5.7 Multivariate results (regression)

This section presents the results of the regression analyses. The relation between the level of compliance (I.a, III.a, I.b and III.b) and several other determinants is tested by means of the linear regression method. The choice for the model is based on the results of the univariate and bivariate analyses above (avoiding endogenous variables) and tested for the levels of compliance I.a, III.a, I.b and III.b. The models to be estimated take the following forms:

Level of compliance = ß1 + ß2(AgePrinciple) + ß3(Self-regulation dummy) + ß4(Non-statutory norms dummy) + ß5(Statutory rules dummy) + ß6(Large caps dummy) + ß7(Mid caps dummy) + ß8(Main index dummy) + ß9(First to main index dummy) + Ɛ

The dependent variable consists of the level of compliance La (Applied)1, level of compliance III.a (Applied and explained sufficiently),2 level of compliance I.b (Applied for comparable code provisions of five countries)3 and level of compliance III.b (Applied and explained sufficiently for comparable code provisions of five countries)4 all expressed in percentages in the linear regression. The explanatory variable AgePrinciple involves the period of time the comply or explain principle has been applicable in a country (varying from 1 to 15 years). Dummies ß3 to ß5 are the judicial corporate governance arrangements (i.e. pure self-regulation, supported by non-statutory norms or facilitation by statutory rules or meta-regulation) or in other words the manner in which the comply or explain principle is embedded in the national corporate governance system. Explanatory variables ß6 and ß7 are measured by the market capitalisation and divided into large caps, mid caps and small caps. Dummies ß8 and ß9 involve the size of the companies under review as well, since next to market capitalisation the stock exchange index is also a commonly used variable in the size hypotheses. Due to the fact that the indexes and their admission criteria differ, the country's stock exchange indexes are divided into the main stock exchange index, first to main stock exchange and second to main stock exchange (see section 5.4.5). The meta-regulation, small caps and second to main index dummies are excluded from the model for several reasons: (i) one less dummy than the recoded groups is necessary for a linear regression model with dummies, (ii) based on the bivariate results these dummies are excluded from the model since they are considered to be of less interest, and (iii) in testing different models the current model shows most significance and the degree of explanation is the highest. Table 5.7 below shows the results of the linear regression.

With respect to the models tested, the market capitalisation and stock exchange index dummies all involve the companies' size but do not show relevant multicollinearity.5. The R² varies from 0.156 to 0.165, stating that between 16% and 17% of the total variance in the level of compliance is explained by the independent variables, i.e. the model. The F-ratios vary from 16.257 to 17.390 and all four are significant (p < 0.01), implying that the level of compliance can significantly be explained and predicted by the model. Hence, for example, model I.a can be formulated as:

Level of compliance I.a in percentages = 91.541 + 0.227'AgePrinciple + 4.130dSelf-regulation +1.981dNon-statutory norms + 2.978dStatutory rules -1.263dLarge caps - 0.725dMidcaps + 2.770dMainindex + 2.442dFirst to main index

(R² = 0.156; ơ²= 13.900; N = 711)

For all four models the explanatory variable AgePrinciple (the period of time the comply or explain principle has been applicable) shows a positive significant correlation (p < 0.01) with the level of compliance expressed in percentages and thus predicts a higher level of compliance. Taking all the variables of the model into account for each year the comply or explain principle has been applicable longer in a country, an increase in the level of compliance is predicted of 0.136 per cent to 0.237. When looking only at the insufficient explanations (levels III.a and III.b), the increase is smaller than when taking all the explanations into account. When taking only the comparable code provisions into account the increase per year is higher than when all the code provisions are taken into account.

With respect to the corporate governance arrangement dummies, meta-regula-tion is used as a baseline, meaning the corporate governance arrangement against which the other four are compared. The four dummies are significant for all the models (p < 0.01 and in model I.b the Non-statutory norms dummy p < 0.05). Compared with a company with meta-regulation as its corporate governance arrangement, companies with the other arrangements are predicted to show a higher level of compliance, since all four dummies show a positive beta value. In models I.a and I.b companies with self-regulation as their corporate governance arrangement (4.130 and 4.910) show the highest level of compliance compared with 'meta-regulation companies'. For models Ill.a and III.b the statutory rules dummy shows the highest positive beta values (2.830 and 3.423). Apparently when taking the quality of the explanations into account, the stricter corporate governance arrangement with statutory rules scores highest and when the quality of explanations is not taken into account self-regulation predicts the highest level of compliance.

Although the explanatory variables |36 and |37 on the market capitalisation contribute to the 16% to 17% of the total variance in the level of compliance that is explained by the independent variables, on a stand-alone basis no significance can be seen. Only the large caps dummy scores at a 0.05 significance level in model I.a and it is remarkable that its beta value is negative. Hence, it is predicted that, compared to the excluded variable - the small caps dummy -,the large caps score -1.263 lower on the level of compliance. It was expected that the large caps would score higher, but on the other hand the bivariate analysis for level of compliance I.a is not significant and only just positive. Nevertheless, the levels of compliance III.a and III.b showed significance in the bivariate analysis, but apparently not in the linear regression model when more independent variables were simultaneously taken into account.

The explanatory variables |38 and |39, i.e. the main index dummy and the first to main index dummy, show more significance (p < 0.01, in model I.b p < 0.05 and no significance for the main index dummy in model III.b). The significant beta values are positive and compared with the second to main index the main index dummy predicts the highest level of compliance (2.770; 0.854 and 2.849) for models I.a, Ill.a and I.b. Hence, for models I.a, Ill.a and I.b it is predicted that, taking all the variables into account, the more important the index the company is listed on (in practice the larger the company), the higher the predicted increase in the level of compliance.