I. Data & Codes
a. Data & libraries
b. Compose variables
# continuous party
d$partyCont <- NA
d$partyCont[d$demStrength == 1] <- -3
d$partyCont[d$demStrength == 2] <- -2
d$partyCont[d$partyClose == 1] <- -1
d$partyCont[d$partyClose == 3] <- 0
d$partyCont[d$repStrength == 1] <- 3
d$partyCont[d$repStrength == 2] <- 2
d$partyCont[d$partyClose == 2] <- 1
# policy group proposer
d$Policy_Group <- NA
d$Policy_Group[d$FL_71_DO == 'proportional_dem' | d$FL_71_DO == 'US1st_dem'] <- "Democratic"
d$Policy_Group[d$FL_71_DO == 'proportional_rep' | d$FL_71_DO == 'US1st_rep'] <- "Republican"
d$Policy_Group[d$FL_71_DO == 'proportional_bipart' | d$FL_71_DO == 'US1st_bipart'] <- "Bipartisan"
d$Policy_Group[d$FL_71_DO == 'proportional_expert' | d$FL_71_DO == 'US1st_expert'] <- "Expert"
d$Policy_Group <- factor(d$Policy_Group, levels = c('Democratic', 'Republican', 'Bipartisan','Expert'))
# policy frame
d$Policy_Frame <- NA
d$Policy_Frame[d$FL_71_DO == 'proportional_dem' | d$FL_71_DO == 'proportional_rep' | d$FL_71_DO == 'proportional_bipart' | d$FL_71_DO == 'proportional_expert'] <- "Proportional"
d$Policy_Frame[d$FL_71_DO == 'US1st_dem' | d$FL_71_DO == 'US1st_rep' | d$FL_71_DO == 'US1st_bipart' | d$FL_71_DO == 'US1st_expert'] <- "US1st"
# party factor
d$party_factor <- NA
d$party_factor[d$partyCont < 0] <- 'Democrat'
d$party_factor[d$partyCont == 0] <- 'Independent'
d$party_factor[d$partyCont > 0] <- 'Republican'
## Order of timing var
d$election_timing <- factor(d$election_timing, levels = c('Pre-election', 'During-election','Post-election'))
d$party_factor <- factor(d$party_factor, levels = c('Democrat', 'Republican','Independent'))
d$anger <- d$emotion_1
## Only Pre-election!
d <- d[d$election_timing != 'Pre-election',]c. Codes & indices
### Partisan framing codes
# Code 1: Left vs. Right
d$fDem_Rep <- NA
d$fDem_Rep[d$Policy_Group == 'Democratic'] <- -.5
d$fDem_Rep[d$Policy_Group == 'Republican'] <- .5
d$fDem_Rep[d$Policy_Group == 'Bipartisan' | d$Policy_Group == 'Expert'] <- 0
# Code 2: Bi vs. Exp
d$fBi_Exp <- NA
d$fBi_Exp[d$Policy_Group == 'Democratic' | d$Policy_Group == 'Republican'] <- 0
d$fBi_Exp[d$Policy_Group == 'Bipartisan'] <- -.5
d$fBi_Exp[d$Policy_Group == 'Expert'] <- .5
# Code 3 = Dem and Rep vs. Exp and Bi
d$fParties_BiExp <- NA
d$fParties_BiExp[d$Policy_Group == 'Democratic' | d$Policy_Group == 'Republican'] <- -.5
d$fParties_BiExp[d$Policy_Group == 'Bipartisan' | d$Policy_Group == 'Expert'] <- .5
