Sum_ManLaiMannie_Homework_4
2024-03-12
Setting up environment
rm(list=ls()); gc()## used (Mb) gc trigger (Mb) limit (Mb) max used (Mb)
## Ncells 526662 28.2 1170034 62.5 NA 669428 35.8
## Vcells 971047 7.5 8388608 64.0 16384 1851813 14.2# List of packages
packages <- c("tidyverse", "infer", "fst", "modelsummary", "effects", "survey", "interactions", "broom") # add any you need here
# Install packages if they aren't installed already
new_packages <- packages[!(packages %in% installed.packages()[,"Package"])]
if(length(new_packages)) install.packages(new_packages)
# Load the packages
lapply(packages, library, character.only = TRUE)## ── Attaching core tidyverse packages ──────────────────────── tidyverse 2.0.0 ──
## ✔ dplyr 1.1.4 ✔ readr 2.1.4
## ✔ forcats 1.0.0 ✔ stringr 1.5.1
## ✔ ggplot2 3.4.4 ✔ tibble 3.2.1
## ✔ lubridate 1.9.3 ✔ tidyr 1.3.0
## ✔ purrr 1.0.2
## ── Conflicts ────────────────────────────────────────── tidyverse_conflicts() ──
## ✖ dplyr::filter() masks stats::filter()
## ✖ dplyr::lag() masks stats::lag()
## ℹ Use the conflicted package (<http://conflicted.r-lib.org/>) to force all conflicts to become errors
## Loading required package: carData
##
## lattice theme set by effectsTheme()
## See ?effectsTheme for details.
##
## Loading required package: grid
##
## Loading required package: Matrix
##
##
## Attaching package: 'Matrix'
##
##
## The following objects are masked from 'package:tidyr':
##
## expand, pack, unpack
##
##
## Loading required package: survival
##
##
## Attaching package: 'survey'
##
##
## The following object is masked from 'package:graphics':
##
## dotchart
##
##
## Registered S3 methods overwritten by 'broom':
## method from
## tidy.glht jtools
## tidy.summary.glht jtools## [[1]]
## [1] "lubridate" "forcats" "stringr" "dplyr" "purrr" "readr"
## [7] "tidyr" "tibble" "ggplot2" "tidyverse" "stats" "graphics"
## [13] "grDevices" "utils" "datasets" "methods" "base"
##
## [[2]]
## [1] "infer" "lubridate" "forcats" "stringr" "dplyr" "purrr"
## [7] "readr" "tidyr" "tibble" "ggplot2" "tidyverse" "stats"
## [13] "graphics" "grDevices" "utils" "datasets" "methods" "base"
##
## [[3]]
## [1] "fst" "infer" "lubridate" "forcats" "stringr" "dplyr"
## [7] "purrr" "readr" "tidyr" "tibble" "ggplot2" "tidyverse"
## [13] "stats" "graphics" "grDevices" "utils" "datasets" "methods"
## [19] "base"
##
## [[4]]
## [1] "modelsummary" "fst" "infer" "lubridate" "forcats"
## [6] "stringr" "dplyr" "purrr" "readr" "tidyr"
## [11] "tibble" "ggplot2" "tidyverse" "stats" "graphics"
## [16] "grDevices" "utils" "datasets" "methods" "base"
##
## [[5]]
## [1] "effects" "carData" "modelsummary" "fst" "infer"
## [6] "lubridate" "forcats" "stringr" "dplyr" "purrr"
## [11] "readr" "tidyr" "tibble" "ggplot2" "tidyverse"
## [16] "stats" "graphics" "grDevices" "utils" "datasets"
## [21] "methods" "base"
##
## [[6]]
## [1] "survey" "survival" "Matrix" "grid" "effects"
## [6] "carData" "modelsummary" "fst" "infer" "lubridate"
## [11] "forcats" "stringr" "dplyr" "purrr" "readr"
## [16] "tidyr" "tibble" "ggplot2" "tidyverse" "stats"
## [21] "graphics" "grDevices" "utils" "datasets" "methods"
## [26] "base"
##
## [[7]]
## [1] "interactions" "survey" "survival" "Matrix" "grid"
## [6] "effects" "carData" "modelsummary" "fst" "infer"
## [11] "lubridate" "forcats" "stringr" "dplyr" "purrr"
## [16] "readr" "tidyr" "tibble" "ggplot2" "tidyverse"
## [21] "stats" "graphics" "grDevices" "utils" "datasets"
## [26] "methods" "base"
##
## [[8]]
## [1] "broom" "interactions" "survey" "survival" "Matrix"
## [6] "grid" "effects" "carData" "modelsummary" "fst"
## [11] "infer" "lubridate" "forcats" "stringr" "dplyr"
## [16] "purrr" "readr" "tidyr" "tibble" "ggplot2"
## [21] "tidyverse" "stats" "graphics" "grDevices" "utils"
## [26] "datasets" "methods" "base"Task 1
Graph the histogram distribution using the populist scale (as we did in the tutorial) for the country of Sweden. What do you note?
