--- title: "Study 3 - original analysis" author: "Matthew Leitao" date: "2/15/2022" output: html_document: #code_folding: hide df_print: paged toc: yes toc_float: true toc_depth: 5 toc_collapse: False editor_options: chunk_output_type: inline --- ```{r setup, include=FALSE} knitr::opts_chunk$set(echo = TRUE) ``` ```{r loading packages} if(!require(haven)) {install.packages("haven")} if(!require(tidyverse)) {install.packages("tidyverse")} if(!require(psych)) {install.packages("psych")} if(!require(ggplot2)) {install.packages("ggplot2")} if(!require(ggdist)) {install.packages("ggdist")} if(!require(gghalves)) {install.packages("gghalves")} if(!require(PupillometryR)) {install.packages("PupillometryR")} if(!require(jmv)) {install.packages("jmv")} if(!require(plyr)) {install.packages("plyr")} if(!require(lme4)) {install.packages("lme4")} if(!require(lmerTest)) {install.packages("lmerTest")} if(!require(modelsummary)) {install.packages("modelsummary")} if(!require(MuMIn)) {install.packages("MuMIn")} if(!require(papaja)) {install.packages("papaja")} if(!require(reghelper)) {install.packages("reghelper")} if(!require(emmeans)) {install.packages("emmeans")} if(!require(lavaan)) {install.packages("lavaan")} if(!require(semPlot)) {install.packages("semPlot")} if(!require(bmlm)) {install.packages("bmlm")} if(!require(sjPlot)) {install.packages("sjPlot")} if(!require(sjmisc)) {install.packages("sjmisc")} if(!require(EMAtools)) {install.packages("EMAtools")} if(!require(report)) {install.packages("report")} if(!require(MplusAutomation)) {install.packages("MplusAutomation")} if(!require(rstatix)) {install.packages("rstatix")} ``` #Health Algorithm - Belief 1 ## Loading the data ```{r} Digital_Belief_Cleaned <- read_sav("Study 1 - Beliefs/Data/Digital Beliefs - Study 1 - Cleaned - July 1, 2021.sav") ``` ```{r} describe(Digital_Belief_Cleaned$Age) prop.table(table(Digital_Belief_Cleaned$Gender)) describe(Digital_Belief_Cleaned$IsMale) ``` ```{r} describe(Digital_Belief_Cleaned$Trust) ``` ## Between Subjects ANOVA ```{r} anovaOneW( formula = Trust + Believe ~ Condition, data = Digital_Belief_Cleaned, welchs = FALSE, fishers = TRUE, desc = TRUE, descPlot = TRUE, norm = TRUE, eqv = TRUE ) HealBelief.1.Trust <- aov(Trust + Believe ~ Condition, data = Digital_Belief_Cleaned) summary(HealBelief.1.Trust) ``` ```{r} HealBelief.1.Trust <- aov(Trust ~ Condition, data = Digital_Belief_Cleaned) summary(HealBelief.1.Trust) TukeyHSD(HealBelief.1.Trust, conf.level = .95) ``` ```{r} HealBelief.1.Believe <- aov(Believe ~ Condition, data = Digital_Belief_Cleaned) summary(HealBelief.1.Believe) TukeyHSD(HealBelief.1.Believe, conf.level = .95) ``` ##T-Tests ###Simplified All #### Trust ```{r} pairwise.t.test(Digital_Belief_Cleaned$Trust, Digital_Belief_Cleaned$Condition, p.adjust.method = "none") #Digital_Belief_Cleaned %>% cohen.d(Trust ~ as.factor(Condition)) ``` #### Believe ```{r} pairwise.t.test(Digital_Belief_Cleaned$Believe, Digital_Belief_Cleaned$Condition, p.adjust.method = "none") ``` ###Planned Comparsons ####Non-Human-Like Vs Doctor ```{r} ttestIS( formula = Trust + Believe ~ Coding_NONvHum, data = Digital_Belief_Cleaned, vars = vars(Trust, Believe), meanDiff = TRUE, effectSize = TRUE, desc = TRUE, plots = TRUE) ``` ```{r} t.test(Trust ~ Coding_NONvHum, data = Digital_Belief_Cleaned) report(t.test(Trust ~ Coding_NONvHum, data = Digital_Belief_Cleaned)) ``` ```{r} t.test(Believe ~ Coding_NONvHum, data = Digital_Belief_Cleaned) report(t.test(Believe ~ Coding_NONvHum, data = Digital_Belief_Cleaned)) ``` #### Human-Like vs Doctor ```{r} ttestIS( formula = Trust + Believe ~ Coding_HumvHum, data = Digital_Belief_Cleaned, vars = vars(Trust, Believe), meanDiff = TRUE, effectSize = TRUE, desc = TRUE, plots = TRUE) ``` ```{r} t.test(Trust ~ Coding_HumvHum, data = Digital_Belief_Cleaned) report(t.test(Trust ~ Coding_HumvHum, data = Digital_Belief_Cleaned)) ``` ```{r} t.test(Believe ~ Coding_HumvHum, data = Digital_Belief_Cleaned) report(t.test(Believe ~ Coding_HumvHum, data = Digital_Belief_Cleaned)) ``` #### Both AI vs Doctor ```{r} ttestIS( formula = Trust + Believe ~ Coding_AIvHum, data = Digital_Belief_Cleaned, vars = vars(Trust, Believe), meanDiff = TRUE, effectSize = TRUE, desc = TRUE, plots = TRUE) ``` #### Non-Human vs Human-Like ```{r} ttestIS( formula = Trust + Believe ~ Coding_AIvAI, data = Digital_Belief_Cleaned, vars = vars(Trust, Believe), meanDiff = TRUE, effectSize = TRUE, desc = TRUE, plots = TRUE) ``` ## Exploritory Analysis ### ANCOVA ```{r} DigitalBeliefs1_ANCOVA <- aov(Trust ~ Condition + Warmth + Competent + IsMale + Age, data = Digital_Belief_Cleaned) summary(DigitalBeliefs1_ANCOVA) ``` ### LM #### Trust ##### Models ```{r} DigitalBeliefs_LM_1 <- lm(Trust ~ as.factor(Condition), data = Digital_Belief_Cleaned) summary(DigitalBeliefs_LM_1) ``` ```{r} DigitalBeliefs_LM_2 <- lm(Trust ~ as.factor(Condition) + IsMale + Age, data = Digital_Belief_Cleaned) summary(DigitalBeliefs_LM_2) ``` ```{r} DigitalBeliefs_LM_3 <- lm(Trust ~ as.factor(Condition) + IsMale + Age + Warmth + Competent , data = Digital_Belief_Cleaned) summary(DigitalBeliefs_LM_3) ``` ##### Exporting Models ```{r} Study3_TrustModels <- list( "Model 1" = DigitalBeliefs_LM_1, "Model 2" = DigitalBeliefs_LM_2, "Model 3" = DigitalBeliefs_LM_3 ) ``` ```{r} modelsummary(Study3_TrustModels, , fmt = 2, estimate = "{estimate}{stars}", statistic = c("std.error"), stars = c("*" = .05, "**" = .01, "***" = .001), gof_map = c("nobs", "r.squared", "adj.r.squared"), align = "lccc", notes = "Note. * = .05, ** = .01, *** = .001", title = "OLS Regression On Trust", output = "Study3_TrustModels.docx") ``` #### Belief ##### Models ```{r} DigitalBeliefs_LM_1 <- lm(Believe ~ as.factor(Condition), data = Digital_Belief_Cleaned) summary(DigitalBeliefs_LM_1) ``` ```{r} DigitalBeliefs_LM_2 <- lm(Believe ~ as.factor(Condition) + IsMale + Age, data = Digital_Belief_Cleaned) summary(DigitalBeliefs_LM_2) ``` ```{r} DigitalBeliefs_LM_3 <- lm(Believe ~ as.factor(Condition) + IsMale + Age + Warmth + Competent , data = Digital_Belief_Cleaned) summary(DigitalBeliefs_LM_3) ``` ##### Exporting Models ```{r} Study3_Belief_Models <- list( "Model 1" = DigitalBeliefs_LM_1, "Model 