Embedded Files
Let's use hierarchical regression to improve an organization's selection process
Let's use hierarchical regression to improve an organization's selection process
Software Used:
R
Methods Used:
Correlation Matrix
ANOVA
Heirarchical Regression
What's the purpose of this project?
HR teams often use numerous methods to identify the best external talent. However, applicants can become disheartened during the application process if they feel they are going through many interviews or completing unnecessary tasks (article). Using statistics, we can determine whether an organization should continue using its three selection methods.
About this data set: This data set contains the scores for conscientiousness, interview, and work samples for about 200 employees. This data is available here.
Making the models
By correlating our variables with a correlation matrix, we can determine the order of entry for the IVs. Looking at the first column of results, we can determine the order entry going from highest to lowest value.
Interview should be first, followed by conscientiousness, followed by work sample when building the hierarchical regression. Model 1 will contain interview, Model 2 will contain interview and conscientiousness, and Model 3 will contain the prior 2 variables and work sample.
Model 2 F(1,173)=7.91, p<.01
To determine which model is the best, we conduct an ANOVA to examine R squared change of the 3 models. The p-value for model 2 compared to model 1 is significant, but model 2 compared to model 3 is not significant. We can also examine the AIC and BIC scores. Model 2 has the same AIC score as Model 3, but Model 2 has a lower BIC score. So, we can assume Model 2 is best.
Checking our data for assumptions
Our VIF score is less than 10 for all 3 variables, so multicollinearity did not occur.
Our graph shows a mostly horizontal line, so we can assume data does not violate the assumption of linearity.
The above QQ plot shows the data points mostly following the straight line, so the normality assumption is not violated.
The residuals appear to be evenly distributed across the range, so we can assume the data does not violate the homoscedasticity assumption.
Find the best model
The F-test result of model 2 indicates model 2 is the best model since its p-value is the smallest F(2,173)=14.04, p<0.01.
To find the beta-weights associated with each predictor of model 2, we transformed the raw data scores into z-scores and conducted regression. The beta-weights are above. Both predictors are significant; interview β=.27, p=.000253, and conscientiousness β=.21, p=.005671.
R squared for model 2= 0.1397 and R squared for model 1=0.1006, so R squared change = 0.03902. This demonstrates both the validity of model 2 and its incremental validity.
Interview is the best predictor because it has the lowest p-value p=.000253.
Overall, the organization should not use all 3 assessment tools for their selection system. Only the interview and conscientiousness measure held significance during hierarchical linear regression, so the work sample should be dropped.
Page updated
Google Sites
Report abuse