Understanding the Consequences of Economic Empowerment in Developing Countries

We conduct multiple lab in the field experiments to investigate the possible underlying mechanisms of backlash.

2023-12-03

2024-01-10

We have two primary outcomes:

1. Willingness to backlash: For experiment 1,3 and 4 we create a variable indicating whether the individual is willing to pay to reduce their partner’s income. This will be constructed as an indicator variable (0/1) where 1= if the DM took the action (i.e., paid to reduce their partners income).
2. Willingness to hide: For experiment 2 we create a variable indicating whether the individual is willing to pay to hide whether the household participated in an empowerment program. This will be constructed as an indicator variable (0/1) where 1= if the DM took the action.

See description above.

Our secondary outcomes include the following:

1. Labelling of policy - difference on respondent’s supportiveness depending on whether a hypothetical program is labelled as about ‘empowerment’ as about a more neutral ‘families and communities’
2. Respondent’s decision to sign petition. Specifically, at the end of the survey, respondents will be offered the chance to sign a letter showing their support for women’s rights and gender equality. We also experimentally vary a prime before the petition. Some respondents are randomly assigned a role model prime, where they are reminded of a famous persons support for gender quality. A random subset are not shown this prime prior to being explained the petition.
3. Non Incentivized willingness to pay - e.g. survey questions about how much, hypothetically, the respondent would be willing to pay for the relevant behaviour, such as overruling a female given leadership. We consider this a robustness test.

In addition, we will conduct analyses, assessing the relationship between the backlash behaviours identified above (the primary outcome), and
a. Partner characteristic information: This is a set of dummies depending on the information given to DMs about their partner. We randomise three different characteristics. This will tell us if different norms impact backlash. This is restricted to the social norms experiment.
b. Gender attitudes index - weighted summary index of the gender attitudes asked in the survey. We will create an index for the husband and a separate index for the wife.
c. Norms about appropriate behaviour in the game.
d. IPV attitudes index - weighted summary index of the IPV attitudes asked in the survey. We will create an index for the husband and a separate index for the wife.
e. Relationship quality index- weighted summary index of the relationship quality questions asked in the survey. This is only reported by women.
f. Controlling behaviours index - weighted summary index of the Controlling behaviours women report
g. Women’s experience of IPV
a. Ever experienced (each of emotional, physical, or sexual IPV)
b. Frequency of experiencing (each of emotional, physical, or sexual IPV) in past 12 months
c. Injuries
h. Men’s likelihood of becoming violent - weighted summary index
i. Decision-making index- weighted summary index of women’s input in decisions in the household, including across domains
a. How much input does woman have
b. Who makes final decision
i. Decision-making aspirations index- weighted summary index of women’s input in decisions in the household, including across domains - Women’s aspiration for greater decision-making.

We also measure a number of robustness checks such as measuring experimenter demand effects, discussion with other survey participants, and social desirability bias. This will be reported in a robustness section.

See above

Our design aims to test and measure multiple possible behavioural explanations for the existence of gender-based backlash.

To systematically study backlash, we conduct multiple experiments. The experimental treatments are explained above. In this section we explain the method to measure backlash/willingness to hide.

In all experiments we use a variant of the take it or leave it method. In our variant, an individual must decide whether they will backlash (i.e., to reduce their partner’s income by Rs 100) or not. The cost of backlash will be randomly assigned at the level of the individual. The cost will either be 10 or 20 Rs. Subjects will only face one cost. In the social image experiment we measure whether they are willing to pay to avoid their partner knowing that their household participated in an empowerment program.



******************************************************
Hypothesis
*****************************************************
Our primary hypothesis is:
All experiments except social image: Decision-makers in the treatment group (given information about their female partner in an empowerment program) backlash at a higher rate than decision-makers in the control group (NOT given information about their female partner being in an empowerment program).

Social Image: DMs in the control group are more likely to avoid their partner knowing their household participated in some female empowerment program relative to DMs in the treatment. Further, DMs are more likely to hide when they are assigned a male partner relative to female.

Differences across experiments:
We don’t have clear expectations on which experiment will have the largest treatment effects. This is primarily because there is no existing evidence testing or hypothesising about these differences.

In addition, we have the following sub-hypotheses:
● The survey measures will follow a similar pattern to the incentivized willingness to pay responses
● Men with greater willingness to pay to backlash in the empowerment program treatment are more likely to have perpetrated IPV and hold more conservative attitudes (see b-i) above.

