Complexity and Social Networks Blog

This is an impressive research project. The one thing I would like to press on here is the concluding framing of the choice. The theory underlying whether a transaction occurred assumed that the key choice was that of the borrower, not the sponsor. Bob Putnam argued at the end that the real decision point likely lay with the sponsor, not the borrower-- ultimately it was up to the sponsor to accede to a loan or not. At the end, I think discussion was pointing to the need to model both sides of the decision-- the decision to ask and the decision to say yes. This brings to mind a vivid metaphor by Lois Verbrugge, of the creation of relationships as a two stage process-- of meeting and of mating (in this case, asking, and sponsoring). What determines who you might ask, in this context? Clearly, the interest rate, where, generally folks will want to pay a lower rate. What about friends, and friends of friends, etc? Actually, it is a little unclear to me. As a borrower, wouldn't you want less constraint in acting opportunistically? Wouldn't you want a completely unconnected stranger to loan you money, so that if things went sour, there were fewer ramifications to reneging on the loan?

Of course, who you ask is presumably related to who might say yes-- no one likes to be turned down. And this is the real reason, I think, why someone is likely to ask friends, etc.

Distance is an interesting variable, since the small size of these communities would suggest that distance should not matter. However, distance is a pretty robust predictor of exchange at many scales, from within offices (e.g., Tom Allen's work) to international trade (gravity models), and everything I have seen suggests that probability of exchange declines very quickly with distance. Not so clear what would drive distance in this context-- norms to help neighbors? Asking the first person you run across? Or, on the sponsor side, better monitoring?

I would also expect that it would be awkward for high status people to ask low status people to sponsor (although the probability of a yes would presumably be higher). One network measure of status is to measure the difference between incoming and outgoing links (see Krackhardt)-- those who run surpluses are likely to be higher status.

On the "mating" side, what is going to drive a potential sponsor to say yes? Assessed probability of repayment is one, where having an ongoing relationship and common third party friends ("relational" and "structural" embeddedness, a la Uzzi and others). Others may include affect (Iris Bohnet suggested this at the talk); assessed value of the proposed expenditure (who has the best business case?); norms of responsibility (the norm may be to be helpful to neighbors, but not necessarily to non-neighbors); competition for your scarce resource; who can pay you back in the future in some fashion.

Ref:

Verbrugge, L.M. 1977 'The structure of adult friendship choice'. Social Forces 56: 577-597.

My point about framing was not that "the real decision" was the sponsor's, but that "trust" (the nominal topic of this very interesting project) was an characteristic of the sponsor, not of the borrower. The underlying model is one in which the lender/sponsor should offer a lower interest rate if s/he "trusts" the borrower more and that trust should be increased (through both information and enforcement) if the network linking the borrower and lender is tighter. But explaining the borrower's choice of whom to borrow from is at best an indirect test of that hypothesis; the more direct test is whether a lender is (at at some fixed interest rate) more willing to lend to a borrower who is "closer" (sociologically or physically). I was merely asking Markus to explain his decision to use this highly indirect test, given that the results of the indirect test (that is, predicting the borrower's choice of lender) is, in effect, assuming that borrowers uniformly and accurately perceive the various lenders' levels of trust, an assumption that seems questionable to me.

After the talk it occurred to me that the math of the statistical model might turn out to be identical either way (since all the key variables in the current model, like physical distance, are identical for borrowers and lenders), but Markus' presentation certainly framed the test in terms of borrowers' decisions.

By the way, my other question might be worth Markus thinking about, too--are there external effects of networks on trust and trustworthiness? In other words, if one of Markus' communities had a much higher network density overall, would that lead to lower implicit interest rates, EVEN for borrower-lender pairs who were not themselves linked. My intuition is "yes," because both general information fidelity and probability of enforcement would be higher in the more networked community. I suspect that we are often willing to make bets about others' trustworthiness because of the network context, even when we don't actually have any known tie, direct or indirect. I'm more likely to loan money to a Harvard faculty member than to a Yale faculty member, even if I don't know either of them, because I'd bet that, if I need to, I can more easily get better information about--and put more pressure on--the Harvard woman, given the relative density of ties that she and I are involved in in this community. And I'm certainly more likely to lend to a total stranger on the Harvard faculty than to a total stranger emerging from the T in Harvard Square. I think that in the contemporary world these "external" effects dominate the direct effects, because we often deal with people whom we don't know, even indirectly, but who are members of some shared community. If I'm right about this, then this is a clear prediction for the many-site study Markus is about to begin. Even after including all dyadic variables in the equation, I predict that the overall density of ties in a given community would itself be a significant predictor of [low] interest rates in a pooled multi-site regression.

Thanks, Markus, for a fine presentation, provocative in the best sense.

Bob

I think the suggestion of measuring the set of sponsors who were approached by a client (ideally in correct sequential order) until a sponsor agreed to cosign the loan is great. I could imagine that clients sometimes indeed try to first approach some less known sponsors to minimize the ramifications of future bad experiences. Actually, some leaders of evangelical churches in at least one community had some reservations about our program because they thought that it might lead to tense relationships between friends.

The significance of geographic distance puzzled us as well. We didn't expect it since we advertised the program widely - therefore, it was not the case that a person didn't know about a sponsor in a different block even if that person was a friend or indirect friend.

It would be interesting to distinguish whether the geography effect is due to better monitoring, the ability to more easily ask sponsors who live closeby or because our social network survey does not fully capture all the social relationships between individuals (which in return are correlated with georgaphic distance). I believe it's the latter but I cannot think of a good test.

Status is a very interesting idea and makes perfect sense.

BTW: A copy of the (very preliminary) paper is available here:

http://www.nber.org/~mmobius/Post/papers/trust-peru2.pdf

Is trust the main thing that is going on here? I don't think so. Trust, in this context, is the cognition of a sponsor that a potential borrower will repay loan, and thus is a good risk. But, following on Bob's example: (1) there are people I would trust (in this sense) that I would not lend money to (because I don't feel the obligation to do so); (2) there are many people that I would trust that would not ask me for a loan; (3) and there are people I do not trust that I would loan money to (because I feel an obligation). I think the observation of a transaction is partly affected by trust, so defined, but I think that there are other things going on, and that the network, in part, captures those other things, and thus one has to be careful in interpreting the network effects as reflecting "trust."

It would be useful to do ask sponsors: what is the probability that that each individual on the roster would renege on their loan-- high/low/dk. How is this related to the network? How is this related to behavior? etc.

Down the road, it would also be useful to look at default rates on these loans to evaluate the reality of whether people were trustworthy, and how that is related to network structure.

"Down the road, it would also be useful to look at default rates on these loans to evaluate the reality of whether people were trustworthy, and how that is related to network structure." (quoting Lazer)

I entirely agree. From a conceptual point of view, the central variable is trustworthiness, not trust. Trust without trustworthiness is merely gullibility. Trust might be a proxy for trustworthiness, but in the end what reputational effects and repeat-play effects should explain is trustworthiness, not trust. And what lowers transaction costs and has all the other magical properties here is trustworthiness, not trust.

One could also then examine the allocative efficiency of the network-- e.g., were the high default people more likely to get the high interest rate loans?

Very interesting! THX!