Content uploaded by William Baah-Boateng
Author content
All content in this area was uploaded by William Baah-Boateng on Jan 01, 2015
Content may be subject to copyright.
The Gender Asset Gap Project
Working Paper Series: No. 15
Measuring Personal Wealth in
Developing Countries:
Interviewing Men and Women about
Asset Values
Cheryl Doss, William Baah-Boateng,
Louis Boakye-Yiadom, Zachary
Catanzarite, Carmen Diana Deere,
Hema Swaminathan, Rahul Lahoti,
Suchitra J.Y.
The Gender Asset Gap Project is a joint initiative of an international research
team that was formed in 2009 with four objectives: 1) to collect individual-
level asset data from three different countries (Ecuador, Ghana and India)
in order to demonstrate the importance and feasibility of collecting data on
women’s access to and ownership of property; 2) to identify the minimal
set of questions on individual level asset ownership that are needed in
multi-purpose household surveys to calculate the gender asset and wealth
gaps; 3) to develop various measures of gender asset and wealth gaps that
can be used by national governments to track progress toward Millennium
Development Goal 3 on gender equality and women’s empowerment; and
4) to identify the critical enabling or constraining social, economic, and
institutional factors affecting women’s asset ownership in order to help
policymakers and others to improve women’s claims to productive assets.
The project is housed at the Centre of Public Policy (CPP) at the Indian
Institute of Management Bangalore (IIMB). The project team leaders are
Hema Swaminathan, IIMB; Abena D. Oduro, University of Ghana; Carmen
Diana Deere, University of Florida; Cheryl Doss, Yale University; and Caren
Grown, American University. FLACSO-Ecuador hosted the field work in
Ecuador.
The Gender Asset Gap Project Working Papers present preliminary results
and have not been formally peer reviewed. They are circulated in order to
stimulate discussion and critical comment. The authors hold the copyrights
to the contents of the papers and any opinions expressed are solely theirs.
Centre for Public Policy
Indian Institute of Management
Bannerghatta Road, Bangalore 560076, Karnataka, India
Ph: 91 80 26993323. Fax: 91 80 26994050
Email: genderassetgap@iimb.ernet.in
Project website: http://www.genderassetgap.org
Website: www.iimb.ernet.in
1
Measuring Personal Wealth in Developing Countries:
Interviewing Men and Women about Asset Values
Cheryl Doss, William Baah-Boateng, Louis Boakye-Yiadom, Zachary Catanzarite, Carmen
Diana Deere, Hema Swaminathan, Rahul Lahoti, Suchitra J.Y.
November 29, 2013
2
Introduction
Why Assets?
While traditional approaches to studying poverty have centered on income or consumption as
indicators of economic well-being, recent research has shown that the study of assets provides a
more comprehensive understanding of poverty. Indeed, the study of assets provides a means of
capturing lifelong patterns of earnings and wealth accumulation among individuals.
The study of asset ownership is evolving, and a recent area of interest is the extent to which
women have fewer and less valuable assets than men. Indeed, studying assets provides a new
way of understanding differing poverty and vulnerability levels between men and women.
Measuring the value of the assets a woman owns can give us insights into her standard of living,
her bargaining power (both inside and outside the household), and her vulnerability to domestic
violence or economic shocks. Establishing a means of comparing men’s and women’s asset
ownership, then, becomes an important step.
Yet, measuring assets has a number of challenges. The question of aggregation is an example:
for the purposes of some research questions, there is no need to aggregate across assets, since
understanding the ownership of specific assets by individuals may suffice. In other cases—such
as in the study of wealth—aggregation is necessary to create a measure of value that can be
compared across people who own different types of assets.
How To Design a Feasible Study of Asset Ownership?
As we move toward analyzing individual asset ownership, a number of questions about survey
design must be answered. First, is it necessary to gather information on all physical and financial
assets and if not, which assets should be prioritized? Second, is it necessary to interview more
than one person in a household? Third, what is the most reliable measure of asset value? And
finally, are there differences in the perception of value between men and women?
Incorporating an asset module or asset questions into a multi-purpose survey does entail costs for
investigators, and these costs are substantially higher if it involves interviewing additional
people. While adding a minimal set of questions has a relatively minor cost, asking various
household members about their individual asset ownership will increase costs substantially.
Thus, it is useful to understand the potential benefits of interviewing a second person.
Interviewing a second person may provide more complete information on all household assets
and their individual ownership. Further, interviewing men and women separately regarding their
own assets allows for a more rigorous gender analysis.
While a complete measure of household wealth and the gender wealth gap would require
information on all the assets owned by individuals, gathering such detailed information might not
be feasible. Which assets should be included will depend, in part, on the nature of the research
questions. To calculate measures of the gender asset and wealth gaps, information on the most
valuable assets in a given context might suffice. For other purposes, it may be important to
3
include assets that contribute to productivity or livelihoods, even if they have a lower monetary
value.
Establishing Value
To aggregate across assets, each asset must be valued. The literature has used a wide range of
value measures. Some surveys simply ask respondents to identify the value of a given asset
without further specification, which is clearly unsatisfactory. Other surveys ask the value of an
asset if it were sold today (“realization” value or potential sales value) or the cost of purchasing
that asset today (“replacement” value) (Davies, 2008). The Living Standard Measurement Study
(LSMS) surveys generally only ask about rental values of real estate, from which the present
value of a property may be estimated. These different approaches can result in significantly
different estimates, particularly if large transaction costs put a wedge between selling and
purchasing prices. Thus far there has been scant research on the reliability of the various
measures of asset value utilized in household surveys. Since, even for real estate, secondary data
is not readily available on the “true value,” we cannot simply compare the true value with that
provided by the survey respondents. We can, however, consider which approaches give us the
most complete information and the most robust answers.
It would be useful to know whether some groups of individuals provide systematically different
estimates of value than others. Our particular interest lies in exploring differences in valuation
between men and women. If there are systematic differences in the reporting of value by men
and women, this will have implications for measures of asset poverty and calculations of the
gender wealth gaps.
There are many reasons why men and women may provide different estimates of value. First,
men and women may have different relations with the market. Participation in the various asset
markets—such as those for land, housing, livestock, or consumer durables—may be very
different for men and women. In some areas, particular markets—such as the large livestock
market—may not be places where many women participate. Yet women may follow the prices in
other markets—such as those for urban housing—more closely than men. A second reason is that
men and women may specialize within the household, each having information only on assets
relevant to the activities within their domain. Thus, the patterns of estimated value might differ
depending on whether men or women are interviewed.
Survey Background and Methodology
The Gender Asset Gap Project (GAGP) is a three-country initiative that carried out both
qualitative and quantitative research on individual and household asset ownership.
Representative household asset surveys were carried out in 2010 in Ecuador, Ghana, and the
Indian state of Karnataka.
In Ecuador, the sample of 2,892 households is nationally representative of rural and urban areas
and the two major regional geographic and population groupings of the country (the highlands
and coast). A total of 4,668 persons completed the individual questionnaire. In Ghana, a total of
2,170 households were surveyed nationally and 3,288 persons answered the individual
4
questionnaire. In Karnataka, a total of 4,110 households and 7,185 individuals were surveyed
across the rural and urban areas of nine districts covering all agro-climatic zones of the state.
Two survey instruments were administered. The first, the household inventory, was aimed at the
household member(s) who knew the most about the assets owned by its members. It collected
data on household demographics, the inventory of physical assets owned by household members
(place of residence, agricultural land, other real estate, livestock, agricultural implements, non-
farm businesses, consumer durables), and the identity of the asset owners. Regarding the second
instrument, up to two adult members of the household were interviewed separately. Detailed
information was collected on the respondents’ ownership of physical assets, financial assets,
household decision-making, rights over assets, and awareness of inheritance laws.
Three valuation measures were used: the price that would be received if the asset were to be sold
at the time of the survey (what we term the “potential sales price”), the replacement or
construction cost, and the rental rate. The potential sales price was asked for all assets; other
valuation questions were asked for assets as appropriate. For the replacement value of the
principal residence, the Ecuador questionnaire included in the replacement cost both the cost of
construction as well as the current value of the lot on which it was situated; whereas the
Karnataka and Ghana questionnaires only asked about the construction cost of the dwelling.1
The wording of the questions about value was very similar in the three countries, except for
minor variations based on insights gained in the qualitative fieldwork and the field-testing of the
instruments. The questions in the Ghana questionnaire, for example, were: “If you were to sell
the dwelling today, how much would you receive?” For the replacement or construction cost,
respondents were asked, “How much would it cost to construct a dwelling like this today?” The
third valuation measure was the rental value. Respondents were required to provide an estimate
of the monthly rental value of a similar structure in the neighborhood. The question asked was,
“For how much could a dwelling like this, in this neighborhood, be rented per month?”2 The
questions on the potential sales price and rental value for agricultural land were similar. If the
agricultural land was leased/rented/sharecropped out, respondents were asked the amount that
was received. If the land was owned and farmed by the household, they were asked how much
would be received if it were to be rented or leased out.
1 For more details on the survey methodology, see Doss et al. (2012).
2 In Ecuador, the questions were: “If you were to sell the dwelling today, how much would you receive, including
the lot?” “How much would it cost to construct this dwelling today including the price of the lot?” “For how much
could a dwelling like this be rented per month?” In Karnataka, the questions were: “If the dwelling were to be sold
today, how much money would be received (in this neighborhood only)?” “How much would it cost to construct a
dwelling like this today?” “For how much could a dwelling like this be rented (in this neighborhood only)?”
5
Where the three surveys differ is in the placement of the valuation questions and who was asked
to respond to them. In the Ecuador survey, the valuation questions were asked in the household
inventory completed by the principal adults of the household (i.e. a couple, or a male/female sole
head). The respondents were asked to list all assets owned by a household member and to
provide a value for each asset. Fieldwork had revealed that the most reliable responses were
obtained when the couple was given the opportunity to discuss the potential asset value together
and arrive at a consensus estimate. This is partly because of the gendered division of labor; each
person was more likely to know the potential sales values for the assets that they directly utilized,
purchased, and/or had sold themselves. It is also related to the fact that men and women often
have different social networks. Women, for example, seemed more likely than men to know
about the price of homes in their neighborhood that had recently sold, while men seemed more
familiar with agricultural land prices. Allowing each to share their knowledge and discuss almost
always produced an answer by consensus. In the case where one member of the couple was not
present for the inventory—but had completed the individual questionnaire and considered him or
herself an owner or co-owner of the asset in question—he/she was then asked to give valuation
estimates. In these latter instances, two measures of value were obtained—one from each
respondent.3
In the Ghana survey, the valuation questions were only asked in the household inventory, and
only one person responded to this section. Thus, there is only one measure of value, and that is
from the primary respondent.4
In the Karnataka survey, the valuation questions were asked only in the individual questionnaires.
So it was never the case that the two respondents in the household provided a joint answer. The
question about potential sales value was asked to both respondents with exactly the same
wording. However, the construction and rental values were only asked of the primary respondent
(the person who completed the household inventory).
Regarding construction value, in all three countries respondents were instructed to value the
dwelling assuming that the materials used would remain the same. One advantage of the
instructions was that it prevented people from reporting how much they would like to spend to
replace their house, given their current budget (which might allow upgrading). In several cases in
Karnataka, respondents said that the materials were no longer available or that different materials
would be used now.
Components of the qualitative fieldwork in each country centered on the issue of markets and
knowledge of prices. They provided insights into how to phrase questions and follow-up
3 Of the 1,734 cases of homeownership in Ecuador, 36% correspond to households consisting of a principal couple
where they completed the household inventory together and thus provide one joint estimate of asset values; 26% to
households consisting of a principal couple where only one of them completed the household inventory but the other
spouse provided separate estimates of values when they were an owner/co-owner; and 6% to households consisting
of a principal couple where only one spouse completed both parts of the questionnaire. The remaining 32% of these
households were headed by a sole male or female where only one person answered both the household inventory
and the individual questionnaire.
4 The one exception is for other real estate, where the value question was asked both in the household inventory and
the individual questionnaire.
6
questions. For example, enumerators were encouraged to probe for a response to the valuation
questions. When respondents did not know how much their dwelling might sell for, they were
asked about the prices of similar homes in the neighborhood that had been sold recently.
What Assets Should be Analyzed?
In the GAGP surveys, information was collected on all physical assets owned by someone in the
household, as well as on all financial assets owned by the respondents. Our results reveal that
four asset categories comprise the majority of household wealth in the three countries. The
categories of principal residence, agricultural land, other real estate, and non-farm business
comprised 81.6% of gross physical household wealth in Ghana, 89.9% in Ecuador, and 92.5% in
Karnataka. Thus, if the focus of the analysis is on measuring total wealth, these four asset
categories would capture the bulk of it. The final choice of assets, however, should be country-
specific. In some countries (or regions of countries) livestock may account for a relatively large
share of household wealth.
