Rpubs link: https://rpubs.com/rpubs11111

Relicated paper

Napier, J. L., & Jost, J. T. (2008). Why Are Conservatives Happier Than Liberals? Psychological Science, 19(6), 565–572, https://doi.org/10.1111/j.1467-9280.2008.02124.x

Workspace setup

Global r chunks setup:

library(knitr)
knitr::opts_chunk$set(
    echo = TRUE,
    fig.align = "center",
    message = FALSE,
    warning = FALSE,
    cache = TRUE,
    cache.lazy = FALSE,
    tidy = TRUE
)

Libraries:

library(haven)
library(plyr)
library(dplyr)
library(ordinal)
library(Hmisc)
library(readr)
library(misty)
library(ggeffects)
library(plotly)
library(sjPlot)
library(ggplot2)

1.Introduction

In their original article, Jost and Napier (2008) drew on the theoretical framework supplied by system-justification theory(see Jost and Banaji, 1994) in an attempt to explain why those holding conversative views are consistently found to be happier than their liberal counterparts (Taylor, Funk, & Craighill, 2006). System-justification theory suggests that certain ‘system-justifying’ ideologies hold a ‘palliative function’ for individuals, providing “a kind of ideological buffer against the negative hedonic consequences of social and economic inequality” (Jost and Napier, 2008: p. 565). The focus of the current replication is the authors’ highly original and insightful third study, which using data from the General Social Survey (1974-2004) reports that greater objective social inequality is associated with diminished life satisfaction, especially for liberals who “lack ideological rationalizations that would help them frame inequality in a positive (or at least neutral) light” (p. 571). In a multilevel regression analysis, Jost and Napier found that increasing economic inequality (measured with the Gini index) in the US was related to falling individual-level happiness, with liberals disproportionately affected compared to conservatives.

The main purpose of the current analysis is to address the methodological shortcomings present in the original study, which have consequently refigured in following replications. The most pressing of these is the treatment of an ordinal dependent variable (life satisfaction) as a continuous, numeric variable. The ‘cardinalization’ (Kalmijn et al 2010) of distinct response categories (“not very happy”, “pretty happy” and “very happy” ) assumes their equidistance; when placed into standard regression methods this results in several statistical disadvantages, such as a skewed distribution, a downward bias on estimates and the prediction of values outside the possible range of outcomes (see Agresti 2010; Hox 2010). Thus, by using the same data with a more statistically appropriate modelling technique, the current analysis will test the robustness of the original study, determining whether reported findings were “merely the result of analytic decisions” (Freese and Peterson, 2017: p. 152).

A further criticism of the author’s methodology is derived from Schlenker et al. (2012), who in their own replication, noticed that the Gini index and the year were interchangeable as indicators of the level of inequality, producing almost identical results when used in a model. Whilst US income inequality has been steadily growing over the last 40 years, in the original analysis it is uncertain whether it is a change in the level of inequality or long-term structural changes responsible for changes in happiness.

By extending the time period of the analysis to include the most recent wave of the GSS (until 2018), this replication will advance theoretical development, confronting the authors’ system-justification hypothesis with new evidence. The results will assess the generalizability of Jost and Napier’s findings across the decade following the financial crisis, whilst retaining an appropriate level of similarity with their statistical procedures.

2.Data and Methods

GSS <- read_sav("~/Desktop/SMI205/data.2/GSS7218_R3.sav")
gss1 <- GSS %>%
  filter(between(YEAR, 1974, 2018))
gss2 = subset(gss1, select = c(ID, YEAR,HAPPY, POLVIEWS, SEX, MARITAL, 
                                WRKSTAT, AGE, INCOME ,HEALTH, DEGREE))        
names(gss2) <- c('id', 'year', 'happy', 'polviews', 'sex', 'married', 
                  'emp', 'age', 'inc', 'health', 'edu')
gss <- na.omit(gss2)

gss$happy <- 4-gss$happy
gss$happy <- ordered(gss$happy, levels = 1:3,
                       labels = c("not too happy", "pretty happy", "very happy"))

gss$sex <- ifelse(gss$sex==2,1,0)
gss$married <- ifelse(gss$married==1,1,0)
gss$emp <- ifelse(gss$emp==4,1,0)
gss$health <- 5-gss$health
gss$age.g <- as.numeric(cut2(gss$age, g=9))
gss$age.sq <- gss$age^2
gss$inc.n <- cut(gss$inc,
                 breaks = c(-Inf, 8, 11, Inf), 
                 labels = NULL, ordered_result = TRUE)
gss$inc.n <- as.numeric(gss$inc.n)

range01 <- function(x){(x-min(x))/(max(x)-min(x))}

gss$age.g <- range01(gss$age.g)
gss$age.sq <- range01(gss$age.sq)
gss$inc.n <- range01(gss$inc.n)
gss$edu <- range01(gss$edu)
gss$health <- range01(gss$health)

