This essay is a commentary on a Research Note by James Trussell, Barbara Vaughn and Joseph Stanford entitled "Are All Contraceptive Failures Unintended Pregnancies? Evidence from the 1995 National Survey of Family Growth."

If we assume that there is a distinct and identifiable entity known as an "unintended pregnancy," we need to ask whether women's retrospective reports provide accurate and unbiased information on their experience with this phenomenon. A growing body of evidence suggests not. A study using the National Longitudinal Survey of Youth to examine the impact of variations in the timing of asking a question about pregnancy intendedness found strong evidence of bias: Asking the question after the birth resulted in a higher proportion of the pregnancies being deemed wanted than when the question was asked during the pregnancy.1 Other research based on a sample of middle-class couples also demonstrated that reports of pregnancy wantedness become more positive between conception and mid-pregnancy and between mid-pregnancy and six months after the birth.2

One reason for this bias may be that a woman feels increased pressure to give a socially desirable response, or may genuinely change her mind, as a fetus develops and, finally, a child is born. This problem suggests three potential courses of action. First, we could combine what we learn from traditional cross-sectional studies with research that measures intendedness and wantedness prospectively, perhaps even embedding a prospective component in a cross-sectional study. Second, we could continue to seek improvements in the measurement value of the retrospectively obtained information, perhaps by changing question wording or by prompting respondents to recall in greater detail how they felt when they discovered they were pregnant. Third, analysts could develop models of the predictors of reporting bias (e.g., time since conception) and use these models to control for error in analyses using retrospective data.

Even if we deal successfully with the problem of retrospective reporting bias, however, we must still address another basic issue. For decades, demographers have spoken of unintended pregnancy as if it were a distinct phenomenon, different from intended pregnancy in the way a blue chair is different from a red chair. However, the research evidence clearly demonstrates that the "intendedness" of a pregnancy (as measured by the NSFG) is a continuum involving at least two dimensions—intentionality or planning plus an affective dimension expressing happiness or dismay over being pregnant. Focus-group research suggests that "wanting" a birth has to do with one's community, one's partner and one's values about childbearing, while planning (closely related to intending) a birth has to do with one's preparation, life goals and education.3 This study and other research have further shown that the idea of planning a pregnancy does not fit into the way some individuals see their lives.

With regard to the affective dimension, the positive and negative extremes may not be located on the same continuum; that is, positive and negative feelings may co-exist,4 thus producing ambivalence. At the ends of these continua, we see "wantedness" and "unwantedness" concretely: Contrast the many U.S. conceptions that end in abortion with the pregnancies achieved by infertile couples through expensive investments in assisted reproductive technologies. However, given the complexity of the underlying phenomenon of "unintended pregnancy," we agree that the measure traditionally used by our field is a case of misplaced concreteness and a failure to differentiate theoretically distinct dimensions.

There are similar problems of nonspecificity with the application of the term "contraceptive failure" to the failure rates calculated from NSFG data. As Trussell, Vaughan and Stanford observe, a pregnancy may be considered to result from a "contraceptive failure" whether or not the woman was actually physically using her method during the act of intercourse that resulted in the pregnancy. Thus, a pregnancy that occurs when a woman skips three oral contraceptive pills, neglects to use a condom "just this once" or has sex four months after her last DMPA injection can be counted as a contraceptive failure. Researchers conducting clinical efficacy trials have usually been careful to distinguish "use-effectiveness" estimates from "method-effectiveness" estimates. The distinction between these two is a reminder of the fact that users do fail, often because of the ambivalence noted above.

Demographers are not the first to overlay categorical (either/or) measurement on a continuous variable: We speak of child abuse, poverty and youth, even though the cutoffs we use to define these terms are at some level arbitrary and varying. We do this because the categories are useful, indeed necessary, to communicate effectively about the phenomenon in which we are interested. The trick is to define cutoffs that are well informed by common sense and science, and, if we are monitoring trends, to assure that the measures are applied consistently across the years. Similarly, using measures that combine distinct but related dimensions can be helpful in providing a summary indicator of an important outcome. The research to date reassures us that the intentionality and affective dimensions of the "intendedness" measure are related—in fact, strongly related. We think this justifies continued use of the traditional measure.

At the same time, the rich and growing body of research that helps us to "unpack" and better understand the dimensions of pregnancy planning and pregnancy wantedness provides important lessons for both service providers and researchers. Service providers should be encouraged to listen more carefully to women's affect as they counsel about birth control choice and use. Researchers should continue their efforts to expand approaches to these concepts and to develop improved ways of measuring them in future studies, both sample surveys and ethnographic studies. The National Center for Health Statistics began this process in the 1995 NSFG, and will surely continue it in their 2001 survey.