7

method is especially useful for comparing various possible models of the observed data

(Bentler, 1980, 1990).

**Model specification **

Latent variable structural equation modeling was employed to examine the determinants

of physical activity. Age, gender, education and income were specified as exogenous

variables while social connections, self-efficacy and physical activity were endogenous latent

variables. Social connections and self-efficacy were specified to mediate the path from age,

gender, education and income to physical activity. All indicators from the latent variables

have loadings, with the latent variable representing the measured construct. One of the

loadings for each endogenous variable has to be constrained at a value of 1.0 for

identification of the structural model (Maruyama, 1998). There was a correlation among the

residuals of the indicators for the latent variables as it is assumed that these indicators share

common variances. The hypothesized model was tested in the samples for Whites, Blacks and

Hispanics respectively.

**Model Fit Statistics **

We used both absolute and comparative fit statistics to evaluate the hypothesized model.

As the absolute index (chi-square, goodness of fit [GFI]) is dependent on the sample size,

there is a higher probability that a correct statistical model with large sample size will be

rejected even when it should not be rejected (Bentler, 1980; Maruyama, 1998). Comparative