Multidimensional Poverty Assessment of Internally Displaced Persons in Iraq

Decades of conflict have contributed to high flows of internal displacement in Iraq. The incidence of these flows on the welfare of internally displaced persons is not well understood. This paper attempts to fill this gap in the literature by investigating the link between internal displacement and multidimensional poverty, using one of the most comprehensive household surveys for poverty analysis in Iraq. The results show crucial differences between internally displaced and non-displaced households with respect to multidimensional poverty. Furthermore, instrumental variable regression analysis suggests that the relationship is causal, that is, the probabilities of multidimensional and monetary poverty are higher because of internal displacement.


Introduction
Iraq has a long-standing history of internal displacement. Before the Syrian refugee crisis beginning in 2010, Iraqi internally displaced populations (IDPs) and refugees were the world's second-largest forcibly displaced population, after Afghanistan (Doocy et al., 2009). As of 2008, the number of IDPs was estimated at 2.8 million (IOM, 2008). However, the number of hundreds of thousands of Iraqis were forcibly displaced due to the war (Cockburn, 2007;Stewart, 2007;Steele, 2008). The bombing of the Samarra Al-Askari mosque in February 2006 set off a new surge of violence and insecurity, causing a massive flow of people -third phase.
Indeed, the peak of pre-ISIS displacement did not occur right after the US-led invasion but between 2006 and 2008, after the bombing of the Al-Askari mosque (Refugee Studies Center, 2012). The final phase of displacement marks the Daesh insurgency period spanning from 2014 to 2018. The post-2014 period represents one of the most important episodes of internal displacement in Iraq (IOM, 2016). However, the current paper focuses on internal displacement before 2012 due to the use of household data that were collected between 2012 and 2013.
Unlike refugees, IDPs are not covered by any protection within an internationally recognized legal framework. The IDPs in Iraq benefit from little support from the Iraqi government through targeted policies unless they volunteer to return to their original location. The Ministry of Displacement and Migration (MoDM), formed in 2003, has implemented new policies towards displacement in July 2008, mostly aiming at promoting the return of IDPs. However, most IDPs wished to settle in and integrate with their host community, mainly because of security concerns in their place of origin and better opportunities at their destination (IOM, 2013).
Evidence shows IDPs live in precarious conditions. The majority of settlements are self-built overcrowded houses or illegally occupied houses, with limited access to basic services and poor infrastructure, which may lead to poor health outcomes (IOM, 2013). Besides, IDPs might have difficulties finding a formal job due to lack of an identity document, language or cultural barriers. Additionally, forced displacement may have damaging consequences on IDPs' family structure. For instance, family members may be separated, with working-age adults, and especially males, unable to return to their place of origin. Displacement may also disrupt children's schooling. This suggests IDPs may be more likely to be deprived in multiple dimensions of well-being; which calls for an in-depth analysis of the deprivation of IDPs to identify the dimensions that demand urgent intervention and to help design policies that are best-suited to the needs of IDPs.
The variation of exposure to violence and insecurity between IDPs and non-IDPs provides a unique quasi-experimental setting, with the war representing an exogenous shock to multiple components of household well-being. The dimensions may be analyzed separately, or combined into a single index of multidimensional poverty, enabling a consolidated analysis of multiple dimensions of deprivation. In the following analysis, we apply the Multidimensional Poverty Index (MPI) methodology developed by Alkire and Foster (2011) 1

