Developing an Optimal Call Scheduling Strategy for a Telephone Survey

We use a Markov decision process to minimize the expected number of telephone calls needed to make contacts for a random digit dialing survey. Our states include information about the history of calls made to a telephone number and the action to be selected is the time of the next call attempt. Transition probabilities are estimated using a polytomous logistic regression model with data collected by the Bureau of the Census during a test of telephone interviewing for the Current Population Survey. We find that the optimal strategy for that survey should reduce the number of call attempts required and improve the nonresponse rate.