floor Composed of Natural Tiles
                                       1

                                     0.8

                                     0.6

                                     0.4

                                     0.2
                                           Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
                                                               Lower Bound- 95% CI
                                                               Upper Bound- 95% CI
                                                               Pre-Conflict Mean
Source: 2014 Household Budget Survey, Author’s Calculation




                                                                             35
Figure 3. Expenditure and Coping Strategies by Month

 Household Expenditure                                                               Share of Expenditures Devoted to Food
                                                                                        0.2
    6000000
                                                                                       0.18

                                                                                       0.16
    5000000
                                                                                       0.14

                                                                                       0.12
    4000000
                                                                                        0.1

                                                                                       0.08
    3000000
                                                                                       0.06

                                                                                       0.04
    2000000

                                                                                       0.02

    1000000                                                                                 0
                  Jun         Jul         Aug     Sep     Oct     Nov     Dec                    Jun      Jul     Aug        Sep    Oct     Nov     Dec

                    Lower Bound- 90% CI                 Upper Bound- 90% CI                            Lower Bound- 90% CI         Upper Bound- 90% CI

                    Sana'a                              Rest of Yemen                                  Sana'a                      Rest of Yemen



 Share of Individuals Residing in Households with a Self-Owned                       Share of Individuals Residing in Households where Women
 Enterprise                                                                          have Control over Food Expenditures
  0.6                                                                                 0.6


  0.5                                                                                 0.5


  0.4                                                                                 0.4


  0.3                                                                                 0.3


  0.2                                                                                 0.2


  0.1                                                                                 0.1


    0                                                                                  0
            Jun         Jul         Aug         Sep      Oct      Nov     Dec                   Jun      Jul     Aug         Sep    Oct     Nov      Dec

                    Lower Bound- 90% CI                 Upper Bound- 90% CI                            Lower Bound- 90% CI         Upper Bound- 90% CI
                    Sana'a                              Rest of Yemen                                  Sana'a                      Rest of Yemen

 Source: 2014 Household Budget Survey, Author’s Calculation




                                                                                36
Figure 4. Coefficient Estimates and Confidence Intervals from the Baseline Specification by Month




                     0.4


                     0.2


                       0
                                  August               September               October       November         December

                    -0.2


                    -0.4


                    -0.6


                    -0.8

                                       Coefficient Estimate             95% CI-Lower Bound       95% CI- Upper Bound

                  Notes: Author’s calculations using the 2014 Household Budget Survey.




                                                                               37