to the 2012-2013 Iraqi
Household Socio-Economic Survey (IHSES-II) 2 to investigate the effect of forced displacement on multidimensional poverty.
Although there have been numerous works on the macro-level implications of wars and conflicts (e.g. Collier, 1999; Barron et al., 2004;Collier and Hoeffler, 2004;Miguel et al., 2004), 1 See also Alkire and Santos (2010). 2 This survey is an outcome from the cooperation between the Central Organization for Statistics of Iraq and the Kurdistan Region Statistics Organization supported by the World Bank, which initiated in 2007 with the first IHSES. The IHSES-II is an update of the first round with additional data and information such as individuals' past migration history. micro-level empirical research on the impact of conflict, and forced displacement in particular, is scant. However, in recent years there has been a growing interest in the literature on how displaced households or individuals are affected (Loaiza et al., 2018), with much of the empirical evidence from African countries (Kondylis, 2008;Ssewanyana et al., 2007). Conflictrelated household displacement has been negatively associated with consumption (Ibáñez and Vélez, 2008;Fiala, 2015), wages and employment (Morales, 2018;Calderón and Ibáñez, 2009), asset ownership (Fransen et al., 2017) and food security (Verwimp and Muñoz-Mora, 2017).
This paper intends to contribute to this growing field of literature by exploring the case of a Middle Eastern country, one of the most understudied regions in this empirical literature.
The rest of the paper is organized as follows. In the next section, we discuss various methodologies of poverty measurement. In the third section of the paper, we present our empirical strategy and the results. We conclude the paper in section four.

Poverty measurement 2.1 Monetary vs. Multidimensional Poverty measurements
Sen (1981) distinguishes between two main methods for measuring poverty: the direct and indirect methods. The direct method captures how deprived individuals are with respect to a set of specific basic needs, rights or functioning. The indirect method shows whether individuals' wealth falls below a certain level that can enable them to meet their basic needs, also known as the poverty line. The indirect method essentially uses monetary poverty measurements such as income, consumption or expenditure; which has the advantage of providing a more straightforward identification of poor individuals.
However, there are some basic limitations to this method, which can be encapsulated in the inability of monetary measurements to fully convey the extent of deprivation, or at least some of the essential needs. 3 Moreover, income can be thought of as means to valuable ends and not 3 Those basic limitations are mainly connected to comparability issues. For example, the accuracy of the indirect method might be compromised by the fact that people may be exposed to different prices; attaining the poverty an end in itself. The ability to convert income into "functionings" also varies from one individual to another, i.e., even when faced with the same price, people's needs and their ability to convert resources into "functionings" might differ, which is why it is preferable to measure the achievements directly. Furthermore, the monetary approach is not suited to capture the value of basic services not obtained through the market such as health, education, and water. Nonmonetary indicators may also better identify those who are unable to fully participate in their society due to a lack of resources (Alkire and Santos, 2014). Hence, the use of the direct method has spread in response to these limitations. Multidimensional poverty measurements, for instance, represent variants of the direct method that have increasingly gained the attention of analysts and policy makers despite some criticisms on the way they are built.
Once it is acknowledged that poverty is multidimensional, it raises several practical questions like which poverty dimensions to consider and how to weigh them. There is no consensus on this question in the poverty literature. The construction of multidimensional poverty measures is also complicated by the fact that numerical indicators may not capture the subjective experience of deprivation and that the dimensions of deprivation themselves may be changing across space and over time. This limits the ability of an analyst to construct a universal measure for all social groups for all time periods. But a measure that fulfills all the desirable theoretical properties would come at the cost of simplicity; it would be difficult to convey the relevant information to policy makers. Therefore, scalar indices that aggregate the information from multiple dimensions of deprivation into one quantitative indicator have been widely used (Alkire and Foster, 2011;Maasoumi and Lugo, 2008). Scalar indices are particularly useful for ranking various population groups such as countries or regions (Ferreira and Lugo, 2013 individual-level data to identify individuals who are deprived in multiple dimensions (Alkire and Santos, 2014). Recently, the World Bank (WB) also proposed a global multidimensional poverty measure, which differs from the MPI in crucial ways. Most importantly, the World Bank's approach to multidimensional poverty measurement is informed by the living standard perspective (The World Bank, 2018). Insomuch as standard monetary measures exclude critical components of welfare, supplementing them with non-monetary components gives a fuller picture of well-being. Our analysis also includes the World Bank multidimensional poverty measure for comparison. We use the MPI methodology proposed by OPHI and the World Bank as they are widely used and they incorporate dimensions and indicators that are globally comparable.