sweden_data <- read.fst("sweden_data.fst")# Populist scale (Norris/Inglehart use low levels of trust in parliaments, politicians and parties as an indicator of populism)
# Note: Schafer does it 2 ways, we will focus now on the most straightforward
# Setting values greater than 10 to NA
sweden_data$trstplt <- ifelse(sweden_data$trstplt > 10, NA, sweden_data$trstplt)
sweden_data$trstprl <- ifelse(sweden_data$trstprl > 10, NA, sweden_data$trstprl)
sweden_data$trstprt <- ifelse(sweden_data$trstprt > 10, NA, sweden_data$trstprt)
# Creating and rescaling the trust variable
sweden_data$trust <- scales::rescale(sweden_data$trstplt + sweden_data$trstprl + sweden_data$trstprt, na.rm = TRUE, to = c(0, 100))
# Here's how you would "flip it" because populist is conceived by N + I as 'anti-trust'
# As Schafer explains there are some issues with that conceptualization, but N + I is also widely applied
sweden_data$populist <- scales::rescale(sweden_data$trust, na.rm = TRUE, to=c(100,0))ggplot(sweden_data, aes(x = populist)) +
geom_histogram(bins = 30, fill = "blue", color = "black") +
theme_minimal() +
labs(title = "Distribution of Populist Scale for Sweden",
x = "Populist Scale",
y = "Count")## Warning: Removed 2503 rows containing non-finite values (`stat_bin()`).
In the histogram distribution, it seems that it is slightly right-skewed, and the mode is under 50. As low populist scale means low levels of trust in parliaments, politicians and parties, most people in Sweden do not trust the government. Also, comparing to 25 to 75 in the populist scale, 0 to 25 and 75 to 100 have fewer counts. On the other hand, 25 to 50 in the populist scale has more counts than 50 to 75.
Task 2
Run a linear regression with populist attitudes as the outcome, educational attainment (recoded as a binary “BA or more” and “No BA”), and using the Swedish data. Print the intercept and coefficients only and interpret. Then do a tidy model table displaying the p-value (with digits = 3) and interpret.
sweden_data <- read.fst("sweden_data.fst")# Populist scale (Norris/Inglehart use low levels of trust in parliaments, politicians and parties as an indicator of populism)
# Note: Schafer does it 2 ways, we will focus now on the most straightforward
# Setting values greater than 10 to NA
sweden_data$trstplt <- ifelse(sweden_data$trstplt > 10, NA, sweden_data$trstplt)
sweden_data$trstprl <- ifelse(sweden_data$trstprl > 10, NA, sweden_data$trstprl)
sweden_data$trstprt <- ifelse(sweden_data$trstprt > 10, NA, sweden_data$trstprt)
# Creating and rescaling the trust variable
sweden_data$trust <- scales::rescale(sweden_data$trstplt + sweden_data$trstprl + sweden_data$trstprt, na.rm = TRUE, to = c(0, 100))
# Here's how you would "flip it" because populist is conceived by N + I as 'anti-trust'
# As Schafer explains there are some issues with that conceptualization, but N + I is also widely applied
sweden_data$populist <- scales::rescale(sweden_data$trust, na.rm = TRUE, to=c(100,0))sweden_data <- sweden_data %>%
mutate(
# Recoding education
educ.ba = case_when(
essround < 5 & edulvla == 5 ~ "BA or more",
essround >= 5 & edulvlb > 600 ~ "BA or more",
TRUE ~ "No BA"
),
# Handle NAs for education levels
edulvla = ifelse(edulvla %in% c(77, 88, 99), NA_integer_, edulvla),
edulvlb = ifelse(edulvlb %in% c(5555, 7777, 8888), NA_integer_, edulvlb),
# Explicitly making 'No BA' the reference category
educ.ba = factor(educ.ba, levels = c("No BA", "BA or more")),
)sweden_pop_model <- lm(populist ~ educ.ba, data = sweden_data)coefficients_sweden <- coef(sweden_pop_model)
print(coefficients_sweden)## (Intercept) educ.baBA or more
## 51.358680 -7.868121Sweden’s Output Interpretation:
tidy(sweden_pop_model)## # A tibble: 2 × 5
## term estimate std.error statistic p.value
## <chr> <dbl> <dbl> <dbl> <dbl>
## 1 (Intercept) 51.4 0.181 283. 0
## 2 educ.baBA or more -7.87 0.350 -22.5 2.58e-110tidy(sweden_pop_model) |>
knitr::kable(digits = 3)| term | estimate | std.error | statistic | p.value |
|---|---|---|---|---|
| (Intercept) | 51.359 | 0.181 | 283.363 | 0 |
| educ.baBA or more | -7.868 | 0.350 | -22.494 | 0 |
summary(sweden_pop_model)##
## Call:
## lm(formula = populist ~ educ.ba, data = sweden_data)
##
## Residuals:
## Min 1Q Median 3Q Max
## -51.359 -14.692 -1.359 11.975 56.509
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 51.3587 0.1812 283.36 <2e-16 ***
## educ.baBA or more -7.8681 0.3498 -22.49 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 19.43 on 15711 degrees of freedom
## (2503 observations deleted due to missingness)
## Multiple R-squared: 0.0312, Adjusted R-squared: 0.03114
## F-statistic: 506 on 1 and 15711 DF, p-value: < 2.2e-16Interpretation: Intercept (51.4): this value represents the average score on the populist scale for the reference group in Sweden, which is people with “No BA”. In simpler terms, the expected average populist scale score would be 51.4 for people who have lower educational attainment. This score indicates a certain level of populism, where higher scores suggest stronger populist attitudes. Coefficient (-7.9): the negative coefficient associated with the “BA or more” group indicates that, on average, people who have bachelor degrees or more in Sweden have populist scale scores that are 7.9 points lower than those who do not have bachelor degrees. For the tidy model table displaying the p-value, the p-value is 0 in both “No BA” and “BA or more.” It indicates it is statistically significant and the educational attainment and the Populist scale in the Swedish data is related.
