• No results found

4.4. Operational definitions and measurement

4.4.2. Strategies and tactics

4.4.2.2. Exploratory factor analysis

As the instrument has not been used extensively in accounting settings and as it has to my knowledge not been used in a Norwegian setting, I use factor analysis to validate the strategy-measures. Following Hair et al. (1998, p. 98), factor analysis is applicable if the sample size is larger than 50 (preferably the sample size should be larger than 100) and the number of observations per variable to be analyzed is larger than 5 (preferably larger). According to Field (2009), the variables included should be approximately normally distributed but this

assumption is mainly important if the factor analysis is conducted to generalize findings beyond the sample collected.

Dimensionality of constructs

Two factor analyses are first run on the items that intend to measure each different negotiation strategy (79 questionnaires included) to investigate the dimensionality of the two scales.

Contending scale items

The 5 items of the contending scale are subjected to principal components analysis (PCA). As the correlation matrix reveals that most of the coefficients are 0.3 and above, the KMO-value is 0.652 and Bartlett’s Test of Sphericity is statistically significant, the data are suitable for factor analysis.

The factor analysis suggests a one-factor solution (based on eigenvalues larger than 1) with an explanation of 53.4% of the total variance. One factor is further supported by the underlying theory.

The factor loadings and the communalities are given in Table 7 below.

Table 7 - Component analysis for PCA of contending scale items

Item Component 1 Communalities

s6 (Use influence as auditor) .899 .808

s9 (Use expertise as in accounting) .741 .549

s4 (Argue with client) .710 .504

s17 (Stay firm) .652 .425

s3 (Threaten to qualify) .618 .381

Conceding scale items

The 5 items of the conceding scale are then subjected to principal components analysis (PCA). As the correlation matrix reveals that many of the coefficients are 0.3 and above, the KMO-value is 0.695 and Bartlett’s Test of Sphericity is statistically significant, the data are suitable for factor analysis.

The factor analysis suggests a two- factor solution (based on eigenvalues larger than 1) with the factors explaining 49.9% and 24.7% of the variance respectively. The two-factor solution explains 74.6% of the total variance. The scree-plot, see Figure 6, indicates a possible break after the second factor.

Figure 6 - Scree plot conceding strategy items

To better interpret the two components, oblimin rotation was performed. The results of the rotation show that the items load strongly on the two components (see Table 8 below).

Table 8 - Structure Matrix for PCA with Oblimin Rotation of Two Factor Solution of Conceding Strategy Items

Item Structure coefficients Communalities

Component 1 Component 2

s1 (Satisfy client expectations) .893 .156 .802

s12 (Satisfy client needs) .887 .235 .787

s7 (Accommodate client wishes) .819 .242 .672

s18 (Gave in for wishes) .098 .870 .769

s13 (Made concessions) .349 .821 .699

A closer study of the items reveals that they are of two types; the first type consists of items that focus on trying to “please” the client if possible (item 1, 7 and 12) and these items are likely to be understood as “positive” by the auditor. The second type (item 13 and 18) consists of items that verbally focus on the compromising part of conceding behavior. These items are likely to be understood less positively by the auditors due to the independence requirements inherent in auditing. As items 1, 7 and 12 measure the extent to which the auditor tried to satisfy the client’s needs, expectations and wishes, there seems to be a good correspondence between what these items measure and the “concern for other” dimension in the Thomas’

model (1976), and items 13 and 18 are removed from the scale.38 It should be noted that in principle, these changes will impede the comparability of findings with previous research findings using these scale items. In practice this is a relatively small problem as only Gibbins et al. (2010) use these items to measure auditors’ choice of negotiation strategy.

38 Question 13 and 18 seem to be susceptible to a response bias, i.e. the respondents have responded using the answer that they think is ”correct” instead of answering how they really behaved in the negotiation. We see in Appendix 2 that few of the respondents “admit” to have behaved as indicated in statements 13 and 18 and the Spearman’s rank order correlation coefficient between statements 13 and 18 and the accounting outcome of the negotiation solution is very low (r=0.057 and r=-0.003, respectively).

It is arguable that leaving out items 13 and 18 moves the measurement from measuring what the auditor did to measuring the auditor’s intention (to what extent did the auditor want to win the negotiation) or goal commitment (the latter term is used in Kadous et al. (2003) as a measure of motivation and is measured with 5 statements on a five-point Likert scale). But as pointed out previously (see section 2.4.4), the two dimensions of the Thomas’ (1976) model also reflect the motivational orientation of the negotiator. Consequently we see this issue more as a question of terminology than of substance, and we will continue to use the term strategy in this thesis.

Factor analysis (8 items)

To make sure the remaining 8 items that measure the contending and the conceding scale items load as intended on the two factors so that these two scales can be used, the 8 items were subjected to PCA. As the correlation matrix reveals many coefficients of 0.3 and above, the KMO-value is 0.668 and Bartlett’s Test of Sphericity is statistically significant, the data are suitable for factor analysis.

The factor analysis suggests a two- factor solution (based on eigenvalues larger than 1) with the factors explaining 34.0% and 30.6% of the variance respectively. The two-factor solution explains 64.5% of the total variance. Inspection of the scree-plot, see Figure 7, indicates a clear break after the second factor. Two factors are further supported by the underlying theory.

Figure 7 - Scree plot contending and conceding items

To better interpret the two components, oblimin rotation was performed. The results of the rotation show that the items load strongly on the two components. One of the items (s4) loads on more than one factor with a larger loading than 0.3 on both factors, see Table 9 below. As this item loads the most on the factor it is supposed to load on and as it reflects an important facet of the contending scale, this item is kept in calculating the contending scale.

Table 9 - Pattern and Structure Matrix for PCA with Oblimin Rotation of Two Factor Solution of Negotiation Strategy Items

Item Structure coefficients Communalities

Component 1 Component 2

s6 (Use influence as auditor) .891 .102 .815

s9 (Use expertise as in accounting) .744 -.106 .558

s4 (Argue with client) .689 .400 .665

s17 (Stay firm) .667 -.272 .502

s3 (Threaten to qualify) .639 -.161 .425

s12 (Satisfy client needs) .023 .875 .769

s1 (Satisfy client expectations) -.129 .869 .762

s7 (Accommodate client wishes) -.126 .811 .665

Note. Major loadings for each item are given in bold

The interpretation of the two components is consistent with the theoretical foundations as the contending items load strongly on component 1 and the conceding items load strongly on component 2. The results further support the use of two separate scales as suggested by previous research.

Scale reliability:

The contending scale has good internal consistency with a Cronbach’s alpha of 0.764.

Removal of any of the items will not improve the scale. It follows from the factor analysis above that item 4 also loads on the conceding scale, but as Cronbach’s alpha for the

contending scale is slightly higher with this item included, item 4 will not be removed from the scale.

The conceding scale also has good internal consistency with a Cronbach’s alpha of 0.833. It will improve slightly to 0.843 with exclusion of item 7. As two items have already been removed from the conceding scale, item 7 will be kept when calculating the conceding scale.

To summarize: we have the following items that will be used when calculating the two scales:

Factor 1 (contend) = item no. 3, 4, 6, 9 and 17 Factor 2 (concede) = item no. 1, 7 and 12