Precision of Category Versus Continuous Economic Data: Evidence from the Longitudinal Research on Officer Careers Survey.

摘要

This research evaluates category versus numeric responses to questions in the U.S. Army Research Institute for the Behavioral and Social Sciences's Longitudinal Research on Officer Careers (LROC) Survey, which examines career intentions of junior Army officers. The assessment is based on the relative efficiency of estimates of regression model parameters. Efficiency is measured by the standard errors of coefficient estimates of models applied to category and numeric response data, respectively. The analysis consists of two parts. First, a Monte Carlo experiment is conducted. It estimates ordered logit models (OL) for category data and an ordinary least squares (OLS) regression model using numerical response data with measurement error. Second, the analysis of the LROC survey data involves estimation of ordered logit and OLS regression models. The categorical career intentions questions provide data for the dependent variables in the ordered logit models. The dependent variable for the OLS model is the numeric response to the intention question. Sixteen explanatory variables that measure career-related variables (e.g., source of commission and branch satisfaction) and socioeconomic variables (e.g., gender) are included in each model. Findings indicate that standard errors of regression estimates are smaller for numeric than for categorical data. Junior officers, Retention, Regression analysis, Category variables, Continuous variables.