Besides developing the rankings which were presented in the previous sections, a correlation analysis has also been conducted in order to examine the relationship strength between the different constructs; and more importantly, to examine the correlation between the constructs and supplier performance. Due to the size of the correlation matrix, it has been enclosed in Appendix F. As it can be seen in the matrix, many of the results are in line with what scientific studies and theories have suggested, especially when looking at correlations between the micro-level constructs. For instance, ’cost’ has a significant negative relationship with ’innovativeness’, while ’innovativeness’ has a significant positive relationship with ’flexibility and agility’, and ’relationship and partnership potential’. Other findings, which can be argued to be quite predictable are the significant relationships between ’CSR’ and ’sustainability’. The reason why the relationship between ’CSR’ and ’sustainability’ was predictable is due to the fact that two aspects often are discussed together; this is, for instance, also the case in Thornton et al. (2013) which is the study from where most of the CSR factors for this thesis have been retrieved.
It is also interesting to observe how the micro-level constructs are related to the macro-level constructs. For instance, the positive significant correlations between ’relationship and partnership potential’ and ’accessibility’, ’economic’, and ’markets’ might indicate that buying firms are most comfortable with partnering with suppliers which are both easily accessible location-wise (and with easy access to tier to suppliers and raw materials), and that the suppliers should be in countries which are economically stable, while the proximity to other markets (i.e., growth markets with future business potential) are not too distant. Beside examining the correlations between the different type of constructs, it is also interesting to examine the correlation between the various types of constructs and the performance of the selected suppliers (i.e., performance as an overall measure based on the satisfaction of the buyer regarding the latest major outsourcing decisions). Three micro- level constructs have a significant positive relationship with the supplier performance, while seven macro-level constructs have a significant positive relationship with supplier performance. The fact that so many macro-level constructs are significantly positively correlated with performance indicate that such constructs are even more important than what the mean rankings in the previous sections show.
A multiple linear regression analysis was also conducted to model the impact that each construct has on the performance. The coefficients table from the multiple linear regression analysis is illustrated in Figure 8.1 below, while further details are added in an extended table in Appendix F together with the correlation matrix.
Figure 8.1. Visualisation of the coefficients table from SPSS
Due to the small sample size (N=42), the regression analysis will only focus on the constructs which were found to be of most importance for practitioners. For this reason, the analysis was narrowed down, thereby taking the five most important micro-level constructs and the five most important macro-level constructs from the overall ranking developed in Section 8.1 into consideration. The reason for this selection is that the factors from those exact constructs will be used for the practical part of the thesis (See Part III), as the author and A. T. Kearney agreed to include factors from the five most important constructs from each domain into the software. This is further explained in Chapter 10. The reason why a regression analysis has not been conducted for each type outsourcing motivation (i.e., cost and innovation as the two motives) is due to the low sample of respondents who scored the constructs based on an innovation-seeking outsourcing engagement (i.e., n=10). If a larger sample size was present, the regression analysis would naturally be conducted for each type of outsourcing motivation. Although it is not used in the multiple linear regression analysis, table that includes the ranking for the two types of outsourcing engagements (See Table 8.3) is still kept in the report due to the interesting results that it portrays.
The results of the multiple regression analysis show some interesting results. First of all, in the model summary it can be seen that R Square equals 0.613, meaning that the independent variables (i.e., the selected constructs) explain 61,6% of the total variation in the dependent variable (i.e., the supplier performance). This can be considered a very good measure. When looking at the coefficient table, it can be seen that ’cost’ has a significant impact on the performance (0.002 < 0.05), but with a Beta value of -.633. This indicates that when the ’cost’ construct (and the factors within the construct) is considered important by the practitioners included in the study, then the performance of the appurtenant selected supplier was considered low, which is why the impact of ’cost’ on performance is negative. This is in line with previous research, which for instance argue that when ’cost’ related factors are used in isolation from other types of factors in
CHAPTER 8. EMPIRICAL RESULTS supplier selection activities, then it might result in low supplier performance due to the short-term focus of a pure cost perspective. This is also insinuated in Chapter 3, where it has been presented how outsourcing practices have moved from being very transaction cost focused to becoming more core competence, agility, and partnership focused, as cost cannot be used as a single measure any longer. ’Cost’ is, of course, an important construct to consider in supplier selection, but the regression analysis shows that it should not be used independently without the inclusion of other constructs. ’Quality’ and ’time and delivery’ also both have a significant impact on performance (.049 < .05 and .011 < .05), but with a positive Beta value. Considering that industries in general are becoming more dynamic and competitive, it can be argued that those two results are quite inherent. Looking at the macro-level constructs, only ’governmental regulations’ has a significant influence on performance (.015 < .05). The fact that this is the only macro-level construct which has a significant positive impact on performance can be considered quite surprising, as ’labour’ and ’economic’, for instance, both had a significant correlation with performance in the correlation matrix. What is even more surprising is that the standardised coefficients of ’accessibility’, ’business climate’, and ’economic’ become negative, although they all have a strong positive relationship with performance in the correlation analysis. The disordered patterns of beta values might be a result of multicollinearity taking place. Multicollinearity is commonly measured via collinearity diagnostics based on tolerance levels and the Variance Inflation Factor (VIF) - the reciprocal of the tolerance value. Menard (1995) state that a tolerance value below 0.20 is a cause for concern, while Neter et al. (1989) and Hair et al. (2006) articulate that the VIF value should not exceed 10. Craney and Surles (2002) suggest that the cut-off value should be as low as 5 for the VIF measure, which is why this threshold is used in this report when measuring the multicollinearity. The results of the multicollinearity analysis, which is included in the extended coefficient table for the regression analysis in Appendix F, shows that all the tolerance values are above 0.20, and that none of the VIF values exceed the threshold of 5. In fact, only ’time and delivery’ exceeds the VIF value of 3, which is why it can be argued that the overall regression model is not affected by significant multicollinearity. Also, since the alphas are very insignificant in the regression when it comes to the macro-level constructs which have a negative Beta, the surprising disordered patterns in the Beta results cannot be generalised. As a function of the multicollinearity results, none of the constructs will be removed from the analysis. Nevertheless, since some of the macro-level constructs have a strong - or even significant - positive correlation with performance in the correlation matrix, and since those same constructs (except for ’governmental regulations’ and ’labour’) have negative Beta values, it is strongly encourage that this research is duplicated with a larger sample size, in order to identify if the disordered patterns persists.
It has to be noticed that the Alpha values presented above are based on a two-sided regression analysis. Since only positive results are expected due to the performance measures, a table with the results of a one-sided regression analysis would conventionally be included; however, since SPSS do not have this function in the latest version, the two- sided table has been included. Looking at the one-sided level of significance can be done by using the t-values values presented in the coefficients table, and then manually look up the level of significance for a one-sided regression in a t-table. After having done so
for all the t-values presented in the coefficients table in Table 8.1 it was identified that the level of significance do not change for any of the constructs, which is why the values presented in the coefficients of the two-sided regression are used in the former paragraph. That is, none of the none-significant Alphas become significant by changing the view from a two-sided to a one-sided regression.