Gender Roles in Risky Household Decisions: A Study in Bangladesh

We elicit risk preferences of married couples in agricultural households in rural Bangladesh under different treatments. We vary the incentive structure and decision-making process across treatments. Using a sequential game structure we try to simulate intra-household decision-making in paternalistic societies where the female household member might have to react to decisions made by the primary male household head.

There are three treatments that sample units/respondents are subjected to: Solo, Shared, and Joint. At the heart of each is a simple risk elicitation task similar to that implemented by Binswanger et al. (1981).

In the Solo treatment a subject is asked to make a choice from among six options each with a unique degrees of riskiness each. Each option has two equally likely non-negative payoffs. In this treatment, the subject gets the entire amount of any resulting payoffs.

In the Shared treatment the subject makes a similar choice from the set of six risky choices but any resulting payoffs is equally divided between the subject and their spouse.

In the Joint treatment the subject and their spouse playing as a team (household) each make a decision from the set of risky decisions. The combined payoffs from the two chosen risky alternatives is equally divided between the two subjects. This treatment has a sequential structure with one subject, the first-mover, making their choice first and their spouse, the second-mover, making their choice conditional on the first-mover’s choice. In order not to divulge the first-mover’s actual choice to the second-mover we ask them which option they would choose for every potential choice of the first-mover. This also allows us to collect the complete strategy followed by the second-mover. Husbands and wives will each take turns, once playing as the first-mover while their spouse in the second-mover and once playing as the second-mover while their spouse plays as the first-mover. The combined payoffs from the two chosen options is equally divided between the two subjects. The first-mover is also incentivized to guess the second-movers strategy.

References:
Binswanger, H. P. (1981). Attitudes toward risk: Theoretical implications of an experiment in rural India. The Economic Journal, 91(364), 867-890.

2024-11-25

2024-12-12

Lottery choice by wife and husband under each treatment, the risk aversion parameters calculated based on the respective lottery choices, and elicited beliefs of respondents about their spouse's choices and through those their risk preferences.

The risk aversion parameters are calculated assuming constant relative risk aversion utility function for individuals.

We also collect data on savings and loan taking behavior of both husbands and wives. Comparing their answers within a household also allows us to test for asymmetric information in the household.

We use a within-subject design. Each household in our sample is subjected to all three treat

In selected households enumerators will ask the primary male adult and his wife to take part in a risk preference assessment game each. A male enumerator will attend to the adult male household member and a female enumerator will attend to his wife. Both household members will be interviewed simultaneously in separate parts of their household. The experiment has a within-subject design, with participants playing a risk preference assessment task under different treatments. Within a household both members will also be asked to play the same game each, though they will do so in isolation from their partner.

Each household member will first play the Solo game/treatment. Then they will each play the Shared game/treatment.

The household members will then play one round of the Joint game/treatment with the husband as the first-mover and the wife as the second-mover and one round of the Joint game/treatment with the husband as the first-mover and the wife as the second-mover.

In order not to divulge the first-mover’s actual choice to the second-mover we ask them which option they would choose for every potential choice of the first-mover. This also allows us to collect the complete strategy followed by the second-mover.

Players will be paid for any one of the four games in addition to a participation bonus. At the end of all the games, the enumerators will randomly decide which game the wife and the husband will be paid for. This is done to limit the possibility of members finding out each other's choices and performance. The exact method used to decide which game to pay for is described later.

1. There is a 20% chance that the household members will be paid for their performances in the Solo game
2. 20% chance that the household members will be paid for their performances in the Shared game: Each subject gets half of the winnings from the lottery they chose in their Shared game plus half of the winnings from the lottery their spouse chose in their Shared game.
3. 20% chance that the wife and husband will be paid for the Joint game in which the wife was the first-mover and the husband was the second-mover: The wife and the husband is each paid half of the combined winnings from the lotteries chosen by the wife as first-mover and the lottery chosen by the husband as the second-mover. The wife will also each be paid an extra 150 BDT if she correctly predicts her husband’s strategy as the second-mover, 100 BDT if she makes only one mistake, and 50 BDT if she makes only two mistakes.
4. 20% chance that the wife and husband will be paid for the Joint game in which the husband was the first-mover and the wife was the second-mover: The wife and the husband is each paid half of the combined winnings from the lotteries chosen by the wife as second-mover and the lottery chosen by the husband as the first-mover. The husband will also each be paid an extra 150 BDT if he correctly predicts his wife’s strategy as the second-mover, 100 BDT if he makes only one mistake, and 50 BDT if he makes only two mistakes.
5. 20% chance that each household member will be paid 250 BDT in addition to the participation bonus equal to 150 BDT

Randomization to decide which treatment will be incentivized in a household is performed on SurveyCTO software in the tablet computers used by the female enumerator dealing with the female respondent in the household. Based on the realized value of a random number with uniform distribution over the following values {1, 2, 3, 4, 5}, the respondents in the selected household are paid:

