Wednesday, April 04, 2007

On Dr. Cameron's Study

I've been reading through this study linked over at Pam's. The buzz of the study is that gay people die younger than hets. It starts by hazzily discussing a 2003 survery finding that ~1.7% of respondents identified as gay or bisexual.

The study faults Statistics Canada for leaving people ages 65 and over out of their 1.7% homosexual or bisexual statistic. Cameron states that an inclusive percentage would be 1.43%, a change of ~16%. This change is not surprising given that the rate of don't knows/refusals to answer (DKRA) increased with age. This is presented in their own Figure 1.

The percentage of DKRA's more than doubles in the 65+ age bracket; the number of respondents in the bracket does not. Based on simple examination of the numbers, it seems appropriate to leave this bracket out of a summary statistic. When you consider potential confounding factors (generational considerations of what being gay is, changes in reluctance to answer questions about sexuality with age, etc.), excluding the age group with a very different DKRA percentage seems even more appropriate.

Second, Cameron's use of cohort life tables seems, to me, inappropriate. It is not explained how Cameron's example of a cohort of deceased Egyptians is relevant to the groups at hand or a justification for using cohort studies here. No complete cohort is defined and by the author's own admission, no "population at-risk" data was available to them. I found a great example of a retrospective cohort study at a Stanford site:

To estimate the association between intrauterine growth and childhood survival, Samuelsen et al (Am J Epidemiol. 1998 Nov 15;148(10):983-91) used records from Norwegian Birth Registry to obtain the birth weights of all children born between 1967-1989. The vital statistics of the children were then determined through linkage to a national death registry. The authors observed that, for children between ages 1 and 5, those with birth weights <>
Cameron's equivalent of a "national death registry" are obituaries from the Washington Blade and Washington Post. From the study:
For U.S. estimates, we examined a series of consecutive obituaries published from 1993 through 2005 in the Washington Blade, a gay newspaper published weekly in Washington, D.C.

Each obituary with enough detail was coded for age-at-death, whether the individual was gay or lesbian, whether the deceased was partnered at time of death, and whether the death was due to HIV/AIDS or some other cause. Some of this series has been reported elsewhere.8 In the current paper, we have re-analyzed all of this data in ways comparable to the sets of Danish and Norwegian deaths.

We gathered four sets of consecutive obituaries from the Washington Post, covering the following time periods: June/July 1988, April 1989, Jan 1999, and Jan 2002. Each obituary was coded for sex, age-at-death, and whether the deceased was ever-married or unmarried.

I think I'll leave that at "nuff said."

If the data were available, wouldn't a survival analysis with censored data be more appropriate here? The report reads like a study in finding and massaging a statistical technique to fit the conclusion you want. It must have been one interesting presentation at March's Eastern Psychological Association meeting. Actually, I did some digging into the EPA's 2007 meeting schedule. Cameron's poster was listed last in its session. This was a poster presented at a meeting. It wasn't a peer reviewed paper. It wasn't even a paper/discussion.

Cameron has also been kicked out of the American Psychological Association and "officially condemned by both the American Sociological Association and the Canadian Psychological Association."

Finally, I'm not a statistician. My undergraduate work was in biomedical engineering and my graduate work was in electrical engineering. My master's thesis utilized maximum likelihood estimation to improve digital images. My PhD work is in organ/cellular physiology. So, I'd love to hear other folks' takes on this.