Welfare measures and the composition of the bottom decile: The example of gender and extreme poverty in South Africa

ABSTRACT

We examine the composition of the bottom decile in South Africa using three alternative measures of socio-economic status (SES): an asset index, household income per capita and household expenditure per capita. We show that the gender composition of the bottom decile is sensitive to the measure used. We discuss possible reasons for these discrepancies, highlighting gender differences in asset ownership and location. This has implications for the use of asset indices for identifying the poorest members of society.

KEYWORDS:

• Poverty measurement
• asset indices
• income
• expenditure
• gender

1. Introduction

This paper highlights discrepancies between alternative measures of socio-economic status (SES) in the tail of the distribution. Specifically, it examines the gender composition of the bottom decile in South Africa, using alternative measures of SES: asset ownership, income and expenditure. Overall, asset indices have been found to correspond well with monetary measures such as expenditure (Filmer & Pritchett, 2001, Filmer & Scott, 2012). However, little is known about assets on the one hand, and income and expenditure measures on the other, at the tail of the SES distribution.

The paper is organised as follows. The next section briefly reviews the literature. Section 3 discusses the methods we use to investigate the composition of the bottom decile. We discuss our results in section 4. We show that the gender composition of the bottom decile is sensitive to the measure used to define the bottom decile. We then examine some possible reasons for the discrepancy between rankings based on the asset index and monetary measures. Finally, we discuss some implications for the use of asset indices for assessing socio-economic status at the tail of poverty.

2. Literature review

Asset indices are commonly used as a measure of socioeconomic status. However, limited attention has been given to the robustness of rankings based on asset indices versus other measures. The research that has investigated the degree of correspondence between asset indices and income and expenditure measures, such as Filmer & Scott (2012), suggests that the welfare measure used does not have a large effect on the relationship between economic status and health, education and labour market outcomes. Moreover, it suggests that asset indices yield similar results regardless of the method used to construct them. The focus has, however, been on correspondence across the whole distribution, or with regards to classifying individuals or households as poor. Our analysis focuses on the correspondence at the tail, a topic that has received little attention.

Previous research has looked at the implications of urban-rural differences in asset holdings and their implications for household welfare rankings and inequality measures based on asset indices. Wittenberg & Leibbrandt (2017) find that differences in the assets owned by rural and urban households can make a difference to household welfare rankings. They show that predominantly rural assets such as livestock can have a negative weighting in the asset index, implying that households that own only these assets would be ranked lower than households that have no assets. This means that asset indices may inflate urban-rural differences in wellbeing (Wittenberg & Leibbrandt, 2017). This point is similar to one we will raise, in the sense that it demonstrates that patterns of asset accumulation differ by sub-group, making comparisons across sub-group difficult.

There is a substantial literature on gender differences in asset holdings (Deere & Doss, 2006). However, to our knowledge there is no published research on the implications of gender differences in asset holdings and accumulation for the robustness of rankings based on asset indices versus other measures. Posel et al. (2016) investigate the implications of the use of adult equivalence scales (AES) for gender differences in income and poverty profiles. They find that the use of AES significantly narrows the gender income gap, but does not make a significant difference to the proportion of women among African adults living in poverty. Rogan (2016) compares poverty rates in male-headed and female-headed households using the multidimensional poverty index (MPI) and income measures, and finds that female-headed households face a higher poverty differential using either measure, but that the poverty differential is larger for income poverty than for multidimensional poverty, suggesting that relying on income poverty may overstate the poverty differential. This research is closely related to our paper, but it differs in that we compare the gender profile of the bottom decile based on an asset index versus income and expenditure per capita.

3. Methods

We investigate whether the composition of the poorest decile differs depending on the measure of socio-economic status used, i.e. asset-, income- or expenditure-based measures, with a particular focus on the gender composition of the bottom decile as an example of the discrepancies that may arise when using a single measure. We also investigate whether the composition is sensitive to the construction of the asset index, income and expenditure measures. After showing that there are discrepancies depending on the measure used, we examine the characteristics of the bottom decile to explore possible reasons for these differences.

We use the 2014 National Income Dynamics Survey (NIDS). We focus on the most recent wave of NIDS currently available because it provides relatively recent data and allows us to explore how the factors associated with membership of the poorest decile differ depending on the measure used. However, we also use NIDS Wave 1, the Income and Expenditure Survey (IES) and the General Household Survey (GHS) to confirm the robustness of the patterns we observe.

