Abstract
This chapter is the centrepiece of the empirical agenda since it pursues a systematic test of the effects of established predictors on the individual and the contextual level on generalized EU regime support. It takes the longitudinal perspective since it strives to compare results of 10 Standard Eurobarometer survey waves from 2006 to 2015 relying on identical measurements of predictors and methodologies. Applying various estimation techniques, the purpose of the chapter is to identify the decisive factors driving generalized EU support and how their effects potentially change during the period under investigation. The heterogeneous and multidimensional repercussions of the Euro crisis can be modelled using multi-level models with cross-level interactions that account for the temporal and geographical structure of the survey data and allow for direct group comparisons. The study reveals the Euro crisis has changed the nature of the presence of European politics in European societies. However, the salience and domain of this change appeared differently in different societies at different points in time. This diagnosis is mirrored in substantial—partly temporary, partly permanent—changes of explanatory power of key attitudinal predictors of generalized EU regime support.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
Instead of cumulating all contextual units at one level (level 2) for each combination of country and time subsamples, the gold standard would be to estimate cross-classified multilevel regression models. Those would specify country units and time units on level 2 and level 3 as well as constructing a level 4 out of the multiplied combinations country*time. While individuals are nested in country and time units respectively, country units are not nested in time units and vice versa. To overcome potential estimation biases at the higher levels, the ideal solution would be to estimate such cross-classified multilevel models see Gelman and Hill (2007, 241 f.), Raudenbush and Bryk (2002, 373 f.). From a practical point-of-view, specifying cross-classified models would extend the number of terms in the regression equation by length. I refrain from this sophisticated approach since I do not expect higher level variance of country and time units to be confounded. Instead, I specify two-level regression models in which level-1 captures the individual respondents and level-2 consists of all existing combinations of countries and survey waves (country*time).
- 2.
The key commands for Study II are logit, melogit, margins, marginsplot, and coefplot Stata Press (2017).
- 3.
The sample sizes vary between 15,275 (2006 Q2) to 18,671 (2013 Q4) fully-responding observations after list-wise deletion of cases with missing data on the relevant variables. Originally, the 2006 Q2 survey consisted of 26,665 observations, the 2013 Q4 survey of 26,829 observations.
- 4.
For a discussion on minimum requirements for maximum likelihood estimation such as observation numbers, see Stegmueller (2013).
- 5.
In model 1, the number of observations was restricted to the full model (M5) to ensure that change in model fit across the models is independent of varying numbers of respondents. That is why the number of respondents is harmonized across the different specifications.
- 6.
The Akaike and the Bayesian Information Criteria are useful instruments for model selection Akaike (1998 [1973]), Raftery (1995). Based on different but related mathematical definitions, they inform about the amount of information lost when comparing the estimated model to the entire observed data. A thorough discussion deliver Burnham and Anderson (2004). The best practice intuition is that to select the model specification that produce the lowest AIC/BIC score. The absolute numbers themselves cannot be interpreted so that the scores are only useful for relative comparisons of fitted regression model relying on the same dataset Long and Freese (2014, 123 f.).
- 7.
The question whether the effect of national political attitudes is either a form of heuristic cue or a substantiated effect will be addressed later in this chapter (Sect. 6.4).
- 8.
Crisis context A serves as baseline category, all estimates of the CLI can be found in the Appendix C1.
- 9.
I refrain from verbally formulating the AMEs of the other IVs since the substantial impact has already been pointed out in the discussion of the logistic models. By design, the interaction effects cannot be displayed here since the AMEs indicate the average effect over all observations—therefore regardless of the different outcome conditions of the crisis contexts.
- 10.
To recall, crisis contexts A–C represent the pre-crisis period, the acute phase, and the recovery period for the donor countries, while the contexts D–F represent the respective phases in the singular crisis countries (including Italy).
- 11.
This method is further explained and proven by Bolsen and Thornton (2014). Following their analysis, a pairwise two-tailed test of two coefficients with alpha = 0.05 corresponds graphically to comparing the 83% confidence intervals of those two coefficients.
- 12.
H13 concerning the role of political involvement is further examined in Sect. 6.4.
- 13.
- 14.
Zero stands for the worst aggregated WGI score within the group of the EU27 from 2006 to 2015 while One stands for the highest score.
- 15.
CMEs were estimated based on a modified version of logistic regression model M5 with an additional interaction between the crisis contexts-variable and the measures of political involvement.
- 16.
I recommend building on previous work by Wagner (2012) for this purpose.
- 17.
