Related Experiment Videos
Using logistic regression to detect item-level non-response bias in surveys
1450 Erickson Hall, Michigan State University, East Lansing, MI 48824, USA. wolfee@msu.edu
Journal of Applied Measurement
|August 9, 2003
Summary
Non-response bias in surveys is often systematic, not random. This study found males and lower-achieving students were more likely to not respond to survey questions, impacting data accuracy.
Area of Science:
- Social Sciences
- Statistics
- Educational Research
Background:
- Non-response bias is a significant challenge in survey research, potentially compromising data validity.
- Understanding the nature and magnitude of non-response is crucial for accurate data interpretation.
- Previous research has highlighted the need for robust methods to assess item-level non-response bias.
Purpose of the Study:
- To introduce and demonstrate a procedure for evaluating item-level non-response bias in questionnaires.
- To identify systematic patterns in non-response for specific survey items.
- To estimate the potential bias introduced by non-random missing data.
Main Methods:
- Utilized logistic regression to analyze non-response patterns.
- Applied the procedure to a question on drug use behaviors from the National Educational Longitudinal Study of 1994.
- Examined demographic and socioeconomic factors associated with non-response.
Main Results:
- Non-responses were found to be systematic rather than random.
- Males and lower-achieving students showed a higher likelihood of non-response.
- Significant two-way interactions were identified between ethnicity and socioeconomic status (SES), and ethnicity and geographic region.
Conclusions:
- The assumption of data missing at random (MAR) can lead to severely biased parameter estimates.
- The developed procedure effectively identifies systematic non-response bias.
- Recommendations are provided for survey researchers to evaluate and mitigate non-response bias threats.