MPI definition
The initial step to estimating multidimensional poverty involves defining a set of indicators that belong to specific poverty dimensions, to which deprivation cutoffs are assigned in the second step. There are three equally weighted poverty dimensions considered in the global MPI analysis: education, health, and standard of living. Ten indicators are defined across these three dimensions, in line with the Millennium Development Goals (MDGs). These indicators are also equally weighted within each dimension as shown in Figure 1. 4 4 The definitions of the indicators are provided in Table B1 in Appendix B. Appendix C explains how the weights are shifted across indicators when the necessary information is missing. Source: OHPI, 2015 The MPI methodology follows the mathematical structure of one measure of acute multidimensional poverty as proposed by Alkire andFoster (2007, 2011): the adjusted headcount ratio or the Mo measure. To build the Mo measure, it is necessary to go through a process that involves multiple steps. First, there is an evaluation of the household's standing relative to each indicator deprivation cutoff. Then the household's deprivations are weighted by the indicator weights. A household is considered MPI poor if the total of its weighted deprivations is 33 percent or more of all possible deprivations. See Santos and Alkire (2011) for a detailed technical description of the MPI, its dimensions, and the indicators.

MPI methodology
The MPI methodology also follows the two main steps for measuring poverty as conceptualized by Sen (1976): identification of poor individuals and the aggregation of the poverty information.
One of the basic requirements of the MPI is to set a cutoff zi for each indicator. A cutoff is intended as the minimum level of attainment associated with each indicator. Assuming xi is the achievement of individual i in an indicator, that person is deprived in that indicator if xi falls below the cutoff (xi<zi.). The next step is to determine the weight of each indicator. The three dimensions of the MPI are equally weighted so that each dimension weight is assigned a value of 1/3. Within each dimension, the indicators p are assigned equal weights wp. The health and education dimensions have two indicators each, therefore each indicator's weight within these dimensions is 1/6. As for the living standard dimension, each of its six indicators receives a weight of 1/18. The weights are used to calculate each individual's deprivation score ci,, a measure of total weighted deprivation which is obtained by a weighted sum of deprivations as follows: (2) In the Alkire-Foster methodology, a poverty cutoff (k) is also defined. It is the threshold of total weighted deprivations above which a household is considered to be poor so that ci ≥ k. The value of k is set to 1/3.
Following the identification of deprived individuals, it is possible to compute the headcount ratio (H), which is the share of individuals experiencing multidimensional deprivation over the total population. The second measure we consider is the Multidimensional Poverty Index (MPI) or the M0 measure, also known as the adjusted headcount ratio. It is the product of the headcount ratio Hand the intensity of multidimensional poverty A: 3. Methodology 3.1 Determinants of household deprivation

The binomial logistic model
To assess the relationship between multidimensional poverty and forced displacement, we formulate the following logit model: Consider that a household j is poor (Y=1) if its deprivation score is 1/3 or more, or non-poor (Y=0) if its deprivation score is below 1/3. The state of deprivation is determined by factors depicting household forced displacement status along with a set of household socio-economic and demographic indicators. These factors are grouped into a vector X so that: Where Xj = Fj + Hj (5) with Yj * being the underlying latent variable capturing the poverty index of household j; Fj represents the displacement dummy, and Hj other socio-economic and demographic indicators; uj is the stochastic error term; β is a column vector of parameters.
We assume the cumulative distribution of uj is logistic (Greene, 1993) and therefore follows a logit model. We posit the probability of being poor as: where β is the raw vector of parameters and ⍺ is a scalar. The logit function to be estimated by the maximum likelihood method is expressed as follows: The natural log of the odds of the household to be MPI poor is the logit variable Dj. The maximum likelihood method used for the estimation of equation (7) does not impose any assumption on the normality or homoskedasticity of errors in the predictors.
The marginal effect of an explanatory variable Xj on the probability of being poor is given by: where βk is the kth element of β.

