The neighbourhood effect in economic voting: the association between local unemployment figures and national economic perceptions and incumbent voting in Belgium, 2009–2019

ABSTRACT

While economic voting theory assumes that voters respond to economic conditions, critics have argued that most voters lack an adequate understanding of key national economic indicators. In this paper, we investigate the occurrence of a neighbourhood effect, where citizens can observe unemployment levels in their own local communities. Using both official statistics and survey data of three recent election studies in Belgium, we assess whether local unemployment levels are associated with the assessment of the national economy, and incumbent voting. While the results show that the local unemployment level is strongly associated with the assessment of the national economy, results regarding a direct association with supporting incumbent political parties are mixed. We argue that the neighbourhood effect is an important mechanism in economic voting, as citizens react to a neighbourhood effect in their assessment of the state of the national economy.

KEYWORDS:

• economic voting
• neighbourhood effect
• unemployment
• Belgium
• Belgian Electoral Study
• retrospective voting

Disclosure statement

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

Notes

1 The results for the regions separately are included in Appendix J. The associations are similar between the regions, and to the results presented here. However, almost none of the associations reaches common levels of statistical significance. However, because of the aggregate nature of the main independent variable of interest, the number of observations at this level of analysis is limited, resulting in low statistical power.

2 The three samples excluded respondents from the capital Brussels because of legal and practical limitations (the National Register is not allowed to keep information about the language of a citizen). However, with the Dutch-speaking and French-speaking regions, the samples cover about 90% of the Belgian population – and this is a common approach used for electoral research in Belgium.

3 Respondents indicating not to have turned out are coded as missing observations. However, note that turning out to vote is compulsory in Belgium, and turnout rates are stable around 90%. In our data, only ca. 8% indicated not to have turned out to vote themselves. One party that had governed since 2014 withdrew from the federal government in December 2018, a few months before the May 2019 elections. A preliminary analysis (Figure B.1 in Appendix B) suggested that voters still considered this party as “incumbent”, and therefore it was classified as such.

4 For the 2019 election, this means we include the change in unemployment levels between 2017 and 2018.

5 Between 2008 and 2009, the average change in unemployment rate was 0.811% (std. dev. 0.585); in 2014 this was 0.149 (std. dev. 0.322); between 2017 and 2018 (last years with available data) this was −0.626 (std. dev. 0.226).

6 Only for the 2009 election, we do not include this control, as the respondent’s income was not asked.

7 As the individual is the unit of analysis, family income is not a very precise indicator. Therefore, for the 2019 data (the only in which household size was measured), we divided the family income (taking the middle point of the interval) by the number of household members, leading to a modified household equivalence scale. The results using this control were substantially the same as those presented here.

8 We also estimated model combining the data of 2014 and 2019. The results, reported in Appendix K, support the findings presented here. When pooling the data, there is no evidence for a significant association between change in unemployment rate and incumbent support.

9 These models might suffer from an omitted variable bias. Therefore, in Appendix F, we show the models including more control variables (i.e. the proportion of highly educated citizens in every municipality, the average age of the citizens in a municipality, and a separate model including the vote share in the last election), and this does not alter our conclusions. Note, however, that the most recent data of these controls come from the 2011 census. While we believe that they can still be used to give a sense of the structural differences between municipalities, they are not recent enough to include them in the main models reported here.

10 The association becomes non-significant when removing the outlier at the lower end of unemployment change in 2019; the results of the other years remain unchanged when removing the outliers.

11 We also conducted the analyses of economic perceptions and incumbent voting combining the data of 2014 and 2019. The results, reported in Appendix L, are in line with those presented here: while there is evidence for an association between unemployment levels and economic perceptions, there is no consistent evidence for such an association with incumbent voting.

12 See Appendix H for the results of these analyses. Note that we cannot estimate all models including the past vote due to data limitations.

Additional information

Funding

This work was supported by the Research Foundation Flanders.