Volunteering and health benefits in general adults: cumulative effects and forms

Three of the five health outcome variables were measured by a single question item with a 4-point or a 5-point scale, and some researchers believe this type of outcome should be treated as ordered categorical variables. Therefore, we re-ran the data with Probit regression models. Results showed that the standardized coefficients of Probit regression models were similar to the findings obtained from multivariate linear regression models. In fact, statisticians have argued that when a dependent variable is measured by 4-point or 5-point rank-ordered categories, there would be little difference between the results obtained from ordinal regression and OLS regression [20, 21]. When an outcome is measured by an ordered categorical scale, e.g. “Overall, how would you rate your mental health at the present time? Would you say it is excellent, very good, good, fair, or poor?” on a 5-point scale (1 = excellent, 2 = very good, 3 = good, 4 = fair, and 5 = poor), a latent continuous variable, y^*, ranging from -∞ to ∞ is denoted. For measurement purposes this is, mapped to an observed variable y. Hence, the above ordinal outcome as an example is in principle related to a continuous, latent variable y^* in indicating a person’s level of mental health by category, ranging from “excellent” to “poor”. Therefore, the observed y is related to y^* in an equation of “y _i  = m if τ _m-1 ≤ y _i ^* <τ _m for m = 1 to Ј”. For easier interpretation, the present study opted for the results of multivariate linear regression models. Readers can contact the first author to obtain the results from Probit regression models for reference.