The general instrumental variable (IV) strategy
From the relationship in equation (6) it is difficult to draw inference regarding the causal effect of forced displacement on the probability of a household being in multidimensional poverty.
Indeed, one could argue that the causality runs in the other directions, i.e., poor households are more likely to be displaced. Also, there may be unobservable factors that affect household deprivation while at the same time serving as the push factors for the household's movement.
For instance, restrictive policies by the state towards certain regions could create uprisings and sectarian violence in these regions and simultaneously contribute to spreading vulnerability and deprivation. These policies might then be indirectly correlated with forced displacement through their strong correlation with conflict and violence. Hence, these issues must be accounted for to avoid attributing any effect of these unobservable factors on household poverty to their forced displacement status. To account for potential endogeneity, we add dummy variables for the interaction between origin and destination governorate fixed effect. 5 Besides, we would need an instrument that explains household forced displacement without being correlated with household deprivation status.
Given that Iraq has a long history of conflict, it is a challenging exercise to find a credible instrument. Nevertheless, we use the Iraq Body Count (IBC), a unique publicly available fatality database compiling detailed civilian casualties from violence and conflict since 2003, to instrument for the probability of forced displacement. These data mainly come from media sources. The total fatality per governorate determines the intensity of exposure to conflict and as such represents a potentially good instrument to the probability of being forcibly displaced at the time of, or right after the violent outbreak. We assume there is no correlation between household current multidimensional poverty status and the intensity of violence at their governorate of residence at the time they migrated for two reasons: first, the dimensions and indicators of multidimensional poverty are less sensitive to any change caused by violence intensity, which may not be the case for monetary poverty; and second, the time lag between violence and the survey period would significantly mitigate any potential direct effect of violence intensity on multidimensional poverty. where Y denotes our dummy for the household deprivation status; Xe is the dummy for the household displacement status which is our endogenous regressor; Xo is the vector of observable exogenous regressors; Xu is the unobservable variable which is correlated with Xe but not correlated with Xo; W is our instrumental variable (total fatalities in the household's governorate of origin two months 6 prior to the migration date); β and α are the vectors of parameters to be estimated; e is the random error term.
In the first-stage regression, we apply a non-linear least squares (NLS) to equation (9) that enables us to obtain a consistent estimate of the parameter α ( �) and compute the residual as follows: In the second stage, we consistently estimate β with an NLS regression on the following equation: Wheree 2SRI is the error term, which is different from e since has been replaced with the residual � in the outcome regression.

Variable definitions
Our main explanatory variable -annotated IDH (Internally Displaced Household) -is a dummy variable taking value 1 when there is at least one IDP in the household and 0 otherwise. 7 Our initial hypothesis is that households with IDPs are poorer. In addition, we add a set of controls capturing households' demographic and geographical characteristics, whose definition and summary statistics are reported in Table 1.
In an alternative model specification, we disentangle the effect of internal displacement based on the three major historical phases of forced displacement. These phases are proxied by variables Pre03IDH, O3_05IDH, and Post05IDH 8 and we expect them to be positively violence outbreak, we assume a two-month time window provides a good balance time and space wise if violence and so its impact were to spread geographically with time. 7 The distribution of IDPs within IDHs is shown in Figure B1 in Appendix B.
8 Note that there are only 53 IDHs with two or more IDPs who have migrated in more than one phase. For these households, only the latest phase will be considered. correlated with household poverty status. Furthermore, we control for several household and household-head characteristics in the regression. 9

