Methodological Experiment on Measuring Asset Ownership from a Gender Perspective (MEXA)Living Standards Measurement Study

Surveys and Methods UnitDevelopment Research Group

The World Bank

Coffee Hour with the LSMS: Better Statistics for Better Outcomes10/01/2015

TALIP KILICSenior Economist

HEATHER MOYLAN Survey Specialist

Motivation• Assets: Key inputs into income generation, capital & credit access &

consumption smoothing

• Research on disparities in personal wealth provide insights into – Intra-household bargaining outcomes, conflicts & ensuing effects– Long-term economic inequities & inter-generational implications– Processes shaping class structures & inter-relationships

• Data on asset ownership largely available at the household-level, even when assets are owned & controlled by individuals – Extensive use of proxy respondents even w/ individual disaggregated data

• Technical guidelines on survey protocols & questionnaire design to capture asset ownership at the individual level do not exist– Persistence of “myths” in the absence of appropriate data & methods– Guidelines key to x-country availability & comparability of sex-disaggregated SDG indicators on asset ownership

Objectives• Assess, through a randomized household survey experiment, the effects of different approaches to respondent selection on measuring individual ownership of & rights to assets• Inform the design of 6 surveys that will be implemented throughout 2015 with support from the UN Evidence and Data for Gender Equality (EDGE) initiative • Inform the UN EDGE guidelines on measurement of individual ownership of & rights to assets

– To be submitted to the UN Statistical Commission for adoption in 2017

Key Questions• Are we doing enough to capture individual asset ownership

patterns by only interviewing the self-identified most knowledgeable household member in household surveys?

• Does reporting change when individuals know that other household members are being interviewed?

• Do married individuals provide different information on their asset ownership when interviewed separately from their partners?

Overview of Treatment Arms

Arm Who? How? What?1 “Most Knowledgeable” Household Member Alone Assets Owned Exclusively/ Jointly by Household Members2 Randomly Selected Member of Principal Couple Alone Assets Owned Exclusively/ Jointly by Household Members3 Principal Couple Together Assets Owned Exclusively/ Jointly by Household Members4 Adult (18+) Household Members Alone, Simultaneous Assets Owned Exclusively/ Jointly by Household Members5 Adult (18+) Household Members Alone, Simultaneous Assets Owned Exclusively/Jointly by Respondent

Scope of Data Collection (1)• Basic Socio-Economic Information• Core Asset Information (Asset Level)– Dwelling & Residential Land– Agricultural Land– Non-Agricultural Land & Other Real Estate– Livestock – Non-Agricultural Businesses – Agricultural Equipment– Consumer Durables– Financial Assets & Liabilities– Valuables

Scope of Data Collection (2)

Type of Ownership/Rights Individual Disaggregation

Reported Ownership Within-Household Identification of Individuals

Outside-Household Identification of Individuals

Capacity to Exercise Right Independently?

Identification of Provider of Consent/Permission

Economic Ownership

Documented Ownership

Bundle of Rights- Bequeath- Sell- Rent Out- Use as Collateral- Make Improvements/Invest

Context: UgandaCountry Attribute Estimate YearPopulation 37.5 mil 2013Rural Population Share 84.6% 2013Annual Population Growth Rate 3.3% 2013National Absolute Poverty Rate 19.5% 2012Rural Absolute Poverty Rate 22.4% 2012GDP Per Capita in Current USD 657 2013Ag Value Added As a Share of GDP 25.3% 2013Share of Ag Employment in Total Employment 65.6% 2013• Legal pluralism in acquisition & enforcement of property rights• Large disconnect – de facto (customary) vs. de jure (legislative) arrangements

Sampling Design• 140 Enumeration Areas (EAs) selected with probability proportional to size across Uganda

– Actual EA coverage: 137• Rural/Urban EA Split: 60/40 percent • HH listing in each EA for random selection of sample HHs• 20 HHs randomly selected in each EA, 4 randomly allocated to each treatment arm in each EA prior to field work

Fieldwork• Implementing agency: Uganda Bureau of Statistics (UBoS)• Fieldwork period: April-August 2014• Used computer-assisted personal interviewing (CAPI) application designed in Survey Solutions

Fieldwork (2)• Interviews in Arms 4 & 5 attempted to be conducted in parallel

Fieldwork (3)• Pursued to achieve gender (& language) match between respondents & enumerators

Challenges – Field Reports• Approaching communities & respondents

– “In some rural areas, misconception regarding the objective of the survey led some respondents to keep away from us on the first day.”– “There was resistance from respondents in Kampala: people are paranoid even after introducing the survey very well.”– “Residents of an EA charged against us on our second day because of the questions on asset values and formal ownership.”

