Experimental Design
In 1987, in cooperation with multiple government agencies, the Self Employed Women’s Association (SEWA) Union, a collection of trade groups with a membership of over 500,000 women in Gujarat, launched a housing lottery for beedi workers. Union members with a monthly income of less than 700 rupees (US$11.28) were eligible to participate, and all 497 eligible women entered the lottery. They came primarily from two caste groups, Koshti (35%) and Padmasali (41%), while Muslims (10%) were the third largest group. SEWA leaders conducted the drawing of the 110 winners at a public gathering on International Housing Day in 1987. After the lottery, the Union worked with the Ahmedabad Urban Development Authority (AUDA) to construct homes. Six years later, AUDA built the houses on vacant government land situated 7.5 miles from the city center. The units were single-story rowhouses of approximately 200 square feet situated back-to-back with a narrow alley running in between (Colony A). Lottery winners received a significant housing subsidy on Colony A units. The construction cost of a unit was 45,000 rupees, and the winner paid an initial deposit of 900 rupees. She then paid 124 rupees (US$2) in monthly rent. This rate was guaranteed for 20 years and was less than half the average rent reported by losers in our survey.
The follow up survey took place between May and October 2007, 20 years after the housing lottery and 14 years after lottery winners obtained possession of Colony A units; the official list of lottery winners – which included participant name and address in Colony A – was available through SEWA Union. Researchers obtained the names and addresses of lottery losers from multiple sources. First, the SEWA Union officially maintained a list of 297 lottery losers (out of 387) who had indicated an interest in entering a future housing lottery. Second, a former SEWA employee provided a participant subset list that included names of an additional 26 lottery losers. In addition to names and addresses, this list of 109 participants also had a handful of baseline characteristics (1987 address, marital status, husband’s occupation, and the incomes of the participant, husband and household) that researchers use in the following subsection as part of a randomization check. Finally, researchers undertook tracking interviews with the listed lottery participants in an attempt to identify the remaining 64 (17%) lottery losers. Ultimately, researchers obtained an additional 30 names (47% of missing) as referrals from women who were in the lottery, and their participation was verified upon contact (from now on, “referrals list”). Hence, the names of only 34 out of 387 lottery losers – or 9% of losers and 7% of all lottery participants – remain unidentified. After constructing the participant list, researchers tracked and surveyed 443 participants (or a family member, in cases of death or mental illness), giving a response rate of 96% of the 463 participants who could be named (89% of the original 497 participants). No one refused the survey. Identical attrition rates of 4% across winners and losers drawn from the 463 listed participants, and similar rates of mortality and proxy surveying among the 443 participants found. Attrition from the set of named participants and inclusion on the participant list are uncorrelated with observable characteristics.
The respondents were asked about household demographics, various socio-economic indicators, and the health, schooling, marital status, and current occupation of their children. Researchers collected detailed data on their residential location and mobility over the last 20 years and obtained a full employment history for the participant and her husband. A neighborhood and networks module asked respondents about their social interactions with immediate neighbors and adult children, risk-sharing mechanisms (in terms of exposure to major city-level shocks in the last six years and the coping mechanisms they used to deal with them), and collective action undertaken over the last three years. Researchers also collected GPS coordinates for participants’ 1987 and 2007 residential locations. In 2011, a real estate agent valued a subset of current lottery winner and loser residences and researchers conducted qualitative fieldwork with randomly selected 21 participants from four strata: five losers, four winners who never moved into Colony A, six winners who moved into Colony A but subsequently moved out, and six winners who still lived there. Researchers used semi-structured interviews to probe respondents on how their housing mobility opportunities affected their socioeconomic wellbeing and their networks.
Our analysis sample encompasses the 443 tracked and surveyed lottery participants, which were spread across 18 neighborhoods, with half of the women living in the inner-city neighborhoods of Amraiwadi (11%), Bapunagar (15%), Dhudeshwar (12%) and Rakhial (13%). To reduce data-mining concerns, researchers group outcomes into three thematic indices: urbanicity, property rights, and amenities. Each index is the simple average of z-scores for component outcomes and is balanced at baseline across treatment and control. The urbanicity index includes distance (in miles) from home to city center, time to walk to nearest school and time to walk to nearest hospital. The average respondent lived 2.3 miles from the city center (measured as a straight line) and a 17-minute walk to the nearest school. The property rights index includes home ownership in 1987, years of home ownership prior to 1987, whether possessed an official title, and whether title was in the participant’s name. Interpretations of reported ownership are ambiguous because many of the residential structures in city slums are illegal, and occupants frequently claim ownership absent official documentation. Sixty-three percent report that someone in their household owns the property (with average ownership being seven years) and 50% report having documentation. Just under 10% state that the documentation is in their names. Finally, the amenities index includes whether the 1987 house had a separate kitchen, a private toilet, and a water connection, and whether it was safe for a woman to walk in the neighborhood after 10 pm. The majority of women (84%) claim to live in a safe neighborhood by this measure.
Researchers also asked respondents their reasons for choosing their 1987 residence. Over 30% report that they chose their location to be near family, and another 12% state that neighborhood resources drove their location choice. Only 3% named price as the main factor. For the arguably random subset list of 109 lottery applicants recovered from a former SEWA employee, we also have baseline income information, which we present in Panel C. Comparing baseline characteristics across winners and losers shows that both household types were similar at the time of the lottery in nearly all observable dimensions. However, Muslims are over-represented in the loser category by 6.8 percentage point. To further investigate whether Hindus were possibly favored in the housing draw, researchers regressed respondent perception of whether the lottery was conducted fairly on respondent religion and find no difference across Hindu and Muslim participants; therefore, researchers interpret Muslim underrepresentation among the winners as a random occurrence.
Researchers assessed the intent-to-treat (ITT) effects - changes in social interactions, informal insurance and collective action - with and without controls such as ethnic identity indicators for whether the household is Muslim, Koshti caste, or Padmasali caste (omitting all other caste groups), a variable indicating whether the participant’s name was referred by another member (rather than gathered from a Union list), and whether the participant’s information was reported by proxy because she had died or was unable to answer due to mental illness. When the unit of observation is a child, researchers cluster standard errors at the participant level. Researchers assess how winning the lottery influenced subsequent residential mobility and program take-up, as well as urbanicity and housing quality, economic wellbeing.
When assessing the ITT results, researchers look at the mobility effect, driven by the two-thirds of winners who moved to Colony A, and in some instances, by the one-third of winners who still live there (program compliers), as well as the income effect associated with selling or renting subsidized housing by non-movers (program non-compliers). This was done to be able to extrapolate results to settings with different compliance rates or enforcement of lease agreements. In addition to running regressions on the entire sample, researchers ran regressions on the set of only winners.