Descriptive statistics
In the present analysis, we use the data gathered in the second round of the IHSES-II which is one of the most comprehensive surveys that provides information on household characteristics and living conditions in Iraq. The fieldwork was conducted between January 2012 and February 2013. The final data cover 25,146 households across 18 governorates of Iraq organized into 2,832 clusters for a total of 176,042 individuals. The respondent in each household is usually the head of the household or an authorized family member, although there are also questions on household members' characteristics. One important feature of these data is the availability of information on household members' migration history. It is this information that enables us to identify IDPs. These are people who have lived in another governorate -or their current governorate of residence but a different environment (rural or urban) -for an uninterrupted period of six months, and who have moved from that previous place for one of the following reasons: -Forced displacement: Individuals who have been deported or forced out of their usual place of residence.
-Conventional armed conflict: People who lived close to battlefronts.
-Civil conflict: People who lived in areas where there is internal insecurity.
According to these criteria, 5,939 IDPs are identified in 2,027 IDHs, 12 for a total of 14,402 people living in these households, including IDPs. It is necessary to account for the latter group in our analysis to get an insight into the impact of forced displacement on the whole IDH unit. Table 2 shows some basic descriptive characteristics of the IDPs in our sample. We also add statistics on all IDH members, since they are the ones on which our analysis will be based. 13 Besides, we include figures on non-IDPs to understand how their characteristics may differ from IDPs rather than for comparison. From the first row, we can see the average age of IDPs is 35 while that of all IDH members is considerably lower (23) and comparable to the one of the non-IDPs population which is 22. This suggests on average IDPs are adults from the working-age population group. Most of the IPDs are male -unlike IDH members and non-IDPs which have a slightly higher proportion of females -and most have not completed the primary level of education. 14 A large majority of IDPs, IDHs members, and non-IDPs get food ration distributed under the public distribution system. In general, the main reason for leaving their place of origin is a general state of insecurity or targeting for persecution. 12 We adopt a broad definition of an IDH, intended as a household unit hosting at least one IDP; since we assume vulnerability embedded in the forced displacement of one member to be a shock with potential externalities to all other members. However, one could think of different approaches to this definition. For instance, a household unit where the head of household is an IDP or more simply, a household unit where all members are IDPs. Following these definitions would reduce our IDH sample, to 1,863 according to the first definition and to 289 as per the second definition. Furthermore, we run the risk of omitting the potential externalities to the entire household and therefore underestimating the real impact of forced displacement on household deprivation status. 13 The IDH may include children born after their parents' migration.
14 However, IDPs have more education than non-IDPs.   Next, we compare the characteristics of IDHs and non-IDHs. Table 4 shows that while both groups are similar in several indicators, they are statistically different on other dimensions as indicated by the results from the t-test 17 in the last column. Displaced households, on average, are slightly older, have an older household head who is also less educated, have lower dependency ratio, and are more likely to be in urban areas. One observation we can draw from these statistics is that internal displacement may not be exogenous to some of the household innate characteristics. These characteristics may drive any empirical relationship we may find between our main variables of interest. Hence, it would be necessary to account for this potential endogeneity with an IV analysis. To get some more insight into the difference in deprivation patterns between IDHs and non-IDHs, we conduct a comparative analysis across dimensions, indicators, and regions. This is discussed in detail in Appendix A. To get further insight into the latter result, we compute the average marginal effect as the partial derivative of the non-linear probability function. We obtain a marginal effect of 0.017, meaning that on average the probability of falling below the poverty line is 1.7 percent higher for IDHs.

Results
This supports our initial argument that IDHs are more vulnerable. Households with highly educated head of household tend to be less deprived. Unsurprisingly, the correlation between multidimensional poverty and per capita consumption is negative. The association between share of working-age men and women is positive and negative respectively. In a context where women's labor force participation is very low and there are low social expectations of women working, a larger share of women in the household working may indicate working by necessity rather than choice.