l2data <- read_csv("~/Desktop/SMI205/data.2/Sheet 2-Table 1.csv")
gss3 <- cbind(gss, l2data)
gss3 = subset(gss3, select = c(id, year, happy, polviews, sex, married, emp,
                                age, health, edu, age.g, age.sq, 
                                inc.n, rate.unp, infl, gini, pres))

gss3$gini <- center(gss3$gini, type = "CGM",group = gss3$year)
gss3$rate.unp <- center(gss3$rate.unp, type = "CGM", group=gss3$year)
gss3$infl <- center(gss3$infl, type = "CGM", group = gss3$year)

v.dat = subset(gss3, select = c(year, happy, polviews))
v.dat$l.c <- ifelse(v.dat$polview > 4, "Conservative", 
                         ifelse(v.dat$polviews<4, "Liberal", "Neutral"))
v.dat$happy <- as.numeric(v.dat$happy)
v.dat <- v.dat[v.dat$l.c != "Neutral",]
v.dat2 <- aggregate(happy~year+l.c, data = v.dat, FUN = "mean")
v.dat2$gini <- c(0.395,0.397,0.398, 0.402, 0.403,0.412,0.415,0.419, 0.426, 0.426,
                 0.431, 0.428,0.428, 0.454, 0.456, 0.455, 0.456, 0.462, 0.462,
                 0.466, 0.47, 0.466, 0.470, 0.477, 0.480, 0.481, 0.486,0.395,
                 0.397,0.398, 0.402, 0.403,0.412,0.415,0.419, 0.426, 0.426,
                 0.431, 0.428,0.428, 0.454, 0.456, 0.455, 0.456, 0.462, 0.462,
                 0.466, 0.47, 0.466, 0.470, 0.477, 0.480, 0.481, 0.486 )
scl <- 4.975

ggplot(v.dat2, aes(x=year)) + 
  geom_line(aes(y=happy, color = l.c)) +
              geom_line(aes(y=gini*scl), linetype = "dashed", colour = "grey") +
  scale_y_continuous(name = "Mean Happiness Score", 
                     sec.axis= sec_axis(~./scl, name = "Gini")) +
  theme(legend.position = "top", plot.title = element_text(hjust = 0.5)) + 
  labs(title = "Figure 1: Liberal vs. Conservative Happiness Over Time",
  x = "Year", colour = "Political Orientation")

gss3$happy.cnt <- as.numeric(gss3$happy)
cor.dat <- as.data.frame(ddply(gss3, .(year), summarise, "corr" = cor(polviews, happy.cnt), 
                                             method = "spearman"))
ggplot(cor.dat, aes(x=year, y=corr)) + 
  geom_area(fill = "blue", alpha = .2) + 
  geom_line() + theme(plot.title = element_text(hjust = 0.5)) +
  labs(title = "Figure 2: Correlation Between Ideology and Happiness Over Time",
       x = "Year", y = "Spearman Correlation Coefficient")

#Control model
mod.0 <- clm(happy ~ polviews + sex + married + health + emp + age.g + age.sq + 
               inc.n + edu, data = gss3)
#Random-intercept model with individual-level predictors
mod.1 <- clmm(happy~polviews + sex + married + health + emp + age.g + age.sq + 
                inc.n + edu + (1|year), data = gss3)
anova(mod.0, mod.1)

## Likelihood ratio tests of cumulative link models:
##  
##       formula:                                                                                   
## mod.0 happy ~ polviews + sex + married + health + emp + age.g + age.sq + inc.n + edu             
## mod.1 happy ~ polviews + sex + married + health + emp + age.g + age.sq + inc.n + edu + (1 | year)
##       link: threshold:
## mod.0 logit flexible  
## mod.1 logit flexible  
## 
##       no.par   AIC logLik LR.stat df Pr(>Chisq)    
## mod.0     11 62566 -31272                          
## mod.1     12 62535 -31256   32.81  1  1.016e-08 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

3. Results

Table 1 displays the results of the final model, with the coefficients in the form of odds ratios (values calculated when all other predictors held constant at their means). Individual-level political conservatism was found to be positively related to happiness, with an odds ratio of 1.04: for a 1 unit shift to the right on the 7-point ideological scale, a respondent was 4% more likely to belong to a more happy response category. The level of inequality was found to have a strong negative association with individual happiness (OR = 0.05, p=0.050). The interaction between political conservatism and the level of inequality (OR = 1.08) was found to be statistically insignificant (p-value of 0.780).

library(ordinal)
#Final model
mod.fin <- clmm(happy~polviews + sex + married + health + emp + age.g + age.sq + 
                inc.n + edu + rate.unp + infl + gini + pres + polviews*gini +
                (1|year), data = gss3)
tab_model(mod.fin,
          pred.labels = c("Not very happy|Pretty happy", "Pretty happy|Very happy",
                          "Political Conservatism","Sex (Male)", "Married", "Health",
                          "Employment Status",
                          "Age", "Age Squared", "Income", "Education","Unemployment Rate",
                          "Inflation",
                          "Gini", "President", "Conservatism*Gini"),
          string.ci = "C.I. (95%)",
          title = "Table 1: CLM Multilevel Regression Analysis of Self-Reported Happiness")

