UCLA

Dental data and the way dental data are collected present interesting and challenging statistical issues that can complicate analyses and if not addressed appropriately lead to misleading conclusions. By their nature, dental data suggest the possibility of spatial correlation between neighboring teeth. However, it is not uncommon for statistical methodology assuming independent observations to be employed, which results in unwarranted precision in the inferences drawn from the model. Another obstacle in the analysis of dental data is that teeth can be missing, a pervasive issue that can have an outsized influence on research conclusions. Additionally, collection of dental data on oral health behaviors (OHBs) occurs during dental visits, making the data subject to recall bias. One proposed solution to mitigate recall bias is the use of ecological momentary assessments (EMAs), administered in real time, through surveys on the phone and asking them about recent OHBs. However, this approach inevitably results non-response and consequent missing data, along with questions regarding the optimal number of survey questions per EMA to minimize subject fatigue. Research in Bayesian spatial data analysis (Banerjee, Carlin, and Gelfand, 2014) and missing data (Rubin, 1987; Little and Rubin, 2002) have shown, through application in medical research, the ability of these methods to control for spatially correlated data and use data imputation methods to draw more accurate conclusions when missing data are present. We endeavor to apply and advance these methods in a dental data setting to answer research questions regarding patterns and underlying mechanisms of dental decay in methamphetamine users.