Dirty Data: The Effects of Screening Respondents Who Provide Low-Quality Data in Survey Research

Abstract

The purpose of this study is to empirically address questions pertaining to the effects of data screening practices in survey research. This study addresses questions about the impact of screening techniques on data and statistical analyses. It also serves an initial attempt to estimate descriptive statistics and graphically display the distributions of popular screening techniques. Data were obtained from an online sample who completed demographic items and measures of character strengths (N = 307). Screening indices demonstrate minimal overlap and differ in the number of participants flagged. Existing cutoff scores for most screening techniques seem appropriate, but cutoff values for consistency-based indices may be too liberal. Screens differ in the extent to which they impact survey results. The use of screening techniques can impact inter-item correlations, inter-scale correlations, reliability estimates, and statistical results. While data screening can improve the quality and trustworthiness of data, screening techniques are not interchangeable. Researchers and practitioners should be aware of the differences between data screening techniques and apply appropriate screens for their survey characteristics and study design. Low-impact direct and unobtrusive screens such as self-report indicators, bogus items, instructed items, longstring, individual response variability, and response time are relatively simple to administer and analyze. The fact that data screening can influence the statistical results of a study demonstrates that low-quality data can distort hypothesis testing in organizational research and practice. We recommend analyzing results both before and after screens have been applied.