IV strategy to identify the impact of forced displacement
In this section, we present the findings from the models formulated to correct for potential endogeneity. The results are presented in Table 6. Since our instrumental variable is only available from 2003 onwards, our sample size is reduced to 23,983. 20 Column (1) displays outcomes from the auxiliary regression of the two-stage residual inclusion (2SRI) model where we instrument for the IDH dummy with the IBC variable -two months fatalities count 21 -and run a generalized linear model (GLM) regression. Results from the first regression show a positive correlation between the total fatality at IDPs' governorates of origin -two months prior 20 We are left with 865 IDHs of the 2,027 IDHs in our full sample. 21 We did some robustness checks using total fatalities 1 month and 3 months prior to migration and the results did not change much -see Tables D3 and D4 in Appendix D. to migration -and household displacement status. The marginal effect of this variable shows the probability of being displaced increases by 0.1 percentage point per additional person killed.
In the second stage, we introduce the residual Xu(hat) in the MP dummy model and run a GLM regression. Results from the second-stage regression are reported in column (2); the point estimate on the IDH dummy is positive and significant. On average, the probability of being deprived is 8.2 percent higher for IDHs, which is much higher than the 1.7 percent in the logit model in Table 5.
Proceeding further, we implement an IV analysis using a Generalized Method of Moments (GMM) model regression, whose results are displayed in column (3). Again, the estimate for the IDH dummy is positive and significant, indicating IDHs are 3 percent more likely to be poor.
The coefficients in columns 2 and 3 in Table 6 are not strictly comparable to those in Table 5 because  Table 5. This negates the argument that the difference in the sample could be responsible for the difference in the relationship between household displacement status and multidimensional poverty between logit and IV regressions. However, it is still possible that this result does not generalize to those displaced before 2003, so we cannot draw a similar conclusion on pre-2003 IDHs as our data do not allow for a wider time coverage.
Columns (5) and (6) show the regression results for monetary poverty from the second stage of the 2SRI and GMM regressions respectively. In column (5), the marginal effect of the IDH dummy on the monetary poverty dummy returns a value of 0.068, meaning on average IDHs have a 6.8 percent higher probability of falling below the poverty line. From the GMM model in column (6) we get a lower marginal effect of 3.9 percent. Both coefficients from columns (5) and (6) are much higher than the 1.4 percent marginal effect in Table 5.
In the last two columns, we perform similar IV analysis with the WB MP dummy. Results from the 2SRI and GMM regressions point to a positive effect of internal forced displacement on that alternative indicator of multidimensional poverty. The second stage of the 2SRI regression returns a marginal effect of 0.071 as displayed in column (7), while the GMM model gives a marginal effect of 0.029 as we can see from column (8). 22   Appendix A: Iraqi household deprivation patterns: IDHs vs. non-

IDHs' members
Following the Alkire-Foster methodology, we compute basic multidimensional poverty indicators for our sample population. Table A1 reports some key measures of the MPI methodology for our sample population, but also those measures are broken into IDH, Pre03IDH, 03_05IDH, Post05IDH and non-IDH members. The headcount ratio or the incidence of poverty (H) for the total population is 7.55, which means 7.55 percent are MPI poor or in acute poverty -also shown in Figure B2 in Appendix B. When distinguishing between IDH and non-IDH members, that percentage does not change much for the latter, while Overall, the analysis by IDHs' and non-IDHs' members suggests forced displacement status differences in general poverty measures are not compelling. However, if we limit ourselves to these broad poverty measures, we are likely to miss some insightful elements that might point to relevant deprivation differences across these two cohorts. That is why further MPI analysis is required. Hence, we proceed to examine the incidence of deprivation in each of the MPI indicators; which is intended as the proportion of the population that is poor and deprived in each indicator, also known as censored headcount ratios. The bar chart in Figure A1 shows these ratios for IDHs' and non-IDHs' members. 23 The top two indicators in which more MPI poor people are deprived are school attendance and sanitation. The proportions of poor people with respect to each of these two indicators are around 6 to 7 percent. School attendance appears on the top of the chart for IDHs' members as the indicator in which more MPI poor people are deprived. For non-IDHs' members, it is sanitation that drives the most deprivation. The dirt floor is the third indicator in which poor IDHs are concentrated; while for non-IDHs it is not having access to clean water. In general, there is no big difference in the proportion of MPI poor people across the two groups -the only notable differences are with the water, floor, and cooking fuel indicators. What differs is the ranking of each of these indicators, which is also a signal of their importance in the deprivation of poor people. Therefore, doing such analysis would help set 23 Figures for the total population are depicted in Figure B3 in Appendix B. We will not comment on those figures here as they are pretty similar to the ones shown in Figure A1 for non-IDHs' members.
forth priorities for operations and policy intervention towards specific groups, IDPs in particular.

