Die relative Bedeutung der Kovariatenwahl, Reliabilität und Art der Datenanalyse zur Schätzung kausaler Effekte aus Beobachtungsdaten

Translated title of the contribution: On the relative importance of covariate choice, reliable measurement and analytical method for estimating causal effects from observational studies

Thomas D. Cook, Steffi Pohl, Peter M. Steiner

Research output: Contribution to journalArticle

Abstract

In evaluation studies it is often not possible to conduct randomized experiments for estimating causal effects of a program. Instead, quasi-experimental designs are frequently used. The causal interpretation of a statistically adjusted estimate is only warranted if the assumptions of the underlying quasi-experiment are met and the statistical analysis is properly conducted. In this article the relative importance of covariate choice, the covariates' reliable measurement, and the choice of analytic method (regression analysis or diffent types of propensity score methods) are investigated using a within-study comparison which compares the effect estimate from adjusted quasi-experimental design to the corresponding effect estimate from a randomized experiment. The results show that the covariate choice is the most important factor for removing selection bias. The reliable measurement of covariates is the second most important factor. The choice of an analytic method does not have a strong impact on bias reduction. To date, the superiority of propensity score methods in comparison to regression methods is not yet supported by empirical evidence.

Translated title of the contributionOn the relative importance of covariate choice, reliable measurement and analytical method for estimating causal effects from observational studies
Original languageGerman
Pages (from-to)203-224
Number of pages22
JournalZeitschrift fur Evaluation
Volume10
Issue number2
StatePublished - Dec 1 2011

Keywords

  • Causal Effects
  • Impact Analysis
  • Propensity Scores
  • Quasi-experimental Designs

ASJC Scopus subject areas

  • Statistics and Probability
  • Communication
  • Education
  • Social Sciences (miscellaneous)
  • Applied Psychology
  • Strategy and Management

Fingerprint Dive into the research topics of 'On the relative importance of covariate choice, reliable measurement and analytical method for estimating causal effects from observational studies'. Together they form a unique fingerprint.

  • Cite this