1. There is a 20% chance that the household members will be paid for their performances in the "Solo" game: In this case each enumerator will pay their player their respective winnings from their "Solo" game in addition to the 150 BDT participation bonus,
2. 20% chance that the household members will be paid for their performances in the "Shared" game: In this case each enumerator will pay their player half of the winnings from the lottery they chose in the "Shared" game plus half of the winnings from the lottery their spouse chose in their "Shared" game. In addition, each player will receive the 150 BDT participation bonus.
3. 20% chance that the wife will be paid for her performance in her "Joint-First" game and the husband will be paid for his performance in his "Joint-Second" game: In this case the female (/male) enumerator will pay the wife (/husband) half of the combined winnings from the lottery chosen by the wife in her "Joint-First" game and the lottery chosen by the husband in his "Joint-Second" game. The wife will also each be paid an extra 150 BDT if she correctly predicts her husband’s strategy in his "Joint-Second" game, 100 BDT if she makes only one mistake, and 50 BDT if she makes only two mistakes. Finally, each household member will also receive the participation bonus equal to 150 BDT.
4. 20% chance that the wife will be paid for her performance in her "Joint-Second" game and the husband will be paid for his performance in his "Joint-First" game: In this case the male (/female) enumerator will pay the husband (/wife) half of the combined winnings from the lottery chosen by the husband in his "Joint-First" game and the lottery chosen by the wife in her "Joint-Second" game. The husband will also each be paid an extra 150 BDT if he correctly predicts his wife’s strategy in her "Joint-Second" game, 100 BDT if he makes only one mistake, and 50 BDT if he makes only two mistakes. Finally, each household member will also receive the participation bonus equal to 150 BDT.
5. 20% chance that each household member will be paid 250 BDT in addition to the participation bonus equal to 150 BDT
The reason for the elaborate randomization attached to the payoffs is so that players are not able to deduce which exact game they are being paid for and thus not be able to deduce their spouse's choices for the remuneration received. The enumerators will also not declare to the players their spouse's choices or which game they are being paid for.

Randomization is used in the experiment to determine which of the treatments will be incentivized in a household. This is done at the individual level. However, every household in the sample is subjected to the four treatment types: Solo, Shared, Joint (Wife as first mover) and Joint (Husband as first mover).

No

Treatment will be provided at the level of the individual (unit of observation) and not at cluster level. So, the number of clusters is same as the sample size that is 1050 households out of which the primary adult male and their wife will be surveyed.

The planned number of households is 1050 that is 1050 pairs of husbands and wives.

The experiment has a within-subject design. So, each household is subjected to the four treatments in successive order. So, the entire sample of 1050 households is used in eacg treatment arm.

0.2 times the assumed standard deviation of the outcome variables. The standard deviations of the outcome variables were based on the dataset for Chowdhury et al. (2022) who conducted a Binswanger (1981) style risk elicitation exercise amongst Bangladeshi households in a rural population similar to ours. We are interested in multiple hypotheses: 1. For the hypothesis that an individual has similar risk preference in the Solo and Shared treatments: MDE is taken to be 0.40 which is equal to 0.2 times the SD (1.70) from the lottery choice variable in Chowdhury et al. (2022). The variable of interest is discrete taking values either 1 or 2 or 3 or 4 or 5 or 6. Errors are assumed to be normally distributed. Sample size for 0.8 power is 300 households. 2. For the hypothesis that husbands and wives have similar risk preference in the Solo treatments: MDE is taken to be 0.40 which is equal to 0.2 times the SD (1.70) from the lottery choice variable in Chowdhury et al. (2022). The variable of interest is discrete taking values either 1 or 2 or 3 or 4 or 5 or 6. Errors are assumed to be normally distributed. Sample size for 0.8 power is 300 households. 3. For the hypothesis that husbands and wives have similar risk preference in the Solo treatments: MDE is taken to be 0.40 which is equal to 0.2 times the SD (1.70) from the lottery choice variable in Chowdhury et al. (2022). The variable of interest is discrete taking values either 1 or 2 or 3 or 4 or 5 or 6. Errors are assumed to be normally distributed. Sample size for 0.8 power is 300 households. 4. For the hypothesis that an individual's risk preference as calculated from their choice in the Shared treatment is equal to the couple's joint risk preference as represented by either intended choice of first-mover: The variable of interest is risk the aversion parameter calculated based on lottery choices. It is discrete and takes values 0.12, 0.16, 0.4, 1.3, 2.51, and 3.03. The SD is 1.26 for Shared and 1.40 for Joint. The MDE is equal to 0.25 which is equal to 0.2 times the average of the two SDs. Errors are assumed to be normally distributed. Sample size for 0.8 power is 420 households. 5. For the hypothesis that a person's guess about the second-mover's strategy is accurate: MDE is taken to be 0.40 which is equal to 0.2 times the SD (1.70) from the lottery choice variable in Chowdhury et al. (2022). The variable of interest is discrete taking values either 1 or 2 or 3 or 4 or 5 or 6. We assumed that beliefs about lottery choices have the same SD as actual lottery choices. Errors are assumed to be normally distributed. Sample size for 0.8 power is 200 households. 6. For the hypothesis that husbands and wives are equally good at predicting their spouse's strategy: The variable is interest is constructed by calculating the absolute deviation of person's guess about their spouse's lottery choice from the spouse's actual lottery choice. The SD of the constructed variable was calculated using the following formula: Var(Choice-Belief) = Var(Choice) + Var(Belief) – 2*Cov(Choice, Belief). We assume that Cov(Choice, Belief) is equal to 0.2. It is reasonable to assume that a person’s choice and their spouse’s belief about it would be positively correlated. Given these assumptions the calculated the SD is 2.14 giving an MDE of about 0.4. Errors are assumed to be normally distributed. Sample size for 0.8 power is 560 households. References: Binswanger, H. P. (1981). Attitudes toward risk: Theoretical implications of an experiment in rural India. The Economic Journal, 91(364), 867-890. Chowdhury, S., Sutter, M., & Zimmermann, K. F. (2022). Economic preferences across generations and family clusters: A large-scale experiment in a developing country. Journal of Political Economy, 130(9), 2361-2410.