Data
quality
The Social Networks Project has paid a great deal of attention to data
quality. Several steps were taken during fieldwork to maximize
the quality of the data collected by all three projects. In addition,
post facto assessments of the validity and reliability of the data
collected have been carried. In particular:
Fieldwork.
At least one of the PIs has been present at every stage of fieldwork,
unlike many other field projects. The
presence of a PI
has three advantages: 1) It has been possible to respond authoritatively
and immediately to the inevitable problems that arise in the field;
2) Team morale is maintained; 3) Interactions with interviewers and
supervisors aided in identifying potential issues of data quality and
interpreting survey responses. We provide a myriad of examples in a
document summarizing Watkins’ field notes for the first round
of the Kenya Diffusion and Ideational Change Project (Watkins
et al. 1995) and in the field report of the pre-test of the Family
Transfers Project
(Weinreb
1998).
Questionnaires. The same basic survey questionnaire has been used
in all Social Networks surveys. The questionnaire was less developed
in the first survey wave in Kenya but, since each round has served
as a pretest for the subsequent rounds, successive refinements of the
questionnaires have meant that there has been decreasing ambiguity
in the questions.
Questionnaires'
checking and data entry. To minimize the effects of
interviewer error on the data, questionnaires were checked and data
entered in each field site. Checking was done first by the supervisors
and then on the next day by a PI and graduate student members of the
field team. This permitted detecting interviewer errors while the team
was still in the survey site, so that interviewers could return to
respondents if there were missing data or other apparent errors. Questionnaires
are stored at the University of Pennsylvania, which permits further
checking in the course of new analyses. .
Quality
assessments. The papers on the survey methodology of the Social
Networks Project are numerous. Among them, there are various assessment
of the quality of the data collected by the project's surveys:
-
The
literature on interviewing in developing countries suggests that
it is desirable to use same-gender interviewers who are strangers
to the community. In some circumstances this may not be practical.
Weinreb
(2000) evaluated responses in cases where the interviewer
knew the family of the respondent and cases where he/she did
not, by gender.
Miller,
Zulu and Watkins (2000) analyzed husband' and wives' responses
on questions about the possession of household goods (e.g.
a pit latrine and a mattress), and questions about current use
of family planning and AIDS
in th MDICP. They found that discrepant responses were systematic
with respect to gender (i.e. when husbands and wives disagree,
it is typically the husband
who says "yes" and the wife who says "no")
and with respect to region. Miller et al. found the same systematic
patterns of husband-wife
discrepancies in the Kenya Demographic and Health Survey and
the Malawi Demographic and Health Survey, suggesting that these
patterns are not particular to our
surveys.
-
Bignami-Van
Assche (2003) takes advantage of the panel nature of the
study as well as of a set of re-interviews conducted in
the
MDICP-2, to examine the consistency of responses
for respondents interviewed
by the two waves of the Malawian survey.
- Schatz
(2003) compares the measurement of women's status and autonomy
in the Malawi household survey with qualitative interviews
she conducted with a sub-sample of the household
sample in Malawi
A
detailed summary of the literature on data quality produced in
the context of the Social Networks Project, and a reasoned discussion
of the more general problem of data quality assessment
in demographic
research in developing countries, can be found in the Introduction
to the monographic volume on "Research on Demographic
Aspects of HIV/AIDS in rural Malawi" (Watkins
et al. 2003).