Being intolerant of the intolerant. The exclusion of Western European anti-immigration parties and its consequences for party choice (original) (raw)

Abstract

In various European countries established parties have responded quite differently to the recent rise of anti-immigration parties. In Italy and Austria these parties entered governing coalitions. In France and Belgium the established parties agreed never to collaborate in any way with anti-immigration parties. In this paper we aim to assess whether this strategy of exclusion affects the electoral support for anti-immigration parties. To answer the research questions, we link expert survey data to individual-level survey data and perform analyses across 11 parties and across 4 time points. We find that the effect of exclusion depends on the institutional context, in particular the threshold for entering parliament, and the influence of parliamentary opposition parties on policy-making. According to our estimates the former Flemish Bloc benefited from being excluded and the Northern League in Italy would have benefited if it had been excluded. The Danish Progress Party, on the other hand, would have been hurt if it had been excluded. The other parties in our analyses are hardly affected. To the extent that the exclusion of anti-immigration parties is meant to change electoral outcomes in favour of the established parties, its success is thus quite mixed.

Access this article

Log in via an institution

Subscribe and save

Buy Now

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Figure 1

The alternative text for this image may have been generated using AI.

Similar content being viewed by others

Notes

  1. We focus on the national level because politics at that level is bound to have the greatest impact on the electorate. In the eyes of the voter, national elections are the most important elections: ‘first order elections’ (for example, Reif and Schmitt, 1980)
  2. In addition to inter-party variations, ostracism can differ among different levels of policy-making: the local, regional, national and international level. Parties may collaborate at the local level, and simultaneously refrain from cooperation at the national level. This might contaminate the national-level effects that we examine in this paper, which makes us less likely to find any significant effects. This makes us more confident about the significant effects that we find in this paper.
  3. Note that we do not consider co-opting policy positions as ‘exclusion’. It is perfectly possible, and even reasonable, for a party to co-opt another party's policy position without cooperating with the other party. Thus, co-opting a party's policy positions does not preclude systematically boycotting it.
  4. Interestingly, Downs (1957) begins from what Simon (1985) would call a ‘substantive’ view of human rationality, and not from a bounded or ‘procedural’ view. According to Downsian spatial analysis, a voter is expected to vote for the party that would yield the highest utility. In practice, this does not make much of a difference in this paper, however.
  5. The ostracism of a particular party may affect targeting parties in similar ways. A party that participates in a cordon sanitaire around a particular other party may be internally divided over the issue of whether or not to continue this strategy. This is clearly the case for the Flemish Liberals and Democrats (VLD), a party that has been systematically boycotting the Flemish Interest (VB) over the last decade although several prominent VLD members have argued for a rapprochement to the VB (for example, Maddens and Fiers, 1998). If both the mainstream right and the anti-immigration party are divided, the net electoral effect of the ostracism of the anti-immigration party may be zero. This is a counterargument that Van Donselaar does not deal with.
  6. It is possible that the exclusion also leads voters to not state their party choice if they voted for an anti-immigration party. This would then lead to an under-representation of anti-immigration party voters in contexts where an anti-immigration party is excluded. Given our findings that the effects of ostracism depend on the political context, we have no reason to assume that this should make us less confident about our results.
  7. Belgium is a federal state with two completely separate party systems, which means that a Flemish voter cannot vote for a Wallonian party and vice versa. For this reason, we treat the two parts of Belgium as two separate systems (see also for example, De Winter and Swyngedouw, 1996).
  8. Because of the small sample of the two Belgian sub-systems, these results are not very reliable, but the other sub-samples do not pose such problems.
  9. The studies consisted of independent cross-sectional surveys fielded in each member country of the EC/EU immediately after each of the EP elections. The European Elections Studies are extensively documented on the European Elections Studies’ web site: http://www.europeanelectionstudies.net/.
