Ausgewählte Kausalitätstheorien im Vergleich

 

 Buy Article Permissions and Reprints

Zusammenfassung

Verschiedene Ansätze zur kausalen Inferenz haben sich in unterschiedlichen Forschungsrichtungen getrennt voneinander entwickelt. Der folgende Beitrag versucht 3 dieser Ansätze kompakt vergleichend darzustellen. Im speziellen werden der „Potential-Outcomes“-Ansatz, die Denkschule von Campbell und die Theorie der kausalen Graphen (DAGs) betrachtet. Jede der 3 Theorien weist Stärken und Schwächen auf. Es ist aber auch möglich, diese Ansätze in der angewandten Forschung miteinander zu kombinieren. Der Potential-Outcomes-Ansatz liefert eine formale und exakte Definition eines kausalen Effekts, bietet aber wenig Hinweise auf Selektion von Kovariaten zur Adjustierung. Der Ansatz von Campbell hat keine mathematische Fundierung eines kausalen Effekts, aber expliziert eine große Anzahl von speziellen Bedrohungen der internen Validität, die in einer Studie beachtet werden müssen. Der DAG-Ansatz gibt Forschern ein Instrument an die Hand, mit dessen Hilfe sich bestimmen lässt, welche Kovariaten zur Adjustierung hinzugezogen werden müssen, aber setzt voraus, dass kausale Annahmen solcher Kovariaten bezüglich ihres Einflusses untereinander bestimmt werden können.

Abstract

Different research traditions have separately developed various theories of causal inference. Three of these approaches are contrasted in this paper. In particular, I will explore differences and similarities of the potential outcomes approach, the generalized causal inference framework of Campbell, and the theory of causal DAGs. Each of the 3 approaches has strengths and weaknesses; however I argue that it is possible to combinethem in applied research. The potential outcomes approach offers an exact and formal definition of a causal effect, however it is not very informative as to which covariates need to be selected for adjustment. The Campbell framework lacks a precise mathematical definition of a causal effect, but offers a long list of threats to internal validity that researchers can try to rule out in their studies. Causal DAGs offer applied researchers a tool to find minimally sufficient adjustment sets of covariates that allow an unbiased estimation of a causal effect, given that researchers are willing and able to encode their causal assumptions of all important covariates.

• Literatur

• 1 West SG, Thoemmes F. Campbell’s and Rubin’s perspectives on causal inference. Psychological Methods 2010; 15 (01) 18-37
  CrossrefPubMedGoogle Scholar
• 2 Rubin DB. Causal inference using potential outcomes. Journal of the American Statistical Association 2005; 100 (469) 322-331
  CrossrefPubMedGoogle Scholar
• 3 Shadish WR, Cook TD, Campbell DT. Experimental and quasi-experimental designs for generalized causal inference. Boston, MA: Houghton, Mifflin and Company; 2002
  Google Scholar
• 4 Pearl J. Causality: Models, Reasoning, and Inference. New York: Cambridge University Press; 2000: 331-358
  Google Scholar
• 5 Heckman JJ. The scientific model of causality. Sociological Methodology 2005; 35 (01) 1-97
  CrossrefPubMedGoogle Scholar
• 6 Dawid AP. Causal inference without counterfactuals. Journal of the American Statistical Association 2000; 95 (450) 407-424
  CrossrefPubMedGoogle Scholar
• 7 Steyer R, Partchev I, Kröhne U et al. Probability and Causality 1. Springer; 2011
  Google Scholar
• 8 Mayer A, Thoemmes F, Rose N et al. Analyzing total, direct, and indirect effects in mediation models. Unpublished manuscript. 2011
  PubMedGoogle Scholar
• 9 Rubin DB. Comment: Neyman (1923) and causal inference in experiments and observational studies. Statistical Science 1990; 5 (04) 472-480
  PubMedGoogle Scholar
• 10 Holland PW. Statistics and causal inference. Journal of the American Statistical Association 1986; 81 (396) 945-960
  CrossrefPubMedGoogle Scholar
• 11 Rosenbaum PR. Observational studies. Springer Verlag; 2002
  Google Scholar
• 12 Rubin D. Author’s reply: Should observational studies be designed to allow a lack of balance in covariate distributions across treatment groups?. Statistics in Medicine 2009; 28 (09) 1420-1423
  CrossrefPubMedGoogle Scholar
• 13 Rosenbaum PR, Rubin DB. The central role of the propensity score in observational studies for causal effects. Biometrika 1983; 70 (01) 41-55
  CrossrefPubMedGoogle Scholar
• 14 Thoemmes FJ, Kim ES. A systematic review of propensity score methods in the social sciences. Multivariate Behavioral Research 2011; 46 (01) 90-118
  CrossrefPubMedGoogle Scholar
• 15 Schipf S, Knüppel S, Hardt J et al. Directed Acyclic Graphs (DAGs) – Die Anwendung kausaler Graphen in der Epidemiologie. Gesundheitswesen 2011; 73: 888-892
  Article in Thieme ConnectPubMedGoogle Scholar
• 16 Frangakis CE, Rubin DB. Principal stratification in causal inference. Biometrics 2002; 58 (01) 21-29
  CrossrefPubMedGoogle Scholar
• 17 West SG, Biesanz JC, Pitts SC. Causal inference and generalization in field settings: Experimental and quasi-experimental designs. In: Reiss H, Judd C. (eds.) Handbook of research methods in social and personality psychology. Cambridge Univ Pr; 2000: 40-84
  Google Scholar