• No results found

Figure 4.8: Significant Paths in the China Structural Model (only significant paths shown)

Statistic 15 Effect Size

H1c: Role Prescription dampens the effects of Institutional Collectivism. Role Prescription X Institutional Collectivism to Relevance USA .0657 1.1497* .025 Germany -.0172 .2159 -.009 China -.0021 .1204 -.009 H2c: Role Prescription moderates the effects of In- Group Collectivism Role Prescription X In-Group Collectivism to Relevance USA -.0117 .1331 .017 Germany .0563 1.2668* .000 China -.0322 .7170 -.008

In evaluating hypothesis H1c, I found that the path is positive and significant only in the USA. It is not significant in Germany or China. In the USA, the single effect of

Institutional Collectivism on Relevance was not significant and assumed to be zero. The single effect of Role Prescription on Relevance was positive. This indicates that the interaction between Institutional Collectivism and Role Prescription is one in which Institutional Collectivism weakens the effect of Role Prescription (Cohen, et al. 2003). The effect size is very low and is considered to be weak to non-existent so I conclude that this weakening effect is very small.

For hypothesis H2c, we found that the interaction between In-Group Collectivism and Role Prescription was significant only in Germany. However, in this case, the effect size is zero which indicates that Role Prescription has no moderating effect on the operation of In-Group Collectivism.

15* = p < .05 level, one –tailed test

16The effect size is calculated as per Henseler, et al. (2008).

Carte and Russell (2003) suggest guidelines to prevent errors in the assessment of moderation effects. I report the results of these guidelines to show that moderation has been reported correctly.

1) Report effect size – This was done in table 20. The effect size was weak to non- existent for the moderating variables.

2) Interpret main effects with moderating effects – This was done. I included the main effects variables with the moderating variables.

3) Check for curvilinear effects – I found that there was a very low correlation between Institutional Collectivism and Role Prescription in the USA (-.0112) and between In-Group Collectivism and Role Prescription in Germany (.1043). Since the correlations were low, the risk of confounding curvilinear effects with

mediation is low.

4) Clearly establish the causal ordering – This was done in the insights from whistle-blowing literature section.

5) Report power analysis and sample size requirements – Following Cohen, Cohen, West and Aiken (2003) we find for the USA Role Prescription and Institutional Collectivism interaction effect that L=f2(n-k-1) = .025(157-3-1) = 3.825 which

yields a power close to .3, a low level. The N required to achieve a power of .8 is n*=(10.8/.025+3+1) = 440. This indicates that there is a possibility that interaction effect might be understated in this study. Replication with a large sample size is needed. Similarly for the German In-Group Collectivism, Role Prescription interaction, we find that since the interaction effect size is zero that the power is zero and the number required to achieve a power of .8 is undefined.

6) Examine scale coarseness – In this study, respondents replied to Relevance, In- Group and Institutional Collectivism and Role Prescription on seven point scales creating a potential 49 point interaction variable compared with a seven point dependent score. Carte and Russell indicate that this exhibits a potential 50% reduction in effect size due to this feature.

7) Report all transformations – no transformations were performed on the data prior to analysis. In calculation of the moderator variables, the predictor variables were centered, a linear transformation which does not affect R2 or path coefficient

values.

8) Report scale reliabilities – Being formative constructs, In-Group and Institutional Collectivism don’t have reported scale reliabilities. Role Prescription was a single item variable, so its reliability is 1.0. Thus the reliability of the interaction cannot be reported.

9) When moderation is tested in PLS by separating samples in groups with different memberships, different factor loadings and weights are computed for each group for each latent variable. It is therefore necessary to test that the inter-item

covariance matrices with scales are equal using Box’s M to determine if the construct contents as determined by item loadings within scales are the same (Carte and Russell 2003). In this study, by using the two step approach

advocated by Henseler et al. (2008), this problem is avoided using the calculated unstandardized latent variable scores for each of the indicator and dependent variables in the analysis thus fixing the values so they are not recalculated. This not only avoids this problem, but also assures correspondence with the main

effects analysis done.

The end result of this analysis is that the sample sizes used in this analysis are too small which results in low power in testing for the moderation effect. Additionally, the

coarseness of the dependent variable results in a reduction of effect size as well. This indicates that to overcome this downward bias the interaction effect detected must be a strong one in order to be detected in such a low power experiment and we have a potential type II error on the moderation analyses.

