Rejoinder to Treharne, Brickell, and Chinn (2011)

Treharne, Brickell, and Chinn (2011) raised concerns about three aspects of our article on sexual orientation (Wells, McGee, & Beautrais, 2011), namely the methodology, the aims, and the languageused.Theyhaveaclearpreference forqualitativestudies and we agree that such studies can provide in-depth information and explore the complexity and fluidity of sexual orientation. However, that is not the type of study we did, so our work should be judged by the standards of survey research. Treharneetal. objected,firstofall, toourhavingsampledone person per household, stating that we assume that ‘‘this facilitates a representative sample in a national survey. In reality, this approach may lead to under-sampling of minority groups living in shared households that diverge from the traditional.’’In the statistical methods section, we stated that all estimates were weighted according to the study design variables. As anyone familiar with survey sampling would understand, when one person has been selected per household, this is dealt with by weighting by the inverse of the number of eligibles within a household, as has been standard in surveys for many decades (Kish, 1965). An extremely detailed account of weighting in the New Zealand Mental Health Survey was provided in a report to theMinistryofHealth, theprimary fundersof thesurvey (Wells, McGee,&OakleyBrowne,2006).This reportwas referenced in our article and we even provided a URL to facilitate easy access as the report ispublicallyavailable.Becauseof theweightingwe used, there is no bias in our estimates from sampling one person per household. Consideration of confidentiality also led us to choose one person per household, as we were concerned that, if multiple peoplewere interviewed,oneperson mightask another how they had responded to specific questions from the interview. The second concern of Treharne et al. is that we did not use stratified sampling to obtain approximately equal numbers in our comparison groups. It is true that for a fixed total sample size, N, power is greatest for detection of differences between two groups if they are of equal size. Hence, most clinical trials have equal numbers in each arm. In population surveys, however, many comparisons will be made and stratification cannot cover all such comparisons. Stratified sampling is possible only if thereareknownstrata tosample from,which isnot thecase for sexual orientation. Even oversampling was not possible for sexualorientation in theNewZealandMentalHealthSurvey.There were no prior data, such as from census data, so that areas with higher density of sexual orientation minority groups could be targeted. Nor was screening feasible: consider asking about sexual orientation at the doorstep to screen sexual orientation minority groups into the survey. Moreover, which of the multiple aspects of sexual orientation would be included in screening? We asked about sexual orientation well into the surveyand had very few refusals. For population estimates, a very large survey provides the bestopportunity for obtaining contact withall groups in the population, not just people who belong to particular organizations or who volunteer to participate. For a given number in a particular group, comparisons are more precise the larger the comparison group whereas Treharne et al. seem to imply that the opposite is the case and that a small group is disadvantaged compared with a large group. They stated: ‘‘We do not wish to downplay those participants’ experiences, but they were compared to many thousands of heterosexual participants who were used as the reference group with a vastly disproportionate denominator in analyses.’’ There were J. E. Wells (&) M. A. McGee Department of Public Health and General Practice, University of Otago, Christchurch, PO Box 4345, Christchurch 8140, New Zealand e-mail: elisabeth.wells@otago.ac.nz