The measure was neglected for the operationalization of involvement in the prior analyses since it would have reduced the level-1 observations to 152,876 and at the same time resulted in the loss of 27 country*time units or one SEB wave respectively.
- 18.
Overall the model fit of M8 has increased compared to M5 due to the inclusion of the additional national attitudes (AIC 140.034 (M8) vs. 156.702 (M5); BIC 142.941 vs. 159.870; R2 MZ 0.543 vs. 0.537).
References
Acock, A. C. (2016). A gentle introduction to Stata (5th ed.). College Station, TX: Stata Press.
Akaike, H. (1998 [1973]). Information theory and an extension of the maximum likelihood principle. In E. Parzen, K. Tanabe, & G. Kitagawa (Eds.), Selected papers of Hirotugu Akaike (pp. 199–213). New York: Springer.
Bauer, G. (2010). Graphische Darstellung regressionsanalytischer Ergebnisse. In C. Wolf & H. Best (Eds.), Handbuch der sozialwissenschaftlichen Datenanalyse (1st ed., pp. 905–927). Wiesbaden: VS Verlag für Sozialwissenschaften.
Bolsen, T., & Thornton, J. R. (2014). Overlapping confidence intervals and null hypothesis testing. Newsletter of the APSA Experimental Section, 4(1), 12–16.
Burnham, K. P., & Anderson, D. R. (2004). Multimodel inference. Sociological Methods & Research, 33, 261–304. https://doi.org/10.1177/0049124104268644.
Faas, T. (2010). Arbeitslosigkeit und Wählerverhalten: Direkte und indirekte Wirkungen auf Wahlbeteiligungen und Parteipräferenzen in Ost- und Westdeutschland. Zugl.: Duisburg-Essen, Univ., Diss., 2008 (Studien zur Wahl- und Einstellungsforschung, Vol. 17). Baden-Baden: Nomos.
Foster, C., & Frieden, J. (2017). Crisis of trust: Socio-economic determinants of Europeans’ confidence in government. European Union Politics, 18, 511–535. https://doi.org/10.1177/1465116517723499.
Gelman, A., & Hill, J. (2007). Data analysis using regression and multilevel/hierarchical models (Analytical methods for social research). Cambridge: Cambridge University Press.
Hox, J. J., & Wijngaards-de Meij, L. (2015). The multilevel regression model. In H. Best & C. Wolf (Eds.), The SAGE handbook of regression analysis and causal inference (pp. 133–152). Los Angeles: SAGE.
Kaufman, D., Kraay, A., & Mastruzzi, M. (2010). The worldwide governance indicators: Methodology and analytical issues (September 2010) (World Bank Policy Research Working Paper No. 5430). https://ssrn.com/abstract=1682130.
Long, J. S., & Freese, J. (2014). Regression models for categorical dependent variables using Stata (3rd ed.). College Station, TX: Stata Press.
Rabe-Hesketh, S., & Skrondal, A. (2012). Multilevel and longitudinal modeling using Stata (3rd ed.). College Station, TX: Stata Press.
Raftery, A. E. (1995). Bayesian model selection in social research. Sociological Methodology, 25, 111–163. https://doi.org/10.2307/271063.
Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear models: Applications and data analysis methods. Thousand Oaks, CA: SAGE.
Stata Press. (2017). Stata: Multilevel mixed-effects reference manual (15th ed.). College Station, TX: Stata Press.
Stegmueller, D. (2013). How many countries for multilevel modeling?: A comparison of frequentist and Bayesian approaches. American Journal of Political Science, 57, 748–761. https://doi.org/10.1111/ajps.12001.
Williams, R. (2012). Using the margins command to estimate and interpret adjusted predictions and marginal effects. Stata Journal, 12(2), 308–331.
Wooldridge, J. M. (2010). Econometric analysis of cross section and panel data (2nd ed.). Cambridge, MA: MIT Press.
World Bank Group. (2017). Worldwide governance indicators. http://info.worldbank.org/governance/wgi/. Accessed 30 Oct 2017.
Author information
Authors and Affiliations
Rights and permissions
Copyright information
© 2020 Springer Nature Switzerland AG
About this chapter
Cite this chapter
Bauer, S. (2020). Study II: EU Support During the Euro Crisis (2006–2015). In: Citizens’ Support for the European Union. Contributions to Political Science. Springer, Cham. https://doi.org/10.1007/978-3-030-16461-4_6
Download citation
DOI: https://doi.org/10.1007/978-3-030-16461-4_6
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-16460-7
Online ISBN: 978-3-030-16461-4
eBook Packages: Political Science and International StudiesPolitical Science and International Studies (R0)