NIDS is a nationally representative longitudinal survey of individuals and their households. There are four waves to date, collected between 2008 and 2014–15. We use Wave 4 (2014–15) (SALDRU, 2016). All our analyses are weighted using the post-stratified weight variable [w4_wgt]. This weight is the design weight (correcting for non-response and reshuffling of household membership) calibrated to gender-age-race population totals using the Statistics South Africa 2015 mid-year population estimates (Chinhema et al., 2016).

The poverty headcount rate in South Africa in 2015 was 40 percent using the lower-bound poverty line, and 25 percent living in extreme poverty (Stats SA, 2017). However, we opted to focus on the bottom decile, in order to consider the very poorest in society. Similar gender patterns to the ones described below are present in the bottom 20 percent, bottom 30 percent, bottom 40 percent and bottom 50 percent, but become progressively weaker as one moves up the distribution (see annex Table A1).

To assess the gender composition of the bottom decile, and how this varies by the measure used, we divide the proportion of men in the bottom decile for a given measure by the proportion of men in the total population. A ratio larger than 1 implies what we will refer to as ‘overrepresentation’, while a ratio smaller than 1 implies what we refer to as ‘underrepresentation’.

NIDS includes data on ownership of a list of consumer durable assets: a radio, hi-fi/CD or MP3 player, television, satellite dish, DVD player, computer, camera, cell phone, electric stove, gas stove, paraffin stove, microwave, fridge/freezer, washing machine, sewing machine, lounge suite, motor vehicle, motorcycle/scooter, bicycle, non-motorised boat, motor boat, donkey/ox cart, plough, tractor, wheelbarrow, and grinding mill. These assets were used to create our asset index, with the ownership of more assets taken as an indicator of wealth. The weight attached to ownership of each asset was determined by principal components analysis (PCA). This method was popularised by Filmer & Pritchett (2001) as a way of creating wealth indices from asset ownership data in the absence of reliable income or expenditure data. The first principal component is used to create the asset index. It is the linear combination of the asset variables that captures the most variation shared by all the variables (Filmer & Pritchett, 2001).

Household income per capita and household expenditure per capita are calculated by taking the total household monthly income or expenditure and dividing it by the number of household residents. Total household income refers to household monthly income minus taxes, and includes imputed rental income from owner-occupied housing. Household expenditure refers to total household expenditure in the 30 days preceding the interview. Like income, it includes imputed rent. Certain components of income and expenditure were imputed to correct for item non-response (Chinhema et al., 2016).

To investigate whether the gender composition of the bottom decile is sensitive to the method used to construct the asset index, and income and expenditure measures, we used several other construction methods for comparison.

The asset index may be sensitive to which variables are included in the index, as well as to the method used to construct the index. PCA is common in the literature, but there is no consensus over which assets should be included. Some authors include only ownership of consumer durable assets in the index, while others add variables such as housing characteristics, electricity and water source. It has been argued that multiple correspondence analysis (MCA) is more appropriate for constructing asset indices using categorical variables (Booysen et al., 2008). For this reason, we create several additional asset indices to assess the robustness of our findings. Firstly, we use housing, energy and water characteristics in addition to consumer durable assets. Secondly, we use MCA in addition to PCA. In line with the literature we do not adjust the asset index for household size. The majority of durable assets typically included in asset indices are public goods at the household level (Filmer & Scott, 2012). In other words, multiple household members can benefit from the assets simultaneously, so it is not necessary to adjust for household size.

Income and expenditure per capita are crude measures that do not take into account economies of scale within the household (a two-person household does not need double the resources of a single-person household) or the fact that children generally need less food than adults. This has implications for rankings based on income or expenditure per capita measures. For example, if certain types of households have more children on average, income or expenditure rankings using the crude per capita measure may make them seem worse off (Posel et al., 2016). Adult equivalence scales can be used to adjust for economies of scale and the fact that children require different resources to adults. There are a number of adult equivalence scales in use, calculated using a variety of methods and with varying degrees of adjustment for economies of scale and the different needs of adults and children. The available literature suggests that the choice of equivalence scale and the parameters chosen makes little difference to overall poverty rates in South Africa. However, in some cases the choice does affect the ranking of various demographic groups (Streak et al., 2009, Deaton & Paxson, 1997). We use two alternative scales – the ‘original’ and ‘new’ OECD adult equivalence scales – to test the sensitivity of the gender composition of the bottom decile of household income and expenditure to these factors. The ‘original’ OECD scale assigns a weight of 1 to the first household member, 0.7 to each additional adult, and 0.5 to each child. The ‘new’ OECD scale assigns a weight of 1 to the first household member, 0.5 to each additional adult, and 0.3 to each child.

