Appendix A: Questionnaire Outsourcing Performance 2007
1. Which service provider is providing outsourced IT services for your organization?
2. Please indicate how satisfied you are with the performance of this service provider, using the scale below.
Very dissatisfied
Very Satisfied
3. Please indicate to what extent you are likely to recommend this service provider.
Certainly not Certainly
4. To what extent do you feel like you are involved in a high quality partnership with the service provider.
Certainly not Certainly
Regarding the section below, please indicate whether you a) totally agree, b) agree, c)
somewhat agree, d) somewhat disagree, e) disagree or f) totally disagree with the statements.
5. The account- and deliverymanagers (operational management) are capable of achieving the objectives and solve problems effectively.
6. The service provider generally meets the service levels as agreed upon in the Service Level Agreement (SLA).
7. The service provider is flexible in bringing changes to the contract.
8. The service provider will shoulder reasonable commercial risk and make necessary investments to reduce that risk.
9. The service provider actively identifies innovation opportunities
The following questions are meant to provide insight into the scope of the relationship with your service provider.
10. What is the age of the relationship with your current service provider?
• Less than 1 year
• 1 to 2 years
• 2 to 4 years
• 4 to 7 years
• Longer than 7 years
11. What are the total annual costs of the services provided by your service provider?
• Less than 0,5 million euro
• 0,5 to 1 million euro
• 1 to 2 million euro
• 2 to 5 million euro
• 5 to 10 million euro
• 10 to 25 million euro
• 25 to 50 million euro
• More than 50 million euro
12. What percentage of the total IT budget of your organization is currently provided by your service provider?
• Less than 2 % of our total IT-budget
• 2-10 % of our total IT-budget
• 11-25 % of our total IT-budget
• 26-50 % of our total IT-budget
• 51-75 % of our total IT-budget
• More than 75 % of our total IT-budget
13. What is the total IT-budget of your organization?
• Less than 5 million euro
• 5-10 million euro
• 10-25 million euro
• 25-50 million euro
• 50-100 million euro
• More than 100 million euro
Appendix B: Outsourcing in an IS context
In this section, the particularities of IS Outsourcing are presented, such as the associated benefits and the vendor selection process. In the outsourcing-related academic literature, particular emphasis has been given to IS outsourcing (Kakabadse, 2002). In 2008, the global market for IT Services was valued at 770 billion dollars, growing at an annual rate of 6.5%1. Relative to BPO outsourcing (HR, F&A and CRM), IS outsourcing is most popular, especially in Europe.
According to Grover, Cheon and Teng (1996), IS outsourcing success can be assessed in terms of benefits. They identify three types of benefits from IS outsourcing:
- Strategic benefits, which refer to the ability of a firm to focus on its core business activities and on the strategic uses of IT;
- Economic benefits, which refer to the ability to utilize expertise and economies of scale in human and technological resources of the service provider;
- Technological benefits, which refer to a) the ability to gain access to leading-edge IT and b) the avoidance of the risk of technological obsolescence, resulting from dynamic changes in IT.
Grover, Cheon and Teng (1996) argue these benefits should be weighed against the increased transactional costs, decrease in flexibility and conflicting objectives of the firm versus the service provider2. These considerations are reflected in a business case, which is the first step in the outsourcing process. Following the approval of the business case, a shortlist of potential service providers is developed. The service providers on the shortlist are selected on the basis of internal (strategy) as well as external (capabilities) requirements. To gain better inside into how service providers could fit these requirements, clients sometimes issue a Request for
Information (RFI)3. Next, the shortlisted service providers (typically between four and six) are
1 Data from Datamonitor.com
2 This weighing of benefits and costs is not in scope of this thesis, as it concerns the initial decision to outsource or not.
3 The RFI generally outlines the client company’s present position and objectives, the operations, services and functions it is seeking to outsource its staffing position and anything else of relevance. However, most future buyers think they already have a good sense of the market for IT services beforehand, so they do not issue an RFI (Michell and Fitzgerald,1997).
sent a Request for Proposal (RFP)4. The proposal responses are then evaluated using a set of formal criteria5.
Integrating the definitions of IS outsourcing and interorganizational relationships leads to a specific definition of IS outsourcing relationships: “an ongoing linkage between a service provider and buyer, arising from a contractual agreement to provide one or more
comprehensive IT activities, processes, or services with the understanding that the benefits attained by each firm are at least in part dependent on the other” (Goles and Chin, 2005).
