A latent variable modeling procedure that can be used to evaluate intraclass correlation coefficients in two-level settings with discrete response variables is discussed. The approach is readily applied when the purpose is to furnish confidence intervals at prespecified confidence levels for these coefficients in setups with binary or ordinal outcome measures and nesting of subjects within higher order units. The method can aid educational and behavioral researchers in their study of sources of observed outcome variability and model choice considerations in multilevel settings, and is illustrated with empirical survey data.