As outlined in the following section, we find that the gender composition of the bottom decile is sensitive both to whether an asset index, income or expenditure is used to identify the bottom decile, and to the construction of these measures. To explore possible reasons for this, we examine the characteristics of the bottom decile defined using these measures. We use various descriptive statistics to examine the composition of the bottom decile as measured using an asset index, household income per capita and household expenditure per capita. We look at differences by various demographic and household characteristics: gender, employment status, education level, geographic location, household size, gender of household head, presence of an adult female in the household, grant access in the household and the average number of assets owned by households. We then used logistic regression to investigate the factors associated with membership of the bottom decile by these different measures. We controlled for a set of demographic and household variables: gender, being in a traditional area, age (allowing for a non-linear relationship), household size, grant access in the household, and gender of the household head.

4. Results

4.1. Rankings are sensitive to the measure used

The composition of the bottom decile is sensitive to the choice of SES measure used to determine membership. Whether one uses household income per capita, expenditure per capita or an asset index as a welfare measure paints a different picture of the poorest. Table 1 shows that men are underrepresented (ratio below 1) in the bottom deciles of household income per capita and household expenditure per capita. However, men are slightly overrepresented (ratio above 1) in the bottom decile of the asset index.

Table 1. Ratio of share of adult males in bottom decile to share in whole population.

	Asset index	Household income per capita	Household expenditure per capita
Male	1.09 (52)	0.72 (35)	0.78 (37)
P-value (H0: ratio = 1)	0.005	0.000	0.000

Note: Percentages in parentheses. Proportion of men in sample treated as the population value.

To ensure that this pattern was not unique to the NIDS Wave 4 dataset, we also looked at the gender composition of the bottom decile by these three measures in several other South African datasets. Table 2 shows that similar patterns were found in NIDS Wave 1, the Income and Expenditure Survey 2010/11 and the General Household Survey 2014 (SALDRU, 2016, Statistics South Africa, 2012, 2015).

Table 2. Distribution of males in the bottom decile in various datasets.

	Bottom decile of asset index	Bottom decile of household income per capita	Bottom decile of household expenditure per capita*
NIDS Wave 4
Male	1.04	0.85	0.87
NIDS Wave 1
Male	1.04	0.88	0.92
IES 2010/11
Male	1.11	0.98	0.93
GHS 2014
Male	1.06	0.95	1.07

*Note: In the GHS the expenditure variable was categorical rather than continuous. The figures for the GHS thus refer to households with an expenditure per capita of less than R800, which made up 9% of the population – a rough approximation of the bottom decile.

The ranking using an asset index is sensitive to the method of construction of the index. Males are overrepresented in the bottom decile using the original asset index (using only consumer durable assets) (Table 3). However, if one constructs an asset index including certain housing/energy/water characteristics in addition to durable assets, using PCA or multiple correspondence analysis (MCA), men are slightly but insignificantly underrepresented in the bottom decile (Table 3).

Table 3. Sensitivity of gender distribution of adults in bottom decile to construction of asset index.

	Asset index (PCA)	Asset index including dummies for specific housing/energy/ water characteristics (PCA)	Asset index including all categories of housing/energy/ water characteristics (MCA)
Male	1.09 (52)	0.98 (47)	0.98 (47)
P-value (H0: ratio = 1)	0.005	0.553	0.354

Note: Percentages in parentheses. Proportion of men in sample treated as the population value.

The composition of the bottom decile of expenditure is sensitive to the use of adult equivalence scales. Males are underrepresented in the bottom decile of household expenditure per capita, but are overrepresented after adjusting using the original and new OECD adult equivalence scales (see Table 4). The picture that emerges using income data is not substantially changed by the use or choice of weights. The pattern becomes slightly less strong but men are still underrepresented in the bottom decile defined by household income.