• Scheduling interviews– “It is hard to find the principal couple at home together. People go to the gardens and take care of other responsibilities.”– “ID Project: Most people are mobilized away from home.”– “Kampala: It is almost impossible to find people at home. It is even more difficult to find the principal couple together at home.”– “Arms 4 & 5: Interviewing respondents simultaneously is a challenge – especially during weekdays.”

Challenges – Field Reports (2)• Asset valuation

– “Difficulty in verifying the figures respondents give us. When a respondent is asked the size of a parcel that is located 20 kilometers away and that is claimed to be 20 even 40 acres, it is hard to refute or prove what we are told since we cannot physically see this land.”– “Declaration of assets by couples particularly land and bank accounts requires significant amount of explanation/probing.”– “Some respondents not willing to provide information initially but would later provide after more sensitization.”

Sample CompositionTable 1. MEXA Households Interviewed

    Interviewed w/ a Couple  Initial

Allocation Expected Interviewed % of Expected Any More than

1 InterviewBoth Members of

Couple InterviewedTA #1 548 490 495 100% 325 -- N/A

TA #2 548 299 304 100% 304 -- N/A

TA #3 548 299 272 91% 272 -- 272

TA #4 548 490 475 97% 303 188 161

TA #5 548 490 481 98% 319 190 169

Total 2,720 2,068 2,027 98% 1523 378 602

Sample Composition (2)Table 2. Distribution of Treatment Arm 4 & 5 Households

According to # of Adults InterviewedTA #4 TA #5

Total % Total %Households Interviewed 475 481All Eligible Adults Interviewed 295 0.61 286 0.59

4 adults 14 0.03 15 0.033 adults 20 0.04 23 0.052 adults 137 0.29 133 0.281 adults 124 0.26 115 0.24

Subset of Eligible Adults Interviewed 180 0.38 195 0.413 out of 4 15 0.03 12 0.022 out of 4 20 0.04 21 0.041 out of 4 11 0.02 12 0.022 out of 3 26 0.05 23 0.051 out of 3 8 0.02 12 0.021 out of 2 100 0.21 115 0.24

Average # of Adults Interviewed 1.62 1.61

Interview Dynamics• Gender Match-up : 81.6% of female respondents paired

w/ female enumerators & 74.6 % for male respondents

• Simultaneous Interviews (Arms 4 & 5): 63% - based on the paradata collected by the CAPI platform

• Alone Interviews: 90% - Across Modules & Arms

• Duration of Interviews (minutes) – Average for Entire (Household & Individual) Interview: 60– Average for Individual Interview Across Arms: 34

39 vs. 29 in Arms 3 vs. 5

Interview Dynamics (2)Table 3. Selected Module Interview Durations (Minutes)

Conditional on Module Being Administered  Overall TA #1 TA #2 TA #3 TA #4 TA #5

Household Interview 36 35 36 36 38 35

Individual Interview

Dwelling 10 9 9 10 10 11

Agricultural Land 19 18 18 23 20 18

Livestock 5 5 5 5 5 3

Non-Farm Business 12 13 13 14 11 10

Financial Assets 4 5 4 5 5 3

Interview Dynamics (3)Table 4. Selected Ownership & Rights Question Durations (Seconds)

Conditional on Questions Being Administered

 Module Reported Economic Bequeath Sell Collateral Rent Out Improve

Dwelling 16.4 14.3 11.2 7.4 8.2 6.6 6.8

Agricultural Land 7.0 5.5 8.1 6.2 5.8 5.8 5.5

Non-Farm Enterprises 9.4 7.2 8.9 6.8 6.0 5.5 5.2

Empirical Analysis• Balance tests indicate that randomization was successful• Unit of analysis: Adult individual• Focus on all 13 asset types• Pooling individual reporting to identify owners in Arm 4/5 HHs with multiple respondents• Comparison across Arms 1 through 5 using HHs with a couple

– Arm 4 & 5 effects robust in comparisons across Arms 1, 4 & 5 using all HHs• Alternative sample specifications in Arms 4/5

Empirical Analysis (2)• Core specification:

– Probit (OLS) regressions for binary (continuous) dependent variables– Standard errors clustered at the EA-level, take into account stratification – i & h represent individual & household– represent constant & error term– represent identifiers for Arms 2 through 5– Comparison category: Arm 1 (i.e. standard of practice)– Tests of equality of for complete inter-arm comparisons– C represents a vector of control variables– Regressions estimated on the whole & by gender

