Mode Effects Between Telephone and Web Interviews in the Post-COVID-19 Questionnaire Survey CoVerlauf: Exploratory Study

TL;DR

This study compared how data collection methods (web vs. phone) affected responses in a post-COVID-19 survey, finding differences in symptom reporting but no major impact on overall results.

Contribution

The study identifies selection and measurement effects between web and phone interview modes in a post-COVID-19 survey.

Findings

Participants with higher education and more symptoms were less likely to choose phone interviews.

Phone interviews led to more free-text responses and higher symptom counts compared to web interviews.

Selection and measurement effects were observed, but overall study findings were not significantly affected.

Abstract

Concurrent mixed mode designs for data collection may introduce mode-specific biases. This study investigated mode effects between computer-assisted website interviewing (CAWI) and computer-assisted telephone interviewing (CATI) in the concurrent mixed mode study CoVerlauf.

This secondary analysis of previously published results explored associations between the choice of interview mode and socioeconomic characteristics, the number of reported post-COVID-19 symptoms, and response patterns in symptoms collected via multiple-choice (MC) and free-text items.

Associations were calculated using multiple logistic regression analyses, mixed effects logistic regression analyses, and gamma generalized linear models. Selection and measurement effects in symptom counts were analyzed with the back-door method.

Among 1779 participants, 77.8% (1384/1779) chose CAWI, and 22.2% (395/1779) chose CATI. Odds of selecting CATI were lower with a higher education (odds ratio [OR] 0.46, 95% CI 0.34 to 0.63), for those represented by another person (OR 0.28, 95% CI 0.13 to 0.56), and with each additional symptom experienced at the time of the interview collected from MC (OR 0.86, 95% CI 0.77 to 0.96). Compared with those aged 50 years to 59 years, the odds increased with age (60‐69 years: OR 2.26, 95% CI 1.53 to 3.36; 70‐79 years: OR 4.98, 95% CI 3.12 to 7.97; ≥80 years: OR 17.29, 95% CI 9.34 to 33.11) and each additional symptom experienced at infection collected via MC (OR 1.09, 95% CI 1.04 to 1.15) and free-text items (OR 1.94, 95% CI 1.61 to 2.37). Odds of responding to free-text items were higher with CATI (OR 5.22, 95% CI 4.25 to 6.42), for women (OR 1.57, 95% CI 1.30 to 1.89), for those with ≥3 pre-existing conditions (OR 1.44, 95% CI 1.07 to 1.96), with missing information on educational attainment (OR 1.95, 95% CI 1.11 to 3.42), and for each additional MC symptom (OR 1.15, 95% CI 1.12 to 1.19). Odds of collecting additional symptoms from free-text items were lower with CATI (OR 0.58, 95% CI 0.41 to 0.82) but increased with each additional MC symptom (OR 1.08, 95% CI 1.02 to 1.14), ≥2 pre-existing conditions (2: OR 1.8, 95% CI 1.02 to 3.00; ≥3: OR 2.0 95% CI 1.18 to 3.43), and ages 40 years to 49 years (OR 2.0, 95% CI 1.13 to 3.55). Selection effects were found for the number of symptoms at infection collected via MC (difference –0.63, 95% CI –0.95 to –0.31) as were measurement effects for symptoms collected via MC at infection (difference 0.97, 95% CI 0.47 to 1.5), via free text at infection (difference 0.40, 95% CI 0.29 to 0.52), and via free text at the time of the interview (difference 0.15, 95% CI 0.05 to 0.25).

Our results suggest that interviews via CATI resulted in higher numbers of reported symptoms collected via MC and free text and that both selection and measurement effects contributed to this increase. However, no evidence was found that the interview mode affected the overall findings of the CoVerlauf study.

Figures