  10. Only the results concerning the nine parties that still existed at the time that the expert survey was conducted (June 2005) could be cross-validated. However, given that no errors were discerned in the codings of other parties, we do not see any reason to question the coding of the other two parties, the Danish FrP and the Dutch CD.
  11. This variable is sometimes referred to as the ‘probability to vote’ question. It must be emphasized that the variables do not measure probabilities in the statistical sense of that word, as their sum is not – and should not be – constrained to 1 (Van der Eijk, 2002). Tillie (1995) has demonstrated that answers to it are not constrained by the intention to vote for a specific party in the upcoming election. In other words, voters who are certain to vote for a particular party may still give a very high score to the party that is their second preference.
  12. In the 1999 European study, more than 93 per cent of the respondents gave the highest utility to the party they would vote for in a national election at that time. Similar percentages were obtained in 1989 (Van der Eijk and Franklin, 1996a,1996b, Chapter 20) and for 1994 (Van der Eijk et al, 1999, p. 168).
  13. Our focus on ideology and the left–right scale ignores that other dimensions may be at play as well when it comes to vote choice. However, we have no way of dealing with this issue, as we do not have data at our disposal that would completely solve this problem.
  14. Most importantly, the conclusions concerning our four hypotheses do not change when excluding the party size variable. We may seem to have a theoretical argument not to include the party size variable. The size of the party is theoretically expected to be affected by exclusion before the measurement. Taking the size of a party into account would therefore enforce a focus on the short-term effects of exclusion, whereas we expect exclusion to have an impact in the long run rather than in the short term. As a result, we make it very difficult to find any effects of exclusion, which may lead us to commit Type-II errors. However, not taking into account party size may lead us to attribute pragmatic considerations to exclusion. Not surprisingly, the effect of exclusion on the anti-immigration party becomes significantly negative. It is impossible to disentangle whether this is because of the exclusion of these parties, or simply because they are small (for other reasons).
  15. And perceived ideological closeness is important for the vote for any party, as has been repeatedly shown (for example, van der Eijk and Franklin, 1996b).
  16. It would be interesting to also include the interaction between these variables. Such an interaction variable would tap the effect that the more moderate its position and the more moderate that of its main rival, the more votes the anti-immigration party will receive. However, this interaction variable is highly correlated with its lower-order effect of the anti-immigration party's ideological position (_r_=0.88, significant for _P_=0.001, one-tailed, _N_=21 509).
  17. The interaction variables are not included in Table 2, as their descriptive statistics are difficult to interpret.
  18. With his permission, this appendix relies largely on a paper by Van der Eijk (Van der Eijk, 2002).
  19. The original regression equation is y i =_b_0+_b_1*x i +e i . In this equation the predicted value ŷ i =_b_0+_b_1*x i . By substituting _b_0+_b_1*x i for ŷ i in the equation, the new regression equation (using the ŷ i as predictors of party utility) becomes: y i =ŷ i +e i . If one were to estimate a new regression using the predicted value as the single predictor of the propensity to support that party, the estimate of the intercept and slope become 0 and 1, respectively, and e i (which forms the basis for the computation of explained variance) is unaltered. When stacking the _y_-hats on top of each other in the stacked matrix, the actual variable added to the stacked matrix is not the predicted value (_y_-hat), but the deviation of the _y_-hats from their mean for each party. This still encapsulates the variance in party preferences caused by the independent variable, but prevents differences among parties in the average level of party preferences from being incorporated in the newly created independent variable. Such differences among parties in average preferences are caused by other factors besides the predictor variable x, and should hence not contribute to the variance in the newly created predictor (_y_-hat). For an elaborate discussion of this procedure, see (Van der Eijk and Franklin, 1996a,1996b, paper 20).

References

Download references

Author information

Authors and Affiliations

  1. European University Institute, Italy
    Joost Van Spanje
  2. ASCoR, University of Amsterdam, The Netherlands
    Joost Van Spanje
  3. ASSR, University of Amsterdam, The Netherlands
    Wouter Van der Brug

Authors

  1. Joost Van Spanje
  2. Wouter Van der Brug

Corresponding author

Correspondence toJoost Van Spanje.

