Is Income Inequality Related to Tolerance for Inequality?

For general information on the ISSP, see the webpage http://​www.​issp.​org. For information on the administration of the national surveys and their response rates, see Gendall (2011).

The ISSP also asks about the fair wage for a cabinet minister. I do not use this “fair” wage estimate as an indicator of a typical high salary, as it would introduce a distinction between private and public sector employees. However, I repeated all analyses by measuring “fair high incomes” not by “(fair income chairman + fair income doctor)/2,” but by “(fair income chairman + fair income doctor + fair income minister)/3.” This does not change the results presented below. The ISSP also asks what a shop assistant should earn. I did not use this as a proxy for people’s estimate of a fair low wage, because the item does not exist in the ISSP 1987, which would make it impossible to compare the 1987-ISSP wave to later waves. However, where data are available, I repeated all analyses by calculating fair low incomes by “(fair income unskilled worker + income shop assistant)/2.” This also does not change the results below significantly. Additionally, what people say is a fair income for an unskilled worker and a shop assistant is highly correlated in the first place (r = 0.986, p < 0.001).

I have also used the polity2 variable of the Polity IV dataset, which similarly measures how democratic a country is (for details, cf. Marshall, Jaggers and Gurr 2011: 16f.), but results are very similar. I therefore use Freedom House data, as it covers more countries.

These are Australia, Austria, Belgium, Chile, Czech Republic, Denmark, Estonia, Finland, France, Germany, Hungary, Iceland, Italy, Japan, Norway, Poland, Portugal, Slovakia, Slovenia, South Korea, Sweden, Switzerland, Turkey, the UK and the USA.

I have computed all regressions using maximum likelihood and restricted maximum likelihood models. The AIC for maximum likelihood models consistently outperformed restricted maximum likelihood models, so I use maximum likelihood models throughout, which is also appropriate due to the relatively large sample size.

Note that most figures have logged y-scales, as the “fair income variable” is logged for the reasons mentioned in the data section. Taking the geometric mean of non-logged values yields essentially similar results, while circumventing a logged scale. But the logged scale is more appropriate for the reasons mentioned in the data section.

I have classified Denmark, Finland, Norway and Sweden as social democratic; Austria, Belgium, France, Germany and Italy are conservative; Australia, the UK and the USA are liberal; Chile, the Czech Republic, Estonia, Hungary, Iceland, Japan, Poland, Portugal, Slovakia, Slovenia, South Korea, Switzerland and Turkey are classified as a fourth category of unclear cases.

Note that the ISSP is not an individual-level panel dataset. One can make it a country-level panel dataset, however, as countries have participated in more than one wave. I therefore pool data on the country-level and use multilevel regressions that cluster people in both countries and years.

Appendix Fig. 2 shows the inequality that an average respondent in each country and year tolerates. Note that the scale is adjusted for each country, as changes within countries over time are more important than differences between countries. Also, note that Appendix Fig. 2 does not use logged y-scales, because this would have made it impossible to display all graphs in one figure. However, results look virtually the same when using a logged scale.

Note that the y-axis is again log-scaled. In this case, this makes sense because a change from 10 to 20% more tolerance for inequality is more consequential than, for example, a change from 80 to 90%, although both constitute an increment of 10.

For the following cases, there are measures of inequality for four years before and after a single time point: Australia (1992, 1999), Austria (1999), Bulgaria (1992, 1999), Chile (1999), Cyprus (1999), Czech Republic (1999), France (1999), Germany (1992, 1999), Hungary (1992, 1999), Israel (1999), Italy (1992,), Japan (1999), Latvia (1999), New Zealand (1992, 1999), Norway (1992, 1999), Philippines (1992, 1999), Poland (1992, 1999), Portugal (1999), Russian Federation (1992, 1999), Slovak Republic (1992, 1999), Slovenia (1992, 1999), Spain (1999), Sweden (1992, 1999), the UK (1992, 1999) and the USA (1992, 1999).

I could lag average tolerated inequality of the last wave in each country. But then neither inequality, nor lagged tolerated inequality would vary between individuals of a country and ISSP wave. It would then make no sense to use a multilevel regression.