2" = DigitalBeliefs_LM_2, "Model 3" = DigitalBeliefs_LM_3 ) ``` ```{r} modelsummary(Study3_Belief_Models, , fmt = 2, estimate = "{estimate}{stars}", statistic = c("std.error"), stars = c("*" = .05, "**" = .01, "***" = .001), gof_map = c("nobs", "r.squared", "adj.r.squared"), align = "lccc", notes = "Note. * = .05, ** = .01, *** = .001", title = "OLS Regression On Belief", output = "Study3_Belief_Models.docx") ``` ```{r} Chp2_St3_Mediation <- 'Trust ~ Coding_HumvHum Trust ~ Warmth Warmth ~ Coding_HumvHum' Chp2_St3_Mediation_Model <- sem(model = Chp2_St3_Mediation, data = Digital_Belief_Cleaned, estimator = "REML") summary(Chp2_St3_Mediation_Model, standardized = TRUE, fit.measures = TRUE, rsq = TRUE) ``` ```{r} Chp2_St3_Mediation <- 'Trust ~ Coding_HumvHum Trust ~ Competent Competent ~ Coding_HumvHum' Chp2_St3_Mediation_Model <- sem(model = Chp2_St3_Mediation, data = Digital_Belief_Cleaned, estimator = "REML") summary(Chp2_St3_Mediation_Model, standardized = TRUE, fit.measures = TRUE, rsq = TRUE) ``` ## Better Plots ### Trust ```{r} Digital_Belief_Cleaned$Condition Digital_Belief_Cleaned <- Digital_Belief_Cleaned %>% mutate( Condition_Fix = case_when( Condition == "Non-Humanlike AI" ~ "Standard AI", TRUE ~ Condition ) ) Digital_Beliefs_3_Longer$Condition <- factor( Digital_Beliefs_3_Longer$Condition, levels = c(1,2,3), labels =c("Humanlike AI", "Standard AI", "Doctor") ) Digital_Belief_Cleaned$Condition ``` ```{r} Digital_Belief_Cleaned %>% filter(!is.na(Condition_Fix)) %>% mutate(Condition_Fix = fct_reorder(Condition_Fix, Trust, .fun = "mean")) %>% ggplot(., aes(x = Condition_Fix, y = Trust, fill = Condition_Fix, col = Condition_Fix, na.rm = TRUE)) + geom_flat_violin(position = position_nudge(x = .2, y = 0), adjust = 4)+ geom_point(position=position_jitter(width = .15),size = 2, alpha = 1/2, color = "black")+ stat_summary(fun.data = "mean_cl_normal", geom = "errorbar", width = 1/4, size = 1.25, color = "white") + stat_summary(fun = "mean", geom = "point",shape = 23, stroke = 1.25, size = 4, color = "white") + scale_fill_manual(values = c("#F8737F", "#FCBB6D", "#475C7A"))+ scale_colour_manual(values = c("#F8737F", "#FCBB6D", "#475C7A"))+ guides(fill = FALSE, col = FALSE, scale = "none") + geom_boxplot(aes(x = as.numeric(Condition_Fix)+ 0.25, y = Trust),outlier.shape = NA, alpha = 0.4, width = .1, colour = "BLACK") + ggtitle("Trust In Recommendation by Recommender") + ggtitle("Trust In Recommendation by Recommender") + ylab('Trust In Recommendation')+ xlab('Recommender') + theme_apa() + theme(plot.title = element_text(face = "bold", size = 16), axis.title = element_text(face = "bold", size = 14), panel.background = element_rect("transparent"), plot.background = element_rect("transparent", color = NA), legend.background = element_blank(), legend.box.background = element_rect("transparent", colour = 'white'), axis.text = element_text(colour = "white", size = 14, face = "bold"), axis.ticks = element_line(colour = "white"), axis.line = element_line(colour = "white")) + scale_y_continuous(breaks=seq(-3,3,.5)) ggsave("Study 3 - Trust Means.png", width = 11, height = 