Participation in Corporate University Training: Its Effect on Individual Job Performance

Canadian Journal of Administrative Sciences, Dec 2004 by Morin, Lucie, Renaud, Stéphane

Measures

Participation in corporate university training. Our independent variable, participation in corporate university training, was measured using the total number of courses completed by an employee in the corporate university program from January 1, 1996, to December 31, 1998. The data set provided by the bank did not contain any information about when the courses were taken. Thus, it was not possible for us to know the cumulative number of courses completed by each employee at the end of 1996 or 1997.

Post-training job performance. To measure employees' job performance after training, we used the 1998 supervisory ratings of job performance from the data set. Each rating was recorded on a l-to-5 scale with 1 indicating that an employee had not achieved his/her 1998 objectives and 5 indicating that the employee had surpassed all the 1998 objectives (2: partial achievement of the 1998 objectives; 3: achievement of the 1998 objeclives; 4: employee had surpassed some of the 1998 objectives). Because the data set did not provide us with details about when courses were completed, as mentioned above, we chose the 1998 supervisor scores to ensure a post-training measurement of performance. Given this, it was impossible to measure performance after a three-month to one-year time lag, as is common practice in empirical research on training.

Control variables. Our statistical analyses included many control variables. Age was expressed in years, level of education was measured using a five-point ordinal scale, tenure within the firm was expressed in years, scheduled weekly work was measured in hours, and hourly wage rate was expressed in dollars. For job category, we used four dummy variables capturing the five broad job categories at the bank (junior auxiliary, senior auxiliary, junior, intermediate, and senior manager). Pretraining job performance was also explicitly controlled for in this study. It was measured by 1996 supervisory ratings. However, for some participants who might, for instance, have completed one course in 1996, this measure would not be a pure measurement of pre-training job performance. All control variables were included in the model because the literature on performance and human resources management indicated that they might have an influence on job performance (Renaud, St-Onge, & Magnan, 2000, 2004), hence confounding any evidence about the impact of corporate university training. For instance, in the training domain, results have shown that: (a) age is negatively associated with motivation to learn (Colquitt et al., 2000); (b) education level is negatively associated with training offered by the employer (Veum, 1993); (c) tenure is negatively associated with participation in training (Shearer & Steger, 1975); and (d) employees in a management position receive more training than employees in a non-management position (Saari, Johnson, McLaughlin, & Zimmerle, 1988).

Results

Results from a descriptive analysis indicated that 317 employees out of 1,484 completed at least one course in the corporate university program between January 1996 and the end of 1998. On average, these employees completed almost 4 courses (M = 3.68, SD = 2.31) with the number of courses ranging from 1 to 9. For our dependant variable, results showed that on average both employees who participated in corporate university training (M = 3.37, SD = .546) and employees who did not participate (M = 3.31, SD - .549) modestly surpassed their 1998 job performance objectives with individual scores ranging from 2 to 5. Among all the performance scores, 96% had either the value 3 or 4, indieating a restricted variance in the dependant variable. OLS coefficients obtained from such a restricted variance are less efficient then those resulting from a normally distributed variable. Table 2 provides a correlation matrix of all variables. The correlations among the control variables are generally low, suggesting that multicollinearity is not likely to impact the results.