The models presented here include general linear models, the standard applications of which assume that the endogenous variable is equal-interval. Examples of solutions to the statistical difficulties presented by endogenous variables with upper and or lower bounds are given, exemplified by randomized experiments in connection with inmate vocational training programs. The complications produced by using truncated and censored endogenous variables are exemplified by a study of hourly wages earned by ex-prisoners. The use of multiple-equation models is described through an example involving the impact of different police patrolling practices. Unobserved variables and the critical empirical role they play in statistical procedures are addressed; they have recently become a subject of interest to statisticians. Confirmatory factor analytical models are examined in terms of three possible applications to criminal justice evaluations: to address the classic underadjustment problem in nonequivalent control group designs, when the researcher has multiple indicators of some underlying variable of interest, and to study measurement error in criminal justice evaluations. Viable alternatives for longitudinal data can be time-series models (e.g., to model the impact of multiple interventions or to consider the relationships between two or more stochastic time series). Other procedures which, although currently less visible in criminal justice evaluations may become significant in the future, are mentioned under the broad headings of robust and resistant estimation, Bayesian techniques, ridge regression for solving multicollinearity, and nonrecursive (multiple-equation) models in which one or more of the endogenous variables is subject to truncation and censoring. Figures, graphs, notes, and 56 references are included.