Key IT outsourcing contractual issues include service levels, transfer of assets, staffing, pricing and payment, warranty and liability, dispute resolution mechanisms, termination, intellectual property matters, and information security (Lee, 1996).
Although similarities exist, relationships in IS outsourcing are fundamentally different from other type of relationships. Information technology is a unique type of asset; it is generally nor
strategic nor differentiating from competitors, but in many modern firms there are IT components which are considered to be business critical and IT in general is pervasive throughout the firm’s operations (Dibbern et al. 2004). It does not just serve one homogenous function, but rather it is interrelated with practically all organizational activities (Lacity and Wilcocks, 1996). Moreover, information technology is both complex and rapidly evolving in terms of a company’s IS needs and technology leaps (Lacity et al. 2003).
4 The RFP includes requests for a detailed technical solution and the related financial offering.
5However, the ultimate service provider selection often takes place on the basis of a set of factors that are much softer and more perceptual than the formal evaluation, such as the perception of cultural fit (Michell and Fitzgerald,1997).
Appendix C: Assumptions and Diagnostic Tests
The first assumption of the multiple regression model is that each random error has a probability distribution with zero mean. In relation to model 1, this assumption says the average value of SAT changes for each observation and is given by regression (1). Thus, the assumption implies the model is, on average, correct.
The second assumption is that each random error has a probability with variance σ2. This is an unknown parameter and it measures the uncertainty in the statistical model. With regard to model 1, it asserts some observations on SAT are not more likely to be further from the regression function than others.
The third assumption is that any two observations on the dependent variable are uncorrelated.
In other words, the size of an error for one observation has no bearing on the likely size of an error for another observation.
The fourth assumption is that the random errors have normal probability distributions.
Diagnostic tests are conducted to check whether problems commonly associated with the multiple regression model are present. The outcomes of these tests are summarized in the table below.
Test for… Test Statistic Value / Probability for Model 1
Value / Probability for Model 2 Heteroskedasticity White (F-statistic) 2.390 / 0.000 1.747 / 0.004
Autocorrelation Durbin-Watson 1.915 1.894
Functional form Ramsey RESET (F-test) 0.795 / 0.372 0.021 / 0.885 Normality of residuals Jarque-Bera 68.636 / 0.000 11.991 / 0.002 Table 1. Diagnostic Tests and results
The values of the Durbin-Watson statistic are used to test for autocorrelation. For model 1, the sample size is n = 874 and the number of regressors is K = 11, which corresponds to a critical values of dL = 1.864 and dU = 1.9111. For this model, d = 1.915, which is > dU, so we can conclude there is no autocorrelation (at a significance level of 0.05). For model 2, the critical values (n = 487, K = 5) are dL = 1.837 and dU = 1.869, while d = 1.894, so there is no evidence of autocorrelation in this model either.
The White test is used to detect heteroskedasticity, which occurs when variances for all observations are not the same, as is often encountered when using cross-sectional data.
Heteroskedasticity Test: White, Model 1
F-statistic 2.390047 Prob. F(39,810) 0.0000
Obs*R-squared 87.72036 Prob. Chi-Square(39) 0.0000 Scaled explained SS 144.6637 Prob. Chi-Square(39) 0.0000 Heteroskedasticity Test: White, Model 2
F-statistic 1.746911 Prob. F(39,447) 0.0044
Obs*R-squared 64.40921 Prob. Chi-Square(39) 0.0064 Scaled explained SS 68.82117 Prob. Chi-Square(39) 0.0022 Table 2
The value of the F-statistic is below the critical value of 0.05 for both models, meaning that the null hypothesis is rejected. The null hypothesis of the White test is ‘no heteroskedasticity’, so we conclude that there might be heteroskedasticity. This means a violation of the standard
assumption of constant variance in linear regression models. The variances of the coefficients tend to be underestimated, and might in some cases cause a variable to appear significant while it is not. However, despite the presence of heteroskedasticity, the least squares estimator is still linear and unbiased.
OLS provides consistent parameter estimates in the presence of heteroskedasticity, but the usual OLS standard errors are incorrect and should not be used for inference. White’s (1980) heteroskedasticity consistent covariance estimate does not change the point estimates of the parameters, only the estimated standard errors.
The White covariance matrix is given by:
It has become common practice in applied work to report standard errors from the White heteroskedasticity-consistent covariance matrix when one suspects that there might be heteroskedasticity (Cribari-Neto and S.G. Zarkos, 1999). The table below shows the White Heteroskedasticity-Consistent Standard Errors & Covariance.
