Table 3 shows the odds ratios with logistic regression for the predictors of the different outcomes of interest.
To additionally test whether the regression coefficients of the predictors were robust, we cross-validated each adjusted regression model by using the Chow-test [29]: We split the total sample into two random parts, amounting to 75 and 25% of the total sample, and created a new variable that indicated to which sub-sample each case belonged.
The total standardised and nonstandardised regression coefficients of the predictors together with the results of the significance tests are provided (Table 6) and represent the sum of direct and indirect effects.
The coefficient of determination (R) and the standardised and nonstandardised (b) regression coefficients of the predictors together with the Pearson correlation coefficient (r) and squared semipartial correlation (sr) are reported.
The results of logistic regression analysis of the predictors for using oral health care facilities are summarised in Table 4. Various predisposing factors significantly increased the probability of using oral health care facilities, including Christian affiliation and moderate or high material living conditions.
Furthermore, to investigate if the regression coefficients of the predictor variables differed between models, we used the approach by Cohen32 in which the difference between the predictors' regression weights is divided by the standard error of the difference and the resulting z-score is tested for significance.
In case, variables are standardized beforehand, the regression coefficient of the predictor variable equals Pearson's correlation coefficient.
* p < 0.10, ** p < 0.05, # p < 0.01 Because the dependent variable (personal or indoor endotoxin) was log transformed, we exponentiated the regression coefficient of the predictor, thus yielding the proportional change in endotoxin exposure and 95% confidence interval (CI).
Estimating the partial regression coefficients of the predictor variables by ordinary least squares (OLS) requires that the sample size exceed the number of coefficients, which in the GWAS context, may be of order 10 or even 10.
The results from binomial logistic regression of predictors of the likelihood of sitting the UMAT twice or more are outlined in Table 5.
