Assessing the Accuracy of Self-Reported Health Expenditure Data: Evidence from Two Public Surveys in China

This paper utilizes Benford's law, the distribution that the first significant digit of numbers in certain datasets should follow, to assess the accuracy of self-reported health expenditure data known for measurement errors. We provide both simulation and real data evidence supporting the validity assumption that genuine health expenditure data conform to Benford's law. We then conduct a Benford analysis of health expenditure variables from two widely utilized public datasets, the China Health and Nutrition Survey and the China Family Panel Studies. Our findings show that health expenditure data in both datasets exhibit inconsistencies with Benford's law, with the former dataset tending to be less prone to reporting errors. These results remain robust while accounting for variations in survey design, recall periods, and sample sizes. Moreover, we demonstrate that data accuracy improves with a shorter time interval between hospitalization and interviews, when the data is self-reported as opposed to proxy responses, and at the household level. We find no compelling evidence that enumerators' assessments of respondents' credibility or urgency to end interviews are indicative of data accuracy. This paper contributes to literature by introducing an easy-to-implement analytical framework for scrutinizing and comparing the reporting accuracy of health expenditure data.