As noted above, data on financial assets was only collected for the respondents, not for all
household members. We would expect that the financial assets of the respondents (the principal
adults) would comprise a significant proportion of the household financial assets, especially in
smaller, nuclear households. Moreover, the data indicate that financial assets of the respondents
comprise a relatively small proportion of the total value of gross household wealth. In Ecuador,
financial assets are only 2.4% of total gross household wealth; for Karnataka, it is 4.5%; for
Ghana it is 5.3%. This is consistent with what we would expect to see in developing countries
(Davies, et al. 2008). Thus, even if we are missing the financial assets of other members of the
household, we would expect these to be a relatively small proportion of total household assets.
Researchers may wish to include other assets in their survey because of their specific
contributions to livelihoods, rather than their monetary value per se, or because of what they
might reveal in terms of gender gaps. For example, ownership of agricultural equipment may be
critical for agricultural productivity, and owning a mobile phone may provide both income-
generating opportunities and social networking.
Should More than One Person be Interviewed?
Studies on labor force participation (Bardasi et al., 2011), income (Fisher et al., 2010), and
financial information (Fletschner and Mesbah, 2011) suggest that responses may differ
depending on who within the household is interviewed—this provides an important justification
for interviewing more than one person. Another reason for interviewing multiple people in an
asset module is that one person may not know about all of the assets owned by everyone in the
household. This might be the case, for example, if productive assets are related to sex-specific
roles or responsibilities. Or it may be that certain assets—such as financial assets—are concealed
from the others. Thus, if additional assets were added to the household inventory when
interviewing a second person, this would indicate the utility of interviewing multiple respondents.
The GAGP surveys were not designed specifically to test hypotheses about who should be
interviewed, though they provide some insights into this issue. As described above, the interview
7
strategies in the three countries differed based on the findings of the qualitative phase of the
study. In Ghana and Karnataka, one person was asked about all household assets in the
household inventory. Later, that individual and, where possible, one other person of the opposite
sex were each asked about their own assets. In Ecuador, the aim was to interview the principal
couple together and they responded to the household inventory questions together whenever
possible. Only spouses who were not present for the household inventory were subsequently
asked about their own ownership of assets in the individual questionnaire.
In all three of the GAGP surveys the second respondent was asked if they were an owner of each
of the assets previously reported in the inventory. They were then asked if they owned any
additional assets that had not already been listed. In the three surveys, relatively few major assets
were added to the inventory by interviewing a second respondent, as Table 1 shows.
Table 1. Major Assets Added by Interviewing a Second Respondent (% added to inventory)
Asset Ecuador Ghana Karnataka
Principal dwelling 0.40 NA NA
Agricultural parcels 1.36 0.35 0.15
Other real estate 3.91 0.41 0.35
Non-farm businesses 0.52 1.04 0.12
With one exception, less than 1.5% of the total number of these assets was added by the second
person across the three surveys. This suggests either that the primary respondent does know
about the vast majority of major assets and few assets are concealed, or that the second
respondent does not often reveal hidden or missing assets to the enumerator. In Ecuador, almost
4% of the other real estate (a category which includes non-agricultural land, such as housing lots,
other dwellings, such as rental properties, and commercial establishments or buildings) was
added by the second respondent.5 Thus, if the second person had not been interviewed, couple
and household wealth would have been somewhat underestimated.
In the GAGP surveys the second respondent was asked about additional assets in reference to a
previous listing. A different approach would be to interview two people per household, ask each
to fill in a household inventory, and then compare the results. While more time and resource
intensive, each person would list all of the assets that they knew about rather than just adding to
a list, potentially uncovering a greater number of assets.
Another reason that it might be important to interview a second person is if men and women
differ in their understanding of whom each asset belongs to. Men and women may differ in their
5 The reason the category of “other real estate” resulted in such a relatively high share of additional assets in
Ecuador may be related to the prevailing marital regime of partial community property. Other real estate may be
more likely than the principal dwelling or agricultural parcels to be individually owned property, acquired by
someone prior to marriage or inherited, and thus not part of the community property of the couple.
8
understanding of the legal rights of ownership, especially as they relate to the rights of spouses.
Gender socialization may also affect their ability or willingness to claim particular assets.
The interview strategies pursued in the GAGP surveys allow us to compare the responses of the
primary and secondary respondents with respect to who owns each asset. In Table 2 we present
inconsistencies in responses relating to asset ownership within couples interviewed separately for
the four major asset categories discussed above. The answers are considered inconsistent if they
do not provide the same responses about whether each of them is an owner. (Inconsistencies
regarding whether they report ownership by a third person are not included.)
In Ecuador among couples who were interviewed separately, there were relatively high rates of
inconsistency; in 31% of these interviews, the husband and wife gave different responses
regarding whether the agricultural parcel was owned by him or her or both. Different responses
were given by 28% regarding the ownership of the dwelling. The answers regarding ownership
were much more consistent in Ghana; nonetheless, 8% of couples still gave different answers
regarding ownership of the dwelling and other real estate. The inconsistency in Karnataka was
the highest, especially for immoveable property; more than 90% of the secondary respondents
listed themselves as owners when the primary had not included them as owners in the asset
listing.
The above suggests that interviewing two respondents in a household often yields different
responses regarding who owns the assets. However, there are at least two additional factors that
might be in play in explaining these differential responses, and which make it difficult to
attribute the inconsistencies only to the sex of the respondent. First, the wording of the questions
was different in the household inventory and the individual questionnaire. The person answering
the inventory was asked to identify the owners of each asset with the question, “To whom does
this asset belong?” In the individual questionnaire, the respondent was asked, “Do you consider
yourself the owner or one of the owners of this asset?” It is very probable that the latter would
elicit a more inclusive response on ownership while the former would elicit a more detached
listing of owners. Certainly in the case of Karnataka, the differential wording is considered to
have played a key role in the inconsistent responses on ownership (although it is hard to estimate
Table 2. Inconsistencies Within Couples Over Asset Ownership
Ecuador Ghana Karnataka
Asset N
(assets)
%
inconsistent
N
(assets)
%
inconsistent
N
(assets)
%
inconsistent
Dwelling 443 28.0 510 7.7 1,818 97.3
Agricultural
land
94 30.9 873 3.3 1,786 91.3
Other real
estate
132 21.2 413 7.8 506 91.1
Non-farm
business
532 20.9 641 1.6 467 53.5
9
to what extent). Also, in Karnataka only those secondary respondents who were not listed as
owners by the primary respondent were asked the ownership question in the individual
questionnaire. In contrast, in Ecuador and Ghana all second respondents were asked whether
they consider themselves an owner of the asset.
A second factor is that the asset inventory/listing could have been undertaken in the presence of
any or all other household members; whereas the individual interviews almost always took place
in complete confidentiality.
Another reason for the cross-country differences may be differences in the marital regimes. Both
Ghana and Karnataka are characterized by separation of property while Ecuador is governed by
partial community property (Deere et al., 2013b). In Ghana, under the separation of property
regime all property acquired after marriage is individually owned and thus both the husband and
wife are very likely to share a similar understanding of who owns it. The cultural norms also
clearly recognize the individual property of husbands and wives; i.e., that marriage does not lead
to community property. In Karnataka, gender norms rather than the nature of the marital regime
might explain the large number of secondary respondents (mostly women) adding themselves as
owners. Men reporting themselves as property owners to the exclusion of their wives is
culturally acceptable in a patriarchal society, and it does mirror reality. However, it also seems to
be the case that women, when asked a pointed question about their claim over the assets of their
spouses, assert some degree of ownership as well. This again reflects the social context where
women typically construe what belongs to the husband as belonging to them as well (even
though they may not have a formal claim on the asset).
In Ecuador, property acquired by either spouse after marriage (other than through inheritance) is
legally joint property, while property brought to the marriage or inherited remains individual
property. The focus groups revealed a strong sense of joint ownership of major property within
marriage (Deere et al., 2013a). Thus, some of the inconsistencies may be due to a tendency to
consider all property to be jointly owned, irrespective of how it was acquired. If one spouse is
referring to the social norm and the other to legal rights, the answers may be inconsistent.6
An additional potential reason for collecting individual level asset data is to understand the
relationship between empowerment and property ownership. If this is the purpose, then it may be
necessary to ask individuals about their own property ownership, rather than relying on a proxy
by someone else in the household. While a number of recent studies have examined how the
intrahousehold allocation of assets affects the outcomes of household decisions, they use
ownership as reported by one member of the household. For example in Ghana, Doss (2005)
finds that the share of land, businesses, and savings owned by women in the household affects
the outcomes of household decisions. Allendorf (2007) uses DHS data that reports whether
6 For Ecuador, a detailed gender analysis was carried out on the inconsistencies among couples over who among
them was the owner of the principal residence. In the 119 cases of such disagreements over ownership, it was more
frequent for wives to consider that the home was jointly owned while the husband considered himself the only
owner (32%) than where husbands considered the home to be jointly owned while the wife considered herself the
only owner (24%). Wives also appear to be more self-effacing; in 19% of the cases the husband reported the home
as jointly owned but the wife reported that he was the only owner, whereas in only 8% of the cases the wife reported
the home as jointly owned but the husband reported that she was the only owner.
10
women say that they own land and finds that women’s land rights affect child health and
nutrition. To the extent that we are interested in how asset ownership is empowering for women,
we might want to identify women’s understandings of their ownership and rights over assets. But
that is beyond the scope of this paper.
Table 2 only considers multiple respondents who are couples. In both Karnataka and Ghana, the
second person interviewed was not necessarily the spouse of the first. While most intrahousehold
analyses that use a bargaining framework focus on interactions between the couple (Doss, 2013),
other intrahousehold analyses could provide additional insights. It may be that couples are more
likely to know about each other’s assets than two other adults in the household, such as a parent
and adult child. Thus the choice of multiple respondents will depend on the research question.
Of those households where the respondents were a couple (as opposed to a single man or a single
woman), 95.8% of the physical wealth of Ghanaian households is owned by that couple. The
share is slightly lower in the other two countries: 88.1% in Ecuador and 83.0% in Karnataka.
What Measure of Value is Most Reliable?
Three types of issues arise when considering measures of asset values. The first is whether
respondents can provide a measure of value. There are a number of reasons, discussed in detail
below, why they may not provide a value and the response appears as “missing.” Second is the
question of whether respondents are better able to report one measure of value than another.
Finally, we are interested in which of the measures is the best measure of value.
Missing Values
Considerable attention was given in the qualitative fieldwork to the issue of whether respondents
would be able to estimate the value of their assets. One of the conclusions reached was that, in
many cases, markets for certain assets do not exist. This is the case in some rural areas where
there is no demand for rental housing or where dwellings are rarely sold. In these locales, the
predominant way of acquiring a dwelling, if such is not inherited, is to build it oneself on a lot
which has been inherited or purchased. Similarly, markets for land may be thin or nonexistent in
some locales. It was also not unusual to find that, in certain areas, there is no market for used
appliances, small agricultural tools, etc.
Thus to deal with the issue of missing markets, respondents were given the option in the
questionnaire of reporting that “there is no market” in addition to other options (including that
they did not know, or refused to divulge the value). While this helps capture the situation on the
ground, it presents a problem in terms of the analysis of household and individual wealth,
particularly if one group of individuals are more likely to report that they do not know the value
of their assets. Missing observations for potential sales prices would result in an underestimation
of the total wealth of households, unless these values were later imputed.7
7 The wealth estimates included in Doss et al. (2011) utilize imputed values. These are not used in the present paper
since our interest in this section is in examining the consistency of the reported value observations.
11
We expect to find differences between urban and rural areas since the existence of markets
differs by locale; we expect rural areas to be more likely to report missing values than urban
ones. Table 3 provides the percent of missing valuation responses for two key assets, the
principal residence and agricultural land. For this table, only the responses of the primary
respondent (in Ecuador this may be an individual or the couple together) are reported.
Table 3. Percent of Missing Values by Form of Valuation, Principal Residence and
Agricultural Land, by Urban/Rural (Primary Respondent Only)
Principal Residence Agricultural Land
N Sales value Const-
ruction
value
Rental
value
N Sales value N Rental
value
Ecuador
Urban 1,045 2.7% 3.3% 1.4% 130 0.8% 109 11.9%
Rural 682 3.7% 4.4% 10.0% 382 5.0% 368 15.8%
Total 1,727 3.1% 3.7% 4.8%*** 512 3.9%** 477 14.9%
Ghana
Urban 193 20.7% 16.6% 12.4% 929 81.9% 288 49.0%
Rural 664 12.8% 11.3% 26.1% 2,433 70.5% 1,560 48.8%
Total 857 14.6%*** 12.5%* 23.0%*** 3,362 73.6%*** 1,848 48.8%
Karnataka
Urban 791 11.4% 5.6% 5.7% 313 17.6% 184 20.7%
Rural 2,380 5.3% 1.7% 28.0% 2,368 4.9% 2,090 10.1%
Total 3,171 6.8%*** 2.7%*** 22.4%*** 2,681 6.3%*** 2,274 10.9%***
Note: *** p<.01, ** p<.05, * p<.1 for a chi-squared test of significant differences between urban and rural
respondents. “Missing” includes responses where no market was indicated, as well as “don’t know”, and where no
response was provided.
