Should we care about - PowerPoint Presentation

1. Should we care about pay ratios?Renzo CarrieroDepartment of Cultures, Politics and SocietyUniversity of Torino – Italy& Collegio Carlo Alberto, Torino1st ISSP User Conference – Social Inequality – 12 December 2022

2. MotivationQuestions on actual and ideal wages of various occupations are distinctive features of ISSP Social Inequality modulesThey represent meaningful ways to quantify respondents’ perceived and tolerated wage inequality, by (ex-post) computing pay ratiosHowever, recent research (Pedersen & Mutz 2019; Heiserman & Simpson 2021) raised heavy criticism to pay ratiosQ1: Is it possible to salvage pay ratios for modelling individual attitudes toward income inequality? Q2: How do pay ratios fare compared to standard word-based items about inequality?

3. Criticism to pay ratios Pedersen & Mutz (2019): pay ratios suffer fromAnchoring bias: people guess actual wages and anchor ideal wages to their prior guessesRatio bias: people tend to perceive ratios as larger when they are expressed in terms of large numbers (eg.: 100,000 to 1,000) rather than small numbers (100 to 1)Heiserman & Simpson (2021): pay ratios suffer from(Partial) Inconsistency with word-based (Likert-type) inequality attitudes scaleLow correlation with inequality attitudes scale (i.e. low concomitant validity)Low correlation with political affiliation (i.e. low construct validity)

4. Research questionsIs it possible to handle actual and ideal pay ratios in such a way that they provide meaningful and accurate measures of individuals’ attitudes toward wage inequality? How should we adjust ideal pay ratios in order to control for anchoring bias?Should we replace ideal pay ratios with word-based inequality attitudes when we investigate about individual and contextual determinants?Do individual party preferences and class location predict (adjusted) ideal pay ratios worse than word-based inequality attitudes? Do country differences in (adjusted) ideal pay ratios match differences in word-based inequality attitudes?

5. Data and measuresISSP 2019 Social Inequality module Actual pay ratio: ln(estimated wage ratio) between chairman of a large national corporation and unskilled factory workerIdeal pay ratio: ln(«should earn» wage ratio)  main DV (with several adjustments)Attitudes toward (i.e. evaluation of) income inequality is the alternative DV, a composite index of 3 items:Differences in income in [COUNTRY] are too large How do you feel when you think about differences in wealth between the rich and the poor in [COUNTRY]? (0-not angry at all / 10-extremely angry)How fair or unfair do you think the income distribution is in [COUNTRY]?

6. Data and measuresIndividual predictors: Party voted in last election, recoded as left, centre, right, other/not declaredClass: higher grade service class, lower grade service class, small business owners, skilled workers, unskilled workers (Oesch’s schema via iscogen)Controls:AgeSexEducation levelSelected 8 countries: Denmark, Finland, France, Sweden, Switzerland, USA, Australia, New Zealand (lowest missingness on actual and ideal pay ratios & highest Cronbach’s alpha on income inequality attitudes index)

7. Dealing with anchoring biasAcross the 8 countries, 22% of variance in perceived actual pay ratios explained by country dummies alone  national context is a likely source of anchoringActual and ideal pay ratios are highly correlated: 0.61 on average (cross-country range: 0.44-0.71)Need to adjust ideal pay ratios to perceived actual pay ratios. Three options:«Individual adjustment»: computing ln(ideal/actual pay ratio), «Regression adjustment»: controlling for ln(actual pay ratio) in regression models«Prediction adjustment»: computing residuals from the regression of ideal ratio on actual ratio in each country

8. Analysis planComparison of party and class coefficients from regressions of five DVs in each country: unadjusted ideal pay ratio«individual adjusted» pay ratio (ideal/actual ratio)«regression adjusted» pay ratio (ideal ratio controlling for actual ratio)«prediction adjusted» pay ratio (residuals)Income inequality attitudesFor this analysis, all DVs are within-country standardized to allow easy comparison of coefficients across modelsPooled data analysis: regressions of the 5 DVs (not standardized) on individual predictors and country dummies

9. Summary of results: single country analysesRegressions of ideal pay ratios (several adjustments):Party voted is best correlated with prediction adj. ideal ratiosClass is mostly not significant, whatever the adjustment, except for SE, FI, CHRegressions of attitudes towards income inequality:Party voted is strongly correlated with income inequality attitudes, effects are at least twice as large as the effects on pay ratiosClass moderatly correlated with attitudes, but only in SE, FI and CH

10. Coeff. from pay ratios regressions (pooled data)Party differences do not change muchClass differences are mostly insignificant when using individual adj. ratioCountry differences change a lot across different adjustmentsIn particular: cross-country variance completely explained out when using prediction adj. ratio. Country-level variance is absorbed by the predicted effect of actual ratio on ideal ratio

11. Coeff. from attitudes regressions (pooled data)Country differences in attitudes do not match the patterns obtained with any version of ideal pay ratios, though overall more similar to individual adj. ratios

12. ConclusionsTo investigate party differences, «prediction adjusted» ideal pay ratios are the safest choiceTo investigate class differences, ideal pay ratios, however adjusted, do not perform well because class matters only in a few countriesCountry comparisons on ideal pay ratios may be tricky and very sensitive to adjustments. Differences in pay ratios across countries do not match differences in attitudesParty voted and class more strongly associated with word-based income inequality attitudes  Should we abandon pay ratios? Maybe yes, but…Word-based items do not reveal the magnitude of inequality that respondents are willing to tolerateTo solve the anchoring problem, a solution is to offer the same anchors to all respondents (see Pedersen & Mutz)

13. Thank you!renzo.carriero@unito.it

14. Party coefficients from pay ratios regressionsParty voted best correlated with prediction adj. ideal ratios

15. Class coefficients from pay ratios regressionsClass mostly not significant, whatever the adjustment, except for SE, FI, CH

16. Party and class coeff. from attitudes regressionsParty voted strongly correlated with income inequality attitudesClass moderatly correlated with attitudes only in SE, FI and CH