(1)

Empirical Analysis (3)• (6) Primary dependent variables – All binary

– Reported Ownership – Overall; Exclusive; Joint (w/ anyone)– Economic Ownership – Overall; Exclusive; Joint (w/ anyone)

• (6) Secondary dependent variables – Continuous in italics– Right to Bequeath – Right to Sell – Right to Rent Out– Right to Use as Collateral – Right to Invest/Make Improvements– PCA-based rights index across 5 rights

Key Findings• The following synthesis focuses on

• Effects statistically significant at least at 5 percent level• Female adults in households with a couple, for reported & economic ownership indicators, pooling individual reporting

• No statistically significant effects of Arm 2 – across all asset types• Arm 3 exerts statistically significant positive effects on only overall & joint dwelling reported ownership incidence

– No similar effect for other assets & for economic ownership• Arm 4 exerts statistically significant positive effects on reported & economic ownership (through joint ownership) across all asset types

– Larger effects for assets w/ higher ownership incidence– Largely indistinguishable from Arm 5 effects

Treatment Arm 4 Effects• Sizeable in magnitude, significant at the 1 percent level

Table 5. Selected Arm 4 Marginal Effects on Female Incidence of Ownership

 TA1 Mean

Average T4 Effect

Effect As a % of TA1 Mean

Reported OwnershipDwelling 0.110 0.090 82%Agricultural Land 0.120 0.077 64%Large Livestock 0.182 0.165 90%Small Livestock 0.276 0.099 36%Financial Assets 0.182 0.088 49%Durables 0.476 0.162 34%Small Agricultural Equipment 0.500 0.170 34%Economic OwnershipDwelling 0.271 0.130 48%Agricultural Land 0.271 0.122 45%

Treatment Arm 4 Effects (2)• Stronger in magnitude if the Arm 4 sample is restricted to

– Adults in households with both principal couple members interviewed– To principal couple in households with both members interviewed

• Robust if sample includes all households, whether or not with a couple

Table 6. Selected Arm 4 Marginal Effects on Female Incidence of OwnershipTA4 Sample: Adults in HHs with Both Principal Couple Members Interviewed

 TA1 Mean

Average T4 Effect

Effect As a % of TA1 Mean

Reported OwnershipDwelling 0.110 0.119 108%Agricultural Land 0.120 0.147 123%Economic OwnershipDwelling 0.271 0.197 73%Agricultural Land 0.271 0.214 79%

Untangling the Knot• Intra-household variation in reporting on ownership in Arm 4

Table 7. Agreement on Individual Ownership in Arm 4 HHs w/ 2+ RespondentsLevel of Analysis: Individual Owners

  Dwelling Ag Land NFE FinancialAverage Share of Respondents in Unanimous Agreement on an Individual's Reported Ownership 0.61 0.48 0.37 0.26

Average Share of Respondents in Unanimous Agreement on an Individual's Economic Ownership 0.51 0.45 0.26 N/A

Untangling the Knot(2)• Intra-household variation in reporting on ownership in Arm 4• Respondents’ reporting on their own ownership & rights

– Arm 4 effects disappear – true for both male & female respondents, & for respondent subsample that form a couple– Given the disproportionate Arm 4 effects in previous slides among females vis-à-vis males, the effect is in part driven by males reporting higher rates of joint ownership in Arm 4 vis-à-vis Arm 1

Untangling the Knot (3)• Intra-household variation in reporting on ownership in Arm 4• Respondents’ reporting on their own ownership & rights• Hidden assets

Table 8. Hidden (Selected) AssetsPooled Data Across Treatment Arms

 # of Respondent

Owners# of Owners Reporting a

Hidden Asset

ModuleOverall

#Male Female Overall Male Female

# % % %Parcels 833 62.3% 37.7% 25 3.0%    Non-Farm Enterprises 536 42.5% 57.5% 1 0.2%    Financial Accounts 795 46.9% 53.1% 111 14.0% 16.4% 12.8%Loans Given Out 287 56.4% 43.6% 78 27.2% 25.3% 29.6%Loans Taken Out 410 51.1% 48.9% 93 22.7% 24.6% 17.7%

Concluding Thoughts• Clear value addition of implementing Arm 4

– Robust & (statistically & economically) sizeable impacts across modules– Individual interviews would serve to alleviate data gaps more broadly– Not unreachable given constraints shared by MEXA & other HH surveysBut need...– Careful questionnaire design & piloting– Agile, gender-balanced, mobile teams– Re-thinking fieldwork management, scheduling interviews