The question on the potential sales value of the principal residence and agricultural land was
asked for all those that were owned. For residences, the question on the rental value was also
asked for all those owned. For agricultural land, however, some plots of land were owned and
cultivated by the household members themselves while others were owned by the household but
leased out/sharecropped out to others. The surveys obtained a hypothetical lease/rental value for
the former category and the actual lease/rental value for the latter. The results in the table above
consider only the plots where the hypothetical value was reported; therefore, the total number of
plots is different for the sale and lease values.8
Generally, there are statistically significant differences between respondents in urban and rural
areas, although it is not the case that respondents in rural areas are less likely to provide a
measure of valuation. For the principal residence in both Ghana and Karnataka, respondents in
urban areas are less likely than those in rural areas to provide an estimate of the potential sales
value or construction cost. However, in all three countries rural respondents are more likely than
8With respect to the plots of land that were leased/rented out: of 467 such parcels in Karnataka, the actual rental
values were provided for all but 17. In Ghana and Ecuador, very few parcels fell into this category.
12
urban respondents to have missing values for the rental value of the dwelling. The patterns for
agricultural land across countries are not consistent. In Ecuador, urban respondents are more
likely than rural ones to provide an estimate of the sales or rental value;9 yet the opposite pattern
is found for Karnataka and Ghana (only for potential sales value in Ghana).
Overall, in Ecuador and Ghana there is generally a higher percentage of missing responses for
agricultural land than there is for housing. For Karnataka this pattern does not hold with respect
to rental values. Moreover, irrespective of the form of valuation, the percent of missing
responses is much higher for Ghana than for either of the other two countries.10
It is worth decomposing the reasons behind these missing values, since they have different
implications for survey design. For example, if the predominant reason is that the primary
respondent does not know the value of the asset, or was unwilling to estimate it, then it may be
useful to interview a second respondent. Another argument in favor of multiple respondents is
that respondents might disagree on whether there is a market for certain assets based on their
personal experience.
For the principal residence, both Ecuador and Karnataka distinguished between missing values
due to “don’t know” and “no market”. For the sale value, 49% of the missing values in Ecuador
are reported as being missing because there is no market and 42% because the respondent does
not know the value. In Karnataka, far fewer report that there is no market (9%), while 65% did
not know the potential sale value. Other reasons for missing values were not coded. For the
rental value, the majority of missing values (86% in both Ecuador and Karnataka) are due to the
absence of a rental market. Ghana did not distinguish between the two possible responses.
For Ecuador, the majority of missing values for agricultural land are due to the response that
there is no market for land in the area; this is true for both the potential sales value and the
rental/lease value. In Karnataka, very few said that there was no market when asked about the
sale price of land, while more reported that there was no rental market. In Ghana, respondents
were not given an option to say that there was no market for land in this section; however, many
reported that there was no market for land in the section of the survey where they were asked
about their rights over land. This information can be used to infer that 38% of those with missing
values for agricultural land consider that there is no market for land.
The majority of respondents do provide values for the major assets. For the principal residence,
the valid response rates for potential sales value range from a high of 97% in Ecuador to a low of
85% in Ghana. The percentage of missing values for land and the main dwelling are about the
same for Ecuador and Karnataka, and much higher for land in Ghana. It is more common in
Ecuador than in Karnataka for respondents to say that there is no market and that they cannot
provide a potential sales value. Overall, rental values show a higher incidence of missing values
than sale values for both dwelling and land in Karnataka (although there are some differences
across rural and urban areas).
9 This may be because urban owners of agricultural parcels own land closer to urban areas where it is more likely
that markets for land exist.
10 Family land in Ghana (which pertains to the lineage) is not included in the analysis since it legally cannot be sold.
13
Statistical Properties of Value Measures
Identifying the “best” method of valuation is challenging because we do not know the true value
of the principal assets. Thus, we have to consider other characteristics of a good measure of
value, such as whether respondents are at all able to report a value (as we did above). It might
not be unreasonable to conclude from the preceding analysis that the type of value that is
characterized by the lowest incidence of missing values is the most reliable, since the majority of
the respondents are able to provide some positive response to it. This approach, however, does
not consider the distribution of the positive values. Therefore a second approach is to consider
the descriptive statistics of the different measures of value. Examining the nature of these
distributions—specifically the extent of the scatter as indicated by the coefficient of variation,
skewness, and kurtosis—might suggest the value category characterized by the least
heterogeneity within a given area as being the better measure.
Tables 4 and 5 provide the descriptive statistics for the different valuation measures for residence
and land, respectively. Only observations where all three measures for residence and both
measures for land were positive values—i.e., not missing—are included in these tables.
14
Table 4. Statistical Properties of the Valuation Measures for the Principal Residence, by
Urban/Rural.
Urban Rural Total
Sales
value
Const.
cost
Rent Sales
value
Const.
cost
Rent Salesval
ue
Const.
cost
Rent
Ecuador (US$)
N 992 992 992 594 594 594 1,586 1,586 1,586
Mean 34,581 40,731 149 15,299 17,758 69 27,359 32,127 119
Median 25,000 30,000 100 8,000 10,000 45 15,530 20,000 80
S.D. 37,170 53,317 168 18,447 21,981 82 32,837 45,628 147
Std error 1,180 1,693 5.3 757 902 3.4 825 1,146 3.7
Coef Var 1.07 1.31 1.13 1.21 1.24 1.19 1.20 1.42 1.24
Skewness 3.24 6.41 7.46 2.14 2.32 3.55 3.46 6.88 7.54
Kurtosis 17.7 68.5 100.7 4.8 5.6 18.1 21.2 85.2 113.4
Ghana (Cedis)
N 141 141 141 461 461 461 602 602 602
Mean 17,845 13,410 73 5,285 3,561 21 8,227 5,868 33
Median 6,000 4,000 20 2,000 1,500 5 2,500 2,000 6
S.D. 26,751 21,042 191 11,587 7,665 71 17,257 12,866 113
Std error 2,253 1,772 16.1 540 357 3.3 703 524 4.6
Coef Var 1.50 1.57 2.64 2.19 2.15 3.37 2.10 2.19 3.43
Skewness 2.33 2.24 7.73 5.63 6.24 9.04 4.12 4.28 10.96
Kurtosis 8.6 7.5 76.0 43.0 54.2 104.4 23.0 23.7 168.0
Karnataka (Rupees)
N 684 684 684 1,663 1,663 1,663 2,347 2,347 2,347
Mean 825,598 595,514 1,850 193,248 208,529 475 377,537 321,311 875
Median 300,000 300,000 1,000 100,000 150,000 300 150,000 150,000 400
S.D. 1724,786 985,141 3491 255,713 241,900 741 997,508 595,767 2,080
Std error 65,948 37,667 133 6,270 5,931 18 20,590 12,297 42
Coef Var 2.09 1.65 1.89 1.32 1.16 1.56 2.64 1.85 2.38
Skewness 8.92 4.7 9.12 4.80 4.67 8.16 14.65 7.47 13.93
Kurtosis 130.1 34.9 128.0 41.1 41.8 117.8 363.65 89.0 320.9
Note: Rent is reported on a monthly basis.
Consider first the statistical properties for the samples as a whole, reported in the last three
columns. In both Ecuador and Ghana, the coefficient of variation (a measure of the signal-to-
noise ratio, or the standard deviation as a proportion of the mean) is lower for potential sales than
construction or rental values. Therefore, in theory, the potential sales value is a more robust
measure in these countries as it provides values with less deviation from the mean.11 In
Karnataka, in contrast, the potential sales value has the highest coefficient of variation and
construction value has the lowest.
11 In practical terms, this indicates a greater likelihood of seeing significant results with the use of the market value
since the standard deviation is lower relative to the mean. (The standard deviation appears in the denominator of
many significance tests.)
15
Examining the measures of skewness and kurtosis we see that, again, these are lower for
potential sales value than construction or rental values in Ecuador and Ghana. Therefore, the
sales value questions tend to result in a more normal distribution of responses than the rental
value questions- an important consideration since normality is often an assumption in statistical
analysis, particularly with Ordinary Least Squares (OLS) methods. In Karnataka, again,
skewness and kurtosis are all smaller for construction value than for potential sales value. In part,
this may be because of the impact of a few extraordinarily high outliers in sales value.
As expected, there are important differences in the reported values between urban and rural areas,
whether potential sales, construction or rental, such that urban areas report significantly higher
values for residences than rural areas. Only some of the aforementioned country patterns tend to
hold when disaggregating by rural and urban locales In Ecuador and Ghana the coefficient of
variation is lowest for potential sales value in urban as opposed to rural areas and performs better
than other value indicators in urban areas. In rural areas the lowest coefficient of variation differs,
being that for rental values in Ecuador and construction values in Ghana. In Karnataka, in
contrast, the lowest coefficients of variation are found for rural areas, with the very lowest
indicator being for rural construction costs.
With respect to skewness and kurtosis, these are lower for all three measures of value in urban as
opposed to rural areas in Ghana, while Ecuador reveals the opposite pattern, and Karnataka, yet
another, mixed pattern.
16
Table 5. The Statistical Properties of The Valuation Estimates for Agricultural
Parcels, by Urban/Rural
URBAN RURAL TOTAL
Sales Annual
Rent
Sales Annual
Rent
Sales Annual
Rent
ECUADOR
N 96 96 309 309 405 405
Mean 14802 891 11077 343 11960 473
Median 8000 200 4000 120 5000 180
S.D. 19901 2813 24128 590 23227 1477
Std error 2031 287 1373 34 1154 73
CoeffVar 1.34 3.16 2.18 1.72 1.94 3.12
Skewness 2.50 7.19 5.80 4.75 5.28 12.43
Kurtosis 6.92 58.59 42.60 32.80 37.96 193.12
GHANA
N 118 118 534 534 652 652
Mean 10,669 3,881 5892 460 6,756 1,079
Median 3300 205 1,375 100 1,500 100
S.D. 23,435 24,914 27,079 1,670 26,504 10,751
Std error 2,157 2,294 1,172 72 1,038 421
CoeffVar 2.20 6.42 4.60 3.63 3.92 9.96
Skewness 5.42 9.47 14.15 10.02 13.00 21.90
Kurtosis 39.58 96.58 235.63 127.36 213.27 519.66
KARNATAKA
N 134 134 1838 1838 1972 1972
Mean 913014 32367 398740 13884 433686 15140
Median 300000 10000 150000 5000 160000 5000
S.D. 1774078 72743 1017979 42401 1093145 45325
Std error 153257 6284 23744 989 24616 1020
CoeffVar 1.94 2.25 2.55 3.05 2.52 2.99
Skewness 4.73 4.83 16.01 24.45 13.57 20.13
Kurtosis 33.08 28.79 409.54 826.68 303.48 603.01
Overall, agricultural land potential sales values have the lowest coefficient of variation, skewness,
and kurtosis for all three countries. When we consider rural respondents only, we see an
important difference. For both Ecuador and Ghana, the three statistical indicators perform better
for rental values than for potential sales values of land. In contrast, in Karnataka the three
indicators perform better for sales in rural areas.
The analysis of the statistical properties of the various measures of value of housing and land
suggests that it is difficult to draw straightforward conclusions across countries, particularly for
real estate in both urban and rural areas. In our subsequent analysis of gender differences we will
17
focus on potential sales values. Sales values have the practical benefit of being a common way
for lay people to understand value.
Gender Differences in Reporting Asset Values
One of the important challenges in doing an individual level asset ownership survey or module is
determining who needs to be interviewed within the household. Above, we discussed the
question of whether interviewing additional people increased the number of assets reportedly
owned by the household. Another reason is if there is a gender bias in the valuation measures. It
would be useful to know whether there is a systematic bias in reporting of values based on
gender.
Table 6 indicates the percentage of values that are missing by sex of the respondent. For Tables 6
and 7, only the primary respondents are included since this allows for greater comparability
across the countries. For Ecuador, there are three categories of primary respondents: individual
men, individual women, and the couple answering jointly. For Ghana and Karnataka, the primary
respondents were all sole individuals.
