interventions. First, it was mentioned that experiments are not ethical because control groups do not benefit from interventions. It was quickly pointed out, however, that if a study were being done to test the effectiveness of the implementation (heretofore unknown), then that ethical issue would not be relevant. Second, it was pointed out that governments often interfere with controlled experiments and tamper with comparison groups. Third, it was noted that members of comparison groups themselves begin to adopt aspects of the interventions that are under study, thus contaminating the experiment. Finally, it was mentioned that there is a reluctance among social scientists to conduct experimental studies for fear that the results may be disappointing and that no program effects will be found.
HOUSEHOLD/POPULATION-BASED SURVEYS
John Casterline discussed the contribution of household surveys to program intervention evaluation. Data collected by household surveys, such as contraceptive attitudes and behaviors, are of great importance in the evaluation of the effectiveness of family planning programs. These data, especially valuable when collected by several surveys over time, are even more valuable when the same respondents are reinterviewed in successive surveys, allowing the causes of behavior change to be investigated. They are further enhanced if linked with other types of data. The most common link is that between the household and the family planning service environment. Such cross-sectional linkages have weaknesses, however. First, contraceptive services change over time, creating problems
with interpretation; it is likely that the service environment at the time of the household survey would differ from the one that led to the practices observed in the household survey. Second, people migrate, so recent migrants interviewed in a household survey may have adopted contraceptive practices that were not engendered by the service environment in the survey area. For these reasons, among others, Casterline pointed again to the value of longitudinal studies. The linking of longitudinal family planning service and household data (including geographic location of households by geocoding) for identical areal units would enable researchers to analyze the impact of program intervention over time.
Mary Overpeck presented an example that illustrated the importance of linking survey data with other types of records. By analyzing medical records and child care records, the conclusion could be drawn that children who do not participate in child care programs have fewer medical injuries because they seek medical attention less frequently than children who are in child care. Linking these records to survey data with socioeconomic indicators, however, reveals that mothers with relatively high levels of education are more likely than less educated mothers to (1) have children in child care and (2) have health care coverage for their children. Moreover, having children covered by health care is positively related to seeking medical care when needed. By linking data sets it becomes apparent that children in child care seek medical attention more frequently than children not in child care because they are more likely to have health care coverage.
Although surveys provide data for the calculation of prevalence levels, which are important indicators of what is happening in a country, during the discussion participants reiterated the importance of supplementing surveys with other data, such as client records.