Studies of the causes and correlates of crime and deviance make great use of multiple-item measures of self-reported offending. However, these measures of offending, whether recidivism, count of arrests, or self-report inventories have created responses that have highly skewed distributions that are poorly suited to conventional methods of scaling and analysis. This is the second article dealing with these methodological problems and concerns statistical models for substantive analyses that relate those scores to explanatory variables. Specifically, this article attempts to inform criminological researchers about tobit regression, an alternative regression model intended for continuous data that are censored or bounded at a limiting value. This paper intends to establish the magnitude of the problems when standard ordinary least squares (OLS) methods are applied to self-reported offending and the degree of improvement offered by tobit regression. The paper begins with choosing a statistical model. This is followed by the use of three empirical analyses to test the appropriateness of OLS and tobit regression for analyzing self-reported offending. It was shown that the common ad hoc practice of OLS analysis of summative scores was ill suited to the data, and improvement was offered by transformations that altered the metric of self-reported offending and by the tobit regression model. References