My question concerns estimation of “peer effects“ or “neighborhood effects” in a multilevel framework. The idea of such an effect is that the behavior of a household (on level-1) is influenced by the behavior of others in the same cluster/neighborhood (level-2) who are perceived as peers. Manski (1995:127) termed this an “endogenous effect”, i.e. that “the propensity of an individual to behave in some way varies with the prevalence of the behavior in the group’’, and pointed out that identifying such an effect is difficult if not impossible. When going through literature and searching the net, I have found many different examples of how people try to estimate this, but also criticism of some of the same strategies.
The reference for Charles F. Manski (1995) is “Identification Problems in the Social Sciences”, Harvard University Press, but the main arguments are also included in the following article: Charles F. Manski (1993) “Identification of Endogenous Social Effects: The Reflection Problem, The Review of Economic Studies, Vol. 60, No. 3 (Jul., 1993), pp. 531-542”.
In a multilevel framework where the outcome is a binary choice, say to buy product A or not, it would concern how the outcome on the household level is influenced by the outcome of others on level-2. The data I am using is a representative sample of 32.000 Indian households, from 1450 villages or neighborhoods if urban. It is cross-sectional observational data. The villages/neighborhoods are primary sampling units used by the Census of India. People in the same village or neighborhood are close enough to be potential influencers.
Manski (1995, 1993) discussed three hypotheses to explain how groups (or societies, or environments) may affect individuals: endogenous, contextual, and correlated effects.
Endogenous effects: The individual behavior tends to vary with the prevalence of that behavior in the group. In my example, this effect arises when households’ choices of purchasing product A tends to vary with the average choice of all households in the group (i.e. village or urban neighborhood).
Contextual effects: The household behavior tends to vary with the distribution of background characteristics in the village/neighborhood. There is a contextual effect when the propensity of an individual to behave in some ways varies with the mean of exogenous variables.
Correlated effects: Individuals tend to behave similarly because they have similar individual characteristics in the group (i.e., household education or wealth) or become part of similar institutional environments (i.e., households are part of the same social community – “caste” – or the same social clubs, etc.).
My main hypothesis concerns an endogenous effect, as I described in the original posting, but I also hypothesize contextual as well as correlated effects.
Manski referred to the so-called reflection problem that arises when “a researcher observes the distribution of behavior in a population and wishes to infer whether the average behavior in some group influences the behavior of the individuals that compose the group”. This ‘‘reflection problem’’ arises out of the presence of the mean outcome of the group as a regressor in an equation that also includes the effects of context (e.g. the mean wealth of households in the village/neighborhood) as well as individual characteristics or other group-level variables (for example, institutional environments). Manski showed that the endogenous effects (in Manksi’s terminology) are unidentifiable (in particular it is not possible to distinguish endogenous and contextual effects, although in some situations it may be possible to determine whether an overall social effect (defined as endogenous plus contextual effects), is present.
One way I can think of dealing with this would be to calculate group means but subtract the values for each household and include these as level-2 variables, but I have a feeling that there will still be endogeneity problems that come along with doing so that I am not seeing yet.
My question is: how I can best identify whether the propensity of a household to buy product A varies with the prevalence of buying product A in the village/neighborhood?
KML