Dependent Variable: SAT Method: Least Squares Sample: 1 874
Included observations: 847
White Heteroskedasticity-Consistent Standard Errors & Covariance
Coefficient Std. Error
IND -0.316985 0.215191
SPDOM -0.012471 0.205483
BDOM 0.398401 0.263390
SQUAL 0.675781 0.030554
PQUAL 0.283761 0.025964
CUL 0.023314 0.052985
Table 3
A common solution for decreasing the degree of heteroskedasticity is the transformation of the variables into logarithms. To facilitate this procedure, the value of the dummies was increased by 1, because it is impossible to take the logarithm of zero. The results of this transformation was that the heteroskedasticity did not disappear from model 1 (F-value = 10.66, p-
value=0.0000), nor from model 2 (F-value=3.35, p-value=0.0009).
Heteroskedasticity Test: White, Results after transformation of variables Model 1
F-statistic 10.66259 Prob. F(8,848) 0.0000
Obs*R-squared 78.32706 Prob. Chi-Square(8) 0.0000
Scaled explained SS 203.7432 Prob. Chi-Square(8) 0.0000
Table 4
Heteroskedasticity Test: White, Results after transformation of variables Model 2
F-statistic 3.354476 Prob. F(8,479) 0.0009
Obs*R-squared 25.88958 Prob. Chi-Square(8) 0.0011
Scaled explained SS 36.33309 Prob. Chi-Square(8) 0.0000
Table 5
The outcome indicates the presence of heteroskedasticity cannot be overcome by transforming the variables into logarithms.
The Ramsey RESET test is conducted to detect omitted variables and incorrect functional form.
Ramsey RESET Test, Model 1
F-statistic 0.795194 Prob. F(1,838) 0.3728
Log likelihood ratio 0.806199 Prob. Chi-Square(1) 0.3692 Ramsey RESET Test, Model 2
F-statistic 0.021110 Prob. F(1,475) 0.8845
Log likelihood ratio 0.021643 Prob. Chi-Square(1) 0.8830 Table 6
The test results indicate an acceptance of the null hypothesis at 5% significance level. Null hypothesis is ‘no model misspecification’, hence we conclude there is no evidence of model misspecification.
0 40 80 120 160 200
-2 -1 0 1 2 3 4
Series: Residuals Sample 1 874 Observations 850 Mean 1.74e-16 Median 0.088847 Maximum 3.985955 Minimum -2.528760 Std. Dev. 0.728774 Skewness 0.068510 Kurtosis 4.385346 Jarque-Bera 68.63599 Probability 0.000000
Figure 1
Model 1
0 5 10 15 20 25 30 35 40
-2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5
Series: Residuals Sample 1 534 Observations 487 Mean -1.11e-16 Median 0.055363 Maximum 1.618304 Minimum -2.039514 Std. Dev. 0.635570 Skewness -0.365655 Kurtosis 3.236908 Jarque-Bera 11.99113 Probability 0.002490
Figure 2
According to the Jarque-Bera statistics, the residuals in both models are not normally
distributed. The reason for this is that in model 1, the level of kutosis (‘peakedness’) is too high (above the critical value of 3), and for model 2, the level of skewness (‘symmetry’) is too low (lower than the critical value of 0). However, from observing the histogram figures, we can conclude there are no significant outliers.
Model 2
SAT Variable Distribution
0 50 100 150 200 250 300 350
1 2 3 4 5 6
Series: SAT Sample 1 874 Observations 873 Mean 4.089347 Median 4.000000 Maximum 6.000000 Minimum 1.000000 Std. Dev. 1.133837 Skewness -0.624326 Kurtosis 2.935468 Jarque-Bera 56.86492 Probability 0.000000
Figure 3
Mean and median values of the Kraljic Factor measures
a) VALBU b) VALRE c) VALUE d) CLI Mean 0.268282 0.007570 5.561613 32.20366 Median 0.166667 0.000425 3.000000 28.00000 Table 7
Appendix D: List of variables
Variable Name Type Data Source
Buyer Dominance BDOM Independent (Dummy)
Combined Kraljic Matrix (see Error! Reference source not found.
Service Provider Dominance
SPDOM Independent (Dummy)
Combined Kraljic Matrix (see Error! Reference source not found.
Independence IND Independent (Dummy)
Combined Kraljic Matrix (see Error! Reference source not found.
Interdependence INT Independent (Dummy)
Combined Kraljic Matrix (see Error! Reference source not found.
