A normal P-P plot of standardised residuals (Figure 4) indicated that data points did not strictly lie on the line, but were sufficiently close to determine the normality of distributed error; the data also met the assumption of homoscedasticity as shown in the scatterplot of standardised predicted values (Figure 5) (Dart, 2013).
Another assumption of multiple regression is that no single data point has a high residual value (such data points are referred to as outliers), and nor a high value on a predictor variable (such data points are referred to as having high leverage) (Lund, 2013). With regards to outliers, there were no studentized residuals which exceeded 2 (outlined as problematic, by Blantá, 2014)
Leverage is determined by the equation:
Where v refers to the number of predictor variables and n refers to the number of data points.
Throughout several multiple regression analyses, exclusion of data meant changes to the number of data points. The leverage equation for the final multiple regression analysis was:
None of the remaining 27 data points surpassed this threshold (highest value = .24). Cook’s distance test – one that is used to determine how the individual removal of variables would influence the model – considers values over 1 to be influential (Field, 2013).
Another assumption is that all three predictor variables must also not correlate with each other. A collinearity analysis produced Variance Inflation Factor (VIF) values which indicated some multicollinearity (Pay, VIF = 3.13; Worked Hours, VIF = 1.75; Socialising, VIF = 3.9) between predictors, but none that could strictly be considered problematic (i.e. values of 10 and beyond; StatisticsSolutions, 2018), and thereby violations of this assumption.
Furthermore, to ensure an absence of correlation between residuals, a Durbin-Watson test was conducted (producing a value of 1.8), and by the conservative parameters of Field (2013; where a value close to 2 indicates residuals are uncorrelated), was not initially indicative of any issues. Upon comparison to the precise upper bound value (1.75) provided by Savin and White (1977) the null hypothesis (no correlation) was conclusively accepted.
Using the enter method, pay, worked hours, and job satisfaction was found to explain a significant amount of variance in job satisfaction (F(3, 23) = 16.29, p =