Errors in autobiographical memory and their effects in time-to-event analysis (Project in the Priority Programme "Survey Methodology and Statistics")
Description
- The aim of this project was the evaluation of the nature and extend of memory errors in retrospectively collected datings in a sociological and an epidemiological survey. The effects of memory errors on parameter estimation in hazards models were analyzed and correction methods were evaluated.
The panel 'Labour Market and Social Security' (PASS) served as the data source for the socialogical example, which was analyzed in collaboration with the Institute for Employment Research (IAB) in Nuremberg. Reasons for the end of unemployment benefit were analyzed with time-to-event models. The epidemiological application was based on a health survey of insured persons of a statutory health insurance, where the focus was on the evaluation of the influence of the anticoagulant phenprocoumon on the time to hospitalization for a severe bleeding. The survey data was compared with register data in both applications and dicrepancies were interpreted as memory error in the survey.
Typical memory errors such as forward/backward telescoping, omission and joining could be observed. Important events, e.g. treatment with phenprocoumon or the end of the unemployment benefit, were reported with small errors. The error-proned data yielded biased effect estimates in the time-to-event analyses in both applications. The direction and magnitude of the bias varied between the variables. While an error-proned event status was mainly responsible for the bias, the contribution of an error-proned time-to-event to the bias was negligible.
Methods to correct for measurement errors were implemented in the epidemiological application. It turned out that the quality of the method depends on the assumed error model. Standard methods did not yield a good correction of the biases, because a heteroscedastic error model was present in the data. SIMEX as well as regression calibration were extended to take this measurement error model into account. The extension of regression calibration yielded good corrections compared to the standard method, while the extension of SIMEX did not improved the results. It was hypothesized that this was caused by the sensitivity of SIMEX to misspecification of the error model.
Funding period
- Begin: March 2008
End: February 2010
Sponsor
- German Research Foundation
Contact
Link
Selected project-related publications
- Behr S, Andersohn F, Garbe E. Risk of intracerebral hemorrhage associated with phenprocoumon exposure: A nested case-control study in a large population-based German database. Pharmacoepidemiology and Drug Safety. 2010;19(7):722-730.
https://doi.org/10.1002/pds.1973 - Pigeot I, Behr S. Pharmacoepidemiological databases: Strengths, limitations, methodological challenges. 2nd Conference of the Central European Network (CEN), 12-16 September 2011, Zurich, Switzerland.
- Behr S, Schill W, Pigeot I. Erinnerungsfehler in einer pharmakoepidemiologischen Studie - Vergleich von Registerdaten mit Befragungsdaten. Kolloquium "Statistische Methoden in der empirischen Forschung", Freie Universität Berlin, 16. November 2010, Berlin.
- Behr S, Schill W, Garbe E, Pigeot I. Missklassifikation des Zielereignisses in Datenbankstudien. "Biometrie, Epidemiologie und Informatik - Gemeinsam forschen für Gesundheit". 56. Jahrestagung der Deutschen Gesellschaft für Medizinische Informatik, Biometrie und Epidemiologie (GMDS) und 6. Jahrestagung der Deutschen Gesellschaft für Epidemiologie (DGEpi), 26.-29. September 2011, Mainz.
- Behr S, Eggs J, Müller G, Pigeot I. The effect of recall error in survey data on time-to-event-models. XVII ISA World Congress of Sociology der International Sociological Association (ISA), 11.-17. Juli 2010, Göteborg, Schweden.
- Behr S, Eggs J, Müller G, Pigeot I, Schill W, Schnell R, Trappmann M. Studying recall error in survey data using administrative records. Joint Statistical Meetings (JSM), 29. Juli - 7. August 2010, Vancouver, Canada.
- Behr S, Eggs J. Erinnerungsfehler in PASS. 3. Verbundtreffen Priority Programme on Survey Methodology (PPSM), 7. Mai 2009, Bremen.
- Enders D, Schill W. Die Performance von Regression Calibration und SIMEX in einer Ereigniszeitanalyse mit Messfehlern in den Kovariablen unter einem komplexen Fehlermodell. Biometrisches Kolloquium, 13.-15. März 2012, Berlin.