Jumping on the Bandwagon? Explaining fluctuations in party membership levels in Europe

ABSTRACT

This paper explores how party-specific contextual factors explain variations in membership levels. Based on a subset of MAPP data that includes 2898 yearly membership data points for 262 parties in 24 countries over a period from 1990 to 2014, it examines three sets of explanations: the lifecycle model (party age), the bandwagon model (electoral performances and governemental participation), and the competition model (effective number of parties). Our results confirm that membership ratios present an overall decreasing trend across parties over time. At the same time, we show that this trend is flattening and that there are important variations across parties. Fluctuations of membership are part of a party’s lifecycle. Our results also point toward a bandwagon effect whereby party membership levels increase or decrease according to electoral performances, and to a patronage boost linked to governmental participation. Finally, we show that party system fragmentation decreases individual parties’ membership ratios. Overall, our findings complement the story on party membership decline and calls for further investigations of party-specific explanations.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 32 parties were excluded because of limited longitudinal data points in the MAPP database (LIF-Austria, NEOS-Austria, LDD-Belgium, PC-Belgium, HSP AS-Croatia, HL-Croatia, HSD-Croatia, ANO 2011-Czech Republic, RP-Estonia, EPE-Estonia, DK-Hungary, SEL-Italy, DK-Lithuania, 50+-Netherlands, PSL-Poland, KLD-Poland, UW-Poland, LPR-Poland, PEV-Portugal, PPV-Portugal, PTP-Portugal, PAE-Romania, PUNR-Romania, FD-Romania, PRN-Romania, FD-Romania, UPSC-Romania, PCM-Romania, PUER-Romania, Lipa-Slovenia, SMS-Slovenia, CDS-Spain), Observations from Icelandic parties were also excluded as their membership levels constitute a clear outlier in the data distribution.

2 We followed Allison’s recommendation and computed the VIF (Variance Inflation Factor) for the variables included in the models. While authors differ in their interpretation of the cut-off value of the VIF, most consider that a VIF < 5 is acceptable. When we include the centred version of the variables “time” and “age” in the models, the VIF remains around 4.1, which would be considered acceptable. However, the quadratic version of the models that include the variables “time square” and “age square” display a higher VIF for these variables. It is relatively normal to have a collinearity between a variable such as “time” and “time square”. The problem was that when we include the highly correlated quadratic version of continuous variables “time” and “age”, then the VIF of most variables is >5 which makes the model difficult to interpret. We therefore decided to group parties according to three age categories and thus transform the continuous variable in a categorical one.

3 As we have more missing values for younger parties, we have 25% of observations corresponding to old parties, 37% corresponding to medium old parties and 38% corresponding to young parties.

4 We have run the analyses with a fixed measure of vote share between electoral cycles and obtained similar results.

5 We also have computed a panel data model with fixed effect and reached the same results.

6 The Akaike information criteria and the Bayesian Information criteria both indicate a lower fit of Model 2 compared to Model 1.

7 AIC and BIC are based “on the likelihood of the data given a fitted model penalized by the number of estimated parameters of the model” (Nakagawa and Schielzeth 2013).

8 Given that our data is unbalanced, we also checked for potential attrition or selection bias through the variable addition test developed by Nijman and Verbeek (1992). We did not find any significant attrition bias.

Additional information

Funding

This work was supported by the FNRS [Grant 19483403].