### Timing codes
## Contrast
d$tDur_Post <- NA
d$tDur_Post[d$election_timing == 'During-election'] <- -.5
d$tDur_Post[d$election_timing == 'Post-election'] <- .5
## Dummy
# During!
d$tDur <- NA
d$tDur[d$election_timing == 'During-election'] <- 0
d$tDur[d$election_timing == 'Post-election'] <- 1
# Post!
d$tPost <- NA
d$tPost[d$election_timing == 'During-election'] <- 1
d$tPost[d$election_timing == 'Post-election'] <- 0
### Party Factor
## Contrasts
d$pDem_Rep <- NA
d$pDem_Rep[d$party_factor == 'Democrat'] <- -.5
d$pDem_Rep[d$party_factor == 'Independent'] <- 0
d$pDem_Rep[d$party_factor == 'Republican'] <- .5
d$pInd_Not <- NA
d$pInd_Not[d$party_factor == 'Democrat'] <- .33
d$pInd_Not[d$party_factor == 'Independent'] <- -.67
d$pInd_Not[d$party_factor == 'Republican'] <- .33
## Dummy codes
d$pDemR[d$party_factor == 'Democrat'] <- 0
d$pDemR[d$party_factor == 'Republican'] <- 1
d$pDemR[d$party_factor == 'Independent'] <- 0
d$pDemI[d$party_factor == 'Democrat'] <- 0
d$pDemI[d$party_factor == 'Republican'] <- 0
d$pDemI[d$party_factor == 'Independent'] <- 1
d$pRepD[d$party_factor == 'Democrat'] <- 1
d$pRepD[d$party_factor == 'Republican'] <- 0
d$pRepD[d$party_factor == 'Independent'] <- 0
d$pRepI[d$party_factor == 'Democrat'] <- 0
d$pRepI[d$party_factor == 'Republican'] <- 0
d$pRepI[d$party_factor == 'Independent'] <- 1
d$pIndD[d$party_factor == 'Democrat'] <- 1
d$pIndD[d$party_factor == 'Republican'] <- 0
d$pIndD[d$party_factor == 'Independent'] <- 0
d$pIndR[d$party_factor == 'Democrat'] <- 0
d$pIndR[d$party_factor == 'Republican'] <- 1
d$pIndR[d$party_factor == 'Independent'] <- 0
### centering
d$ownvote.c <- d$ownvote_conf - mean(d$ownvote_conf, na.rm = T)
d$overallvote.c <- d$overallvote_conf - mean(d$overallvote_conf, na.rm = T)
# Media Measures - All, including Fox
d$mediaExposure_all <- rowMeans(d[,c("mediaExposure_1","mediaExposure_2","mediaExposure_3","mediaExposure_4","mediaExposure_5","mediaExposure_6","mediaExposure_7","mediaExposure_8","mediaExposure_9","mediaExposure_10","mediaExposure_11","mediaExposure_12","mediaExposure_13","mediaExposure_14","mediaExposure_15")], na.rm = T)
#psych::alpha(d[,c("mediaExposure_1","mediaExposure_2","mediaExposure_3","mediaExposure_4","mediaExposure_6","mediaExposure_7","mediaExposure_8","mediaExposure_9","mediaExposure_10","mediaExposure_11","mediaExposure_12","mediaExposure_13","mediaExposure_14","mediaExposure_15", "mediaExposure_5")])
# alpha is .92
d$mediaTrust_all <- rowMeans(d[,c("mediaTrust_1","mediaTrust_2","mediaTrust_3","mediaTrust_4","mediaTrust_5","mediaTrust_6","mediaTrust_7","mediaTrust_8","mediaTrust_9","mediaTrust_10","mediaTrust_11","mediaTrust_12","mediaTrust_13","mediaTrust_14","mediaTrust_15")], na.rm = T)
#psych::alpha(d[,c("mediaTrust_1","mediaTrust_2","mediaTrust_3","mediaTrust_4","mediaTrust_6","mediaTrust_7","mediaTrust_8","mediaTrust_9","mediaTrust_10","mediaTrust_11","mediaTrust_12","mediaTrust_13","mediaTrust_14","mediaTrust_15", "mediaTrust_5")])