First, we consider the principal residence (Table 6). For Ecuador, there is no difference in the
percentage of missing responses based on the primary respondent for either potential sales or
construction value. For Ghana, women are more likely to have missing values for construction
costs, but not for potential sales value. In Karnataka, women are more likely to have missing
values for both construction costs and potential sales values.
For rental rates, there are statistically significant differences in the number of missing
observations by the sex of the respondent for all three countries. In Ecuador, couples are most
likely to have a missing response, followed by women. Men are the most likely of the three
groups to report rental values. In Ghana and Karnataka, women are more likely than men to
provide a rental value, since a much higher proportion of men report a missing value.
A second way to consider the gender differences is to ask which measure has the fewest missing
values for men and women. In Ecuador, market values have the fewest missing values,
regardless of who answers. In Karnataka, construction costs have the fewest, again regardless of
who answers. But in Ghana, for men the construction costs have the fewest missing values while
for women rental values have the fewest missing values.
18
Table 6: Percent Missing Values by Form of Valuation, Principle
Residence, by Sex of Primary Respondent
N Sales value Construction
value
Rental
Value
Ecuador
Male 273 1.8% 2.2% 2.2%
Female 832 2.9% 4.0% 4.1%
Couple 622 3.9% 4.0% 6.9%
Total 1,727 3.1% 3.7% 4.8%
Chi-squared p=0.248 p=0.355 p=0.004
**
Ghana
Male 631 13.5% 11.1% 25.8%
Female 226 17.7% 16.4% 15.0%
Total 857 14.6% 12.5% 23%
Chi-squared p=0.122 p=0.039
**
p=0.001
***
Karnataka
Male 2,442 4.5% 1.7% 24.1%
Female 729 14.7% 5.9% 16.3%
Total 3,171 6.8% 2.7% 22.5%
Chi-squared p=0.000
***
p=0.000
***
p=0.000
***
Table 7 shows similar information for agricultural parcels. For agricultural land, in Ecuador there
is no difference in the percent of values that are missing, based on the primary respondent(s). In
Ghana, the value is more likely to be missing for rental value if men are responding; as noted
above, Ghana has a high percentage of missing responses for potential sales value for
agricultural land. In Karnataka, for both potential sales value and rental value the value is more
likely to be missing if women are responding; and the gender gap in terms of missing responses
is quite high.
Thus, we do see some gendered patterns whereby women are more likely to provide rental values
than potential sales values for residences in Ghana and Karnataka, and agricultural land only in
Ghana. Overall, a greater share of women respondents in Ecuador are likely to provide sales,
construction, and rental prices than in Ghana or Karnataka, where women are less likely to
participate in the markets for both residence and agricultural land.
19
Table 7: Percent Missing Values by Form of Valuation, Agricultural
Parcels, by Sex of Primary Respondent
Sales value Lease/Rental value
Ecuador N Percent N Percent
Male 66 1.5% 58 6.9%
Female 201 3.5% 189 14.3%
Couple 245 4.9% 230 17.4%
Total 512 3.9% 477 14.9%
Chi Squared p=0.418 p=0.128
Ghana
Male 2,347 73.5% 1,361 50.2%
Female 1,014 73.8% 487 45.0%
Total 3,361 73.6% 1,848 48.8%
Chi-sq test p= 0.891 p= 0.048
Karnataka
Male 2,291 4.4% 1,996 9.8%
Female 390 17.7% 278 24.1%
Total 2,681 6.3% 2,274 11.5%
Chi-square p=0.000
***
p=0.000
***
Overall, for both types of property there are gendered differences in the response rates. In
Karnataka, women are less likely to provide responses for the sales and construction values. The
one instance in which women in Karnataka are more likely to respond is for rental values for the
house. There are no gender differences in response rates for Ghana for potential sales value for
either place of residence or agricultural land, but there are for rental values, with more missing
values for men. The only significant gender difference in response rates for Ecuador are that men
are more likely than women and much more likely than couples to provide rental rates for
dwellings.
Gender Differences in Reported Values
In addition to the response rates for men and women, we are also interested in whether the values
provided differ systematically by gender. Table 8 compares the means of the various measures of
value for men and women. Again, only those who provided responses to all three measures are
included so that the sample size is consistent.
20
Table 8: Mean Value by Sex of Principal Respondent by Form of Valuation,
Principal Residence
Sex of
respondent
N Sales value Construction
value
Monthly
rental value
Ecuador (US$)
Male 259 37,734 43,975 156
Female 768 28,264 33,747 123
Couple 559 21,309 24,412 97
Total 1,586 27,359 32,127 119
F-test p = .000
***
p = .000
***
p= .000
***
Ghana (Cedis)
Male 435 8,784 6,385 29
Female 167 6,774 4,521 42
Total 602 8,227 5,868 33
t-test p=0.201 p=0.111 p=0.212
Karnataka (Rupees)
Male 1,805 384,042 327,847 885
Female 542 355,874 299,542 843
Total 2,347 377,537 321,311 875
t-test p=0.564 p=0.332 p=0.677
For Ecuador, the values differ significantly depending on who responds. For each variable of
interest—potential sales value, construction cost, and rental value—the post hoc multiple
comparisons reveal significant mean differences among the type of respondent. Men report
significantly higher values than women and couples, and women report significantly higher
values than couples. There are no significant differences by sex in reported values for Karnataka
and Ghana (although the pattern of men reporting higher values largely holds).
Table 9 provides similar information for agricultural parcels. For Ecuador, the F-test reveals no
significant difference in potential sales value for land, but does reveal significant differences in
rental value for land. Post hoc multiple comparisons reveal that male respondents report
significantly higher values than female respondents and couple respondents; female respondents
do not report significantly different values than couple respondents. For Ghana and Karnataka,
again there are no statistically significant differences in mean values for men and women.
21
Thus, in both Karnataka and Ghana the mean values by the sex of the respondent are not
statistically significantly different for either housing or agricultural land. This could be in part
due to the fact that the sample under discussion is that of the primary respondents, who were
selected for interview on the basis of their being the best informed individual regarding the
economic circumstances of the household. Therefore, it could be the case that the sex of the
respondent does not significantly affect the positive values reported, so long as that respondent is
better suited than anyone else in the household to answer the value questions. In Ecuador,
although the respondents were selected in the same way, the results differ. For housing,
irrespective of the measure men report significantly higher values; and, for land, significantly
higher rental values than women or couple respondents.
The section above compares values provided by male and female respondents but does not
control for the fact that the characteristics of the properties being reported on may be different.
For example, tables 8 and 9 pool responses from urban and rural areas, yet as shown earlier
urban housing prices usually exceed rural prices. It may also be that the properties in households
in which women are the primary respondents are systematically different from those in which the
men are the primary respondents. Thus, in the following section, we undertake further analysis to
understand the differences in valuation between men and women.
Table 9: Mean Values by Sex of Principal Respondent by Form of
Valuation, Agricultural Parcels
Sex of
respondent
N Sales value N Annual
rental/lease
value
Ecuador (US$)
Male 54 11,454 54 1,067
Female 161 9,688 161 366
Couple 190 14,029 190 395
Total 405 11,960 405 473
F-test p=0.215 p=0.006
***
Ghana (Cedis)
Male 621 7,484 678 1,054
Female 266 5,319 268 292
All 887 2,165 946 838
t-test p=0.228 p=0.238
Karnataka
(Rupees)
Male 1,775 439,971 1,775 15,359
Female 197 377,050 197 13,170
Total 1,972 433,686 1,972 15,140
t-test p=0.444 p=0.520
22
Sources of Gender Differences in Asset Valuation
In this section, we use the Blinder-Oaxaca decomposition to explore potential differences in
valuation between men and women controlling for characteristics of the assets in question. The
Blinder-Oaxaca decomposition, put forward by Blinder (1973) and Oaxaca (1973) is a procedure
that divides the differential of the mean value of the dependent variable between two groups into
a portion that is “explained” by group differences in objective characteristics and the residual
portion that cannot be accounted for by such differences. This decomposition technique is often
used in the study of labor market outcomes (mainly wages) by groups (e.g. sex or race) to
analyze the mean difference in log wage on the basis of linear regression models in a
counterfactual manner. In the labor economics literature, the residual (or unexplained) part in the
wage decomposition (referred to as the wage remnant) is often used as a measure of wage
discrimination (see Stanley and Jarrell, 1998; Altonji and Blank, 1999; Weichselbaumer and
Winter-Ebmer, 2005).
Model Specification
In line with the original Blinder-Oaxaca decomposition approach, consider a model of asset
valuation conditioned on the characteristics of the assets for male and female respondents as:
mmm XV ' (1)
fff XV ' (2)
ε ~ N(0,1)
where Vrepresents the mean value of the asset, X is a vector of asset characteristics, β denotes a
vector of coefficients of (or returns to) these characteristics, and ε represents a random error term
assumed to be normally distributed with a mean of zero and constant variance. The superscripts
m and f denote males and females, respectively.
Subtracting (2) from (1) yields
ffmmfm XXVV
'' (3)
Subtracting and adding mf
X
', we get
mfmfffmmfm XXXXVV
''''
''' ffmmfmfm XXXVV
(4)
Equation (4) is the Blinder-Oaxaca decomposition equation. The left-hand side of the equation
represents the difference in the mean value of assets reported by male and female respondents.
The first term of the right-hand side of the equation reflects the differences in mean value due to
23
differences in the observable or explained characteristics of the assets owned by them. The
second term denotes the difference in the mean value of assets reported by men and women due
to the shift in the coefficient βi; i.e. the differential returns to the observable characteristics of the
assets reported between males and females. A positive and statistically significant unobservable
difference could suggest an undervaluation of assets by women or an overvaluation by men on
subjective grounds, and vice versa. In other words, the unexplained difference is likely to be
primarily due to membership in the group in question.
The share of differences in the mean value of explained or observable characteristics of the
assets and unexplained or subjective valuation sum to 100% if the direction of the differences is
the same. However, if the direction of the mean differences from the two sources is different or
opposite, then one of the sources would produce more than a 100% share in the mean difference
in the value of the assets.12
In estimating equation (4) we recognize the potential for intragroup correlation as a result of the
effect of location or multiple assets reported by the same household. With intragroup correlation
the assumption of independence of observations is violated even though the observations may be
independent across groups. Ignoring potential intragroup correlation could adversely affect the
standard errors of the coefficients. Our model thus accounts for the possibility of this correlation
in the estimation process.13 In addition, the estimation approach uses the raw or non-log values
of assets as a measure of the dependent variable which may potentially violate the normality
assumption. Consequently, we adopt the bootstrapping estimation approach to address this
concern.
12 For example, if the differences in the observable characteristics of the assets results in a difference in the mean
value in favor of male respondents and differences in the mean value from unexplained or subjective sources also
favor male respondents, the share of the differences in the mean value of assets from the two sources would sum to
100%. On the other hand, if the difference in the mean value of the assets from explained sources favors male
respondents but the mean difference attributable to unexplained or subjective valuation favors female respondents,
the direction of the net difference in the mean assets value would depend on the magnitude of the difference from
the two sources. A stronger difference in the mean value from the explained source in favor of male respondents
than the unexplained source in favor of female respondents would yield more than 100% from the explained sources
with the unexplained sources producing a negative share equivalent to the excess of the 100%.
13 For instance, in the case of dwellings there is only one report per household but households that are in close
proximity (i.e. within the same geographic sampling unit) are more likely to share similar characteristics and thus be
more likely to have similar estimations of value. In the case of lands we lack information on what lands might be in
close proximity to one another (since the land may not necessarily be located close to the principal residence). In
this case, however, there are some households that report multiple land parcels and therefore these parcels are not
independent of one another; we would expect them to be more likely to share similar characteristics and potentially
have more similar values. Hence, in the case of dwellings, we adjust for this potential dependence using the smallest
geographic sampling unit and in the case of lands we adjust for this dependence by using the household within
which the lands are nested.
24
Analysis of Results
Two sets of results (the simple and the extended versions) on gender differences in asset
valuation are presented for each country.14 In the simple version, controls are introduced only for
the characteristics of the dwelling or agricultural parcel. For the decomposition of the gender
difference in the value of the dwelling, the characteristics include the size of the dwelling, the
type of materials used, and the amenities available (water source, electricity, and sewage or
sanitation facilities) as well as whether there is a document for the dwelling and its rural or urban
location. The exact list of characteristics differs slightly among the three countries (see Appendix
1 for the full list of variables and results). The decomposition of the value of agricultural land
includes measures of the size of the parcel, location (measured in terms of distance from the
main road), and whether the parcel is irrigated (for Ecuador and Karnataka) and has an
ownership document (for Ecuador).15 It is expected that these characteristics of the dwelling and
land parcel would directly affect the value of the property and thus approximate “objective”
measures of asset value. In the extended version of the decompositions the characteristics of the
respondent are also included. These include measures of education, occupation, age, and marital
status (and religion and caste in Karnataka) and the wealth status of the household (in the case of
Ecuador and Karnataka). Each of these characteristics might affect how the respondent perceives
the value of the property and his or her ability to estimate the value.
