Quantitative country comparisons: turnover time

In this paragraph the differences and similarities in turnover time for the different country groups will be discussed. It will be recalled that the hypotheses presented in chapter 3 were based on the assumption that more similar countries would entertain better performing surrender relationships with the Netherlands. It is therefore interesting that some subtle differences between the three country groups (similar, dissimilar and intermediate countries) can be observed with regard to turnover time, although the differences do not seem to be extremely large. While the data mainly shows values for surrender times between roughly 45 and 80 days, but also contained a relatively high amount of high values over 80 days while containing very few values below 30, the data seems to be skewed somewhat. Performing a Shapiro-Wilk test to ascertain whether or not the data show normality confirms this face value suspicion, as the test confirms at the 0,01 significance level that all three country-groups do not form normally distributed samples.155 _{Furthermore, subjecting the tree country-groups to 1,5 IQR procedures shows that}

there are several outliers.156 _{As parametric tests such as ANOVA tests assume normalcy and are sensitive}

to outlier values, it was more appropriate to utilize non-parametric tests, in particular the Kruskal-Wallis test, for the analyses. The tests were run with and without the signaled outliers. The Kruskal-Wallis looks at the extent to which the medians of two samples are comparable or whether there has been a shift in medians (Spurrier, 2003)157_{. The null hypothesis is designed to examine the chance of the differences in}

the observed sample distributions arising randomly. When accepting the alternative hypothesis, conversely, the researcher assumes that the differences in the rank distributions of the different samples did not arise by chance.

155 _{All three results yielded a rounded p=0,00}

156 _{For the similar countrygroup (Sweden, Finland and Germany) two cases with a turnover time t>113 were outliers:}

case 13/706674-10 and case 13.497.374. For the intermediate sample (Spain, UK, Belgium) all cases above 110,375 were outliers, which were cases 13/4977468-14, 13-751229-14, 13.737355-13, 13.737.292-13 and 13.737.142-13. Furthermore, case 13.706264-11 fell below the lower 1,5 IQR bound with a value of 9 days. Finally, the dissimilar countrygroup (Bulgaria, Poland, Romania) showed several high value outliers in cases 13/497.127-06, 13.497.470- 200

157 _{J.D. Spurrier (2006), on the null distribution of the Kruskal-Wallis statistic, nonparametric statistics, 15-6, pp.685-}

92

Number of cases (N) Mean rank (not corrected for

outliers)

Mean rank (>1,5 IQR outliers removed) Similar countrygroup (Sweden, Germany, Finland) 31 48,88 44,16 Intermediate countrygroup (Spain, UK, Belgium 38 59,10 53,14 Dissimilar countrygroup (Bulgaria, Poland, Romania) 36 65,96 60,46

Table 6.1: mean country-group ranks on turnover rates with and without outliers

At face value a difference between the ranks for the three groups of countries is visible, as can be seen in table 6.1. As suggested in chapters 3 and 4, EAW’s from countries similar to the Netherlands on the basis of culture, corruption and centralization/decentralization show a lower mean rank than those EAW’s from a group of countries which is more dissimilar. The Kruskal-Wallis test for the samples not altered by removing outliers shows that the differences are significant at the p<0,1 significance level, the test yielding a p=0,098 value.158 _{Using data that excludes the outliers found on the basis of the >1,5 IQR method yields}

a slightly stronger indication of a difference between the sample distributions, with a p value of 0,092.159

The similar results for both the altered and the unaltered samples illustrates that outliers did not have a substantial effect on the Kruskal-Wallis procedure. However, the fact that sampled data does not reach significance at the p<0,05 level warrants some caution in interpreting these results as conclusive evidence to support the statement that there is a structural difference between the distribution of the sampled data. The conclusion is therefore that there is some indication that the Amsterdam court performs slightly better with regard to turnover time when confronted with EAW’s from more similar countries than when confronted with EAW’s from more dissimilar countries, but that the effect found in the sample data would require subsequent research to confirm its existence.

Moreover, the sampled data strongly supports the intuitive proposition that postponed cases have a relatively higher turnover time. Separating the data for postponed cases into a binomial variable in which a case is either postponed or not postponed, and subsequently performing a Mann-Whitney-U non- parametric test for two samples, generates support for the test’s alternative hypothesis that the populations have a different distribution with p values of 0,00 for both the turnover data including outliers and the turnover data excluding outliers. Therefore, the results are significant at the p<0,05 level. The substantial differences in turnover time for postponed and non-postponed cases have been illustrated in table 6.2, which is based on both data including outliers and data excluding outliers.

158 _{Degrees of freedom (Df) 16, N=116} 159 _{Df 2, N=105}

93

Postponed Non-postponed Mean turnover time (outliers

included)

142,3 63,4

Median turnover time (outliers included)

103 59,5

Mean turnover time (outliers excluded)

100,23 60,29

Median turnover time (outliers excluded)

88,5 59

Table 6.2: Mean and median turnover times for postponed and non-postponed cases