regressors: the number of years the respondent has been working in the group (and its square), and
the number of years he or she has been living in the current place of residence. The coe∏icieιιts
on the former variable are positive but not statistically significant: in these groups there seem to
be returns to ‘seniority’ in the sense of age, but not in the sense of experience. As for the latter
variable, residential stability has a negative and significant effect on individual earnings. This is
not surprising if we consider that the areas of study are probably the poorest and most degraded of
urban Nairobi, so having resided somewhere else probably means higher ‘unobserved ability’. This
interpretation may also help explain while the experience variable is not significant. On the one hand,
spending more years in the group increases job-specific skills; on the other hand, to the extent that
high ability workers may be the first ones to quit for formal jobs, length of work within the group
may end up capturing low ability. In an attempt to control for this, column 3 introduces among
the regressors individual earnings before joining the group (in log). The previous results remain
virtually unchanged and past earnings are insignificant, indicating that this variable is a poor proxy
for unobserved ability.7
In column 4 a set of language dummies is introduced to control for whether the respondent is
Kikuyu, Luo, or Kamba (the omitted category is Luhya and other ethnicities). Furthermore, the
dummy ‘Dominant language’ takes value 1 if the individual speaks the same language as the majority
of other group members, and zero otherwise. This is aimed at capturing possible ethnic favoritism,
as found for example by Collier and Garg (1999) in the Ghanaian labor market. As can be seen
from the table, the coe∏cient on this variable is not statistically different from zero. A possible
explanation is that when the decision process is not democratic, what counts is not belonging to
the majority, but being the same ethnicity as the leaders. The last column introduces a dummy for
whether the respondent speaks the same language as the chairperson of the group when decisions are
reported to be made by leaders as opposed to democratically, but the conclusion remains the same.
7 This is not too surprising, given the heterogeneity of the sample in terms of age: for young respondents low pre-
group earnings may simply reflect an out-of-the-labor-force status during school years, while for older ones the same is
not true.
10