Estimation of Survey Cost Parameters Using Paradata

Abstract

In many survey situations, detailed cost parameters are difficult to estimate. This is especially true in surveys involving interviewers. Overall costs may be easily estimated since interviewer hours, materials, and incentives are relatively easy to track. But costs at a more granular level – for example, hours spent travelling, identifying non-sample units, or engaged in other activities -- may be difficult to track. This occurs for a number of reasons. Often, cost information and paradata are collected in separate systems; or the cost information that is collected may not be at a sufficiently detailed level in order to evaluate the costs of particular subtasks. It might be possible to gather these cost data via a special study, but this is usually a very expensive approach. It may also be possible to ask for more detailed reporting from interviewers and other staff. However, this approach might lead to reduced efficiency. This paper proposes the use of regression models estimated from paradata as a method for estimating detailed cost parameters related to interviewer effort. An example of this method is shown from the National Survey of Family Growth 2011-2018. This method was used to evaluate the costs of two treatment arms in an experimental study. The method is also used to monitor interviewer effort over the course of the field period.