📊 What Was Studied:

Surveys are a primary source for studying voter turnout, but they tend to overestimate turnout because of two distinct problems: nonresponse bias (people who do not take surveys differ in voting behavior) and overreporting (respondents claiming they voted when they did not). This study disentangles those two sources of error using linked survey and administrative records for Danish elections.

🧾 What Data Were Used:

• Validated turnout indicators from administrative records for both survey respondents and nonrespondents.
• Respondents’ self-reported voting from the Danish National Election Studies (DNES).
• Administrative covariates (available for respondents and nonrespondents) used to model and compare predictors of turnout.

🔍 How the Analysis Was Done:

• Linked survey responses to administrative turnout to measure actual versus reported voting.
• Decomposed the overall overestimation of turnout into contributions from nonresponse bias and overreporting.
• Used administrative covariates common to respondents and nonrespondents to assess how each source of bias changes estimated relationships between predictors and turnout.

📈 Key Findings:

• Both nonresponse bias and overreporting make substantial contributions to survey overestimates of turnout.
• Both sources of bias materially distort the estimated predictors of turnout when only survey data are used.
• Specifically, combined nonresponse and overreporting mask a gender gap of 2.5 percentage points in favor of women.
• They also obscure a much larger gap—about 25 percentage points—in favor of ethnic Danes compared with Danes of immigrant heritage.

💡 Why It Matters:

• Reliance on unvalidated survey data can lead to incorrect conclusions about who votes and why, understating inequalities by gender and, especially, by ethnic background.
• Linking survey responses to administrative records and separating nonresponse from overreporting is crucial for producing accurate turnout estimates and reliable turnout models.