4. Six Sigma Projects as Avenues of Knowledge Creation
4.7.2. Regression estimation and results
Prior to estimating multiple regression equations to test our hypotheses we assessed the kurtosis and skewness of the six independent variables – two control
variable and four knowledge creation variables, and one dependent variable – Six Sigma project performance. Two variables - team size and Black Belt experience - had
significantly high kurtosis (5.15 and 10.90) and skweness (3.29 and 1.85); we therefore used log transformations for both variables in subsequent regression analyses. The absolute value of skewness and kurtosis for all other variables is less than 0.60 and 0.72 respectively.
4.7.2.1 Hypotheses 1 and 2: We proposed that the four knowledge creation mechanisms would have a direct and positive impact on Six Sigma project performance,
as outlined in hypotheses 1 and 2. For the first hypothesis to be supported, all four knowledge creation mechanisms must explain significant variance in Six Sigma project performance. The second hypothesis would be supported if there is significant
incremental variance in Six Sigma project performance being explained by socialization (tacitặtacit) and externalization (tacitặexplicit) i.e. knowledge creation mechanisms utilizing tacit knowledge, after accounting for variance explained by combination
(explicitặexplicit) and internalization (explicitặtacit) i.e. mechanisms utilizing explicit knowledge.
The first two hypotheses are tested through a common hierarchical regression analysis (Cohen et al., 2003) by entering two knowledge creation variables, one pair at a time, in two steps – (1) combination (explicitặexplicit) and internalization
(explicitặtacit), followed by (2) socialization (tacitặtacit) and externalization (tacitặexplicit). These two steps are referred to as steps 2 and 3 in Table 4.12.
Variation inflation factor (VIF) scores are computed for all coefficients to assess multicollinerity; scores ≥5 are considered unacceptable (Hair et al., 1998).
The R2 and the F statistic for the complete regression model in the third step and the coefficients for the independent variables are the metrics of interest to test the first hypothesis. The change in R2 and the F statistic for the change between step 2 and 3 provide tests for the second hypothesis. Table 4.12 shows the results obtained from the hierarchical regression analysis. Predictors entered into the regression equation in the first step – controls for team size and Black Belt experience – do not explain a significant amount of variance in the dependent variable – Six Sigma project performance (b = 0.06,
ns, and b = - 0.12, ns for log-team size and log-BB experience respectively). The
addition of the two knowledge creation variables in the second step results in a significant amount of additional variance explained (change in R2 =0.18, F for the step = 9.31, p ≤.
001). The F statistic for the equation is significant and the R2 value is 20% indicating variance in the dependent variable explained by the four independent variables (Table 4.12: column 2, F = 5.18, p ≤ 0.01). Of the two independent variables of interest, internalization (explicitặtacit) is positively and significantly associated with Six Sigma project performance (b = 0.29, p ≤ 0.05) while combination (explicitặ explicit) is not, although its effect is in the right direction (b = 0.20, p = 0.11).
In the third step of the equation (Table 4.12: column 3), the addition of
socialization (tacitặtacit) and externalization (tacitặexplicit) as independent variables results in significant additional variance explained in Six Sigma project performance (change in R2 =0.04, F for the step = 2.39, p ≤. 10). One of the knowledge creation mechanism variables added in the third step has a significant coefficient (socialization: b
= 0.23, p ≤ 0.05), while the other does not have a significant impact (externalization: b = - 0.16, ns) and the effect is in the opposite direction. The coefficients for combination and internalization do not change substantially between the second and the third steps.
The overall R2 is 24% (F = 4.40, p ≤ 0.01) and adjusted for number of parameters estimated, is 19%. The highest VIF score among the six predictors is 1.75, substantially lower than the recommended cutoff of 5, indicating that multicollinearity is not a problem.
Thus, from the results shown in column 3 of Table 4.12, hypothesis one is partially supported. A significant amount of variance in Six Sigma project performance is explained by the four knowledge creation variables and higher values of socialization and internalization are associated with higher project performance. Hypothesis two, regarding the incremental effect of tacit knowledge utilizing knowledge creation mechanisms – socialization (tacitặtacit) and externalization (tacitặexplicit) – is also partially supported with significant amount of incremental variance explained by the addition of the two variables to the regression equation. The coefficient for socialization (tacitặtacit) is significant as expected; however, the coefficient for internalization (tacitặexplicit) is not significant and has a negative sign.
4.7.2.2 Hypotheses 3 and 4: Our third and fourth hypotheses propose that the impact of knowledge creation mechanisms on Six Sigma project performance is contingent upon the levels of contextual variables. In order to test these hypotheses of
‘fit as moderation’ (Venkatraman, 1989) we needed to compute multiplicative interaction terms of the ‘causal’ knowledge creation variables and the ‘moderator’ contextual
variables. Computing such interaction terms we estimated two different regressions equations, one for each contextual variable – standardized process and related processes.
Before computing multiplicative interaction terms the scores for knowledge creation and the two contextual variables were centered by subtracting the means to ameliorate multicollinearity (Aiken and West, 1991; Irwin and McClelland, 2001). The resulting deviation scores were then used as independent variables to assess the main effects of the four knowledge creation mechanisms as well as to compute the
multiplicative interaction terms. For testing hypothesis three, two multiplicative interaction terms were computed as cross-products of (1) standardized process * combination (explicitặexplicit) and (2) standardized process * internalization (explicitặtacit). Similarly, for hypothesis four, scores for (1) related processes * socialization (tacitặtacit) and (2) related processes and externalization (tacitặexplicit) were computed.
Table 4.13 shows the results of the two hierarchical regression equations estimated for assessing the effects of the two sets of interactions. In each of the two equations the addition of the interaction terms as predictors in the third step does not result in a significant amount of incremental variance being explained (Table 4.13, Column 3, F for the step = 0.98, ns; Column 6, F for the step = 0.28, ns). Thus, our third and fourth hypotheses regarding moderating effects of standardized process and related processes are not supported.