# Exercise 4.2: Correlations between attitude change and context

We use correlation analysis In order to test whether the attitude changes are driven by changing levels of actual competition. Calculate correlation coefficients between ‘REJECT_CHANGE’, on the one hand, and the three contextual variables, on the other.

*Correlation between ’REJECT_CHANGE and the contextual variables.

CORRELATIONS
/variables reject_change flowcap gdp_growth unemp_change.
EXECUTE.

### SPSS Output

Table 4.1. Correlation between 'REJECT_CHANGE' and the contextual variables
reject_change flowcap gdp_growth unemp_change
reject_change Pearsons Correlation1
Sig. (2-tailed)
N17
flowcap Pearsons Correlation.293 1
Sig. (2-tailed).271
N16 16
gdp_growth Pearsons Correlation-.243 -.115 1
Sig. (2-tailed).348 .671
N17 16 17
unemp_change Pearsons Correlation.241 .027 -.589 1
Sig. (2-tailed).368 .924 .016
N16 15 16 16

Weighted by design weight.

### Questions

1. Do the correlation coefficients have the expected signs?
2. Look at the size of the correlation coefficients. Are the correlations between attitude change and change in actual competition substantial?
3. Are the correlation coefficients statistically significant?
4. What do you conclude from this analysis?

1. Based on group conflict theory, we expect anti-immigration attitudes to become stronger in countries where immigration flows are strong, unemployment is increasing and economic growth is low. Positive values for the variable ‘REJECT_CHANGE’ indicate that anti-immigration feelings have become stronger. In consequence, we expect ‘REJECT_CHANGE’ to correlate positively with ‘flowcap’ and ‘unemp_change’, and negatively with ‘gdpgrowth’. The SPSS output shows that the correlation coefficients have the expected signs.
2. The correlations between attitude change and context variables are all medium size, namely between 0.20 and 0.30 (in absolute value).
3. The reported p-values show that, although they are substantial, none of the correlation coefficients is statistically significant. This is to a large extent due to the fact that this is an aggregate analysis performed at country level. Consequently, the correlations are based on a very small N, and the test lacks statistical power. Therefore, one should not blindly rely on the statistical tests, but also look at the effect sizes.
4. It is hard to draw definitive conclusions based on this analysis (this may not be the most satisfactory of situations, but far from uncommon in the life of a social scientist!). On the one hand, we find three medium-size correlations in the expected direction, giving some support to the group-conflict explanation for attitude change. On the other hand, the evidence is far from conclusive: the size of the correlations is not overwhelming, and there is no statistical basis for concluding that a relation is present 1.

Go to next page >>

#### Footnotes

• [1] In the study this module is based on (Meuleman et al., 2009), the correlation coefficients found were both stronger and statistically significant. Specifically, attitude change was found to be significantly (with alpha=0.10) related to changes in unemployment levels and immigration flows. These different findings are probably due to the fact that, in the article, the analysis was performed on latent means, while we worked with sum scales here. And, as explained in Chapter 2, the latent means provide purer measurements of attitude change, because they correct for the inequivalence in one of the three items.