Accounting enforcement in a national context: an international study

Abstract

The purpose of this study is to investigate whether the national characteristics of culture, religion and political factionalization are associated with the strength of accounting enforcement. The study uses data on percentages of religious adherents in a sample nation, the Hofstede cultural dimensions and political factionalization. National legal code (e.g., Common Law or Civic Code) and market liquidity are controlled for. Factor analysis is used to generate factor scores from the data. The dependent variable, accounting enforcement, is drawn from Brown et al. (J Bus Finance Account 41(1/2):1–52, 2014). The findings demonstrate that this set of national characteristics is strongly associated with national accounting regulatory enforcement. The implications of this research are that national characteristics should be taken into account in considering the impact of accounting standards on accounting comparability across nations. The limitation of this study is that, like much international research, the sample size is limited, here to 42 nations. The authors collectively have many years of research examining/studying domestic and international regulation, its determinants and consequences. This study importantly extends previous research on the determinants and consequences of regulation in the auditing and accounting arenas. This study provides an important contribution to the literature by helping establish that national characteristics do affect accounting enforcement efforts cross-nationally. This helps researchers and regulators better understand whether international standards can provide the link in comparability across nations that proponents are seeking. It does so by focusing on the variation in enforcement across nations rather than on the standards themselves.

Appendix A

This appendix describes how we addressed the methodological issues presented by having a limited data set—although a respectably sized one for international research using nation states as the observation units, and highly correlated variables (see Ali and Hwang 2000). The number of variables in our data set, even in the absence of multicollinearity, may result in overfitting of the model. Babyak (2004, p. 411) defines overfitting as “asking too much from the data.” He goes on to say that, “Given a certain number of observations in a data set, there is an upper limit to the complexity of the model that can be derived with any acceptable degree of uncertainty. Complexity arises as a function of the number of degrees of freedom expended (the number of predictors including complex terms such as interactions and nonlinear terms) against the same data set during any stage of the data analysis.” The problem with this, he notes, is that “findings” that appear in an overfitted model don’t really exist in the population and hence will not replicate.” Given, as Babyak (2004) notes, that if “you put enough predictors in a model you will get a result that is impressive, but lacks substance,” we chose to use a data reduction routine. Absent that routine, our initial results showed an adjusted r² of .40, with only the constant term in the model showing significance. Given that we only have 42 cases with 12 independent variables, our output statistics might be suspect.

Draper and Smith (1998; see also Babyak 2004) state that there should be 10 observations for each independent variable, something that may also be true in principal components analysis.¹⁷ In this regression, that is not possible since we only have a complete set of data on 42 nations, thus we only have 3.5 observations per independent variable. In addition, an inspection of the correlation matrix (see Table 4) reveals a pattern of high correlations among study variables. While a formal multicollinearity statistic might not demonstrate a problem, such high correlations are a concern since some regression coefficients might have their significance levels altered due to the high correlations. Accordingly, we employed data reduction techniques to explore whether the variables reflected underlying latent variables that in themselves are meaningful in terms of our research objectives. Our use of data reduction techniques is consistent with Ali and Hwang (2000) who also employed it in their study of government versus private standard setting in 16 countries. Ali and Hwang (2000) noted that variables within a country tend to be very highly correlated, rendering the use of individual variables in a regression problematic. Preliminary analyses of our data indicate that Ali and Hwang’s (2000) concerns hold true here as well.

Di Stefano et al. (2009, p. 1) note that exploratory factor analysis may be used “for a variety of purposes such as reducing a large number of items to a smaller number of components, uncovering latent dimensions underlying a data set, or examining which items have the strongest association with a given factor.” Rummel (1970) sees several uses for factor analysis. These include (a) interdependency and pattern identification; (b) parsimony or data reduction; (c) uncovering the basic structure of a domain; (d) scaling; (d) data transformation; (e) mapping and (f) hypothesis testing. We clearly are interested in data reduction, as noted above. Given the high intercorrelation of many of our variables, it is also important to use factor analysis to uncover the underlying structure of the domain we explore. Uncovering the basic structure of the domain(s) involved is also important to provide greater insight into the environments that give rise to greater or lesser regulation. Thus, the use of factor analysis helps generate rich insight into, and resources to derive from, the varied environments within which regulation may occur. Further, given that the process of generating the factor structure also gives rise to the ability to generate factor loadings, the output of the factor analysis enables us to see how different countries load on the different factors found if we wish. We also can, and will, use the resultant factor scores to test the hypotheses developed earlier in the paper. While the variables do not exist in isolation, as presented in the hypotheses, the loadings of each variable on the factors enable us to see which factor the variables are most highly correlated with. The different facets that describe a nation are more like syndromes than individual symptoms and the value of exploratory factor analysis to the researcher is that he/she can explore how collections of symptoms (a.k.a., syndromes) affect the research questions of interest. In a way, then, the use of factor scores provides a more realistic look at the forces that affect regulatory effort than does the use of variables in a regression alone.

The data reduction technique that we used was Factor Analysis with a principal components extraction. The scree plot indicated that we had three factors or components. Using all the variables listed, we found that we had a Kaiser–Meyer–Olkin (KMO) Measure of Sampling Adequacy of .49. We dropped the CHRST_OTH variable from the factor analysis and found that the KMO measure increased to .78. This measure varies between 0 and 1, and values closer to 1 are better. We chose CHRST_OTH to drop first because our earlier attempts to conduct a regression analysis of the data showed that, of the religion variables, its presence in the regression equation alone triggered a VIF score much greater than 10, a typically cited threshold for multicollinearity (Kennedy 2000). A KMO measure of .6 is considered the minimal threshold (see http://statistics.ats.ucla.edu/stat/spss/output/principal_components.htm); therefore, the results of the factor analysis that included the other variables but not CHRST_OTH were deemed acceptable. The three extracted components accounted for 66.5% of total variance. The first component accounted for 37.1% of total variance, the second component accounted for an additional 17.1% of total variance, and the third component accounted for 12.3% of total variance. In addition, Bartlett’s test of sphericity rejects the null hypothesis that the correlation matrix is an identity matrix (p < .01). Rejection of Bartlett’s test of sphericity is important in accepting the results of the factor analysis threshold (see http://statistics.ats.ucla.edu/stat/spss/output/principal_components.htm). These results indicate the validity of the three factors, and related factor scores, generated.