There is some literature that has tried to establish how large the stigma of participating in welfare programs are. Indeed, the fact that not all those that can obtain welfare benefits actually take them is an indication that feel bad about it, as long as they are aware of the relevant programs.Just how bad do people feel about this? The problem is that another factor comes in: time costs. Getting welfare benefits typically takes time: getting to the welfare office, waiting, filling forms and getting interviewed. This implies that the econometrician has a hard time figuring out the psychological costs of welfare participation. Indeed, time and psychological costs have very different policy implications. Time costs have a good reason: to select against welfare cheaters who are more likely to be discouraged by lost time. Psychological costs, however, are lower for the cheater. Also, discovering that the latter costs are high for the intended recipients is detrimental to the success of a program.
Colleen Flaherty Manchester and Kevin Mumford build a model of labor supply with an additive, one-time cost to welfare program participation in the utility function. Using structural estimation, they infer what the time cost as well as the psychological cost are, using the example of food stamps and WIC (healthy food checks for small children and pregnant mothers-to-be) in the US. They conclude that time costs amount to 0.5 hours a week for food stamps and 3 hours a week for WIC. The psychological costs are larger, about 3 hours for each program, but incurred only in the first week of participation, by assumption.
I am a big fan of structural estimation because it gives us the right quantitative framework for policy experiments. But the inferences in this case seem rather heroic to me. Indeed, there is no direct observation of the time cost. It is inferred from estimating a labor supply equation on a population where many do not work, for reasons beyond observables, and for them virtual wages are inferred. And the psychological cost amounts to a residual, and accordingly is estimated with a very large standard error. With all these caveats, I am surprised that there is no data that would measure the actual time lost through participation in those programs. This would considerably tighten estimates.