Valuation of Agricultural Land
Table 10 reports the decomposition results (both simple and extended versions) of the potential
sales value of agricultural land for the three countries. The predicted mean values of male and
female respondents in all the three countries are statistically significant. Only for Karnataka is
there a significant gender difference in valuation, with men providing higher mean values of
agricultural land than women for both of the estimations, simple and extended.
In Karnataka about 55% of the difference in the mean value of agricultural land reported by male
and female respondents is significantly explained by objective characteristics—parcel size,
distance from main road, and whether the parcel is irrigated. While 45% of the difference in the
mean value is unexplained, this gender difference is not statistically significant. In the extended
version that includes the characteristics of the respondents, about 104% of the difference in the
mean value of agricultural land in favor of male respondents in Karnataka is significantly
explained by the variables that capture the characteristics or quality of the land and the
characteristics of the respondents. Only about 4% of the difference in the mean value of land is
unexplained, and again this gender difference is not significant. In sum, while men provide
higher mean values of land than women, the differences are primarily explained by objective
factors relating to the characteristics of the parcel and the respondents.
14 Since Ecuador has three types of respondents (men, women, and couples) rather than just two, in the first two sub-
sections below only male and female sole respondents are compared to keep the analysis parallel to that of the other
two countries. In a subsequent sub-section, male and female respondents are compared to couples.
15 Ghana and Karnataka did not include a measure for document ownership because of the lack of variation in the
variable, with only 5% of agricultural plots having documents in Ghana and a similar percentage not having
documents in Karnataka.
25
Table 10: Oaxaca Decomposition of Potential Sales Value of Land, by Sex
Valuation
differential
Ecuador Ghana Karnataka
Simple
version
Extended
version
Simple
version
Extended
version
Simple
version
Extended
version
Prediction
(Male)
11,453.8*** 11,453.8*** 8,343.1*** 8,343.1*** 443,345.9*** 443,345.9***
Prediction
(Female)
9,687.9*** 9,687.9*** 7,694.4*** 7,694.4*** 33,8162.5*** 338,909.1***
Difference 1,765.9 1,765.9 648.7 648.7 10,5183.3** 104,436.8**
Explained –1,005.6
(–57%)
1,007.2
(57%)
3,064.0***
(472%)
3,848.2**
(593%)
5,7686.7**
(55%)
108,834.3**
(104%)
Unexplained 2,771.4
(157%)
758.7
(43%)
–2,420.5***
(–372%)
–3,199.5***
(–493%)
4,7496.7
(45%)
–4,397.5
(–4%)
Note: Level of significance: p>|z|; *** 0.01 ** 0.05 * 0.10
For Ghana, while the mean gender difference in land valuation is not significant, the
decomposition analysis produces significant results. Both explained and unexplained factors are
significant in both the simple and extended models. The characteristics of agricultural land (i.e.
size of the parcel and distance from main road) significantly explain 472% of the higher mean
value of land reported by men compared to women, while 372% of the difference is due to
subjective gender differences in valuation. When the individual characteristics of the respondents
are included in the estimation, the role of both objective and subjective factors is amplified.
The decomposition results for Ecuador show neither a statistically significant difference in the
mean value of agricultural land reported by male and female respondents nor in the source of the
differences. These results are consistent in the two versions of the estimation.
Valuation of the Dwelling
The gender decomposition results for the valuation of the dwelling are reported in Table 11 and
show statistically significant gender differences in the mean value of the dwelling in favor of
male respondents in both Ecuador and Ghana, but not in Karnataka. The sources of the
difference in the mean value vary across the three countries.
For Ecuador 40% of the mean gender difference in the value of dwellings is explained by
differences in the quality of the dwelling, with the remaining 60% emanating from subjective
gender differences; both explained and unexplained differences are significant. Both components
remain significant when the characteristics of the respondents are included in the estimation.
This raises the explained source of the difference to 53%, and reduces the unexplained gender
difference to 47%, which could be due to either men overvaluing or women undervaluing the
dwelling.
26
Table 11: Oaxaca Decomposition of Potential Sales Value of Dwelling by Sex
Valuation
differential
Ecuador Ghana Karnataka
Simple
version
Extended
version
Simple
version
Extended
version
Simple
version
Extended
version
Prediction
(Male)
37,733.7*** 37,733.7*** 8,585.9*** 8,615.1*** 336,456.8*** 336,456.8***
Prediction
(Female)
28,355.4*** 28,355.4*** 6,042.8*** 6,042.8*** 330,414.3*** 330,994.4***
Difference 9,378.3*** 9,378.3*** 2,543.0** 2,572.2*** 6,042.5 5,462.4
Explained 3,716.9**
(40%)
4,945.1***
(53%)
–645.2
(–25%)
–354.9
(–14%)
36,215.4*
(586%)
61,725.6
(1110%)
Unexplained 5,661.4**
(60%)
4,433.2*
(47%)
3,188.3**
(125%)
2,927.2*
(114%)
-30,172.9
(–486%)
-56,263.2
(–1010%)
Note: Level of significance: p>|z|; *** 0.01 ** 0.05 * 0.10
In Ghana, where the gender differences in valuation are also significant, the higher quality of
dwellings reported by female respondents relative to their male counterparts explains 25% of the
difference but is not statistically significant. The unexplained component, however, is
statistically significant and explains 125% of the gender difference. Once individual
characteristics are taken into account, 114% of the difference in the mean value of dwellings in
favor of male respondents is unexplained, while the explained component (which again is not
significant) falls. These results suggest that for Ghana, like Ecuador, the higher mean value of
dwellings reported by men is due in part to significant subjective gender differences in valuation.
In Karnataka, 586% of the gender difference in the valuation of dwellings is significantly
explained by differences in the quality of the dwelling, while the unexplained differences are not
significant. In the extended version, neither component is statistically significant.
Comparisons of Values between Male/Females and Couples for Ecuador
In Ecuador over half of the households with married couples or those in consensual unions were
interviewed jointly and were thus excluded from the male/female comparison reported above.
Here we evaluate the differences in asset valuation between individual males and couples as well
as individual females and couples. Table 12 presents the results of the Oaxaca decomposition of
the sources of the differences in the mean values of agricultural land and dwellings for the simple
estimation only, since couples would have two sets of individual characteristics. For land, the
gender difference in the mean value is not significant, and neither are explained or unexplained
components.
27
Table 12: Oaxaca Decomposition of Potential Sales Value of Land and Dwelling
by Male/Female and Couple Respondents in Ecuador (simple version)
Valuation
differential
Land Dwelling
Male &
couple
Female &
couple
Male &
couple
Female &
couple
Prediction (Male) 11,453.8
***
--- 37,733.7
***
---
Prediction
(Female)
--- 9,687.9
***
--- 28,355.4
***
Prediction
(Couple)
14,028.5
***
14,028.5
***
21,392.4
***
21,392.4
***
Difference –2,574.8 -4,340.6 16,341.3
***
6,963.1
***
Explained –2,780.0
(–108%)
-147.8
(3%)
9,851.9
***
(60%)
6,139.4
***
(88%)
Unexplained 205.2
(8%)
-4,192.8
(97%)
6,489.5
***
(40%)
823.7
(12%)
Note: Level of significance: p>|z|; *** 0.01 ** 0.05 * 0.10
With regard to the dwelling, in contrast, a statistically significant difference in the mean value is
estimated between male and couple respondents as well as female and couple respondents. The
difference in the characteristics of the dwelling explains 60% of the difference in the mean value
between male and couple respondents, while the remaining 40% is due to subjective factors in
valuation, with both components being significant. While differences in the characteristics of the
dwelling underscore 88% of the difference in the mean value of the dwelling between female and
couple respondents and is statistically significant, the remaining unexplained difference is not
significant. This suggests that the quality of the dwelling largely explains the differences in both
men’s and women’s valuations compared to couples; however, an important subjective element
is apparent in men’s higher valuations compared to couples.
Turning to the implications of the above comparative analysis, there is in effect no systematic
pattern of gender difference in the responses regarding asset values across assets and countries.
While a significant gender difference in the value of agricultural land is shown for Karnataka, in
Ecuador and Ghana there is no significant difference in the mean value of land reported by male
and female respondents. The reverse is the case regarding gender differences in the mean value
of the dwelling, these being significant in Ecuador and Ghana but not in Karnataka.
The sources of the gender difference in mean asset value also differ across countries. In the case
of land, the significant gender difference in mean land value found in Karnataka is largely
attributable to objective characteristics of the land parcels owned by men and women. For Ghana,
where the gender difference in the value of agricultural land is not significant, the analysis
reveals significant unexplained differences in land values. Neither objective nor subjective
factors play a significant role in explaining gender differences in the value of land in Ecuador.
With regard to the value of dwelling, both the asset characteristics and subjectivity in the
responses on housing value account for the gender differences favoring men in the value of the
dwelling in Ecuador. In Ghana, by contrast, housing characteristics do not explain the higher
28
values reported by men; the gender difference in the mean value of the dwelling is due to
subjective differences in asset valuation. In Karnataka, men respondents do not report
significantly higher values than women respondents for dwellings; whatever little difference
there is in the simple model is due to differences in the dwelling characteristics.
In Ecuador significant differences in the mean value of the dwelling is reported when comparing
male and female respondents (Table 11) and male and female respondents compared to couple
respondents (Table 12). In the case of the female versus couple comparison, this difference is
almost entirely accounted for by the characteristics of the dwelling, with subjective valuation
behavior contributing marginally to the difference in the mean value. In contrast, in the male
versus couple comparison, subjective factors do appear to be important in explaining the higher
male mean values. Taking both sets of decompositions together suggest that in Ecuador when
men are interviewed alone they may tend to overvalue their dwelling as compared to when they
are interviewed as part of a couple.
Comparisons of Values within Couples
For some of the households, we have two estimates of value for the same property and thus can
compare male and female responses for assets with the same characteristics. This is the case in
Ecuador when both members of the couple were not interviewed jointly. In the individual
interview, the second respondent was also asked about the value of the property if s/he was an
owner. In Karnataka, for all assets, the potential sales value was always asked of the second
respondent as well.
Table 13: Comparison of Means Reported by Husbands and Wives, Principal Residence
and Agricultural Parcels.
Principal Residence Agricultural Parcels
Ecuador
Male 32,435 13,683
Female 31,040 11,002
t-test p= .230 p=0.064
*
N 338 79
Karnataka
Male 298,071 407,690
Female 297,864 328,000
t-test p=0.995 p=0.006
**
N 1,718 1,528
29
In both Karnataka and Ecuador, men and women within the same household reporting on land
provide statistically significant different values for agricultural land, with men’s valuation
exceeding that of women. However, they do not do so for housing.16
Conclusions
This paper examines four methodological questions regarding the collectipn of individual level
asset data.
First: Which assets should be included in a survey? Our results suggest that immoveable wealth
(residence, agricultural land, and other real estate) and nonfarm businesses account for more than
80% of gross physical household wealth. Thus, only including these assets may suffice if one is
interested in capturing a measure of total wealth. As always, the country context and the
particular focus of the research should guide the final choice of assets to be included in any
survey.
Second: Is it necessary to interview more than one person in a household when collecting asset
data at the individual level? There are two potential reasons to interview multiple people within
a household. The first is that one person may not have full information and thus it would be
necessary to interview a second person to fill in the information gaps. Our results indicate,
however, that few additional assets are identified by interviewing a second person within the
household. Therefore, from the point of view of acquiring a complete listing of assets owned by
all household members, it might suffice to interview one respondent.
The second reason to interview more than one respondent in a household is that perceptions on
the ownership of assets may differ among household members and some of these differences
may be important for gender analysis. Even husbands and wives may not provide consistent
answers regarding who owns the major assets within the household. In particular, to identify the
empowering effects of property ownership for women, these results suggest that one may want to
interview both the man and woman in a couple to identify whose understanding of property
rights affects the outcomes of household decisions.
The third question is: Which type of value measure for immoveable property is the most reliable?
Two methods have been employed to answer this question. The first is the incidence of reporting
missing values—such as the absence of a market, not knowing the value, and respondents’
refusal to divulge a value. The results broadly suggest that rental value for residences in all three
countries is characterized by a higher incidence of missing values, primarily on account of
missing rental markets. There are some differences in the specific type of missing value across
rural and urban areas, and by sex of the respondent within each country. Potential sales value has
the lowest incidence of missing values for residences in Ghana and Ecuador while in Karnataka
construction value has the least missing.
