Profesora: Raque Acero Cacho
3. Objetivos y alcance del proyecto
The data analysis was conducted following the approach of conducting item analysis to ensure consistency to identify the key factors that explain the variance, testing for reliability, and construct validity. These are explained in the data analysis chapter in greater detail. Factor analysis was conducted using satisfaction „before‟ scores on various items as the variables (see Appendix B for the list of variables and their descriptions). Once the constructs were identified / validated, the focus of the analysis shifted to comparison of the means between various outsourcing scenarios termed OS Status types in this narrative (i.e. CDA, IS, NUC, OS and UC).
In the comparison of the means, analysis was done at both the overall construct level and at the item / dimension level across OS Status types. Since each construct has multiple items, the average score on each comes close to the continuous distribution assumption needed to conduct analysis of variance. Pair wise comparisons (such as between IS and OS, which alone have both the „after‟ and „expected‟ scores), were done using the two sample t-test. Comparisons within each OS Status type (e.g. „expected ‟compared to „before‟ for within IS
alone) were done using paired t-tests since the independence assumption is not as strong as for other comparisons. (In effect, analysis of variance to compare OS Status type combinations extends the two-sample t-test for testing the equality of two population means to a more general null hypothesis of comparing the equality of more than two means, versus them not all being equal).
Normality was tested using Anderson-Darling‟s test and it was found that wherever the distribution was not normal, the difference in the median values using the Kruskal - Wallistest (a non-parametric test) was at p-levels higher than p-levels achieved for Tukey‟s family wise comparisons. In other words, Tukey (at 90 percent simultaneous confidence level) was the most conservative (comparisons were also done using Fisher‟s method at95 percent individual confidence level but Tukey was more conservative in rejecting the null hypotheses). The comparison between OS Status types as an aggregate of the comparisons at each item level, provides powerful evidence not only of the significant inequality of the means, but also, the pervasiveness of the inequality. After all, as data points increase, even small differences can appear significant; the aggregation approach of providing a score on what percentage of items rejected the null hypothesis, individually, provides a way of comparing support for a broad based difference in the means, across the various items within the construct for different OS types. The comparison of the means requires an assumption that the variance is equal or nearly equal and for this the Levine test was performed and the examples are provided in the next chapter.
A level of creativity was needed in managing with limitations in the number of responses. The 52 responses received from respondents representing firms that have outsourced (OS) and eight representing firms that have in-sourced (IS), together represented a very high response rate from among those that could be targeted as representative of these groups. This is explained in the next chapter under the heading, “Sample Description.” CDA and UC responses were somewhat lower than expected – one would have thought that with all the literature indicating the ushering in of the „rapid growth‟ phase, many more would have outsourcing under consideration or already considered but decided against. This may indicate that outsourcing is not as prevalent or popular as an agenda, when seen from the point of view of industry as a whole, not just the large firms (which have received most attention in available studies conducted by consulting firms).
In any case, with the number of responses received, and without reducing the number of variables or items which really capture the full breadth of the constructs within the study, it
was necessary to make do with the approach adopted in this research. Once factor analysis revealed the constructs based on satisfaction „before‟ scores („before scores had the maximum number of data points since all OS types could respond to them), other comparisons were built with that as the reference point: „expected‟ minus „before‟ brought into focus the responses from all OS types except NUC and „after‟ minus „expected‟ concentrated on just IS and OS.
Comparison of the means provides powerful information about the state of indirect procurement outsourcing. Differences in the predictor variables (i.e., constructs) help shed light on the characteristics that are witnessed for each of the OS types. This is discussed in Chapter 6.
4.6 Summary
This chapter described the data collection and study population. A total of roughly 500 respondents were picked for receiving the mail survey questionnaires from a set of members of a professional association known to have been more inclined to indirect procurement outsourcing. The sampling procedure and data sample are discussed and the contents of the survey questionnaire highlighted. The manner in which the data analysis is conducted (in the next chapter) is summarized.
CHAPTER 5
DATA ANALYSIS AND RESULTS
5.1 Introduction
In this chapter, analysis of data for indirect procurement outsourcing, collected through the survey instrument, will be documented. To begin with, the sample is described in terms of the characteristics of the respondents. This is followed by testing for non-response bias. Next, items, dimensions and constructs are validated using item analysis and factor analysis. Lastly, various combinations of means are compared to identify significant differences that go to prove or disprove the hypothesis laid out in the previous chapter.