Modelling Optimum Response in a Longitudinal Survey

Non-response rates in surveys have been recognized as important indicators of data quality since they introduce bias in the estimates which increases the mean square error. In order to reduce this error, previous studies have examined the effects of response predictors on response rates. There is dearth of information about models which focus on the interaction effects of response predictors on response rates. The study was therefore designed to develop and validate a model which would reduce non-response and achieve optimum response by the introduction of interaction effects of the response predictors that have been broken down into levels. A two-stage stratified random sampling scheme was used in selecting 750 households in Oyo town. Households were interviewed in five waves. An interviewer-administered questionnaire was used to collect data on demographic characteristics and response predictors including age, gender, educational qualification, religion, employment status, family size, and duration of interview. Demographic characteristics were analyzed using summary statistics. Incidence Rate Ratio was used to examine the response rate at various levels of response predictors. Odd ratio was used to examine the relationship between response rate and each of the response predictors. A model was developed by breaking the predictors of response into levels and their interaction effects were introduced into Denise and Lan model. The respondents' mean age and modal family size were 51.8 6.9 and 3 respectively, 64.8% were females, 52.8% were muslims and majority (88.9%) were employed. The family size, duration of interview, education, number of visit, Language of interview, familiarity, gender, house ownership, Nationality and duration of residence in a community are positively related to the response rate. Age is negatively related to the response rate and there is no association between employment status and response rate. The developed model showed that family size (x1), duration of interview (x2), and their interaction (x1x2) significantly (p < 0.05) determined the response rate. The developed model established that both main and interaction effects of response predictors play key roles in improving response rate in a longitudinal survey.