This paper presents the methodology used by DHS for the editing and imputation of data. Producing good quality demographic and health data and making it accessible to data users worldwide is one of the main aims of the Demographic and Health Surveys program. However, large scale surveys in developing countries, particularly those collecting retrospective data, are prone to poor reporting. Survey data have suffered traditionally from incomplete and inconsistent reporting. To handle these problems, DHS performs extensive data editing operations. In addition, imputation procedures were established to deal with partial reporting of dates of key events in the respondent's life.
General techniques for handling incomplete and inconsistent data are considered. The paper then presents the DHS approach to data editing. The major focus is on the editing of dates of events and the intervals between events. The editing and imputation process starts with the calculation of initial logical ranges for each date, and gradually constrains these ranges to produce final logical ranges. Inconsistent data are reported in error listings during this process. Dates are imputed for events with incomplete reporting within these final logical ranges. The levels of imputation required in the DHS-I surveys are presented.
Various problem areas involved with the imputation of incomplete dates are explained. These include biases caused by questionnaire design, miscalculation of dates by interviewers and ancillary data biases. Problems relating to fine temporal variables and to unconstrained ranges for dates are also reviewed.