Outsourcing Success
SAT Dependent OP6, Question 2
Relational Norms RN Dependent (H3) OP, Question 4, 7, 8, 9 Partnership quality PQUAL Moderating (H2) OP, Question 4 Service Quality SQUAL Control OP, Question 6
Cultural Similarity CUL Control (Dummy) Service Provider has its HQ in the Buyer’s country or has at least 1000 employees based in the Buyer’s country Table 8
6 Outsourcing Performance 2007 Questionnaire
Appendix E: Cluster Analysis
The TwoStep Cluster Analysis (in SPSS 16.0) procedure is an exploratory tool designed to reveal natural groupings (or clusters) within a data set that would otherwise not be apparent.
This procedure allows for the creation of clusters on the basis of a categorical variable.
In order to handle categorical and continuous variables, the TwoStep Cluster Analysis procedure uses a likelihood distance measure which assumes that variables in the cluster model are independent. Further, each continuous variable is assumed to have a normal distribution and each categorical variable is assumed to have a multinomial distribution. In this research, the variables in the cluster model are not independent (since contract value is the common determinant of 3 of the variables). However, empirical testing has indicated that the procedure is fairly robust to violations of both the assumption of independence and the distributional assumptions (SPPS 16.0). The two steps are:
1) Pre-cluster the cases into many small sub-clusters;
2) Cluster the sub-clusters resulting from pre-cluster step into the desired number of clusters.
The clusters are created based on the ‘power distribution’ categorical variable. It takes on the values 1-4, each number representing a relationship type (1=‘Independence’, 2=‘Buyer
Dominance’, 3=‘Service Provider Dominance’ and 4=‘Interdependence’). The TwoStep Cluster Analysis procedure is used to group outsourcing relationships according to the power attributes.
Cluster Distribution (Two-Step Analysis)
N % of Combined % of Total
Cluster 1 165 18,9% 18,9%
2 28 3,2% 3,2%
3 179 20,5% 20,5%
4 92 10,5% 10,5%
5 410 46,9% 46,9%
Combined 874 100,0% 100,0%
Total 874 100,0%
Table 9
The cluster distribution table shows the frequency of each cluster. Of the 874 cases assigned to clusters, 165 were assigned to the first cluster, 28 to the second, 179 to the third, 92 to the fourth and 410 to the fifth.
Centroids
Value/Revenue Clients Served Value/Budget Contract Value
Mean
Std.
Deviation Mean Std.
Deviation Mean Std.
Deviation Mean Std.
Deviation Cluster 1 ,0098613 ,02047264 15,49 8,712 ,2611 ,25127 6,35 5,924
2 ,1141578 ,30066629 25,32 20,762 ,9004 ,64759 58,81 47,731 3 ,0087017 ,02483776 24,68 21,599 ,3457 ,29720 ,77 ,181 4 ,0001055 ,00008993 52,86 9,797 ,0539 ,04743 ,87 ,254 5 ,0005498 ,00086394 38,05 18,400 ,2423 ,24415 4,75 5,884 Combined ,0075701 ,05827016 32,20 20,452 ,2683 ,30036 5,56 13,845 Table 10
The Centroids table displays the mean and standard deviation for the cases in each cluster. The mean values of the operational measures of relative power distribution together characterize the relationships, based on the Kraljic Matrix. The outsourcing relationships in clusters 1 and 2 have a relative power distribution indicating ‘Interdependence’. Cluster 3 indicates ‘Buyer
Dominance’, cluster 4 ‘Independence’ and cluster 5 ‘Service Provider Dominance’. This is confirmed by the table below. Furthermore, it follows from the distribution of the data that cluster 5, which indicates ‘Service Provider Dominance’, represents roughly 50% of the relationships.
RELTYPE
1 2 3 4
Frequency Percent Frequency Percent Frequency Percent Frequency Percent
Cluster 1 0 ,0% 0 ,0% 0 ,0% 165 89,7%
2 0 ,0% 1 ,6% 8 1,9% 19 10,3%
3 0 ,0% 179 99,4% 0 ,0% 0 ,0%
4 92 100,0% 0 ,0% 0 ,0% 0 ,0%
5 0 ,0% 0 ,0% 410 98,1% 0 ,0%
Combined 92 100,0% 180 100,0% 418 100,0% 184 100,0%
Table 11
Using the TwoStep Cluster Analysis procedure, five out of the sixteen cells in the Kraljic matrix have been revealed. This means the data in the sample is not clustered such that all cells are well represented. However, each of the relationship types are represented by means of at least one cluster.
Figure 4. Identified clusters (coloured) within the Kraljic Matrix