Table 4. Ratio of adult males in the bottom decile by various measures to adult males in whole population.

	Household income per capita	Household income per capita using original OECD AES	Household income per capita using new OECD AES	Household expenditure per capita	Household expenditure per capita using original OECD AES	Household expenditure per capita using new OECD AES
Male	0.72 (35)	0.78 (37)	0.82 (39)	0.78 (37)	1.09 (53)	1.08 (52)
P-value (H0: ratio = 1)	0.000	0.000	0.000	0.000	0.002	0.015

Note: Percentages in parentheses. Proportion of men in sample treated as the population value.

4.2. Who is in the bottom decile?

Table 5 shows that several characteristics are common to those in the bottom of deciles of the asset index, income and expenditure. Those in the bottom deciles of the asset index, household income and household expenditure per capita are more likely to live in traditional areas. This is particularly the case for the bottom household income and expenditure per capita deciles, and even more so for those who are in both the bottom asset index decile and bottom household income per capita decile.

Table 5. Characteristics of the bottom decile.

	Whole pop.	Bottom decile of asset index	Bottom decile of household income per capita	Bottom decile of household expendi-ture per capita	Bottom decile of both asset index and house-hold income per capita
Adult male*	1 (31)	1.09 (34)	0.52 (16)	0.59 (18)	0.56 (17)
Adult female*	1 (33)	0.92 (31)	0.91 (30)	0.92 (31)	0.93 (31)
Child*	1 (36)	1.00 (36)	1.50 (54)	1.42 (51)	1.45 (52)
Male-headed household	1 (45)	0.92 (42)	0.58 (26)	0.66 (30)	0.65 (30)
Adult female in household	1 (89)	0.90 (81)	1.09 (97)	1.11 (99)	1.07 (96)
Has matric (25 and over)	1 (42)	0.41 (18)	0.41 (17)	0.42 (18)	0.21 (9)
Employed	1 (46)	0.82 (38)	0.30 (14)	0.47 (22)	0.31 (15)
Traditional	1 (34)	1.60 (55)	1.77 (61)	1.91 (65)	2.06 (70)
Urban	1 (62)	0.61 (38)	0.59 (36)	0.51 (32)	0.39 (24)
Grant access in household	1 (61)	1.02 (62)	1.25 (77)	1.34 (82)	1.04 (64)
Age (mean)	28	27	20	22	20
Household size (mean)	5	5	7	9	7
Asset count (mean)	8	2	6	5	2

Note: Percentages in parentheses.

*Note that the figures for adult males and adult females in the bottom decile differ from those in Tables 1, 3 and 4. This is because Tables 1, 3 and 4 compared only adult males and adult females, while Table 5 compares adult males, adult females and children.

**A note on interpreting these figures: some of the characteristics are individual (e.g. employment and education), but the deciles are determined at the household level. In other words, individuals in households in the bottom decile of the asset index and income per capita are on average less educated and are less likely to be employed.

Conversely, the poorest by all measures are less likely than average to be in urban areas. This is to be expected as urban areas provide greater access to economic opportunities. Unsurprisingly, a smaller proportion of those in the bottom decile by any measure is employed than in the whole population. This is particularly the case for those in the bottom income decile. Those in the bottom decile by all three measures are less likely to have matric. Those in the bottom deciles are less likely to be in a male-headed household, particularly when using the income and expenditure per capita measures. Those in the bottom decile are younger on average than the total population, particularly in the bottom decile of income and expenditure. Those in the bottom decile have fewer assets on average – unsurprisingly, particularly those in the bottom decile (by definition) of the asset index.

There are several differences in the characteristics of those in the bottom decile depending on which measure is used. The proportion of men in the bottom asset index decile is slightly higher than that in the whole population, while the proportion of men in the bottom income and expenditure deciles is significantly lower than that in the whole population. The poorest in terms of assets are less likely to have an adult female in the household, while the poorest in terms of income and expenditure are more likely to be in a household with an adult female. The average household size in the bottom asset decile is similar to that of the whole population, but those at the bottom in terms of household income and expenditure per capita live in larger households on average. This is partly by construction: the asset index is not adjusted for household size, while household income per capita is. Lastly, households in the bottom income and expenditure deciles are more likely to receive a social grant than those in the bottom asset decile (if households with more children are more likely to be assigned to the bottom income and expenditure per capita deciles by definition, then this may be because these households are more likely to receive the Child Support Grant).

