Experimental Design Details
To ensure that there is indeed "plausible deniability" in group (a), we first randomise whether an eligible woman gets a ticket at all (with 90% assigned to get tickets). The fact that there are 10% of eligible women that don't get tickets gives a plausible excuse for those spouses in group (a) that want to hide the fact a ticket has been received (they can say to their spouse "unfortunately you didn't win the lottery"). Without this, the fact that we survey all spouses might "give the game away".
In the first stage, we stratify the randomisation on whether married, village-cluster and caste. In the second stage, we will stratify the randomisation on how the eligible woman was selected in the first stage - in particular, whether the woman was assigned to receive the job ticket directly, or whether her spouse was assigned the ticket. Then we can test for heterogeneous effects of the job by how the woman was selected in.
Unmarried women will also be included in the experiment, but primarily to increase the chances of over-subscription in the second stage. For the first stage, unmarried women will be randomly assigned to either receive the job ticket directly, or have their head of household receive it. In both cases, these "pairs" are treated as if in group (a) (with the "non-ticketed" pair not knowing that the other has the ticket).
Update (December 2016): As of writing this update, we have reached at most the halfway point of our study. We have completed the recruitment and opening for 3 centers and are in the process of recruitment and opening of 3 additional centers. If export demand allows the firm to open additional centers in the new year, we hope to do recruitment and opening of more than 6 centers.
Though noisy, our enrollment data at this point suggests a counter-intuitive finding. In particular, enrollment falls as women get more information in the treatment arm where the husband gets the ticket. This is despite the fact that our data consistently confirm our priors regarding gender differences in preferences for women’s employment (e.g. women are less likely to think that weaving is an "inappropriate" job for women to take).
At this point it seems then that there is value in adding some survey questions to shed light on potential mechanisms that might explain this counter-intuitive effect (if it remains once all data is in). Though we already have some questions in the Endline survey that can help (e.g. "who had most influence over the enrollment decision?" to test whether giving information to the wife reduces her influence over the decision), we decided extra questions would be useful, with two hypotheses in particular in mind:
(a) Backlash/nagging wives: discussions become more argumentative when the wife has more information and is more likely to initiate discussions. For this, we are adding questions relating to how argumentative discussions are.
(b) Magnanimous husbands: husbands like to be generous (e.g. for the reciprocity generosity instigates), and so are more likely to allow enrollment when they can take the credit for it. Husbands can take credit for enrollment when the wife does not have information. For this, we are adding questions relating to whether reciprocity is expected if enrollment is allowed.
In addition to the counter-intuitive effect of giving wives information, we find a large positive effect of giving the ticket to the wife in the Discussion group. We have two hypotheses for explaining this and have added questions to address them:
(a) Endowment effects: whoever receives the ticket gets extra influence over the decision due to an endowment effect and only in the discussion group is this effect not confounded by asymmetric information. We added questions about who holds the ticket now to understand endowment effects.
(b) Signalling effects: participants may feel that ticket receipt is a signal that they are the person who should make the decision. This effect may be particularly strong in the discussion group and/or confounded by asymmetric information in the other treatments. To study this, we added questions on why people think we gave the ticket to one spouse instead of the other.
All these survey changes were to the Endline (phone) survey, and took effect for the first time in the field on December 5, 2016 (for the Endline for the 4th center).