Putting the Pieces Together: Salary Models
The real magic happens when we combine these variables into separate models for CIOs, managers, and staff. When we do this, some factors that seem to be associated with salary at the bivariate level are no longer significant predictors after controlling for the others. And the effect of some variables that may have been weakly or unassociated may only pop out when considered in the multivariable environment. On the basis of precedent, we began by creating three separate models, one for each of the organizational-level categories: CIOs, managers, and staff. Our general approach to selecting an appropriate multiple regression model was to first build a full model containing all potential predictors of salary and then reduce that model to obtain more precise estimates of the relationships between salary and the significant predictors (table 1).1 In the following sections, we unpack each of the respective models, providing a more detailed explanation and analysis of the multivariable models for CIOs, managers, and staff.
Table 1. Factors tested in multivariable salary models
Category | Factor | CIO | Manager | Staff |
---|---|---|---|---|
Individual demographics | Gender | ✘ | ✘ | ✓*** |
Generation | ✘ | ✓** | ✓**** | |
Ethnicity | ✘ | ✘ | ✘ | |
Individual career path | Education level (highest achieved) | ✘ | ✓** | ✓** |
Years in current position at current institution | ✘ | ✘ | ✘ | |
Extra years at current institution in different position(s) | ✘ | ✓** | ✘ | |
Extra years in higher education at different institution(s) | ✓*** | ✓*** | ✓* | |
Characteristics of the current position | Cabinet membership (CIOs only) | ✓**** | N/A | N/A |
Number of direct reports in current position | ✘ | ✓** | N/A | |
IT sector (managers and staff only) | N/A | ✓**** | ✓**** | |
Characteristics of the current institution | Institution type | ✓**** | ✓**** | ✓**** |
✓ indicates the factor has a significant association with salary ✘ indicates no significant association was found *p < .05 **p < .01 ***p < .001 ****p < .0001 |
Note
-
Statistical testing of this full model allows us to identify and remove factors that do not have a significant association with salary in the presence of the other ones. The simplified, reduced model then gives more accurate estimates of the relationship between salary and each significant predictor, controlling for the effects of the other significant predictors. Especially in cases of marginal statistical significance, further combinations of predictors were explored, including adding predictors back to the reduced model to reevaluate their potential contribution in the absence of other nonsignificant predictors. Here the art of quantitative research meets its science. Finding consistency in statistical conclusions among these additional models lends additional confidence to the appropriateness of the predictors included and excluded in each of the final models.
↩︎