# asks if mediaTrust_5 was reverse scored; alpha still .95
# Media Measures - Excluding Fox
d$mediaExposure <- rowMeans(d[,c("mediaExposure_1","mediaExposure_2","mediaExposure_3","mediaExposure_4","mediaExposure_6","mediaExposure_7","mediaExposure_8","mediaExposure_9","mediaExposure_10","mediaExposure_11","mediaExposure_12","mediaExposure_13","mediaExposure_14","mediaExposure_15")], na.rm = T)
#psych::alpha(d[,c("mediaExposure_1","mediaExposure_2","mediaExposure_3","mediaExposure_4","mediaExposure_6","mediaExposure_7","mediaExposure_8","mediaExposure_9","mediaExposure_10","mediaExposure_11","mediaExposure_12","mediaExposure_13","mediaExposure_14","mediaExposure_15")])
# alpha is .93
d$mediaTrust <- rowMeans(d[,c("mediaTrust_1","mediaTrust_2","mediaTrust_3","mediaTrust_4","mediaTrust_6","mediaTrust_7","mediaTrust_8","mediaTrust_9","mediaTrust_10","mediaTrust_11","mediaTrust_12","mediaTrust_13","mediaTrust_14","mediaTrust_15")], na.rm = T)
#psych::alpha(d[,c("mediaTrust_1","mediaTrust_2","mediaTrust_3","mediaTrust_4","mediaTrust_6","mediaTrust_7","mediaTrust_8","mediaTrust_9","mediaTrust_10","mediaTrust_11","mediaTrust_12","mediaTrust_13","mediaTrust_14","mediaTrust_15")])
# alpha is .97
#psych::alpha(d[,c("mediaTrust_1","mediaTrust_2","mediaTrust_3","mediaTrust_4","mediaTrust_6","mediaTrust_7","mediaTrust_8","mediaTrust_9","mediaTrust_10","mediaTrust_11","mediaTrust_12","mediaTrust_13","mediaTrust_14","mediaTrust_15","mediaExposure_1","mediaExposure_2","mediaExposure_3","mediaExposure_4","mediaExposure_6","mediaExposure_7","mediaExposure_8","mediaExposure_9","mediaExposure_10","mediaExposure_11","mediaExposure_12","mediaExposure_13","mediaExposure_14","mediaExposure_15")])
# alpha is .96
### Composite Media variables
#average own and overall vote
d$voteLegit <- (d$ownvote_conf + d$overallvote_conf)/2
d$voteLegit.c <- d$voteLegit - mean(d$voteLegit, na.rm = T)
#recenter media/fox trust to match media/fox exposure scale
d$mediaTrust5 <- d$mediaTrust_5 + 3
d$mediaTrust <- d$mediaTrust + 3
d$foxPerception <- (d$mediaExposure_5 + d$mediaTrust5)/2
d$mediaPerception <- (d$mediaExposure + d$mediaTrust)/2
d$foxPerception.c <- d$foxPerception - mean(d$foxPerception, na.rm = T)
d$mediaPerception.c <- d$mediaPerception - mean(d$mediaPerception, na.rm = T)
### Indices
During <- d$election_timing == 'During-election'
Post <- d$election_timing == 'Post-election'
# mean center fox exposure
d$foxExposure.c <- d$mediaExposure_5 - mean(d$mediaExposure_5, na.rm = T)
# mean center other media exposure
d$otherMediaExposure.c <- d$mediaExposure - mean(d$mediaExposure, na.rm = T)
#create difference score
d$diffFoxOther <- d$mediaExposure_5 - d$mediaExposure
#reverse code biden to trump rating win --> trump definitely win = 1; biden definitely win = 9
d$electPredict_T_B2 <- 10 - d$electPredict_B_T
d$electPredict_T_B3 <- 10 - d$electPredict_B_T.1
#combine to make 1 column
d$electPredictTB <- ifelse(!is.na(d$electPredict_T_B), d$electPredict_T_B,
ifelse(!is.na(d$electPredict_T_B2), d$electPredict_T_B2,
ifelse(!is.na(d$electPredict_T_B3), d$electPredict_T_B3, NA)))
d$electPredictTB.plot <- d$electPredictTB - 5 (1) questions from Leaf on 2/10/2021
Regression estimating anger from partisanship and media exposure