4.8

DISCUSSION

This study replicated the study reported in Chapter 2 in the USA, Germany and China. The results of this study have shown that the pattern of association in the structural model demonstrated in previous studies in the USA is found within Germany and China. An exception to this is that in Germany, Role Prescription was not significant on

Relevance. While this might be due to German attitudes toward internal auditors, additional research is required to understand why this should occur. However, while the structural model is similar, this study shows that the effect sizes of Relevance of the report of bad news and Risk Perception are different in the western cultures from that of China. In China, the effect size of Relevance of the message was much smaller than that of Risk Perception, indicating that the Chinese subjects consider the report of bad news less important than their perception of the risk in the project.

The reason for this exchange in effect size is not evident from this study, however, I can speculate that it might have to do with the difference in political cultures. In the western societies with many individual freedoms, no history of government oppression in the subjects’ lifetimes and no fear of serious personal retaliation for defying authority, perceived risk may be held as less important than in China where the opposite conditions hold. In this study, since we did not achieve complete metric invariance between the subjects, we cannot make any comparison between the Risk Perception measures between the cultures. Given that, I am left to hypothesize that in the mental model of the Chinese students, they must be more risk averse or attach more

importance to perceived risk in their decision processes than western students as a result of their political environment.

This study also showed that the decision maker’s perceptions of ambient culture have little to no effect on the Deaf Effect. This is contrary to the deductions based on the definitions of the cultural values described in House, et. al. (2004). In the USA, power distance showed a negative effect on willingness to continue the current course of action. In Germany, perceived uncertainty avoidance in the culture caused a willingness to discontinue the current course of action. In China, perceived uncertainty avoidance reduced the Credibility of the bad news reporter while In-Group Collectivism reduced the Relevance of the report of project problems. The effect sizes of all of these effects were small to non-existent. It is possible that the generally low power for all of the paths of this experiment resulted in failure to achieve significance for these hypotheses. In which case, a larger sample size might generate significance for the cultural values.

Alternatively, there might be a substantive reason for the failure of these hypotheses. Per our model, I would argue that there are other aspects of the mental model that drive the reactions to perceived cultural values.

What these aspects are is uncertain at this point, however, I can speculate several reasons as to why this study achieved no significant different in response based on perceived national culture. First, it is possible that the subjects were composed of

several different cultural subgroups that resulted in a “cancelling out” of their differences. Following Archer (1988, 1995), I recognize that while a society may have a dominant culture, it is composed of several agencies (groups of actors sharing a similar

experience that have a similar viewpoint) which based on their positions in society can have different perspectives of the ambient society. Thus it may be that the members of the different agencies have divergent viewpoints of the culture and this has resulted in a “regression to the mean” when their responses are aggregated together resulting in a similarity of response across the different geographies. To resolve this issue would require separation of the different geographies into the different agencies, which would have differing responses. The instrument used here did not contain sufficient information on which to make such segregation.

An alternative explanation is that other aspects of the cultural environment are more important to the individual’s values makeup than the ambient national culture. One possibility here is that the professional culture of information systems is the same across the world resulting in a commonality of response. There is some evidence for this. Couger (1986) in a study of Singaporean and American programmers and analysts found that the two different cultural groups had substantial similarities in motivational characteristics despite the different geographies and significant demographic

differences. While the level of professional enculturation to be found in students at the sophomore level is likely to be low, another form of cultural homogeneity can perhaps be found in the common experiences as young people. Given the spread of western ideas via the internet and popular culture, it may be that college sophomores of the US and China may share a great deal in terms of values.

A third reason may have to do with nature of measures themselves. The GLOBE framework, while it appears superior to that of the Hofstede framework in terms of content validity, appears to need additional work on cross-country validation. The constructs they have identified are clearly formative as opposed to reflective as they have described them. The items, in this study, are beset by a systematic uniform item bias in different cultures leading to an inability to compare across countries. This creates a concern about the validity of the framework the validity of the framework and results reported when using the framework as described by the GLOBE consortium unless rigorous equivalence testing has been done to prove the equivalence of the metrics across countries. Similarly, the constructs used to measure the Deaf Effect also possess uniform item bias. It might be that the bias found or other deficiencies in the measures has resulted in a suppression of the cultural differences between the cultures.

4.9

CONCLUSIONS

I began this chapter with the question as to whether societal collectivism would change the causal linkages in the model of the Deaf Effect or whether in fact the Deaf Effect occurs. In a replication of the study in chapter 3 across the USA, Germany and China, I found that the respondents’ perspective of societal practices as operationalized by the GLOBE framework had very little effect on the occurrence of the Deaf Effect. The Deaf Effect model was shown to replicate in Germany and China much as it is in the United States. The implication of this is that the mental models associated with evaluating reports of bad news within projects are similar in the three geographies examined.