it is important to investigate what determines individual earnings and the ‘dependency rate’ on group
income, as well as group performance. And even when the attention is on the latter, we may expect
the effects of characteristics such as heterogeneity to be amplified -and possibly different- in this
context. Secondly, as will be described in the next section, the setting in which these groups operate
is as precarious and socially disrupted as one may conceive, due to the virtual absence of property
rights in the squatter communities and to the high degree of in and out-migration. This can also be
expected to have an impact on group functioning. Finally, compared to some of the above studies
in which the degree of heterogeneity and other group characteristics were estimated based on the
leaders’ evaluation, a key ingredient of the survey methodology in this study was to interview all
members of each group and to compare their ‘subjective’ assessments with ‘objective’ measures of
group composition. This reveals some interesting differences, for example, in the case of wealth and
income inequality.
Before turning to the description of the setting in which the data was collected, it is useful
to compare this study to the earlier work by Abraham, Baland and Platteau (1998). The authors
surveyed 510 households from one of the most populated informal settlements of Nairobi, Kibera,
collecting data on participation in different types of groups and on the socio-economic background
of the respondents. The groups are then divided into several categories, such as rotating savings
and credit associations, burial societies, health groups, etc. and their main characteristics in terms
of organizational structure, contributions, and composition are examined. While similar for the
context, the present paper differs from the work of Abraham et al. in three respects. First, their
approach is purely descriptive, while this paper employs multivariate analysis. Second, among their
respondents some are members of a group and some are not, so they could estimate participation
rates but not know who else is in the same group of a respondent. In this paper no participation
regression can be run, because every respondent is by definition member of some group, but the
identity and characteristics of all other group members are known. Finally, the data of Abraham et
al. has a broader coverage, including groups with and without investment or production activities,
while in this paper the focus is on groups with income generating activities.