their samples in sites where lesbians obviously congregated, such as bars or feminist and lesbian activist meetings. Not surprisingly, scientists found in those samples that lesbians consumed alcohol and, parenthetically, were predominantly feminists and lesbian activists. This confounding of outcome variables with convenience sampling strategies undermines confidence in the prevalences of particular behaviors, health risks, and health and mental health conditions reported in these studies.
Much of the existing research on lesbians has obtained research subjects from lesbian community groups (e.g., mailing or membership lists from lesbian or gay organizations). Other researchers have obtained subjects at regional or national gay or lesbian events that draw people from a large geographic area. These nonprobability methods of obtaining lesbian research subjects yield samples that are not representative of the general population of lesbians since they are limited to lesbians who are open enough about their sexual orientation to attend community events or who subscribe to lesbian and gay newspapers and magazines.
These sampling biases may not be fatal flaws, however, if the researcher's interest is in a particular subset of the lesbian population. For example, a sample drawn at a widely attended lesbian event may be relatively representative of lesbians who have come out, a potentially important group for researchers to study because they are the most visible to the public and so may affect how heterosexuals view lesbians. Therefore, nonprobability samples can generate population estimates that may be generalizable to restricted subpopulations of lesbians. It is critical, however, that the limits of generalization be acknowledged and that the findings observed in these restricted groups are not assumed to apply to all lesbians. For many purposes of health research, however, such as measuring the prevalence of disease conditions or risk factors in lesbians, more representative sampling designs will be preferable to nonprobability samples.
A variety of sampling techniques have been designed to identify members of "rare" population subgroups for research studies (i.e., sub-