Cluster and Latent class analyses were used to examine within Stage profiles of individuals at-risk for three behaviors based on their perceived Pros, Cons, and Self-efficacy. In addition, Structural equation and Latent variable modeling were used to examine the predictors of 12 months exercise participation. The results of Cluster and Latent class analyses did not replicate the results of previous studies, which delineated at least four distinct profiles within the Stages. Only two to three profiles emerged within each Stage as well as within the full dataset. Most two-cluster/class, and three-cluster/class solutions consisted of at least one cluster with extremely small sample size suggesting instability. External validation was only partially supported. Within Contemplation, the results for a two-class solution showed that there are differences between two classes on mild physical activities. An ANOVA also indicated that for the two-class solution within the Preparation sample there were differences between two classes on strenuous physical activities. For a three-cluster solution performed on the combined dataset, an ANOVA indicated that mean levels of moderate physical activities differed across clusters.

The best fitting Structural equation and Latent variable models suggest that Self-efficacy does not act as a mediator between independent variables and the outcome. Direct effect models with age and health being the strongest direct predictors of 12 months exercise behavior within each Stage as well as within combined dataset fit the data well. The same model fit equally well across different Stages and was invariant across the Stages when structural paths and covariances were held equal across groups. The fit of the model was further examined using Latent variable modeling. The model could not be tested separately across Stages due to small sample sizes in each Stage relative to the number of parameters; therefore, the model was tested on the total sample. The Pros, Self-efficacy, and Physical activities were entered into the model as latent variables with several manifest indicators. The resulting model fit the data well. Again, a direct effect model with age and health as the strongest direct predictors of 12 months exercise behavior fit the data well.