Main Analysis:
Note: Below is the analysis for the main outcome. We focus only on experiments 1,3 and 4 unless otherwise specified. We estimate all models with and without controls.

Analysis 1: y= B0 + B1 Treatment
Here we merge the experiments and include a dummy =1 if the DM was given information about their female partner in an empowerment program and zero if NOT given information about their female partner being in an empowerment program. The dependent variable is equal to one if the DM was willing to backlash. This analysis will inform us as to whether backlash towards women in female empowerment programs exists across all experiments on average. We do not include the social image decision in this analysis.

Analysis 2: y= B0 + B1 Instrumental Violence Exp + B2 Status Violence Exp
Here we include dummies for each experiment. The dependent variable is equal to one if the DM was willing to backlash. This analysis will inform us as to whether backlash is higher across our experimental settings. The norms experiment is the omitted category.

Note: If the effects of the information sets A, B and C differ in the social norms experiment we will include these as controls and conduct a robustness where we focus on information set C.

Analysis 3: y= B0 + B1 Instrumental Violence Exp*Treatment + B2 Status Exp*Treatment + B4 Social Norms*Treatment + B5Treatment + B6 Instrumental Violence Exp + B7 Status Exp

Here we include dummies for each experiment, the treatment within each experiment and also the interaction between them. The dependent variable is equal to one if the DM was willing to backlash. From this equation we are interested in the following
• The comparison of the controls (i.e., when not provided information about ones partner being in an empowerment program) across experiments. Our focus is on comparing the control of the social norms experiment with the other experiments. This tells us whether there are differences in backlash across experiments when people are not informed that their partner participated in an empowerment program. This is because the setting itself may lead to backlash for instance, giving women status may increase backlash.
• We are also interested in the difference in difference (i.e., the interaction between the experiment and treatment), this informs us as to whether the treatment effect is different across experiments. In other words, do people backlash more against women who are more empowered and does this differ across experimental settings.

Analysis 4: Social Image Concerns
Analysis 4a:
y= B0 + B1 Treatment

Here =1 occurs when the DM is not informed that their partners household participated in a female empowerment program

Analysis 4b:
y= B0 + B1 Male partner*Treatment

In this case we will interact gender of ones partner with the treatment. This will tell us whether social image concerns are generated by other males or females or both.

Analysis 4c:
y= B0 + B1 Backlash occurs in social norms

Here y=1 occurs when the DM decides to not inform their partner that their household participated in a female empowerment program and B1 is whether backlash occurs in the social norms control. This tells us whether people who backlash in the social norms experiment also have social image concerns regarding empowerment programs.

Randomisation will take place via computer.

The unit of randomisation is the individual. That is within each village, we will conduct all experiments and treatments. Within each area the experiment they receive and whether it’s a control or treatment will be randomly assigned.

No

There are two forms of treatment allocation. Subjects will be randomly allocated to one of three experiments (excluding social image), within that experiment they will then be again randomly allocated to a treatment or control. This is all done at the level of the individual.

In total we will collect data from 960 households in 40 villages, which equates to 24 households within each village. This will comprise 960 males and 960 females. As the experiment focuses on the decision of the male household member we focus on the 960 observations of the male here. Althought, if possible, we will try and collect slightly more than 960 around 1000.

In total we have 4 experiments. Everyone will be allocated to one of three- social norms, status and instrumental. All those assigned to social norms will also take part in the social image experiment. Within each experiment we have a treatment and control which means we have 6 variations.

The status and instrumental experiments will each contain 240 observations. This will comprise 120 in the control and 120 in the treatment. In the social norms and social image experiments, we will collect 480 observations. Further, in the social image experiment DMs will be randomly assigned a partner who is male (120) or female (120) in both the control and treatment. In all other cases DMs are assigned to a female partner only. In the social image experiment, we will match DMs with a partner taken from a small sample of respondents who undertook the social norms experiment with either a male partner or were a female DM with a male partner. As this is a small sample we do not study it.

Within each village, we will collect 3 observations for the status and instrumental experiment for each experimental treatment/control variation. This will be doubled for the social norms and social image experiment. (Social norms/social image: 12 obs per village, 6 T and 6 C; Instrumental: 6 obs, 3 T and 3 C and Status: 6 obs, 3 T and 3 C).

Here we calculate the MDE of our main variable of interest—whether an individual has a willingness to reduce another’s income >0. We treatment this as a binary outcome. Focusing on the experiments with less observations (status and instrumental) and assuming a significance level of 0.05, power of .80 and a control rate of willingness to pay of 5%, the MDE is 0.079.