Kenyan Shillings (Ksh), which amounted to 99.6 US dollars in August 1999 prices. Groups differed
substantially in wealth, though, with the average member of the ‘poorest’ group having 2,450 Ksh,
and that of the richest having ten times as much.
In what follows we turn to multivariate analysis to investigate how individual and aggregate
characteristics affect the economic performance of these groups.
4 Empirical results
4.1 Income generation
The first aspect to be considered is the potential for income generation provided by these groups.
One can estimate the following earnings function:
yij — Xij β + Dj ° + "ij
(1)
where yij is the log of hourly earnings of individual i in group j , Xij is a vector of individual
characteristics, Dj if a set of group dummies, and "ij is an error term. Individual characteristics,
which should capture labor productivity, include sex, marital status, age, education, and experience
(proxied by the number of years the respondent has been in the group), as well as language dummies.
A non-linear specification is chosen for age, education and experience, to allow for nonlinearities in the
returns to human capital. Groupfixed effects are introduced to control for differences in remuneration
schemes across activities.
[Insert table 2]
Table 2 reports coe¢cient estimates for (1), with White-robust standard errors in parenthesis.
Group dummies are omitted from the table, but a standard F test in all cases rejects the null that
their coe¢cients are jointly equal to zero. When a small set of demographic controls is included
(column1), the only significant determinant of hourly earnings is age: within each group there are
positive but decreasing returns to age. Surprisingly education is not significant (the omitted category
is people with no formal education), nor are sex and marital status. Column 2 introduces two more