4.3. Factors associated with membership of the bottom decile: regression results

The regression results largely accord with the patterns seen in the descriptive analysis (see Table 6). Some variables show similar associations with membership of the bottom decile regardless of the measure used, while others have contrasting effects. Living in a traditional area is strongly associated with membership of the bottom decile across all three measures. Being an adult male is positively associated with being in the bottom asset decile but is negatively associated with being in the bottom household income per capita decile. It has no significant association with being in the bottom household expenditure per capita decile. Children are significantly less likely to be in the bottom asset index decile, but there is no significant effect of being a child for the bottom income or expenditure deciles.

Table 6. Logistic regressions predicting membership of the bottom decile by different measures.

	Asset index	Household income per capita	Household expenditure per capita
Child	−0.26***	−0.02	−0.09
	(0.010)	(0.792)	(0.377)
Adult male	0.40***	−0.34***	−0.13
	(0.000)	(0.001)	(0.240)
Traditional area	1.18***	0.94***	1.08***
	(0.000)	(0.000)	(0.000)
Adult male#traditional area	−0.35***	−0.01	−0.01
	(0.003)	(0.954)	(0.934)
Age	−0.02***	−0.01*	−0.01*
	(0.007)	(0.082)	(0.054)
Age2	0.00*	0.00	0.00
	(0.071)	(0.302)	(0.211)
Household size	−0.03	0.12***	0.31***
	(0.408)	(0.000)	(0.000)
Household receives grant	−0.12	0.05	0.02
	(0.474)	(0.784)	(0.932)
Male head	−0.14	−0.60***	−0.39**
	(0.237)	(0.000)	(0.031)
Constant	−2.08***	−2.75***	−4.33***
	(0.000)	(0.000)	(0.000)
N	36216	36924	36924

Note: p-values in parentheses.

***indicates significance at the 1% level, **indicates significance at the 5% level, and *indicates significance at the 10% level.

After controlling for adult/child differences, there are no significant nonlinear effects of age (except for the bottom asset decile, but only at the 10 percent level). Larger households are significantly more likely to be in the bottom household income and expenditure per capita deciles, but being in a larger household has no significant effect on membership of the bottom asset decile (again, this may be explained by construction). The interaction between being an adult male and being in a traditional area is significant and negative in the asset index regression. Receipt of a government grant in the household is not significant for any measure. Being in a male-headed household significantly decreases the probability of being in the bottom decile of income or expenditure per capita, but has no significant effect on membership of the bottom asset decile.

5. Possible reasons for discrepancies between asset, income and expenditure measures

Our results show that the gender composition of the bottom decile is sensitive to whether the bottom decile is defined using an asset index, income or expenditure, and to the construction of the measure used. We now explore some possible reasons for these differences.

5.1. Location, migration and remittances

The gender differences in membership of the bottom decile may be influenced by gender differences in residency patterns. There is a significantly higher percentage of adult women than men in traditional areas, and a higher percentage of men in urban areas: 32 percent of adult women live in traditional areas, compared to 27 percent of adult men (see annex Table A2). As seen in Table 5, those living in traditional areas are overrepresented in the bottom decile, while those living in urban areas are underrepresented in the bottom decile, regardless of whether the bottom decile is defined in terms of assets, income or expenditure. The relationship is, however, strongest for the money measures. Understanding why traditional areas are poorer in money terms than assets may then help explain why when extreme poverty is defined in terms of the former, men are underrepresented, yet proportionately are overrepresented when defined in terms of the latter.

There is the possibility that men in urban areas may use some of their income to support family in traditional areas rather than using it to accumulate durable assets in their own households, particularly if their residence in urban areas is viewed as temporary (i.e. migrant workers). Women in traditional areas who receive remittances from men in other areas may then be able to accumulate assets. The NIDS data provides some support for this hypothesis. Men are significantly more likely to give remittances, and significantly less likely to receive remittances (see annex Table A3). Those in urban areas are most likely to send remittances, and those in traditional areas are most likely to receive remittances (see annex Table A4). Those in the bottom household income per capita decile are significantly less likely to give remittances than those in the bottom asset index decile (see annex Table A5).

