Continuing in my theme from last week on how to ask better questions, this week I want to focus on measuring financial services. There is a nice paper by Robert Cull and Kinnon Scott in the World Bank Economic Review based on experiments from Ghana (gated version, ungated version). It’s a similar approach to the labor paper I talked about last week – they vary the amount of detail and who responds to see which types of questions work better. And, as with last week, the headline of this research is that whom you ask and how you ask the questions matters – but the details are somewhat different.
The setup: There are two main approaches out there to capturing financial access (at least in terms of getting some kind of cross-national comparable data). The products based approach that one finds with Finscope surveys (website) and the institutions based approach of the LSMS surveys. The Finscope surveys are more detailed, asking about access to various products. So to measure whether someone is banked, questions about whether the individual has an ATM card, debit card, savings plus account, current account, savings account at a bank, post office account, bank loan, or overdraft facility would be asked (i.e. 8 separate questions). The simple, institutional approach of the LSMS is something like “Some people like to keep their money in an account with a bank. Do you or any member of your household have a bank account?” So the first axis of variation is to compare the more detailed product approach to the institutional approach (to me this was instructive in the sense that in some surveys I lean more towards the product approach, while in others I tend to the institutional) .
The second axis of variation is who responds. There are three options here: the head responds for the whole household, every household member responds for him or herself, or some random selected adult responds for the whole household.
So what do they find? Comparing 7 indicators (banked, indirect access to a bank account, formal non-bank savings, formal credit, informal savings, informal credit, and insurance) they find no significant difference in the reports between the head of the household versus enumerating everyone in the household for six of these measures. The only one where there is a significant difference is insurance where, for some strange reason, the head report nets you higher levels of household insurance than asking each person individually (I don’t quite understand this). On the other hand, asking a randomly selected adult to report gets you unambiguously worse results – they get the informal stuff pretty much right, but the reports for banked, indirect access, and formal credit are significantly lower.
In terms of the product or institution based approach, the results suggest that the more complicated the instrument, the better it is to use the product based approach. They get similar results for banked and formal savings. However, for formal credit (2.8 v 0.8%), informal savings (18.8 v 8.9%) and insurance (16.3 v 5.7%) the product approach gets you much higher usage. The informal savings results, for instance, help me understand why I hardly ever see anyone saving in my (institutions approach) surveys. This is important because you can imagine that working through the more extensive product list can eat up a lot more interview time – so here we learn that we should probably do it, but not for everything.
They then try to use some of the household characteristics to see if these reporting differences seem to be more pronounced in certain kinds of households. For most part, in terms of whom you ask, there is no big explanatory variable. There are a couple of exceptions, though. First, for those random respondents, households with more youngsters fare worse – perhaps suggesting that the youth know less about what their elders are doing. So don’t use a randomly selected respondent and really don’t use a randomly selected young respondent to report on household activities. A numerate head and a greater fraction of household members being employed both increase the reported use of banking services when everyone in the household is enumerated relative to the head. The product vs. institutional examination on the other hand shows nothing significant vis a vis household characteristics – it seems product based cues are good for everyone.
So, taken together these are some interesting results for how we might think about including financial access variables in our surveys. We can potentially save time by asking the household head to report on almost all of the main indicators for others in the household (especially if you weren’t going to do a full household enumeration in the first place). It’s probably better not to use a randomly selected other informant to report on the household. For more complicated financial services, go with a longer list of product related cues (and pilot well to get these right) for a more complete picture (and likely higher rates) of what is going on.