Data and methodology
3.5 Data quality and limitations
Matlab area was chosen by the ICDDR,B as a field site for population policy
research because it is among the few localities in the Third World where
demographic data are known to be accurately collected and complete (D’Souza
1981). Despite the advantages of Matlab as a research site, certain limitations are
contiguous. Diffusion of project-related services occurs across project boundaries
through intervillage migration, mainly through marriage. In this connection Phillips
et al. (1988) argued that contaminating factors tend to understate the treatment
effect, rather than lead to spurious project results.
Data gathered through the DSS system are cleaned routinely both for logical and
between-records errors. DSS data cleaning is so thorough (see section 3.2.1) that
those who have used these data have found few errors. For the present study few
unmatched cases were identified (less than 2 per cent).
3.5.1 Reporting of births
Previous experience from developing countries has shown that birth history data
contain two kinds of error: errors of omission and errors in reporting dates of
occurrence. According to Chidambaram, Cleland and Verma (1980), displacement of
date of birth was more prevalent than omission of births in WFS data. Older women
tended to displace the dates of their early births towards the survey date. The present
data set is free from this type of error because it has been collected through a vital
registration system. As recording of events relies on fortnightly household visits, the
method of data collection does not allow errors of displacement to occur. Moreover,
other studies also make field visits, and errors detected by them are considered
discreditable to DSS workers. Evaluation reports are prepared by the DSS manager
at the end of each year after considering omissions and misreporting of events
detected by DSS supervisory staff and by other studies; omission and misreporting
3.5.2 Reporting of contraceptive use
Underreporting of contraceptive use is not uncommon in Bangladesh (Green 1969;
Stoeckel & Choudhury 1973). Using data from a panel study in Dhaka, Green (1969)
estimated underreporting of contraceptive use at 13 to 22 per cent for males, and 26
to 35 per cent for females. On the basis of tape-recorded transcripts of Bangladesh
Fertility Survey data, Thompson, Ali and Casterline (1982) indicated that the
contraceptive knowledge and use section of the interview was distinguished by a lack
of discussion. There was evidence that items in this section caused considerable
discomfort for both the interviewer and the respondent, which might have caused
underreporting of contraceptive use. There has been no assessment made of the
extent of underreporting of contraceptive use in the present study. However,
contraceptive prevalence rates calculated for the treatment area in 1984 using the
service records and the IDS survey data were found to be close (44 compared to 39
per cent); this difference may be due to sampling error. It is not known to what
extent underreporting exists in the comparison area, but it is believed that as ICDDR
workers visit these households often, respondents would be less embarrassed than
Bangladeshis in other parts of the country when reporting contraceptive use.
3.5.3 Reporting of reproductive preferences
As the ICDDR has been operating the FPHSP in the treatment area, critics might
have thought about contamination of reproductive preference data in this area.
Because of this presence, when ICDDR field workers inquired about reproductive
preferences, respondents may have been tempted to give responses they thought
would satisfy the interviewers. However, this is not the case. Reproductive
surveys were found to be similar in the two areas, indicating that respondents in the
treatment area did not tailor their preferences to satisfy interviewers.
The present study used three concepts to measure reproductive preferences, desired
family size, ideal family size and desire for more children. According to Lightboume
(1984), the quality of data on reproductive preferences can be ascertained to some
extent through non-response levels, test-retest reliability studies and inter-item
consistency analysis. However, the Matlab data do not allow test-retest reliability
studies.
Over the period between 1975 and 1990, the non-response rate for desired family
size declined dramatically, probably mainly because people became more aware that
family size could be controlled. In the 1975 CDP data the non-response rate for
desired family size was 17.1 per cent in the treatment area and 27.2 per cent in the
comparison area, while comparable non-response rates for the 1990 KAP were 1.7
and 3.4 per cent. In the 1984 IDS data the non-response rate for desire for more
children was 3.2 per cent in the treatment area and 4.0 per cent in the comparison
area, while comparable non-response rates for the 1990 KAP were 1.2 and 2.3 per
cent.
An inter-item consistency check was made on the 1990 KAP data; responses to the
question on desire for more children were compared with a variable indicating
whether desired family size exceeded, equalled or was less than the actual number of
living children. Of those who desired more children, 82 per cent in both areas
reported a desired family size in excess of actual family size; 11-12 per cent reported
a desired family size equal to actual family size and 5 per cent reported a desired
family size less than actual family size and were therefore inconsistent (Table 3.2).
Of those who wanted no more children, 55 per cent in the treatment area and 61 per
size, while 41 per cent in the treatment area and 37 per cent in the comparison area reported equal desired and actual family sizes; only 3 per cent in the treatment area and 2 per cent in the comparison area were inconsistent, reporting desired sizes that exceeded actual sizes.