8) ``` ### Believe ```{r} Digital_Belief_Cleaned %>% filter(!is.na(Condition_Fix)) %>% mutate(Condition_Fix = fct_reorder(Condition_Fix, Trust, .fun = "mean")) %>% ggplot(., aes(x = Condition_Fix, y = Believe, fill = Condition_Fix, col = Condition_Fix, na.rm = TRUE)) + geom_flat_violin(position = position_nudge(x = .2, y = 0), adjust = 4)+ geom_point(position=position_jitter(width = .15),size = 2, alpha = 1/2, color = "black")+ stat_summary(fun.data = "mean_cl_normal", geom = "errorbar", width = 1/4, size = 1.25, color = "white") + stat_summary(fun = "mean", geom = "point",shape = 23, stroke = 1.25, size = 4, color = "white") + scale_fill_manual(values = c("#F8737F", "#FCBB6D", "#475C7A"))+ scale_colour_manual(values = c("#F8737F", "#FCBB6D", "#475C7A"))+ guides(fill = FALSE, col = FALSE, scale = "none") + geom_boxplot(aes(x = as.numeric(Condition_Fix)+ 0.25, y = Trust),outlier.shape = NA, alpha = 0.4, width = .1, colour = "BLACK") + ggtitle("Belief In Recommendation by Recommender") + ggtitle("Belief In Recommendation by Recommender") + ylab('Trust In Recommendation')+ xlab('Recommender') + theme_apa() + theme(plot.title = element_text(face = "bold", size = 16), axis.title = element_text(face = "bold", size = 14), panel.background = element_rect("transparent"), plot.background = element_rect("transparent", color = NA), legend.background = element_blank(), legend.box.background = element_rect("transparent", colour = 'white'), axis.text = element_text(colour = "white", size = 14, face = "bold"), axis.ticks = element_line(colour = "white"), axis.line = element_line(colour = "white")) + scale_y_continuous(breaks=seq(-3,3,.5)) ggsave("Study 3 - Belief Means.png", width = 11, height = 8) ``` ```{r} Digital_Belief_Cleaned %>% filter(!is.na(Condition)) %>% ggplot(., aes(x = Condition, y = Believe, fill = Condition, col = Condition, na.rm = TRUE)) + geom_flat_violin(position = position_nudge(x = .2, y = 0), alpha = .2, adjust = 4) + geom_point(position=position_jitter(width = .15),size = 1, alpha = 1/3)+ scale_fill_manual(values = c("#AF1B3F", "#B0A61A", "#144252"))+ scale_colour_manual(values = c("#5B2C38", "#333221", "#1B3842"))+ guides(fill = FALSE, col = FALSE, scale = "none") + geom_boxplot(aes(x = as.numeric(Condition)+ 0.25, y = Believe),outlier.shape = NA, alpha = 0.4, width = .1, colour = "BLACK") + stat_summary(fun.data = "mean_cl_normal", geom = "errorbar", width = 1/4, color = "black") + stat_summary(fun = "mean", geom = "point",shape = 23, stroke = 1, size = 1.5, color = "black", fill = "#FCBB6D") + ggtitle("Belief In Recommendation by Recommender") + ylab('Belief In Recommendation')+ xlab('Recommender') + theme_apa() + scale_y_continuous(breaks=seq(-3,3,.5)) #ggsave("Study 3 - Belief Means.png", width = 12, height = 8) ``` #Health Algorithm - Belief 2 - Voiced Disclaimer - For some reason 87 more participants took the study since last time, we did not compensate them in any way and yet they still have an "aid" number meaning they came from Cloud Research. I opted to keep them in, but also verified that nothing changed as a result of this extra power.