ONLINE APPENDIX FREYBURG, Tina (2011) Transgovernmental Networks as Catalysts for Democratic Change? EU Functional Cooperation with Arab Authoritarian Regimes and Socialization of Involved State Officials into Democratic Governance. Democratization 18(4): 1001-1025 This appendix contains additional results, which are not reported in detail in the paper due to space constraints. Exploratory Factor Analysis (EFA) I use the robust mean and variance-adjusted weighted least squares (WLSMV) extraction procedure, which is insensitive to non-normal distribution of categorical items and appropriate for small samples (Brown 2006: 388). The oblique rotation method Oblimin is applied because I theoretically expect factor inter-correlation: All three dimensions belong to the overall concept of democratic governance. Careful analysis of the correlation matrix reveals that democratic governance items highly correlate with each other and can be retained in the analysis. The Kaiser- Meyer-Olkin test of sampling adequacy meets the benchmark established by Worthington and Whittaker with.676; only item 6 misses the benchmark but is retained due to its theoretical importance (Worthington and Whittaker 2006: 832). Bartlett s Test of Sphericity supports the selection of these items (p =.000, df = 28.000, approx. χ 2 = 139.938). Model fit The generated model has a good fit (cf. Hooper et al. 2008; Hu and Bentler 1999; McIntosh 2007): The absolute fit indices such as an insignificant Chi-square value at a.05 threshold (χ 2 = 6.313; df = 6; p =.3890) and a standardized root mean square residual (SRMR) of.035 meet the required standards. The Comparative Fit Index (CFI) of.999 as an incremental fit index is also very close to the 1.0 benchmark. The total variance explained by the three factors is 71.03 per cent. The findings are robust across alternative methodologies. Replications with the oblique rotation Geomin and the orthogonal rotation Varimax produce the same pattern of factor loadings. Scree tests and replication of the factor analysis with randomly split sub-samples verified the existence of three latent variables (Fabrigar et al. 1999; Thompson 2004: 31-6). 1
Unidimensionality and correlation between factors Correlation between the factors demonstrates that although the three dimensions of democratic governance present factors on their own, they are interdependent. In particular, a positive attitude toward participation implies at least a partially positive attitude toward accountability (r =.554; p =.000). This is intuitively plausible because some of the accountability mechanisms imply involvement of the public (item 8). Factor inter-correlation of transparency with participation (r =.284; p =.110) and with accountability (r =.036; p =.816), respectively, is not significant. Nevertheless, loadings other than expected and (though insignificant) item cross-loading point to the relatedness of participation and transparency. Eventually, transparency as access to information is necessary to enable meaningful participation (item 2). Conversely, participation enhances transparency. Item 5 loads on both dimensions though less high and insignificantly on participation (r =.299; p =.126). In principle, however, the three dimensions form single factors as shown by relative unidimensionality, that is, the squared loading/squared communalities ratio of the two items that load highest on a factor differs to at least.25 (Fürntratt 1969: 66). The only exception is item 5, for which the difference is 22 per cent due to the mentioned cross-loading. 2
Descriptive Statistics of the Dependent Variables Table A1. Descriptive statistics All officials state Participation Transparency Accountability Democratic Governance P NP P NP P NP P NP Max. value 5 5 5 5 5 5 4.92 4.5 Min. value 4 2.5 2.5 3.5 2 2 3.75 2.83 Mean 4.63 4.13 4.1 4.1 3.84 3.82 4.24 3.99 Median 4.75 4.25 4 4 3.67 4 4.14 4.11.371.689.446.338.656.755.307.417 Skewness -.403 -.494-1.062.857 -.511 -.607.711-1.516 Non-responses (%) 16.7 10.53 21.4 7.89 23.81 13.16 N 42 38 Environment Max. value 5 5 4.75 4.75 4.67 5 4.81 4.44 Min. value 4 2.5 2.5 3.5 2 2 3.75 2.83 Mean 4.68 3.94 4.07 4.02 3.74 3.747 4.21 3.90 Median 4.75 4.0 4.0 4.0 3.67 4.0 4.14 4.1.372.711.456.336.650.777.303.444 Skewness -.655 -.266-1.484.932 -.775 -.637.679-1.276 Non-responses (%) 15.6 15.38 25 11.54 31.25 3.85 N 32 26 Competition Policy Max. value 5 5 5 5 5 5 4.92 4.5 Min. value 4 3.5 3.5 4 3 3 3.89 3.92 Mean 4.47 4.46 4.19 4.23 4.07 4.04 4.30 4.24 Median 4.38 4.5 4.0 4.13 3.83 4.17 4.17 4.25.339.520.429.3100.644.677.331.188 Skewness.294 -.533.439 1.558 -.005 -.224 1.012 -.413 Non-responses (%) 20 0 10 0 0 33.33 N 10 12 Note: Values range between 1 (non-democratic) to 5 (democratic); P = participating officials, NP = non-participating officials; cases with missing values excluded listwise. Non-responses cover blank and don t know answers. 3
