Controlling whether and when a pregnancy occurs is a human right. Yet, despite more than 40 years of the U.S. Department Health and Human Services prioritizing the reduction of undesired pregnancies, rates remain high [49]. This is at least in part because despite a great deal of research on this topic, we still have relatively little understanding of why some women are able to get what they want in terms of pregnancy while others are not. This study takes an innovative approach to understanding this puzzle in two ways, which allow me to overcome two persistent barriers to our understanding of undesired pregnancies. First, I apply a unique and innovative theoretical framework—the Traits-Desires-Intentions-Behavior (TDIB) framework—that has been largely neglected by demographers. The TDIB has great potential to contribute to our understanding of undesired pregnancies because it was specifically designed for this purpose; it focuses on the potential mismatches between desires and intentions, and between intentions and behaviors. Second, I use a unique longitudinal, mixed-method study of 18- and 19-year-old women, the Relationship Dynamics and Social Life (RDSL) study, which includes both prospectively measured pregnancy desires and semi-structured interviews (n=75) with subsequently pregnant and non-pregnant respondents. The semi-structured interviews, particularly the non-pregnancy interviews, remain almost entirely un-analyzed, and were designed specifically to generate innovative new hypotheses and evidence about undesired pregnancies. First, we interviewed a group of 40 RDSL respondents, (distributed evenly across white/Black and poverty/non-poverty groups) who experienced pregnancies during the 2.5-year study period. (The vast majority were undesired.) Second, we interviewed a comparison group (n=32) who were similarly distributed across race and poverty groups, but who avoided pregnancy during the study period. Using a systematic anomalous case analysis strategy [75], an abductive approach to generating new hypotheses with “surprising” (i.e., in this case, unpredicted by statistical models) cases, we selected respondents with high model-based propensity for pregnancy, based on a hazard model using the 2.5 years of survey data. We distributed these interviews across three additional groups, based on survey responses: those with zero or non-zero pregnancy desire, one or multiple intimate partners, and perfect or imperfect contraceptive use. I propose to use NVivo software and qualitative analysis techniques to analyze the semi-structured interviews by comparing the pregnant and non-pregnant respondents who differ in terms of three domains: pregnancy desires, intimate relationships, and contraceptive use. I will also compare pregnant and non- pregnant respondents who match in terms of these domains, to generate and evaluate new hypotheses that do not focus on these domains. Finally, I will use these analyses to further explicate and expand the...