Kausaalisuuden arviointi sosiaali- ja terveystieteellisessä tutkimuksessa
Keywords:
kausaalisuus, tilastolliset menetelmät, havainnoiva tutkimus, luonnolliset koeasetelmatAbstract
Causality has scientific and social policy importance. Double-blinded clinical trials offer the best evidence on causality but are rarely available in behavioral sciences. Intervention studies utilize the experimental design, but they are often compromised by too short intervention period and follow-up time. Ecological and individual level statistical associations can indicate causality, but they rarely can show it even if it is possible to adjust the results for confounding factors. The follow-up design can sometimes offer more solid information on causality, but it needs dense measures before and after an exposure. Sibling comparisons allow adjusting the results for family environment indirectly and if monozygotic twins are included also adjusting the results for genetic factors. Natural experiments also include the difference-in-difference design and the regression discontinuity design. Recently, designs utilizing instrumental variables, especially Mendelian randomization, have become more widely used. Showing causality is challenging, but it is often possible to find a study design offering information beyond simple statistical associations. Critical discussion of causality should be an elementary part of all research in social and health research.
