Chapter 5: Developing a typology of business relationship strategies relationship strategies
5.5. Phase2: Quantitative validation of the 2indings
5.5.1. Interaction feedback from interviewees
5.5.2.3. Analysis and results
Three sets of analysis are reported in this section. First, the two samples (i.e. the data from the full time MBAs and the data from the international executive MBAs) are examined for any possible differences based on three control variables. Next, a one-way repeated analysis of variance (ANOVA) and a multinomial logistic regression are performed to test the validity of the findings.
Details of these analyses are described below.
5.5.2.3.1. Independent samples test
Since the data were collected from two sources (full time MBAs and international executive MBAs), a number of t-test analyses were carried out to compare the two groups based on two control variables (i.e. the number of employees and the number of years for which their companies had been active in the market). In these tests, the signi<icance level was set at .05. Since the number of responses for each group is considerably larger than the minimally accepted number of 30, it is assumed that the distributions are normal (Huizingh 2007). Thus, the t-test can be performed. In addition, Levene’s test (Field 2005) for the equality of variances indicates insignificant results, suggesting that the variances are equal.
This is done to choose the right formula for performing the t-test. For both control variables, the t-test reported insigni<icant values of 1.469 and .838, with 309 and 306 degrees of freedom, respectively. These results indicate that these two samples have no differences in terms of the two controlled variables. As such, it can be inferred that there are no significant differences between these groups and, hence, it can be assumed that the rest of the analysis can be conducted on the basis of the combined dataset, which comprises 311 responses in total.
5.5.2.3.2. ANOVA
Following this, a one-way repeated ANOVA was conducted to compare the scores on the five single-item resource acquisition strategy type Likert scales, in order to test for significant differences amongst the five resource acquisition strategy types. This technique is widely used to compare responses to two or more different questions (Pallant 2007).
Using this technique permits a comparison of the scores on the five resource acquisition strategies and hence it helps to identify if there is a significant difference somewhere among the sets of strategies. The means and standard deviations are presented in Table 5.6.
Table 5.6: Descriptive statistics for the five single-item resource acquisition strategy type Likert scales
RAS type N Mean Standard
Deviation
Money bonds 311 4.97 1.628
New market bonds 311 5.64 1.365 Utilisation bonds 311 5.53 1.322 Intellectual bonds 311 5.06 1.562 Credibility bonds 311 5.87 1.303
The results of the multivariate tests based on the one-way repeated ANOVA reveal a significant difference among these five resource acquisition strategy types (Wilks’ Lambda= .658, F(4, 307)= 39.953, p< .0005; with Partial Eta
Squared= .342). Using the guidelines proposed by Cohen (1988, p. 284-287) (.01= small, .06= moderate, and .14= large), these results indicate a very large effect size. In the next step, a pair-wise comparison among these resource acquisition strategy types was conducted (Table 5.7). These comparisons were based on estimated marginal means. In addition, the adjustment for multiple comparisons was based on the Bonferroni method of correction (Harris 2001).
The pair-wise comparison indicates that there is no significant difference between the ‘money bonds’ strategy and the ‘intellectual bonds’ strategy, or between the ‘new market bonds’ and ‘utilisation bonds’ strategies. These findings provide initial evidence of existing hybrid strategies.
Table 5.7: Pair-wise comparison of resource acquisition strategy types
RAS
Intellectual bonds -.090 .121 1.000 Money
5.5.2.3.3. Multinomial logistic regression
In the next step, a multinomial logistic regression was performed to assess how well the five single-item Likert scales predict the resource acquisition strategy types obtained from the self-typing method. This test would give an indication of the adequacy of the model through assessing the goodness of fit. The model contained five independent variables (money bonds, new market bonds, utilisation bonds, intellectual bonds and credibility bonds) and one categorical dependent variable (RAS type based on self-reported responses). Table 5.8 shows the frequencies of each resource acquisition strategy type, based on the self-typing measure.
Table 5.8: Frequency of RAS based on self-report measure
RAS type Frequency Percentage
Money bonds 67 22%
New market bonds 93 30%
Utilisation bonds 47 15%
Intellectual bonds 20 6%
Credibility bonds 84 27%
The results of the multinomial logistic regression indicate that the final model, described below, is statistically significant, χ2= 138.686 with 20 degrees of freedom, p< .001, which indicates that it was able to signi<icantly distinguish the correct type of resource acquisition strategy based on the five single-item Likert scales. Pseudo R-squared results indicate that the final model explained between 36% (Cox and Snell R2) and 37.8% (Nagelkerke R2) of the variance in resource acquisition strategy types.
In this model the money bonds strategy is treated as the reference variable.
This was done for reasons of simplicity, given that it was the first category in the pertinent question (see question number 5 in Apendix C and question number 71 in Appendix D). Therefore SPSS estimated four models: (1) new
market bonds strategy relative to the money bonds strategy; (2) utilisation bonds strategy relative to money bonds strategy; (3) intellectual bonds strategy relative to money bonds strategy; and (4) credibility bonds strategy relative to money bonds strategy. For each of these given models, it is expected that only the corresponding single-item Likert scale will significantly predict the model.
As shown in Table 5.9, only the related independent variable made a statistically significant contribution to the dependent variable, resource acquisition strategy type.
Table 5.9: Multinomial logistic regression results
RAS type Β Standard a. The reference category: Money bond
These results show that for Model 1 (new market bonds relative to money bonds), the Wald test statistic for the predictor ‘new market’ is 14.097 with β=
.763, for Model 2, the Wald test statistic for the predictor ‘utilisation’ is 8.488 with β= .685, for Model 3, for the predictor ‘intellectual’ it is 14.824 with β=
1.421, and for Model 4, for the predictor ‘credibility’, it is 17.885 with β= .843, all with an associated p-value of less than .05. This indicates that the five single-item Likert scales can significantly predict the pertinent type of resource acquisition strategy reported through the self-type paragraph approach. In other words, for every chosen resource acquisition strategy type, respondents scored the pertinent single-item Likert scale relatively higher. As such, this implies that each company in this study has a dominant resource acquisition strategy type.