Regression models to predict corrected height, weight, and obesity indicators among university students in Beijing, China

Objectives: Whereas data collection on subjective anthropometric measures is inexpensive and sometimes may be the only feasible option for large-scale population-based surveys, self-reported height and weight can be susceptible to measurement error and social desirability bias. In this study, we aimed to assess the level of discrepancy between self-reported and device-measured height, weight, and obesity indicators, and to construct regression models to predict corrected anthropometric measures using self-reported data. Methods: Paper-and-pencil-based health surveys were administered to all freshmen enrolled in Tsinghua University in Beijing, China. Freshmen's height and weight were measured by trained staff using stadiometer and digital scale within one week following survey completion. Robust regressions were performed to predict corrected height, weight, body mass index (BMI), and overweight and obesity prevalence using self-reported data (N = 16,675). Results: Male freshmen over-reported both height and weight, whereas female freshmen over-reported height but under-reported weight. Both resulted in underestimation of BMI and overweight prevalence. The predicted values based on robust regressions substantially reduced the discrepancy between self-reported and objectively-measured height, weight, BMI, and overweight prevalence. Conclusions: Parsimonious regression models could be useful in obesity surveillance by predicting corrected anthropometric measures using self-reported data.