Wednesday, November 28

Beware of the difference between correlation and causality

I was emailing a friend of mine who was talking about how bizarre he found a suggestion that there is a link between suicides and religion. I agree, see here for a previous essay. Also see my comment back to him here.


Just to pick up on your last point, AAA. See this story


Now see what the authors’s said in their original research study (which I have a copy if anybody is interested)

Several limitations of our study should be acknowledged. First, we excluded women with limited ability to consent to the study, as well as women who had a severe mental illness. A consequence of these exclusion criteria is that women who may be most vulnerable to abuse could not participate in the study. It is possible that excluding these women decreased the prevalence estimates that were obtained. Second, because of our reliance upon self-report, it is possible that some women were too embarrassed to report IPV. Alternatives to self-report may not be ethical or feasible, with perhaps one exception, namely, male partners; future research might assess men to obtain their perspectives. Such data will enrich our understanding of IPV. Third, our interest was primarily in understanding depressive and PTSD symptoms, rather than full-blown clinical disorders. Thus, we did not conduct structured diagnostic interviews, and our data do not allow us to speculate about psychopathology. Fourth, our study did not assess anxiety or other possible psychiatric consequences of IPV. Given that anxiety is perhaps the most common manifestation of stress in women, future research should include a formal assessment of anxiety. Fifth, we recruited women from one city in Southern India. Use of an urban sample from one region of India suggests caution when generalizing from our results to the country as a whole pending replication. Finally, because of our desire to recruit a large sample, and because of concerns about respondent burden, we did not conduct a detailed assessment of the possible antecedents of intimate partner violence. Such in-depth investigation is clearly necessary to guide the development of preventive interventions. Each of these limitations reveals opportunities for investigators to explore in future research on this critically important topic.

I home into the actual reported paragraph, and this is what it says:

Second, by exploring the correlates of IPV, we learned that IPV was more common among families that were of Hindu faith, younger age, and higher household income, and when alcohol abuse was present in the male partner. The association between IPV and both Hindu faith and family income were unexpected. Future research might re-examine these links to see if they replicate, and explore possible explanations for the relationships if found again.

Now if one was not familiar with statistics, one would expect that people would say that correlation implies causality. But correlation does NOT imply or indicate causality, that is the most basic mistake people make. There are very many other issues with the statistical analysis, but as this is a journalism list, I would hasten to caution against picking up correlated factors and implying causality.

To tie it back to your point, what is this supposed to prove? That it is Hinduism which causes men to beat up their pregnant wives? Or what?

No comments: