The overall aim of this study was to examine a broad range of potential adolescent risk factors for developing purging behaviors in a nationally representative sample of U.S. young adult women. Demographic variables did not differentiate risk for the development of purging behaviors. In univariable analyses, several childhood variables were identified to increase risk for purging. Higher BMI, suicide attempts, overweight self-perception, parental poverty, depression symptoms, hyperactivity-impulsivity, delinquent behaviors, low self-esteem, and never having lived with one (or both) parent (s) were associated with purging at Wave II. Higher BMI and depressive symptoms were associated with purging at Wave III. Finally, lower levels of self-esteem were associated with an increased likelihood of purging at Wave II compared to Wave III.
Of the demographic variables controlled for in our study, purging at Wave I was the only variable that was associated with purging at either follow-up time point. Neither participants’ age, race/ethnicity, nor highest level of parental education was found to be associated with purging at Wave II or III. The lack of age differences across the two groups is not particularly surprising given the fact that the Add Health sample has a restricted age range. Consistent with a growing literature that reports that girls or women from various racial/ethnicity backgrounds are equally likely to develop disordered eating, [26, 27] our results suggest that purging may be a problem across various racial or ethnic groups. Our finding that highest level of parental education, an indicator often used in the literature as a proxy for parental socioeconomic status, did not differ when comparing purging groups warrants comment in light of our result that family poverty did differentiate the two groups. Both variables were measured using parent/guardian self-report. While parental education does not appear to increase risk for purging, experiencing economic hardship may elevate such risk as will be discussed in more detail below.
Regarding the purging variable, it was found that those who reported purging at Wave I were less likely to also report purging at Wave II or III, and purging at Wave I was not associated with a difference in the likelihood of purging between Wave II and III. It is possible that this inverse relationship is a byproduct of how purging was measured in this study, given that the variable only reflected purging behavior over the last seven days. More frequent longitudinal assessments might better capture the extent to which purging behavior is stable between adolescence and young adulthood.
Additionally, our study confirmed several risk factors for purging that have been shown to predict risk for eating disorders in previous studies, and identified previously understudied time varying effects for several of these variables. For example, self-perception of being overweight significantly predicted purging at Wave II but not at Wave III. Research consistently has found that body image concern (operationalized in a variety of ways in different studies) contributed to increased risk for eating disorder symptoms [2, 28] and, therefore, has been targeted as a key risk factor to be addressed in prevention programs . However, our results suggest that this may only be a risk factor over a certain time range. Given the stability of these risk factors over time, it’s surprising that some Wave I predictors that showed significant relationships at Wave II did not predict Wave III purging behaviors. Again, more intensive longitudinal data collection would be useful to more clearly determine how risk factors may affect the likelihood of purging between adolescence and young adulthood.
Similarly, consistent with previous studies, low self-esteem,  depression, [28, 30, 31] and impulsivity [32–34] were significantly associated with increased risk for purging. In addition, we found significant relationships between purging and self-reported suicide attempts as well as between purging and having received psychological counseling. In a series of case control studies, Fairburn and colleagues found that “parental separation” (e.g., due to prolonged illness) and “parental loss” (e.g., due to death) were risk factors for developing an eating disorder [35, 36]. In our study, never having lived with one (or both) parent (s) was a risk factor for purging, but having lost a parent to death was not. We cannot answer whether these discrepant findings are a function of methodological differences (e.g., case–control design versus longitudinal design; differences in how the target groups were defined). As noted by Fairburn and colleagues (35), parental separation and loss are also risk factors for other psychiatric problems. Therefore, these parental risk factors may be indirectly associated with the development of eating disorders. For example, parental separation or loss may increase the risk of developing psychiatric problems, which in turn, could increase the likelihood of eating disorders. Future research that examines the nature of the association between these parental factors and eating disorders is important to determine how parental separation and loss might influence eating disorder risk in terms of being mediated or moderated by other risk factors, such as depression. Finally, in our sample, family poverty was found to be predictive of purging onset. This bolsters emerging literature that suggests a relationship between socioeconomic hardship and unhealthy weight control behaviors .
Several limitations need to be acknowledged. Although Add Health is a large epidemiological study, the sample still was not large enough to test hypotheses about risk factors for purging in young adult men. Therefore, our findings should not be generalized to boys or men. As is common in large-scale epidemiological studies, key study variables were measured using self-report questionnaires, which tend to be less reliable than interview based measures. Childhood abuse variables as well as the ADHD-IN and -HI scales were collected retrospectively and were, therefore, subject to recall errors. Additionally, because frequency information was not available for the purging behaviors, we were unable to study severity of the behaviors. Interval censoring may also be an issue, given that questions only addressed purging in the last seven days, which did not allow us to identify whether those reporting purging during those days were new onset purgers or whether they may have purged prior to the assessment period. As a result, purging is likely to be underestimated.
Finally, information was not available regarding binge eating behaviors or other extreme weight control behaviors, such as extreme dietary restrictions or excessive exercise. Further studies would be needed to compare our results regarding purging to these similar unhealthy behaviors.
These limitations were offset by several strengths. The study involved a community sample, thus not featuring the limitations inherent in studies of patient samples. The relatively large sample size afforded us the opportunity to use multivariable modeling of risk factors for purging, a relatively uncommon yet clinically important behavior. As well, Add Health collected data on numerous psychosocial variables, allowing us to explore factors that have been considered in psychopathology research regarding other disorders but less so in studies of eating disorders. Finally, our study covered a considerable period of time (8 years) from adolescence into early adulthood.