A flexible model of food security: Estimation and implications for prediction
提出一种贝叶斯分级响应模型（BGRM），可同时处理二分类、有序多分类和连续变量，用于测量家庭粮食安全水平及其不确定性，并基于NHANES和CPS数据验证了模型效果。
Abstract We propose a novel Bayesian Graded Response Model (BGRM) for measuring household‐level food security and other latent traits. The BGRM produces continuous food security estimates along with household‐level measures of estimation uncertainty. Unlike the USDA's official model, the BGRM accommodates both binary and ordered polytomous items. We further extend the model to allow for any combination of binary, ordered polytomous, and even continuous variables. To demonstrate the model's features, we estimate the BGRM using responses to the 10 adult core Food Security Module questions from the 2017–2018 National Health and Nutrition Examination Survey (NHANES). Results show non‐trivial uncertainty in household‐level food security estimates and overlap across USDA‐defined food security categories. As a robustness check, we estimate the model with Current Population Survey data, finding qualitatively similar results. We illustrate an application of the continuous food security estimates by calculating Foster, Greer, and Thorbecke indices which capture the prevalence, depth, and severity of food insecurity. To demonstrate flexibility in variable selection, we also include a continuous variable, household‐level monthly food spending , capturing both economic access and experiential food security information in a single latent construct. The adaptability of the BGRM positions it as a versatile tool for measuring food security and related latent traits, particularly when measures of uncertainty or a mix of different variable types are required. While the empirical application illustrates model capabilities, the primary contribution of the study is methodological.