Figure A1. Censored headcount ratios
When accounting for each indicator's weight, we get a slightly different picture than above as shown in Figure A2. School attendance appears as the indicator with the highest contribution to the total deprivation of poor people in both IDHs and non-IDHs. 24 Its weight is more than as twice the rate of the second indicator in the chart of contributors to deprivation -nutritionirrespective of an individual's displacement status. However, school attendance has a higher percentage contribution of 38.3 percent in the deprivation of poor IDHs as compared to only 36.9 percent for poor non-IDHs' members. This indicates child education as the major driver of the poor's MPI, is more important for IDHs than for non-IDHs. Similarly, nutrition weight is also higher in the deprivation of poor IDHs -16.0 percent -than in one of poor non-IDHs -24 Again here, we do not comment on the indicators' contribution for the total sample population but report these numbers in Figure B4 in Appendix B as they are very close to what we get for non-IDHs. 13.9 percent. One striking observation from all the above is that the two major contributors to poverty account for over half of poor people's total deprivation. We get a higher percentage of 54.3 percent for IDHs and 50.8 percent for non-IDHs. Moreover, these two major contributors belong to two separate poverty dimensions: education and health. All this points to the necessity of adapting policy interventions to each specific group's needs while considering the intensity of deprivation they face. It is also critical to avoid focusing on one single sector, but rather favor an inclusive approach that will incorporate key dimensions of poverty into poverty eradication plans. Another notable point we can draw from Figure A2 is the third position of sanitation and its weights in the deprivation of poor IDHs and non-IDHs -12.0 percent and 12.8 percent, respectively. Although this indicator's assigned weight is only 1/6 of the total deprivation score, it emerges as one of the major drivers of the intensity of deprivation of the poor. And again here, it belongs to a different poverty dimension than the first two poverty contributors.

Figure A2. Contribution of each indicator to overall poverty
In further analysis, we disentangle multidimensional poverty at the governorate level. This is depicted in the maps in Figures A3 and A4 for the incidence of poverty by the governorate. In Indicators' percentage contribution to the deprivation of the multidimensionally poor IDH non_IDH general, the bulk of the poor is localized in the south and center of Iraq, in governorates such as Muthanna, Wasit and Maysan; although there are some notable differences between the two maps. Some governorates with relatively low poverty headcount ratios for non-IDHs, in fact, host a sizeable share of IDHs' poor -like for instance Najaf, Diyala and Babylon.
Paradoxically, there does not seem to be a correlation between hosting a high share of poor       The IHSES data were not specifically designed for an Alkire-Foster type of multidimensional analysis; hence the data questions and output might not perfectly fit into the scope of each indicator as defined by the MPI methodology. In the latter, the specification of each indicator's deprivation cutoffs is very precise. Our data only provide this information for the past 12 months, causing some potential underestimation of deprivation with respect to the child mortality indicator. Besides, anthropometrics information has only been recorded for a randomly selected subsample -three households per cluster -of the original sample. This means that information is only available for about 34% of our sample. Therefore, in line with Santos and Alkire (2011) recommendations we apply the following strategy as an attempt to account for this limitation: -When we cannot observe the deprivation status about one of the health dimension indicators -nutrition or child mortality -for certain household types -for instance due to the absence of the applicable member in the household -the remaining indicator within the same dimension receives the entire dimension weight of 1/3.
-When there are no children under 5 in the household and no other household member's anthropometric was recorded, we cannot observe the nutritional status of any member.
Therefore, child mortality gets the full dimension weight of 1/3.