16 The numbers reported here include outliers. For the Karnataka sample, when the outliers are excluded for housing,
the sample drops to 1,667 and the mean difference is highly significant (p=0.000).
30
A similar pattern is seen with respect to agricultural land in Ecuador and Karnataka, with
rental/lease value having a higher incidence of missing values than potential sale value.
Therefore, for Ecuador and Karnataka it appears that one might not acquire the best estimates of
value by asking rental values for residences, especially in rural areas and for agricultural plots—
potential sales values in general yield a higher proportion of valid value responses. In the case of
Ghana, no clear pattern emerges with the valuation of the place of residence. Regarding the
valuation of agricultural land, however, potential sales values are associated with a higher
incidence of missing values than rental values.
The second method adopted is to examine certain statistical properties of the values obtained.
Again we find that, by and large, the potential sales value for both types of assets is characterized
by lower coefficients of variation, skewness, and kurtosis than are rental and construction values.
The sole exception to this pattern is residences in Karnataka where the potential sales value
exhibits greater scatter and deviation from the mean.
In sum, potential sales value seems to be the most robust asset value measure in terms of
yielding both the most non-missing values as well as regarding certain statistical properties.
However, further analysis of the relationships between the value measures is needed to conclude
that it is in fact the best valuation measure for all data collection and analytical purposes. Further,
it is unknown how close the potential sales values obtained are to their true values. Such a
comparison is not possible in the absence of secondary data to benchmark the survey values.
The final methodology question is: Are there gender differences in reporting values? The gender
patterns of who is able to provide responses regarding the value of assets are not systematic
across assets and countries. Whether male or female respondents are more likely to report values
for assets varies according to the country, type of value measure used, and the asset.
Oaxaca-Blinder decompositions of the differences in reported values provided by men and
women provide some insights into the gendered patterns of responses. In Karnataka, men
provide higher values for agricultural land on average, but this is explained by the characteristics
of the land. In both Ecuador and Ghana, on average, there are no statistically significant
differences in predicted values based on the sex of the respondent. However, using just the
sample of households in which a man and woman both provide values for the same parcel of
land or dwelling, men do provide higher values for agricultural land in both Ecuador and
Karnataka.
Considering the values provided on the dwelling, in Ecuador and Ghana men on average give
higher values for their dwelling. This is not explained by the differences in housing and
individual characteristics alone. Contrarily, in Karnataka there are no differences in predicted
values based on the sex of the respondent. The analysis of just couples in which both provided
values for the dwelling for Ecuador and Karnataka does not find any gender differences in
reported values.
The results suggest that there are not large systematic differences in values reported based on the
sex of the respondent that hold across assets and countries. The differences are specific to
particular assets and countries. When there are gender differences, it tends to be that men on
31
average report higher values than women. Unfortunately, data is not available to tell us whether
men overvalue or women undervalue assets.
References
Allendorf, K. (2007). Do Women’s Land Rights Promote Empowerment and Child Health in
Nepal? World Development, 35(11): 1975–1988.
Altonji, G. J. and Blank, R.M. (1999). Race and Gender in the Labour Market. In Ashenfelter, O.
C and Card, D. (ed.) Handbook of Labour Economics, Volume 3C, Elsevier.
Bardasi, E., Beegle, K., Dillon, A., and Serneels, P. (2011). Do Labor Statistics Depend on How
and to Whom the Questions Are Asked? Results from a Survey Experiment in Tanzania. The
World Bank Economic Review, 25(3): 418–447.
Blinder, A. S. (1973). Wage Discrimination: Reduced Form and Structural
Estimates, Journal of Human Resources, 8: 436-455.
Davies, James B. (2008). An Overview of Personal Wealth. In J. Davies (ed.), Personal Wealth
from a Global Perspective. Oxford University Press and UNU-WIDER, pp. 1-23.
Davies, J.B., Sandstrom, S., Shorrocks, A. and Wolff, E.N. (2008). The World Distribution of
Household Wealth. In J. Davies (ed.), Personal Wealth from a Global Perspective. Oxford
University Press and UNU-WIDER, pp. 395-418.
Deere, C.D., Contreras, J. and Twyman, J. (2013a). Patrimonial Violence: A Study of Women’s
Property Rights in Ecuador. Latin American Perspectives.
http://lap.sagepub.com/content/early/2013/06/17/0094582X13492133.
Deere, C.D., Oduro, A.D. Swaminathan, H. and Doss, C. (2013b). “Property Rights and the
Gender Distribution of Wealth in Ecuador, Ghana and India.” Journal of Economic Inequality 11
(2): 249-265.
Doss, C., Deere, C.D., Oduro, A.D., Swaminathan, H., J. Y., Suchitra, Lahoti, R., Baah-Boateng,
W., Boakye-Yiadom, L., Contreras, J., Twyman, J., Catanzarite, Z., Grown, C. and Hillesland, M.
(2011). The Gender Asset and Wealth Gaps: Evidence from Ecuador, Ghana and Karnataka,
India. Indian Institute of Management Bangalore. www.genderassetgap.org
Doss, C. (2005). The Effects of Intrahousehold Property Ownership on Expenditure Patterns in
Ghana. Journal of African Economies, 15(1): 149–180.
Doss, C. (2013). Intrahousehold Bargaining and Resource Allocation in Developing Countries.
The World Bank Research Observer, 28(1): 52–78.
Fisher, M., Reimer, J. J., and Carr, E. R. (2010). Who Should be Interviewed in Surveys of
Household Income? World Development, 38(7): 966–973.
32
Fletschner, D., and Mesbah, D. (2011). Gender Disparity in Access to Information: Do Spouses
Share What They Know? World Development, 39(8): 1422–1433.
Oaxaca, R. L. (1973). Male-Female Wage Differentials in Urban Labor Markets. International
Economic Review, 14: 693-709.
Stanley, T. D. and S. B. Jarrell (1998). Gender Wage Discrimination Bias? A Metaregression
Analysis. Journal of Human Resources, 33: 947–973.
Weichselbaumer, D. and R Winter-Ebmer (2005). A Meta-Analysis of the International Gender
Wage Gap. Journal of Economic Surveys 19: 479–511
33
Appendix 1. Full tables of the Oaxaca Decomposition
Ecuador
Appendix Table E1i: Results of Blinder-Oaxaca Decomposition of Market Value of Agricultural Land
Market Value of Ag.
Land
Simple Model with Asset Characteristics Extended Model with Asset and
Respondent Characteristics
Coefficient Std. Err. z P>|z| Coefficient Std. Err. z P>|z|
OVERALL
Prediction (Male) 11453.76 2802.26 4.09 0.000 11453.76 2782.08 4.12 0.000
Prediction (Female) 9687.89 1061.01 9.13 0.000 9687.89 1067.54 9.07 0.000
difference 1765.87 3005.02 0.59 0.557 1765.87 2974.98 0.59 0.553
explained -1005.55 716.50 -1.40 0.160 1007.22 1531.50 0.66 0.511
unexplained 2771.42 3086.99 0.90 0.369 758.65 3400.10 0.22 0.823
EXPLAINED
Hectares -113.94 340.26 -0.33 0.738 -76.54 293.35 -0.26 0.794
Time to road -14.83 157.59 -0.09 0.925 -16.91 186.38 -0.09 0.928
Documented -597.77 530.57 -1.13 0.260 -453.91 523.01 -0.87 0.385
Irrigated -279.00 321.50 -0.87 0.385 -295.29 345.36 -0.86 0.393
Sierra 18.13 187.71 0.10 0.923
Household Wealth Rank
Lower Third as the Base
Middle Third 0.88 186.11 0.00 0.996
Upper Third 6.10 263.31 0.02 0.982
Respondent Owner 364.09 399.16 0.91 0.362
Age 1591.90 851.19 1.87 0.061
Married/Consensual Union as the base
Single/Widowed/ Divorced/ Separated 152.38 373.38 0.41 0.683
Less Than Primary Education as the base
Primary Education 199.79 439.19 0.45 0.649
Some Secondary Education -349.86 428.76 -0.82 0.415
Secondary Education or More 67.39 599.11 -0.11 0.910
Wage Employment as the base
Casual Employment -22.81 739.41 -0.03 0.975
Self Employed 1.96 249.89 0.01 0.994
Domestic Worker -100.20 133.64 -0.75 0.453
No Employment 54.93 378.89 0.14 0.885
UNEXPLAINED
Hectares 1716.43 1112.42 1.54 0.123 3393.37 2390.38 1.42 0.156
Time to road -1508.77 1423.08 -1.06 0.289 -3165.61 2227.03 -1.42 0.155
Documented 3899.78 3725.33 1.05 0.295 1584.73 3864.76 0.41 0.682
Irrigated 2419.65 1878.20 1.29 0.198 2418.68 2166.36 1.12 0.264
Constant -3755.68 3360.40 -1.12 0.264 -15121.56 18016.03 -0.84 0.401
Sierra 25129.55 16861.75 1.49 0.136
Household Wealth Rank
Lower Third as the base
Middle Third -14253.36 21016.46 -0.68 0.498
Upper Third 50.08 615.51 0.08 0.935
Respondent Owner 6171.21 9917.61 0.62 0.534
Age -35686.38 17034.66 -2.09 0.036
Married/ Consensual Union as the base
Single/ Widowed/ Divorced/ Separated -1435.29 4431.78 -0.32 0.746
Less Than Primary Education as the base
Primary Education -10719.71 4305.61 -2.49 0.013
34
Some Secondary Education -1559.23 1091.50 -1.43 0.153
Secondary Education or More -3551.72 3421.48 -1.04 0.299
Wage Employment as the base
Casual Employment -1157.11 3834.54 -0.30 0.763
Self Employed -2827.19 17609.95 -0.16 0.872
Domestic Worker 89.75 100.14 0.90 0.370
No Employment 2471.16 3855.33 0.64 0.522
Number of observations 215
Number of Males 54
Number of females 161
Appendix Table E1ii: Results of Blinder-Oaxaca Decomposition of Market Value of Agricultural Land for
Male vs. Couple and Female vs. Couple
Market value of Ag.
Land
Male Vs. Couple Female Vs. Couple
Coefficient Std. Err. z P>|z| Coefficient Std. Err. z P>|z|
OVERALL
Predicted (male) 11453.76 2699.59 4.24 0.000
Predicted (female) 9687.89 1075.19 9.01 0.000
Group (2) 14028.53 3203.93 4.38 0.000 14028.53 3279.73 4.28 0.000
difference -2574.77 4036.70 -0.64 0.524 -4340.64 3378.51 -1.28 0.199
explained -2779.96 2521.43 -1.10 0.270 -147.84 1450.52 -0.10 0.919
unexplained 205.20 3489.72 0.06 0.953 -4192.80 3081.20 -1.36 0.174
EXPLAINED
Hectares -440.48 2007.31 -0.22 0.826 -279.64 1288.16 -0.22 0.828
Time to road 79.34 537.50 0.15 0.883 121.28 221.53 0.55 0.584
Documented -1301.12 880.47 -1.48 0.139 114.44 368.80 0.31 0.756
Irrigated -1117.71 923.30 -1.21 0.226 -103.92 343.04 -0.30 0.762
UNEXPLAINED
Hectares 1729.76 1751.17 0.99 0.323 -33.57 2775.19 -0.01 0.990
Time to road -2146.86 2802.88 -0.77 0.444 -665.20 2205.55 -0.30 0.763
Documented -2264.79 3985.00 -0.57 0.570 -6982.36 4052.38 -1.72 0.085
Irrigated 627.47 2043.21 0.31 0.759 -2526.97 1523.75 -1.66 0.097
Constant 2259.62 3007.39 0.75 0.452 6015.30 3637.54 1.65 0.098
No of observations 244 351
No. of males 54
No. of females 161
No. of couples 190 190
35
Appendix Table E2i: Results of Blinder-Oaxaca Decomposition of Market Value of Dwelling
Market Value of
Dwelling
Simple Model with Asset
Characteristics Extended Model with Asset and
Respondent Characteristics
Coefficient Std. Err. z P>|z| Coefficient Std.