5.2. Asset ownership and spending patterns

Another possible reason for the sensitivity of the gender composition of the bottom decile to the measure used to identify the bottom decile is that the pattern of asset accumulation and spending may differ for men and women. Men may choose to spend income on other things rather than purchasing consumer durable assets (e.g. entertainment). Alternatively, men and women may purchase different types of assets. If the construction of the asset index gives more weight to assets more commonly owned by poor women than poor men, for whatever reason, this could go part of the way towards explaining the discrepancies between income and the asset index. The NIDS data suggests that this may be the case. Adult women are significantly more likely to be in households that own the six assets (microwave, washing machine, satellite dish, fridge, TV and lounge suite) with the highest loading in the first principal component (used to construct the asset index), except for ownership of a satellite dish, where the difference is insignificant. On the other hand, men are significantly more likely to be in households that own a motor vehicle. Motor vehicles have only the seventh highest loading in the first component, but are more valuable than the assets with higher loadings, but are unlikely to be an issue for the bottom decile. Furthermore, men are more likely to own a motor vehicle themselves (as opposed to being in a household that owns a motor vehicle, where the difference is less pronounced). Men are also significantly more likely to own a motorcycle or computer, but these have a lower loading than assets more commonly owned by women (see annex Table A6), and again, are less important at the lower end of the distribution. Very few (<1%) households or individuals of either gender in the bottom asset decile own motor vehicles, motorcycles or computers, and there is no significant difference in ownership between genders.

Men may also choose to invest in financial assets rather than physical/consumer durable assets (women may be more risk-averse than men – see Charness & Gneezy (2012) for example), but the asset indices popularised by Filmer & Pritchett (2001) are usually based on consumer durable assets rather than financial assets. The average value of financial assets held by men individually in the NIDS sample is significantly higher than that held by women (see annex Table A7).¹

6. Discussion

Our analysis shows that asset indices and income per capita measures yield different household rankings, and therefore paint a different picture of the poorest. The regression results indicate that being an adult male is positively associated with being in the bottom asset defined decile but negatively associated with being in the bottom household income per capita decile. Using expenditure per capita suggests that men are proportionately represented.

Examining the reasons for the difference in gender composition is informative. Our results are similar to the findings by Antonopoulos & Floro (2005) whose Tobit model suggested that women own slightly more real assets than men although the relationship was not statistically significant. In both cases this appears to be related to asset accumulation patterns, and spending on remittances (Antonopoulos & Floro, 2005). Men are more likely to hold financial assets in individual accounts than women and more likely to be working away from home and therefore sending remittances. Our results also indicate that men are overrepresented when expenditure per adult equivalent is used. Expenditure does not include remittances, which suggests that some men, living in small households, may, through their remittances to others, find themselves in the bottom decile.

7. Limitations

NIDS did not collect gender-disaggregated data on most durable assets. Aside from data on ownership of motor vehicles, motorcycles, computers and cell phones, asset ownership data is only collected at the household level. This makes it difficult to conduct a gendered analysis of asset holdings and to unpick the possible reasons for the discrepancies between asset index rankings and money-based rankings.

We acknowledge that any attempt to examine the tails of the income distribution is subject to the limitation that measurement error is likely to be more pronounced at the tails. For example, it may understate the income of the poor (see, for example, Glewwe, 2007).

8. Conclusion

There are a number of methods of measuring socio-economic status. In this paper, we examine three data sources (income, expenditure and assets) and four different ways of constructing a measure of SES from them (per capita, adult equivalence, PCA, MCA). In previous analysis, these measures have been seen to be highly correlated (Filmer & Scott, 2012). This correlation over the entire distribution of households, however, appears to mask discordance in the lower (i.e. left) tail. That is to say, those identified as among the poorest of the poor differ depending on which measure you use to identify them. We focused on the implication of the methods for the gender composition of the bottom decile, but other differences in characteristics were noted across measures. This lack of concordance suggests that there may be a need to use multiple methods to confirm the robustness of analyses, particularly when focused on the tail.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 However, the number of non-missing observations of financial assets is fairly low (10737 of 56699). Furthermore, observations in the bottom decile of the asset index or household income per capita are far more likely to be missing financial asset data than those in the top decile, so measures based on this data may be problematic.

Annex

1. Sensitivity checks

Table A1. Ratio of adult males in bottom five deciles to adult males in whole population.