Kruskal-Wallis Test and Robustness Checks Assumption of multivariate normality is violated as shown by skewness and kurtosis of measured variables and confirmed with significant Shapiro-Wilk test for small sample size; p-values for the three separate dimensions range from.000 to.001 (experimental group) and.001 to.020 (control group), respectively. The Levene s test of equality of variances, which does not require normality of the underlying data, is generally not significant with p-values ranging from.311 to.799 (project on competition) and from.233 to.708 (with the exception of participation with p =.031; project on environment) for the three separate dimensions: A Kruskal-Wallis test is thus a suitable statistical procedure. Table A2. Welch s ANOVA and one-way ANOVA Participation Transparency Accountability Democratic Governance Env. Comp. Env. Comp. Env. Comp. Env. Comp. One-way ANOVA F-ratio 21.515.002.121.047.000.006 6.393.184 p-value.000.961.730.831.984.937.016.675 N total 49 20 47 21 47 18 39 15 Welch s ANOVA F-ratio 19.100.003.122.042.000.006 6.770.171 p-value.000.957.728.840.984.938.013.689 N total 49 20 47 21 47 18 39 15 Note: Cases with missing values excluded case-by-case; df = 1; Env. = Twinning project on the environment, Comp. = Twinning project on competition matters; p-value.05 in bold. If one repeats the analysis with a Welch's ANOVA, which is robust to heteroscedasticity, and with a usual ANOVA to test for differences between group means, similar results are produced by both tests (see Table A2). 4
References Brown, Timothy A. Confirmatory Factor Analysis for Applied Research. New York: Guilford Press, 2006. Fabrigar, Leandre R., Robert C. MacCallum, Duane T. Wegener, and Erin J. Straham. Evaluating the Use of Exploratory Factor Analysis in Psychological Research. Psychological Methods 4, no. 3 (1999): 272-99. Fürntratt, Ernst. Zur Bestimmung der Anzahl interpretierbarer gemeinsamer Faktoren in Faktorenanalysen psychologischer Daten. Diagnostika 15 (1969): 62-75. Hooper, Daire, Joseph Coughlan, and Michael R. Mullen. Structural Equation Modelling: Guidelines for Determining Model Fit. The Electronic Journal of Business Research Methods 6, no. 1 (2008): 53-60. Hu, Li-tze, and Peter M. Bentler. Cutoff Criteria for Fit Indexes in Covariance Structure Analysis: Conventional Criteria Versus New Alternatives. Structural Equation Modeling 6, no.1 (1999): 1-55. McIntosh, Cameron N. Rethinking Fit Assessment in Structural Equation Modelling: A Commentary and Elaboration on Barrett (2007). Personality and Individual Differences 42, no. 5 (2007): 859-67. Thompson, Bruce. Exploratory and Confirmatory Factor Analysis: Understanding Concepts and Applications. Washington: American Psychological Association, 2004. Worthington, Roger L. and Tiffany A. Whittaker. Scale Development Research. A Content Analysis and Recommendations for Best Practices. The Counseling Psychologist 34, no. 6 (2006): 806-38. Notes on contributor Tina Freyburg is a postdoctoral researcher and lecturer in the European Politics team of the Centre for Comparative and International Studies at ETH Zurich. Her main research interest is in International Relations and EU Studies. Current research projects focus on international democracy promotion, notably on socialization into democratic governance through transnational exchange in authoritarian contexts and the impact of EU political conditionality under difficult conditions. Tina Freyburg Swiss Federal Institute of Technology (ETH) Zurich Centre for Comparative and International Studies (CIS) Haldeneggsteig 4 (IFW D ) 8092 Zurich Switzerland. Email: freyburg@eup.gess.ethz.ch Website: www.mwpweb.eu/tinafreyburg 5