Abstract
Administrative records are increasingly being linked to survey records to heighten the utility of the survey data. Respondent consent is usually needed to perform exact record linkage; however, not all respondents agree to this request and several studies have found significant differences between consenting and non-consenting respondents on the survey variables. To the extent that these survey variables are related to variables in the administrative data, the resulting administrative estimates can be biased due to non-consent. Estimating non-consent biases for linked administrative estimates is complicated by the fact that administrative records are typically not available for the non-consenting respondents. The present study can overcome this limitation by utilizing a unique data source, the German Panel Study "Labour Market and Social Security" (PASS), and linking the consent indicator to the administrative records (available for the entire sample). This situation permits the estimation of non-consent biases for administrative variables and avoids the need to link the survey responses. The impact of non-consent bias can be assessed relative to other sources of bias (nonresponse, measurement) for several administrative estimates. The results show that non-consent biases are present for few estimates, but are generally small relative to other sources of bias.
Introduction
Researchers increasingly link survey responses to administrative data—such as tax, social security, or health records—to reduce respondent burden and improve data accuracy. However, ethical and legal standards typically require obtaining explicit consent from respondents to perform this linkage. This requirement introduces the risk of non-consent bias: if individuals who agree to linkage differ systematically from those who refuse, the resulting linked dataset may produce biased estimates.
Historically, quantifying this bias has been difficult because researchers lack administrative data for those who do not consent. This study addresses this "Catch-22" by leveraging a unique German dataset where administrative records were available for the entire sample frame, allowing for a direct comparison between consenters and non-consenters.
Methodology and Data
The study utilizes the German Panel Study “Labour Market and Social Security” (PASS), a household panel drawn directly from administrative registers. This provided the researchers with "gold standard" administrative data (the "Integrated Employment Biographies") for every individual in the sample, regardless of their survey participation or consent status.
The authors employed a bias decomposition approach to isolate four distinct error types:
- Non-contact bias: Differences between reachable individuals and the full sample.
- Refusal bias: Differences between survey respondents and those contacted.
- Non-consent bias: Differences between those agreeing to linkage and all survey respondents.
- Measurement error: Discrepancies between survey-reported values and administrative records.
Key Results: The Hierarchy of Error
The most significant finding is that non-consent bias is often the smallest component of total survey error. While statistically significant for certain demographic variables like age and foreign citizenship, its magnitude was negligible for substantive variables such as income, employment status, and disability.
As shown in the bias decomposition table above, the non-consent bias for mean monthly income was a mere 1.7 Euros. In contrast, the measurement error—the difference between self-reported earnings and administrative records—exceeded 400 Euros. Similarly, for most variables, the "Total Nonresponse Bias" (non-contact plus refusal) outweighed the bias introduced by non-consent.
Reluctance and Consent Propensity
A common assumption in survey methodology is that "reluctant" respondents—those harder to recruit—are also more likely to refuse data linkage. To test this, the authors grouped respondents into quintiles based on their predicted response propensity.
The analysis revealed no consistent relationship between the propensity to respond to the survey and the propensity to consent to linkage. Consent rates remained relatively stable across all quintiles, ranging from approximately 76% to 80%. This suggests that the mechanisms driving survey participation are distinct from those driving data linkage consent.
Conclusion
The study demonstrates that while non-consent bias exists, it is generally a minor concern compared to non-response and measurement errors. Even with a ~20% refusal rate for linkage, the resulting linked data remains more accurate than self-reported survey data, particularly for sensitive topics like income. These findings suggest that the benefits of data linkage often outweigh the risks of non-consent bias, providing a strong justification for the continued use of linked administrative records in social science research.
Summary updated