Error z P>|z|
OVERALL
Prediction (Male) 37733.69 3332.33 11.32 0.000 37733.69 3332.33 11.32 0.000
Prediction (Female) 28355.42 1535.94 18.46 0.000 28355.42 1535.94 18.46 0.000
difference 9378.27 2916.16 3.22 0.001 9378.27 2916.16 3.22 0.001
explained 3716.85 1584.84 2.35 0.019 4945.06 1879.50 2.63 0.009
unexplained 5661.42 2275.05 2.49 0.013 4433.20 2378.79 1.86 0.062
EXPLAINED
Urban 554.56 334.70 1.66 0.098 593.01 361.60 1.64 0.101
Documented 177.77 202.24 0.88 0.379 139.70 169.93 0.82 0.411
Number of rooms 1381.78 663.01 2.08 0.037 1254.74 596.11 2.10 0.035
Square meters 184.02 146.53 1.26 0.209 158.42 133.85 1.18 0.237
Walls 45.79 65.85 0.70 0.487 44.91 67.39 0.67 0.505
Floor 332.89 389.35 0.86 0.393 283.64 329.98 0.86 0.390
Roof 450.91 540.73 0.83 0.404 395.61 482.15 0.82 0.412
Water source -37.26 106.58 -0.35 0.727 -15.25 101.46 -0.15 0.880
Water location 198.12 201.33 0.98 0.325 31.39 150.06 0.21 0.834
Sewage 428.27 241.96 1.77 0.077 246.27 171.34 1.44 0.151
Sierra -121.10 243.37 -0.50 0.619
Household Wealth Rank
Lower Third as base dummy
Middle Third 120.46 116.77 1.03 0.302
Upper Third 20.25 60.00 0.34 0.736
Respondent Owner 159.86 145.82 1.10 0.273
Age 1803.87 581.07 3.10 0.002
Married/ Consensual Union as the base
Singe/ Widow/ Divorced/ Separated 12.16 124.25 0.10 0.922
Less Than Primary Education Ref ref ref Ref
Primary Education 10.39 84.36 0.12 0.902
Some Secondary Education 37.37 154.25 0.24 0.809
Secondary Education or More 942.18 495.87 1.90 0.057
Wage Employment as base dummy
Casual Employment 157.52 200.74 0.78 0.433
Self Employed 3.57 222.85 0.02 0.987
Domestic Worker -284.06 180.25 -1.58 0.115
No Employment -1049.83 567.48 -1.85 0.064
UNEXPLAINED
Urban 233.37 3514.00 0.07 0.947 2283.76 4287.29 0.53 0.594
Documented -4660.71 4381.13 -1.06 0.287 -3507.48 4028.92 -0.87 0.384
Number of rooms 8113.55 11678.01 0.69 0.487 10904.97 10805.40 1.01 0.313
Square meters -120.76 9979.52 -0.01 0.990 -134.90 9090.18 -0.01 0.988
Walls -8766.00 5393.01 -1.63 0.104 -10117.14 6472.65 -1.56 0.118
Floor 2909.45 5852.67 0.50 0.619 942.06 6388.16 0.15 0.883
Roof 9010.71 5359.73 1.68 0.093 13786.39 8496.96 1.62 0.105
Water source -2384.80 5174.18 -0.46 0.645 -2413.77 7670.15 -0.31 0.753
Water location 2627.49 5299.64 0.50 0.620 1808.08 7849.46 0.23 0.818
36
Sewage 3942.35 5585.27 0.71 0.480 5676.65 6159.26 0.92 0.357
Sierra -2761.63 3615.26 -0.76 0.445
Household Wealth Rank
Lower Third as base dummy
Middle Third 9453.75 9103.65 1.04 0.299
Upper Third 202.61 276.14 0.73 0.463
Respondent Owner 219.36 5309.24 0.04 0.967
Age -15434.74 9976.16 -1.55 0.122
Married/ Consensual Union as base dummy
Singe/ Widow/ Divorced/ Separated 7936.48 3215.51 2.47 0.014
Less Than Primary Education Ref ref ref Ref
Primary Education -2749.16 1644.43 -1.67 0.095
Some Secondary Education -703.16 1482.91 -0.47 0.635
Secondary Education or More -2579.50 2894.27 -0.89 0.373
Wage Employment Ref ref ref Ref
Casual Employment 479.42 507.13 0.95 0.344
Self Employed 2482.33 3152.48 0.79 0.431
Domestic Worker 156.64 101.65 1.54 0.123
No Employment 4378.57 1896.50 2.31 0.021
Constant -5243.25 7703.71 -0.68 0.496 -15876.39 18272.06 -0.87 0.385
No. of observations 1,024
Number of Males 259
Number of Females 765
37
Appendix Table E2ii: Results of Blinder-Oaxaca Decomposition of Market Value of Dwelling for Male vs.
Couple and Female vs. Couple
Market Value of
Dwelling
Male Vs. Couple Female Vs. Couple
Coefficient Std. Err. z P>|z| Coefficient Std. Err. z P>|z|
OVERALL
Predicted (Male) 37733.69 3360.04 11.23 0.000
Predicted (Female) 28355.42 1455.29 19.48 0.000
Predicted (Couple) 21392.37 1522.01 14.06 0.000 21392.37 1420.00 15.07 0.000
difference 16341.32 3160.53 5.17 0.000 6963.05 1586.36 4.39 0.000
explained 9851.86 1761.93 5.59 0.000 6139.35 1144.79 5.36 0.000
unexplained 6489.45 2336.29 2.78 0.005 823.70 1138.59 0.72 0.469
EXPLAINED
Urban 2077.15 603.28 3.44 0.001 1470.78 389.86 3.77 0.000
Documented 443.93 288.04 1.54 0.123 765.70 220.36 3.47 0.001
Number of rooms 1661.21 729.23 2.28 0.023 355.87 315.95 1.13 0.260
Square meters 21.95 328.87 0.07 0.947 -165.83 125.96 -1.32 0.188
Walls 43.23 90.73 0.48 0.634 99.22 74.65 1.33 0.184
Floor 1889.04 649.17 2.91 0.004 1205.88 350.38 3.44 0.001
Roof 2248.59 839.64 2.68 0.007 1327.09 452.21 2.93 0.003
Water source -516.04 402.35 -1.28 0.200 -364.47 333.96 -1.09 0.275
Water location 865.97 595.38 1.45 0.146 706.50 393.84 1.79 0.073
Sewage 1116.83 420.33 2.66 0.008 738.60 237.47 3.11 0.002
UNEXPLAINED
Urban 460.91 3579.88 0.13 0.898 279.35 1473.93 0.19 0.850
Documented -1641.37 3835.69 -0.43 0.669 2519.80 1237.11 2.04 0.042
Number of rooms 12444.16 10631.64 1.17 0.242 4254.17 4124.76 1.03 0.302
Square meters -79.19 10348.49 -0.01 0.994 45.32 1875.15 0.02 0.981
Walls -7077.83 5094.78 -1.39 0.165 1586.39 1894.26 0.84 0.402
Floor 719.18 5702.54 0.13 0.900 -1840.00 2588.21 -0.71 0.477
Roof 6872.44 4969.59 1.38 0.167 -1667.68 2514.64 -0.66 0.507
Water source -1635.34 5278.58 -0.31 0.757 635.15 2589.60 0.25 0.806
Water location 2697.60 5321.74 0.51 0.612 31.47 2685.80 0.01 0.991
Sewage 5154.99 5424.00 0.95 0.342 1162.59 2088.40 0.56 0.578
Constant -11426.11 7711.04 -1.48 0.138 -6182.86 3836.08 -1.61 0.107
No. of Observations 815 1,321
No. of Couples 556 556
No. of Males 259
No. of Females 765
38
Ghana
Appendix Table G1: Results of Blinder-Oaxaca Decomposition of Market Value of Agricultural Land
Bootstrap 1000 iteration
(Replications based on 137 clusters in enumeration areas)
Market Value of Ag.
Land
Simple Model with Asset
Characteristics Extended Model with Asset and
Respondent Characteristics
Coefficient Std. error z P>|z| Coefficient Std. error z P>|z|
Differential
Prediction (Male) 8343.07 815.8364 10.23 0.000 8343.07 815.8364 10.23 0.000
Prediction (Female) 7694.36 726.8929 10.59 0.000 7694.35 726.8929 10.59 0.000
Difference 648.71 992.9958 0.65 0.517 648.72 992.9958 0.65 0.514
Explained 1064.02 1057.082 2.90 0.004 1848.24 1512.228 2.54 0.011
Unexplained -415.31 807.6718 -3.00 0.003 -1199.52 1135.44 -2.82 0.005
Detailed Decomposition
Explained
plot size 3271.188 995.9589 3.28 0.001 3220.969 948.7648 3.39 0.001
proximity to main road -31.85046 19.19591 -1.66 0.097 -47.02797 31.39523 -1.5 0.134
Urban -127.4228 179.9955 -0.71 0.479
Age -269.1395 260.2695 -1.03 0.301
currently single 426.234 431.3789 0.99 0.323
Owner -12.74484 86.01114 -0.15 0.882
No education as the base dummy
basic education 2.974625 72.45328 0.04 0.967
secondary education 641.4507 473.5503 1.35 0.176
tertiary education 98.95022 70.70161 1.4 0.162
Contributing family work & domestic employee as the base
Wage employment -92.14368 106.0317 -0.87 0.385
Self-employment 6.140021 64.99968 0.09 0.925
Total 3064.022 1057.082 2.90 0.004 3848.239 1512.228 2.54 0.011
Unexplained
plot size 912.6507 1964.992 0.46 0.642 1075.031 1800.882 0.6 0.551
proximity to main road 236.2203 472.1383 0.50 0.617 378.0651 507.411 0.75 0.456
Urban 33.73397 659.0111 0.05 0.959
Age 7806.347 4570.449 1.71 0.088
currently single 69.60731 1161.027 0.06 0.952
Owner 4412.996 2550.001 1.73 0.084
No education as the base dummy
basic education 1608.874 1039.078 1.55 0.122
secondary education 199.2996 304.7205 0.65 0.513
tertiary education 171.8577 107.2574 1.6 0.109
Contributing family work & domestic employee as the base
Wage employment 278.3376 234.1866 1.19 0.235
Self-employment 6933.11 2729.702 2.54 0.011
Constant -3563.162 2280.133 -1.56 0.118 -26166.78 8908.427 -2.94 0.003
Total -2420.45 807.6718 -3.00 0.003 -3199.524 1135.44 -2.82 0.005
No. of observations 2,051 2,051
No. of males 1,534 1,534
No. of females 517 517
39
Appendix Table G2: Results of Blinder-Oaxaca Decomposition of Market Value of Dwelling
Bootstrap 1000 iterations
(Replications based on 137 clusters in enumeration areas)
Market Value of
Dwelling
Simple Model with Asset
Characteristics Extended Model with Asset and
Respondent Characteristics
Coefficient Std. error z P>|z| Coefficient Std. error z P>|z|
Differential
Prediction (Male) 8585.87 819.9454 10.47 0.000 8615.078 787.0097 10.95 0.000
Prediction (Female) 6042.84 800.3992 7.55 0.000 6042.84 829.828 7.28 0.000
Difference 2543.03 1148.019 2.22 0.027 2572.237 1135.139 2.27 0.023
Explained -645.242 947.9374 -0.68 0.496 -354.9199 1302.377 -0.27 0.785
Unexplained 3188.272 1276.118 2.5 0.012 2927.157 1543.685 1.90 0.058
Detailed Decomposition
Explained
Urban -475.2465 282.7266 -1.68 0.093 -457.8342 291.12 -1.57 0.116
Document -217.0526 203.7853 -1.07 0.287 -164.5957 172.4144 -0.95 0.34
No of rooms 1638.672 467.4521 3.51 0.000 1645.448 432.7356 3.8 0.000
Dwelling size (sq. m) -75.58058 348.0223 -0.22 0.828 -79.0734 333.4823 -0.24 0.813
Wall quality -829.1508 372.856 -2.22 0.026 -761.8348 398.9998 -1.91 0.056
Floor quality 76.52822 158.6763 0.48 0.63 92.2419 128.3034 0.72 0.472
Roofing quality -393.6661 299.0048 -1.32 0.188 -307.7161 273.3925 -1.13 0.26
Electricity -339.5076 173.2006 -1.96 0.05 -292.2111 217.3212 -1.34 0.179
Water facility -18.17555 125.85 -0.14 0.885 -9.178735 120.7646 -0.08 0.939
Toilet facility -12.06229 290.1315 -0.04 0.967 -3.187977 184.8467 -0.02 0.986
Age -60.64093 227.1689 -0.27 0.79
currently single -117.6125 995.946 -0.12 0.906
Owner -63.35922 94.68507 -0.67 0.503
No education as the base dummy
basic education 108.0155 132.9453 0.81 0.417
secondary education 650.3214 360.4756 1.8 0.071
tertiary education 147.2362 178.4399 0.83 0.409
Contributing family work & domestic employee as the base
Wage employment -438.3562 322.1438 -1.36 0.174
Self-employment -242.5824 235.132 -1.03 0.302
Total -645.242 947.9374 -0.68 0.496 -354.9199 1302.377 -0.27 0.785
