We used an indirect exposure measure, employing study questionnaire data weighted by environmental sampling, which is vulnerable to measurement misclassification error. Still, our questionnaire was previously validated and administered face-to-face by a physician. Furthermore, our study hypothesis was unknown to participants. Groundwater arsenic levels tend to be stable over limited time intervals, and we included only women who received an anemia diagnosis while residing at the study address, for who we could therefore collect water samples representing a relevant exposure interval. Non-cases and cases had a similar median duration of time residing at the study address (8 years and 9 years, respectively) and so we expect that any exposure misclassification would have been similar between the two groups and bias towards the null hypothesis. We also did not account for potential arsenic exposures in the workplace or from contaminated foods, introducing an additional limitation into our exposure assessment strategy. However, no women reported employment with potential occupational exposure, and rice consumption, an important source of arsenic exposure in some populations, is not common in this region of Romania (Neamtiu et al., 2015). We used education level and BMI as proxies for socioeconomic and nutrition status in general, but were unable to evaluate the presence of specific concurrent micronutrient deficiencies, and therefore a potential confounding effect from these factors cannot be excluded. We did not adjust for pregnancy loss, as there was no overall association with arsenic exposure in our earlier work (Bloom et al., 2014).