Appendix

Appendix

Method by which analyses were conductedFootnote 18

The dependent variable in our analysis stems from a series of questions – one for each party – which asks respondents how likely it is that they will ever vote for this party. Earlier research (Tillie, 1995) has shown that these variables perform in the way that Downs (1957) imagined party utility to perform. The electoral preferences obtained from these questions yield a large number of variables: one for each party. In order to answer our research questions these should all be analysed simultaneously. This can be achieved by a variant on the technique suggested for regression in time and space (Stimson, 1985), which involves a regression analysis on a ‘stacked’ data matrix. This is a matrix derived from a ‘normal’ survey data matrix, in which the records represent every respondent*party combination. Figure A1 illustrates graphically how the original data matrix is transformed into a stacked matrix.

Figure A1

Figure A1

The alternative text for this image may have been generated using AI.

Full size image

Structure of the data matrix.

The dependent variable in the analyses is the observed propensity to vote for the party that is indicated in the respective respondent*party combination. In this data matrix, the dependent variable thus pertains to each of the parties in turn, and can be considered as a generic measure of party support. In this stacked data matrix, independent variables that pertain to a relation between party and voter have to be specially constructed. A party's position on the left/right scale, for example, will not capture in this design the effect of ideology on propensity to vote, but a distance between the left/right positions of voter and the respective party does.

In the European Elections Studies of 1989−2004, vote propensities were measured for 21 509 individual respondents in the countries that we selected. In the stacked data matrix each respondent is represented by as many ‘cases’ as there are parties for which (s)he indicated a vote propensity. The stacked data matrix has 158 646 entries.

When no party placement scores on specific issues are available, independent variables have to be constructed empirically. This is done by predicting propensities to support a party on the basis of a simple regression analysis for that party in the original data matrix. This is done for each of the parties in turn. These predicted values (_y_-hats) are simply linear transformations of the original independent variables. These are saved and inserted as new predictors of party support in the stacked data matrix.Footnote 19

This procedure was applied for the variables social class, income, religion, EU-approval and issues. For example, to estimate the contribution of social class to the explanation of party choice, we assessed empirically how well respondents’ subjective assessments of their social class predict preferences for each of the parties. So for each of the parties in turn, a regression analysis was conducted with social class as the independent variable (in the form of four dummys) and vote propensities for that party as the dependent variable. The predicted values of each of these regressions (_y_-hats) were saved and inserted in the stacked data matrix as new predictors of vote propensities. The variable, with which the effect of religion is estimated, is constructed in a similar way: it consists of predicted values of separate regressions per party. The independent variables in these separate regressions were five dummys indicating five different religious denominations and church attendance as an additional variable. The same procedure was used to estimate the effect of EU-approval (a linear bivariate regression). In the case of the issues, we used the answers to the question what is the most important problem facing your country? What are the second and third most important problems? On the basis of these answers we constructed a large set of dummy variables, which indicate whether a respondent mentioned a specific problem, or not. These variables were then used as independent variables in the regressions per party.

Since multiple observations in the stacked data matrix refer to the same respondent, these observations are not independent from each other. Moreover, the distribution of the dependent variable is rather skewed, so that the data do not meet some assumptions that OLS regression requires. For this reason we estimated robust standard errors (in STATA), which do not assume a homoskedastic distribution of the error terms. Moreover, multiple observations pertaining to one respondent were defined as dependent (using the option ‘cluster’ in STATA).

Rights and permissions

About this article

Cite this article

Van Spanje, J., Van der Brug, W. Being intolerant of the intolerant. The exclusion of Western European anti-immigration parties and its consequences for party choice.Acta Polit 44, 353–384 (2009). https://doi.org/10.1057/ap.2009.7

Download citation

Keywords