Unexplained
Urban 2201.608 883.2984 2.49 0.013 2407.281 968.5354 2.49 0.013
Document -86.68667 796.9128 -0.11 0.913 -144.7472 868.5324 -0.17 0.868
No of rooms -601.0703 1884.719 -0.32 0.75 -353.1411 1953.005 -0.18 0.857
Dwelling size (sq. m) 253.9322 1020.573 0.25 0.804 253.9355 1150.507 0.22 0.825
Wall quality 10893.17 4935.852 2.21 0.027 10143.54 4643.071 2.18 0.029
Floor quality -843.651 3646.306 -0.23 0.817 -1401.452 3622.977 -0.39 0.699
Roofing quality -612.5189 4712.42 -0.13 0.897 -3106.803 5458.189 -0.57 0.569
Electricity -1038.93 1001.876 -1.04 0.3 -1066.057 1086.584 -0.98 0.327
Water facility -885.6463 685.3544 -1.29 0.196 -759.7788 684.5543 -1.11 0.267
Toilet facility 1593.818 1003.319 1.59 0.112 1831.551 1030.965 1.78 0.076
Age 1006.381 4646.873 0.22 0.829
currently single 58.28209 1554.671 0.04 0.97
Owner -741.1217 4222.565 -0.18 0.861
No education as the base dummy
basic education 2284.041 873.2225 2.62 0.009
secondary education 531.4143 298.0542 1.78 0.075
tertiary education -264.0251 394.642 -0.67 0.503
40
Contributing family work & domestic employee as the base
Wage employment 29.91921 467.3754 0.06 0.949
Self-employment 420.6071 3280.74 0.13 0.898
Constant -7685.755 5788.651 -1.33 0.184 -8202.674 8273.156 -0.99 0.321
Total 3188.272 1276.118 2.5 0.012 2927.157 1543.685 1.9 0.058
No. of observations 684 684
No. of males 515 515
No. of females 169 169
41
Karnataka
Appendix Table I1: Results of Blinder-Oaxaca Decomposition of Market Value of Agricultural Land
Sale Value as Dependent Variable (bootstrap errors)
Simple Model with Asset
Characteristics Extended Model with Asset and
Respondent Characteristics
Coefficient Std. Err z P>|z| Coefficient Std. Err z P>|z|
OVERALL
group_2 (Male) 443345.9 40395.14 10.98 0.000 443345.9 40425.65 10.97 0.000
group_1 (Female) 338162.5 42230.41 8.01 0.000 338909.1 42465.19 7.98 0.000
difference 105183.3 48827.43 2.15 0.031 104436.8 48299.44 2.16 0.031
explained 57686.65 26726.25 2.16 0.031 108834.3 47609.75 2.29 0.022
unexplained 47496.69 51526.44 0.92 0.357 -4397.5 56661.65 -0.08 0.938
EXPLAINED
leasedout 4555.522 6972.343 0.65 0.514 5306.628 8389.876 0.63 0.527
Distance from Road 2172.819 2575.146 0.84 0.399 1486.638 1868.114 0.8 0.426
irrigated 8542.944 10668.16 0.8 0.423 6146.472 8110.659 0.76 0.449
Area of Land 42415.37 22456.08 1.89 0.059 36216.86 22339.63 1.62 0.105
Education Level: Illiterate/Below Primary as Base
Primary/Higher Primary -1985.273 2962.387 -0.67 0.503
Secondary 18732.55 9423.933 1.99 0.047
Higher Secondary and Above 33582.69 16031.05 2.09 0.036
Occupation: Wage Employed as base
Self Employed 66911.23 33299.97 2.01 0.045
Casual Worker -27625.39 17484.8 -1.58 0.114
Contributing Family Worker/Homemaker/Other -17821.64 20424.94 -0.87 0.383
Marital Status: Currently Single -40107.25 31643.86 -1.27 0.205
Respondent owner of Land -4613.979 9559.637 -0.48 0.629
Age -5387.633 6125.082 -0.88 0.379
Religion: Islam as base
Hindu/Others 4416.724 6226.727 0.71 0.478
Caste: SC/ST as base
BC/OBC 265.7372 2910.141 0.09 0.927
Other Caste 2348.6 3497.897 0.67 0.502
Region: Bangalore as base
Northern Maidan 7595.619 16111.49 0.47 0.637
Southern Maidan -8615.357 14912.12 -0.58 0.563
Coastal -2464.654 10365.62 -0.24 0.812
Malnad 27.63292 5896.181 0 0.996
Wealth Groupings: Bottom 20% as base
Middle 20-60% 9758.788 8319.585 1.17 0.241
Top 60-100% 26921.54 15942.53 1.69 0.091
Urban -2262.287 7306.551 -0.31 0.757
Unexplained
leasedout 34208.73 19120.23 1.79 0.074 2872.903 20248.01 0.14 0.887
Distance from Road -12210.13 16535.9 -0.74 0.46 -12394.37 19190.65 -0.65 0.518
irrigated 16951.15 31596.17 0.54 0.592 23144.15 39307.91 0.59 0.556
Area of Land -153096.9 89664.12 -1.71 0.088 -158505.9 91807.45 -1.73 0.084
Education Level: Illiterate/Below Primary as Base
Primary/Higher Primary -20899.73 23753.54 -0.88 0.379
Secondary -44863.68 32460.7 -1.38 0.167
Higher Secondary and Above 23805.48 16656.65 1.43 0.153
Occupation: Wage Employed as base
Self Employed 81338.38 68601.24 1.19 0.236
Casual Worker 108368.8 88421.3 1.23 0.22
42
Contributing Family Worker/Homemaker/Other 25004.04 60541.36 0.41 0.68
Marital Status: Currently Single 29346.92 74893.08 0.39 0.695
Respondent owner of Land 55784.73 99244.73 0.56 0.574
Age 210020.1 183403.8 1.15 0.252
Religion: Islam as base
Hindu/Others 24659.51 326410.1 0.08 0.94
Caste: SC/ST as base
BC/OBC -27948.07 68507.36 -0.41 0.683
Other Caste 17342.34 12441.33 1.39 0.163
Region: Bangalore as base
Northern Maidan 819587.8 291072.6 2.82 0.005
Southern Maidan 603394.5 215516.8 2.8 0.005
Coastal 243249.2 108468 2.24 0.025
Malnad 242724.3 102572.9 2.37 0.018
Wealth Groupings: Bottom 20% as base
Middle 20-60% -66104.61 65287.47 -1.01 0.311
Top 60-100% -81138.91 131152.3 -0.62 0.536
Urban 53778.31 27319.46 1.97 0.049
Constant 161643.8 101417.1 1.59 0.111 -2156964 857320.1 -2.52 0.012
Number of observations 2505 2504
Number of males 2185 2185
Number of females 320 319
43
Appendix Table I2: Results of Blinder-Oaxaca Decomposition of Market Value of Dwelling
Sale Value as Dependent Variable (Bootstrap 1000 Iterations)
Simple Model with Asset Characteristics Extended Model with Asset and
Respondent Characteristics
Coefficient Std. Err z P>|z| Coefficient Std. Err Z P>|z|
OVERALL
group_2 (Male) 336456.8 28750.65 11.7 0.000 336456.8 28750.65 11.7 0.000
group_1 (Female) 330414.3 64202.91 5.15 0.000 330994.4 64399.3 5.14 0.000
difference 6042.5 65128.5 0.09 0.926 5462.4 65316.9 0.08 0.933
explained 36215 19983.92 1.81 0.070 61725.64 42968.47 1.44 0.151
unexplained -30172 65533.4 -0.46 0.645 -56263.24 85025.4 -0.66 0.508
EXPLAINED
Water Source: Piped Water as base
Public Running Water 3054.97 6883.793 0.44 0.657 1223.9 2782.299 0.44 0.66
Well or other Source 3532.08 6044.211 0.58 0.559 1288.312 2531.918 0.51 0.61
Type of House: Non-Durable House
Semi-Durable House 15.63 1597.772 0.01 0.992 70.67148 1759.588 0.04 0.968
Durable House -371.60 1658.724 -0.22 0.823 39.22195 1392.559 0.03 0.978
Sanitation: No Toilet as base
Has Toilet -3068.96 6527.537 -0.47 0.638 -244.9246 1139.857 -0.21 0.830
Electricity: No electricity as base
Has Electricity -1531.192 942.433 -1.62 0.104 -2975.436 1322.709 -2.25 0.024
Number Rooms 21887.56 10433.19 2.10 0.036 21443.47 10112.77 2.12 0.034
Dwelling area 12696.94 7200.168 1.76 0.078 10296.25 5779.555 1.78 0.075
Religion: Islam as base
Hindu/Others -3967.815 5666.906 -0.7 0.484
Caste: SC/ST as base
BC/OBC 1934.3 1942.715 1.00 0.319
Other Caste 6376.366 6142.198 1.04 0.299
Region: Bangalore as base
Northern Maidan -60043.33 38447.58 -1.56 0.118
Southern Maidan -23627.88 29707.96 -0.8 0.426
Coastal 35538.77 25451.89 1.4 0.163
Malnad 8956.788 33239.14 0.27 0.788
Fuel: Solid Fuel as base (Firewood)
Non-Solid Fuel (LPG,
Kerosene) -6665.411 4667.669 -1.43 0.153
Wealth Groupings: Bottom 20% as base
Middle 20-60% 3112.859 4689.323 0.66 0.507
Top 60-100% 21301.7 9018.915 2.36 0.018
Urban -15960.88 6569.893 -2.43 0.015
Education Level: Illiterate/Below Primary as Base
Primary/Higher Primary -1759.113 1339.672 -1.31 0.189
Secondary 1669.682 3116.092 0.54 0.592
Higher Secondary and Above 37580.5 10535.13 3.57 0.000
Occupation: Wage Employed as base
Self Employed -1534.931 22609.42 -0.07 0.946
Casual Worker -6285.042 6793.528 -0.93 0.355
Contributing Family Worker/Homemaker/Other -28718.61 26208.5 -1.10 0.273
Marital Status: Currently single 74713.88 40879.66 1.83 0.068
Respondent owner of Dwelling -7248.304 6575.942 -1.10 0.270
Age of Respondent -4789.36 2738.509 -1.75 0.080
UNEXPLAINED
44
Water Source: Piped Water as base
Public Running Water 115134.5 113973.8 1.01 0.312 87166.26 78781.85 1.11 0.269
Well or other Source 31907.53 27454.94 1.16 0.245 19951.64 20216.3 0.99 0.324
Type of House: Non-Durable House
Semi-Durable House 62762.16 49610.2 1.27 0.206 -35534.77 81238.41 -0.44 0.662
Durable House 56794.17 24966.37 2.27 0.023 5070.334 40593.15 0.12 0.901
Sanitation: No Toilet as base
Has Toilet -13880.22 32153.03 -0.43 0.666 31691.86 29416.07 1.08 0.281
Electricity: No electricity as base
Has Electricity 3202.846 54101.9 0.06 0.953 20967.91 76795.45 0.27 0.785
Number of Rooms 191936.5 122782.9 1.56 0.118 216101.5 127492.1 1.70 0.090
Dwelling area -127109.2 169932 -0.75 0.454 -163975.4 137646.6 -1.19 0.234
Religion: Islam as base
Hindu/Others -402796.1 604553 -0.67 0.505
Caste: SC/ST as base
BC/OBC -18923.39 39689.85 -0.48 0.634
Other Caste -101078.4 113665.8 -0.89 0.374
Region: Bangalore as base
Northern Maidan 205981.7 387935.1 0.53 0.595
Southern Maidan 152695.6 245667 0.62 0.534
Coastal 45825.58 139124.4 0.33 0.742
Malnad 74706.05 138126 0.54 0.589
Fuel: Solid Fuel as base (Firewood)
Non-Solid Fuel (LPG, Kerosene) 66633.28 42834.96 1.56 0.12
Wealth Groupings: Bottom 20% as base
Middle 20-60% -43943.78 48338.76 -0.91 0.363
Top 60-100% -23247.72 45542.82 -0.51 0.610
Urban 60106.46 27072.49 2.22 0.026
Education Level: Illiterate/Below Primary as Base
Primary/Higher Primary 313.9149 13939.07 0.02 0.982
Secondary -27197.45 18278.52 -1.49 0.137
Higher Secondary and Above -76722.06 55201.5 -1.39 0.165
Occupation: Wage Employed as base
Self Employed -65254.78 59236.61 -1.1 0.271
Casual Worker -202616.3 162184.6 -1.25 0.212
Contributing Family Worker/Homemaker/Other -102986 92611.63 -1.11 0.266
Marital Status: Currently Single -2515.201 46085.83 -0.05 0.956
Respondent owner of Dwelling 153173.4 136805.8 1.12 0.263
Age of Respondent -361766.7 292884.9 -1.24 0.217
Constant -350921.2 151910.7 -2.31 0.021 431909.3 592767.1 0.73 0.466
Number of observations 2972 2970
Number of males 2348 2348
Number of females 624 622