m <- lm(anger ~ (pDem_Rep + pInd_Not) * foxPerception.c * mediaPerception.c, data = d)
summary(m)##
## Call:
## lm(formula = anger ~ (pDem_Rep + pInd_Not) * foxPerception.c *
## mediaPerception.c, data = d)
##
## Residuals:
## Min 1Q Median 3Q Max
## -3.4902 -1.7239 -0.3375 1.5668 5.7631
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 2.82527 0.08045 35.119 < 2e-16
## pDem_Rep 0.15957 0.17963 0.888 0.3745
## pInd_Not 0.38765 0.18506 2.095 0.0364
## foxPerception.c -0.02642 0.07388 -0.358 0.7207
## mediaPerception.c -0.08416 0.09644 -0.873 0.3830
## pDem_Rep:foxPerception.c 0.62435 0.13768 4.535 6.34e-06
## pInd_Not:foxPerception.c 0.16776 0.18796 0.893 0.3723
## pDem_Rep:mediaPerception.c -1.05898 0.19825 -5.342 1.10e-07
## pInd_Not:mediaPerception.c -0.29482 0.23394 -1.260 0.2078
## foxPerception.c:mediaPerception.c 0.13952 0.06374 2.189 0.0288
## pDem_Rep:foxPerception.c:mediaPerception.c 0.02073 0.13271 0.156 0.8759
## pInd_Not:foxPerception.c:mediaPerception.c -0.07782 0.15350 -0.507 0.6123
##
## (Intercept) ***
## pDem_Rep
## pInd_Not *
## foxPerception.c
## mediaPerception.c
## pDem_Rep:foxPerception.c ***
## pInd_Not:foxPerception.c
## pDem_Rep:mediaPerception.c ***
## pInd_Not:mediaPerception.c
## foxPerception.c:mediaPerception.c *
## pDem_Rep:foxPerception.c:mediaPerception.c
## pInd_Not:foxPerception.c:mediaPerception.c
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 2.019 on 1201 degrees of freedom
## (29 observations deleted due to missingness)
## Multiple R-squared: 0.06126, Adjusted R-squared: 0.05266
## F-statistic: 7.125 on 11 and 1201 DF, p-value: 7.949e-12b. Is media exposure associated with anger?
No
c. Do polarized emotions get smaller after accounting for media? (i.e., Is there an interaction between partisanship and media exposure on anger?)
There is an interaction between partisanship (Dem vs. Rep) and media exposure/trust (p < .001) as well as partisanship (Dem vs. Rep) and Fox exposure/trust (p < .001).
d. Graph
i.
media.plot <- ggplot(d[!is.na(d$party_factor),], aes(x = mediaPerception, y = anger)) +
geom_smooth(method = 'lm', aes(x = mediaPerception, y = anger, col = "darkgreen")) +
facet_wrap(~party_factor)
media.plot + theme_classic() +
xlab("Media Trust/Exposure") +
ylab("Anger") +
theme(legend.position = c(1000,1000)) +
coord_cartesian(ylim = c(1, 7)) +
scale_y_continuous(breaks = c(1,2,3,4,5,6,7)) +
scale_color_manual(values = c("darkgreen"))## `geom_smooth()` using formula 'y ~ x'## Warning: Removed 15 rows containing non-finite values (stat_smooth).
ii.
fox.plot <- ggplot(d[!is.na(d$party_factor),], aes(x = foxPerception, y = anger)) +
geom_smooth(method = 'lm', aes(x = foxPerception, y = anger, col = "darkred")) +
facet_wrap(~party_factor)
fox.plot + theme_classic() +
xlab("Fox Trust/Exposure") +
ylab("Anger") +
theme(legend.position = c(1000,1000)) +
coord_cartesian(ylim = c(1, 7)) +
scale_y_continuous(breaks = c(1,2,3,4,5,6,7)) +
scale_color_manual(values = c("darkred"))## `geom_smooth()` using formula 'y ~ x'## Warning: Removed 15 rows containing non-finite values (stat_smooth).