#Calculating predicted probabilities

pred <- ggpredict(mod.fin, c("polviews", "gini[-0.04,0,0.04]"), x.as.factor = T)
pred <- as.data.frame(pred)

pred$Response <- factor(pred$response.level, levels = c("1","2","3"),
                              labels = c("Not Very Happy", "Pretty Happy",
                                         "Very Happy"))
pred$Gini <- factor(pred$group, levels = c("-0.04", "0", "0.04"),
                     labels = c("Low Inq.", "Average Inq.", "High Inq."))

ggplotly(ggplot(pred, aes(x=x, y=predicted, colour = Response)) +
  geom_line() + geom_ribbon(aes(ymin = conf.low, ymax = conf.high), 
                            alpha = 0.1,linetype = 2) + 
  facet_grid(~Gini) + theme_bw() +
  labs(title = "Figure 3: Predicted Probabilities of Happiness Responses",
       x = "Political Orientation (Lib - Cons)", y = "Probability"))

4. Conclusions

This is not to say that the current results amount to the falsification of Jost and Napier’s original conclusion that liberals are more affected by inequality. However, the lack of a significant interaction term between political orientation and the Gini index, situates the current analysis among replication attempts (e.g. Stavrova and Luhmann, 2016), that when including more recent GSS waves (2010-) have failed to find a significant mediating effect of inequality on ideology. As corroborated by the sharp fall in the strength of the correlation between ideology and happiness post 2008 (Figure 2), including evidence from the last decade generally constrains rather than bolsters the generalizability of the original study’s conclusions. Whilst the replication was successful in verifying the continued existence of the ideological gap in happiness, the results signal the need for further theoretical or methodological innovation to understand the psychological mechanisms behind such a gap.

Questions remain as to the suitability of the Gini index as an accurate indicator of actual inequality; both its strong correlation with time as well as a recomposition of income inequality away from income towards wealth in the US following the financial crash (Horowitz et al,. 2020) suggest that future research should look for an alternative indicator.

5. References

Agresti A., Kateri M. (2011) Categorical Data Analysis. In: Lovric M. (eds) International Encyclopedia of Statistical Science. Springer, Berlin, Heidelberg. https://doi.org/10.1007/978-3-642-04898-2_161

Christensen, R. (2012). Ordinal: Regression Models for Ordinal Data. R Package Version 2011.08-11.. 2013.

Freese, J., & Peterson, D. (2017). Replication in social science. Annual Review of Sociology, 43, 147-165.

Horowitz, J., Igielnik, R., & Kochhar, R., (2020). ‘Most americans say there is too much economic inequality in the U.S., but fewe than half call it a top priority’. Retreived from [https://www.pewresearch.org/social-trends/2020/01/09/trends-in-income-and-wealth-inequality/#fn-27661-13]

Hox, J. J. (2010). Multilevel analysis: Techniques and applications (2nd ed.). Routledge/Taylor & Francis Group.

Jost, J.T., & Banaji, M.R. (1994). The role of stereotyping in systemjustification and the production of false consciousness. British Journal of Social Psychology, 33, 1–27.

Kalmijn, W & Arends, L. (2010). Measures of Inequality: Application to Happiness in Nations. Social indicators research. 99. 147-162. 10.1007/s11205-010-9573-z.

Schlenker, B. R., Chambers, J. R., & Le, B. M. (2012). Conservatives are happier than liberals, but why? Political ideology, personality, and life satisfaction. Journal of Research in Personality, 46(2), 127-146.

Smith, Tom W., Davern, Michael, Freese, Jeremy, and Morgan, Stephen, General Social Surveys, 1972-2018 [machine-readable data file] /Principal Investigator, Smith, Tom W.; Co-Principal Investigators, Michael Davern, Jeremy Freese, and Stephen Morgan; Sponsored by National Science Foundation. –NORC ed.– Chicago: NORC, 2018: NORC at the University of Chicago [producer and distributor]. Data accessed from the GSS Data Explorer website at gssdataexplorer.norc.org.]

Stavrova, O., & Luhmann, M. (2016). Are conservatives happier than liberals? Not always and not everywhere. Journal of Research in Personality, 63, 29-35. https://doi.org/10.1016/j.jrp.2016.04.011

Taylor, P., Funk, C., & Craighill, P. (2006). Are we happy yet? Retrieved August 19, 2007, from the Pew Research Center [https:// pewresearch.org/assets/social/pdf/AreWeHappyYet]

U.S. Census Bureau (2020). Historical Income Tables: Households, Retrieved from [https://www.census.gov/data/tables/time-series/demo/income-poverty/historical-income-households.html].

World Bank, Development Research Group. Gini index (World Bank estimate), Retrieved from [https://data.worldbank.org/indicator/SI.POV.GINI]