	Asset index	Income	Expenditure
Bottom 10%	1.09	0.72	0.78
Bottom 20%	1.05	0.76	0.80
Bottom 30%	1.05	0.79	0.81
Bottom 40%	1.02	0.82	0.84
Bottom 50%	1.01	0.84	0.85

Table A1 shows that the gender distribution pattern observed in the bottom decile defined by the asset index, income or expenditure is also present in the other lowest quantiles, but becomes progressively weaker as one includes more of the distribution.

2. Migration and remittances

Table A2. Percentage of adult males and females by area type.

	Female	Male	Total
Traditional	32.33	27.16	29.84
Urban	63.80	67.86	65.76
Farms	3.87	4.97	4.40
Total	100	100	100

Table A2 shows the percentage of adult males and females in each area type. Adult females are more likely to live in traditional areas, while adult males are more likely to be in urban areas.

Table A3. Percentage of adults giving and receiving remittances by gender.

	Percentage	S.E.	95% Confidence Interval
Contributions given to non-household residents
Female	9.04	0.61	7.92	10.30
Male	15.91	0.97	14.11	17.90
Contributions received from non-household residents
Female	15.68	0.67	14.42	17.04
Male	8.62	0.57	7.57	9.81

Table A3 shows the percentage of adult males and females giving and receiving remittances. Adult males are more likely to send remittances, while adult females are more likely to receive remittances.

Table A4. Percentage of adults giving and receiving remittances by location.

	Percentage	S.E.	95% Confidence Interval
Contributions given to non-household residents
Traditional	6.43	0.50	5.51	7.47
Urban	14.90	0.82	13.37	16.57
Farms	12.61	3.47	7.22	21.10
Contributions received from non-household residents
Traditional	14.17	0.76	12.76	15.72
Urban	11.95	0.61	10.80	13.21
Farms	6.92	1.87	4.04	11.60

Table A4 shows the percentage of adults in each area type giving and receiving remittances. Those in urban areas are most likely to send remittances, while those in traditional areas are most likely to receive remittances.

Table A5. Percentage of adults giving and receiving remittances in bottom asset index and income deciles.

	Percentage	S.E.	95% Confidence Interval
Contributions given to non-household residents
Asset index decile 1	11.81	1.21	9.63	14.40
Household income per capita decile 1	2.19	0.44	1.47	3.25
Contributions received from non-household residents
Asset index decile 1	12.63	1.17	10.52	15.10
Household income per capita decile 1	9.70	1.00	7.90	11.85

Table A5 shows the proportion of adults in the bottom asset index and income deciles sending and receiving remittances. Adults in the bottom asset index decile are more likely to send and receive remittances, but the difference is more pronounced for remittances given.

3. Asset ownership

Table A6. Ownership of assets with highest loading in PCA (adults only).

Household asset ownership (adults only)
	Loading of asset in principal component	Female	Male	P-value of difference
Microwave	0.32	56.57	53.58	0.009
Washing machine	0.31	42.06	38.73	0.002
Satellite dish	0.30	35.04	34.71	0.761
Fridge	0.29	81.98	78.15	0.000
TV	0.28	84.33	81.73	0.006
Lounge suite	0.27	58.69	54.92	0.001
Individual asset ownership (adults only)
Motor vehicle	0.26	9.49	18.47	0.000
Motor cycle/scooter	0.10	0.06	1.78	0.000
Computer	0.25	12.52	16.59	0.000
Cell phone	0.14	81.43	80.29	0.224
Ownership of above assets at household level (adults only)
Motor vehicle	0.26	25.14	27.09	0.037
Motor cycle/scooter	0.10	2.41	3.34	0.067
Computer	0.25	24.28	25.13	0.396
Cell phone	0.14	91.25	90.47	0.174

The first section of Table A6 shows the percentage of adult males and females in households that own the six assets with the highest loading in the first principal component. The second section of Table A6 shows the percentage of adult males and females who individually own the four assets for which there is data on individual ownership (rather than only at the household level). The third section shows the percentage of adult males and females in households that own these four assets.

Table A7. Mean individual financial assets by gender (adults only).

Mean individual financial assets
	Mean	S.E.	95% Confidence Interval
Female	4546.24	1462.31	1677.73	7414.75
Male	17518.88	5101.25	7512.11	27525.64

Table A7 shows the mean value of financial asset holdings for adult males and females at the individual level.