(2) Mediation Analyses - Baron & Kenny approach
1. Y ~ X
angermed.i <- lm(voteLegit ~ (pDem_Rep + pInd_Not), data = d)
summary(angermed.i)##
## Call:
## lm(formula = voteLegit ~ (pDem_Rep + pInd_Not), data = d)
##
## Residuals:
## Min 1Q Median 3Q Max
## -3.1667 -0.8920 0.1080 0.8333 2.1080
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 3.34407 0.03644 91.78 < 2e-16 ***
## pDem_Rep -1.27466 0.07151 -17.82 < 2e-16 ***
## pInd_Not 0.56142 0.09055 6.20 7.73e-10 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.138 on 1205 degrees of freedom
## (34 observations deleted due to missingness)
## Multiple R-squared: 0.2324, Adjusted R-squared: 0.2311
## F-statistic: 182.4 on 2 and 1205 DF, p-value: < 2.2e-16Yes, p < .001
2. Z ~ X
angermed.ii <- lm(anger ~ (pDem_Rep + pInd_Not), data = d)
summary(angermed.ii)##
## Call:
## lm(formula = anger ~ (pDem_Rep + pInd_Not), data = d)
##
## Residuals:
## Min 1Q Median 3Q Max
## -2.5582 -1.6649 -0.5582 1.4418 4.3351
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 3.11949 0.06528 47.785 < 2e-16 ***
## pDem_Rep 0.42962 0.12880 3.336 0.000877 ***
## pInd_Not 0.67846 0.16182 4.193 2.96e-05 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 2.052 on 1212 degrees of freedom
## (27 observations deleted due to missingness)
## Multiple R-squared: 0.02234, Adjusted R-squared: 0.02073
## F-statistic: 13.85 on 2 and 1212 DF, p-value: 1.129e-06Yes, p < .001
3. Y ~ Z + X
angermed.iii <- lm(voteLegit ~ (pDem_Rep + pInd_Not) + anger, data = d)
summary(angermed.iii)##
## Call:
## lm(formula = voteLegit ~ (pDem_Rep + pInd_Not) + anger, data = d)
##
## Residuals:
## Min 1Q Median 3Q Max
## -3.3962 -0.8446 0.1415 0.8189 2.4980
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 3.67997 0.06055 60.775 < 2e-16 ***
## pDem_Rep -1.22895 0.07049 -17.433 < 2e-16 ***
## pInd_Not 0.63440 0.08949 7.089 2.30e-12 ***
## anger -0.10756 0.01565 -6.874 9.99e-12 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.116 on 1204 degrees of freedom
## (34 observations deleted due to missingness)
## Multiple R-squared: 0.2613, Adjusted R-squared: 0.2595
## F-statistic: 142 on 3 and 1204 DF, p-value: < 2.2e-16No different – b = -1.28 in the Y ~ X model, b = -1.23 in the model including the mediator
4. Mediation function!
d$anger <- d$emotion_1
angermed <- mediate(y = "voteLegit", x = c("pDem_Rep","pInd_Not"), m = "anger", data = d, n.obs = 1242, n.iter = 100)
summary(angermed)## Call: mediate(y = "voteLegit", x = c("pDem_Rep", "pInd_Not"), m = "anger",
## data = d, n.obs = 1242, n.iter = 100)
##
## Direct effect estimates (traditional regression) (c')
## voteLegit se t df Prob
## Intercept 3.68 0.06 61.71 1238 0.00e+00
## pDem_Rep -1.23 0.07 -17.71 1238 1.01e-62
## pInd_Not 0.62 0.09 7.10 1238 2.11e-12
## anger -0.11 0.02 -6.94 1238 6.24e-12
##
## R = 0.51 R2 = 0.26 F = 146.09 on 3 and 1238 DF p-value: 4.81e-81
##
## Total effect estimates (c)
## voteLegit se t df Prob
## Intercept 3.35 0.04 93.63 1239 0.00e+00
## pDem_Rep -1.28 0.07 -18.11 1239 3.04e-65
## pInd_Not 0.55 0.09 6.20 1239 7.75e-10
##
## 'a' effect estimates
## anger se t df Prob
## Intercept 3.12 0.06 48.42 1239 5.54e-288
## pDem_Rep 0.43 0.13 3.38 1239 7.47e-04
## pInd_Not 0.68 0.16 4.24 1239 2.45e-05
##
## 'b' effect estimates
## voteLegit se t df Prob
## anger -0.11 0.02 -6.94 1239 6.12e-12
##
## 'ab' effect estimates (through mediators)
## voteLegit boot sd lower upper
## pDem_Rep -0.05 -0.05 0.00 -0.05 -0.05
## pInd_Not -0.07 -0.06 0.01 -0.07 -0.04