Quantitative Methods (2~3)

Scale reliability analysis and correlations.

In assessing internal consistency, the item-total statistics suggest acceptable reliability among the scale items measuring attitudes towards management, with Cronbach’s alpha values, if any item is deleted, ranging between .875 and .901 (Travakol & Dennick, 2011). These values exceed the commonly accepted threshold of .7, indicating that the items are consistent with one another and appropriately reflect the construct.

As presented in the descriptive statistics, the aggregate scales' means and standard deviations appear reasonable, indicating a normal range of responses across the sample. For instance, ‘Job Satisfaction’ and ‘Trust in Peers’ show high means and relatively low standard deviations, suggesting that these positive workplace factors are consistently perceived across the sample (Judge et al., 2001).

Bivariate correlations provide insights into the relationships between independent variables (IVs) and the dependent variable (DV). The negative correlations between ‘Intent to Quit’ and variables such as ‘Job Satisfaction’ and ‘Trust in Management’ imply potential predictive relationships. However, the strength of these correlations should be interpreted with caution, considering potential confounders that might influence these relationships (Field, 2017, 27-34).

Regarding confounding among IVs, the strong positive correlation between ‘Job Satisfaction’ and ‘Organisational Commitment’ (r = 0.769) could indicate that these variables share a common underlying construct, which may complicate their contributions when modelling their relationship with the DV. Confounding is an issue because it can obscure the genuine relationship between IVs and DV, leading to spurious conclusions (VanderWeele & Shpitser, 2013).

However, confounding only sometimes leads to problems if identified and controlled for, such as through statistical methods like multiple regression or experimental designs that can isolate the effects of the IVs on the DV (Shadish et al., 2002, pp. 75-105). Recognising confounding factors and their management can even enhance understanding of complex relationships within the data (Pearl, 2009, 78-79).

Multiple regression.

The regression analysis's dependent variable (DV) is ‘Intent to quit’. The independent variables (IVs) were entered using a hierarchical regression approach. Model 1 includes ‘Age’ and ‘Total Years in Paid Employment’ as predictors. Model 1 did not explain significant variance in the DV (R Squared = .010), and the F Change was not substantial (p = .161). This suggests that these IVs alone do not significantly predict ‘Intent to Quit’ (ItO).

Model 2 added ‘Organisational Commitment’ (OC), ‘Trust in Peers’ (TiP), ‘Trust in Management’ (TiM), and ‘Job Satisfaction’ (JS) to the regression. This model explained a more significant proportion of variance (R Squared = .536) and was statistically significant (p < .001), indicating a robust predictive relationship with ‘ItO’.

Entering Model 1 first is advantageous as it follows for assessing the incremental variance explained by the additional IVs in Model 2. This sequential entry method demonstrates the relative importance of the other variables beyond the primary demographic factors (Field, 2017, pp. 27-31).

Model 2 should be interpreted in the coefficients table due to its significant prediction of the DV. The significant betas in Model 2 are for ‘JS’ (β = .252), ‘TiM’ (β = .135), and ‘OC’ (β = -.630). The directionality indicated by the betas suggests that ‘JS’ and ‘OC’ have an inverse relationship with the ‘ItO. Conversely, a positive beta for ‘TiM’ indicates that higher trust in management is unexpectedly associated with an increase in ‘ItO. The magnitude of these betas suggests the size of the effect each IV has on the DV when controlling for other factors in the model.

Betas are partial coefficients, meaning they represent the unique contribution of each IV to the prediction of the DV, controlling for other variables in the model. This is distinct from correlation coefficients that quantify the intensity of a direct linear connection between two variables without considering additional variables (Cohen et al., 2013, pp. 452-459).

Data cleaning involves handling missing data, identifying and removing outliers, and ensuring data consistency, which is necessary to prepare the data for analysis and avoid skewed or misleading results (Tabachnick & Fidell, 2013, pp. 61-77).