• Incidence of ownership of & rights to assets among the female adult population remains low even under Arm 4– And at least in Uganda, Arm 4 treatment appears to be nudging male respondents to be more inclusive in their reporting on ownership of & rights to assets

Concluding Thoughts (2)• Further methodological research need

– Specificity to Uganda: Need for validation in alternative settings• Analysis of the data from the surveys supported by UN EDGE will be important

– On-going work to… • Identify the correlates of the data generation process using paradata• Improve our understanding of “joint ownership”

– Need for follow-up qualitative work, cognitive interviews• Move from individual- to asset-level analysis

– Not a straightforward task to build a unique panel of assets in Arms 4 & 5 when each respondent created his/her roster of assets!• Estimate survey treatment effects on personal wealth

– But… valuation remains problematic!

Methodological Experiment on Measuring Asset Ownership from a Gender Perspective (MEXA)Living Standards Measurement Study

Surveys and Methods UnitDevelopment Research Group

The World Bank

Coffee Hour with the LSMS: Better Statistics for Better Outcomes10/01/2015

TALIP KILICSenior Economist

HEATHER MOYLAN Survey Specialist

Table 9: Selected Results from Multi-Level Hierarchical Model of Question Duration (Seconds)

Sample: All Recorded Answers in Dwelling Module, TA4Hierarchy: Enumeration Area, Interviewer

Female Enumerator † 0.067 ***Respondent Age (Years) 0.000  Distance to Nearest Population Center (KMs) 0.000  Enumerator & Respondent Language Match † -0.040 ***Enumerator & Respondent Gender Match † 0.076 ***Total Distance/Day Covered by Enumerator (KMs) 0.046 **Respondent Years of Education 0.010  # of Household Members < 7 Years of Age 0.009 **Length of Field Entry (Digits or Letters) 0.021 ***Timing of Interview: Morning † ‡ -0.033  Timing of Interview: Afternoon † ‡ -0.105 ***Timing of Interview: Evening † ‡ -0.153 ***Note: ** p<0.05, ***p<0.01. † Dummy variable. ‡ Omitted category is Night.

Table 10: Share of Joint Dwelling Owners With a Given Right, by Treatment Arm & GenderSample: Respondents in Couple Households, Reporting Regarding Their Own Ownership & Rights

Overall TA 1 TA 2 TA 3 TA 4 TA 5  Female Male Female Male Female Male Female Male Female Male Female MaleSell 34% 79% 32% 69% 33% 90% 42% 83% 39% 81% 26% 68%Bequeath 35% 90% 36% 81% 38% 100% 37% 86% 35% 92% 33% 90%Use as Collateral 32% 83% 32% 75% 33% 95% 39% 85% 35% 85% 39% 85%Rent it Out 42% 86% 36% 81% 48% 90% 47% 85% 47% 85% 36% 87%Make Improvements/Invest

67% 95% 77% 94% 57% 95% 68% 95% 65% 92% 67% 97%

Table 11: Arm 4 Treatment Effects on Log FEMALE Personal Housing Wealthby Sample Specification, Ownership Construct & Approach to Housing Wealth ValuationResults from Multi-Level (Region &

EA) Hierarchical Model Specification Identical to Equation 1, Respondent Provided Values Averaged in TA4 & TA5 Households

 Reported Dwelling

Market Value

 Reported

Ownership Economic Ownership

Core Sample        Adults in Couple HH 0.062 ** 0.081***

Alternative Samples        Adults in Couple HHs & Including only HHs in TA4 With Principal Couple Members Interviewed 0.121 *** 0.131***Principal Couple Members & Including only HHs in TA4 With Principal Couple Members Interviewed 0.121 *** 0.126***

 Estimated Dwelling Reconstruction Cost

 Reported

Ownership Economic Ownership

Core Sample        Adults in Couple HH 0.053 * 0.064***

Alternative Samples        Adults in Couple HHs & Including only HHs in TA4 With Principal Couple Members Interviewed 0.121 *** 0.131***Principal Couple Members & Including only HHs in TA4 With Principal Couple Members Interviewed 0.121 *** 0.127***

Table 12. Average Within-HH Respondent Value for Dwelling As a % of Presumed Most Knowledgeable Member Reported Value in Arm 4 HHs w/ 2+ Respondents

No Trimming 298%# of Households 195   Trimmed Top & Bottom 1% 134%# of Households 191   Trimmed Top & Bottom 5% 34%# of Households 174