Exploring R {Rpms} to Predict Perceived Respondents Burden in the Consumer Expenditure Interview Survey

• Activity Number: 453 - Recursive Partitioning for Modeling Survey Data and Randomized Trials
• Type: Topic Contributed
• Date/Time: Thursday, August 6, 2020 : 10:00 AM to 11:50 AM
• Sponsor: Survey Research Methods Section
• Author(s): Daniel Yang*
• Companies: Bureau of Labor Statistics
• Keywords: Survey design; Respondent burden; Proxy indicator; Nonparametric; Tree model; Random forest

Abstract

Ambiguous responses to a questionnaire or refusals to continued participation in a survey could introduce bias and degrade total data quality of the survey as a consequence. These negative events could potentially be induced by increased burden. Survey design upgrades could affect respondent burden. In order to reduce respondents’ perceived burden along with its potential bias, interventions may be required to support data quality. Burden measurements would assist survey management to determine when they are needed. In 2012 and 2017 (at the end of the final wave) perceived burden was collected from respondents of the Consumer Expenditure Surveys (CE) Quarterly Interview. In this study, we will illustrate the construction of a composite burden index score based on a multivariate technique, and will investigate respondent burden proxy indicators by using the nonparametric recursive partitioning model under a complex survey design provided by R {rpms} package. We build a model to predict the respondents’ burden index scores based on household demographics hoping the model can provide insights into types of households that may perceive burden to guide future adjustments to the survey.