*2.3. Statistical Analyses*

The descriptive statistics provided frequencies and percentages of participant characteristics and health risk behaviors, which were compared between Latino subgroups using the chi-square test. A sample size calculation was performed to achieve 80% power for overall correlation analyses and for comparison analyses between Latino descent groups.

Statistical analysis models were derived from the study by Baruth et al. [12]. The prevalence rates of (1) current smoking status, (2) physical activity, and (3) fruit and vegetable consumption were computed in the overall sample and in each Latino subgroup. Next, the prevalence of multiple health risk behaviors was calculated by assigning a score of 1 (versus 0) to each for being a smoker, engaging in less than 30 min of physical activity per day, and consuming <5 servings/day of fruits and vegetables. The number of risk factors for each person was the sum of these individual scores (score range 0–3).

A series of hierarchical log-linear models using SAS/STAT CATMOD 15.2 (SAS Institute, Cary, NC, USA) were fit to the observed counts using maximum likelihood estimation. First, the main effects model was used with the three risk behaviors. The model assumed mutual independence among the three risk behaviors. Then, one or more two-factor interactions were added to the model to obtain the model with the least significant likelihood ratio *p*-value (best fit as indicated by the least difference between the observed and expected frequencies). Within this best fit model, the most significant interaction term (i.e., the interaction term with the lowest *p*-value) identified the most important risk factor cluster. Any interaction included in the final model was interpreted as reflecting dependence among the risk behaviors. The models were calculated for all Latinos and for each Latino subgroup. For Puerto Rican models, 0.5 was used in place of the zero frequency where no Puerto Rican participants smoked and consumed at least 5 servings of fruits and vegetables.
