Social desirability reporting leads to over estimations of church attendance. To date, researchers have treated over-reporting of church attendance as a general phenomenon, and have been unable to determine the demographic correlates of inaccuracy in these self-reports. By comparing over eight months of observational data on church attendance (n = 48 services) to self-report in a rural Fijian village, we find that 1) self-report does not reliably predict observed attendance, 2) women with two or more children (≥ 2) are more likely to over-report their attendance than women with fewer children (≤ 1), and 3) self-report of religiosity more reliably predicts observed church attendance than does self-report of church attendance. Further, we find that third-party judgements of church attendance by fellow villagers are more reliably associated with observed church attendance than self-report. Our findings suggest that researchers interested in estimating behavioral variation, particularly in domains susceptible to social desirability effects, should consider developing and employing third-party methods to mitigate biases inherent to self-report.