Imperfect Social Learning Among Kenyan Smallholders

Abstract:

This paper explores passive and active forms of social learning by Kenyan farmers. We document high rates of error and bias in respondents’ proxy reporting of their network members’ behavior. Proxy reporting of peer behavior proves no more accurate than random guesses based on known adoption rates in the broader population, suggesting limited individual- or networkspecific knowledge of behaviors. Moreover, perceptions of others’ behaviors are systematically related to respondents’ own behaviors, confirming Manski’s (1993) concern about reflection problems that might overstate social networks effects on learning in standard regression models. The bias and inaccuracy of reporting on network members’ behaviors calls into question the importance of passive social learning. We then show, however, that these same respondents actively pursue and accurately report information when given a direct incentive to do so. They mobilize social networks, especially to access information available from socially distant people, the weak ties that Granovetter (1973) suggests are the most valuable information sources within a social network. These results suggest a need to revisit how development economists study social learning, as active social learning may be more important and passive social learning less important than implied by prevailing models and empirical tests of social learning.