Task 3
Now using the Italian dataset and the authoritarian values scale (as we did in the tutorial), graph the average by survey year. Interpret.
df <- read.fst("italy_data.fst")
## Creates an authoritarian values scale based on human modules items (can also do libertarian value scale)
df <- df %>%
mutate(behave = ipbhprp,
secure = impsafe,
safety = ipstrgv,
tradition = imptrad,
rules = ipfrule) %>%
mutate(across(c("behave", "secure", "safety", "tradition", "rules"),
~ na_if(.x, 7) %>% na_if(8) %>% na_if(9))) %>%
# Apply the reverse coding
mutate(across(c("behave", "secure", "safety", "tradition", "rules"), ~ 7 - .x ))
# Now you can calculate 'schwartzauth' after the NA recoding
df$auth <- scales::rescale(df$behave +
df$secure +
df$safety +
df$tradition +
df$rules, to=c(0,100), na.rm=TRUE)
df <- df %>% filter(!is.na(auth))
table(df$secure)##
## 1 2 3 4 5 6
## 54 139 626 1856 2876 2881table(df$auth)##
## 0 12 16 20 24 28 32 36 40 44 48 52 56 60 64 68 72 76 80 84
## 7 4 4 9 13 20 27 57 113 155 163 323 367 596 647 789 823 897 992 733
## 88 92 96 100
## 632 465 299 297df$year <- NA
replacements <- c(2002, 2004, 2006, 2008, 2010, 2012, 2014, 2016, 2018, 2020)
for(i in 1:10){
df$year[df$essround == i] <- replacements[i]
}
auth_avg <- df %>%
group_by(year) %>%
summarize(auth_avg = mean(auth, na.rm = TRUE))
plot_auth <- ggplot(auth_avg, aes(x = year, y = auth_avg)) +
geom_point(aes(color = auth_avg), alpha = 1.0) +
geom_line(aes(group = 1), color = "blue", linetype = "dashed") +
labs(title = "Authoritarian Scale Average by Year (Italy)",
x = "Survey Year",
y = "Authoritarian Scale Average") +
theme_minimal() +
theme(legend.position = "none") +
scale_y_continuous(limits = c(0, 100))
print(plot_auth)
Interpretation: the authoritarian scale average by year in Italy is pretty steady. From 2012 to 2020, it does not have significant changes. It slightly decreases from 75 to around 73 during 2012 to 2016, but it slowly increases back to almost 75 during 2016 to 2020. ALso, the authoritarian scale average is not low that it keeps around the 75.
Task 4
Now run a linear regression model, using the modelsummary package, with authoritarian values as the outcome, and generations as the predictor/potential explanatory variable, and using the Italian data again. Rename the coefficients using the coef_rename function. Interpret the coefficients, whether they are statistically significant, and the adjusted R-squared.
df <- read.fst("italy_data.fst")
## Creates an authoritarian values scale based on human modules items (can also do libertarian value scale)
df <- df %>%
mutate(behave = ipbhprp,
secure = impsafe,
safety = ipstrgv,
tradition = imptrad,
rules = ipfrule) %>%
mutate(across(c("behave", "secure", "safety", "tradition", "rules"),
~ na_if(.x, 7) %>% na_if(8) %>% na_if(9))) %>%
# Apply the reverse coding
mutate(across(c("behave", "secure", "safety", "tradition", "rules"), ~ 7 - .x ))
# Now you can calculate 'schwartzauth' after the NA recoding
df$auth <- scales::rescale(df$behave +
df$secure +
df$safety +
df$tradition +
df$rules, to=c(0,100), na.rm=TRUE)
df <- df %>% filter(!is.na(auth))
table(df$secure)##
## 1 2 3 4 5 6
## 54 139 626 1856 2876 2881model_data <- df %>%
mutate(
# Recoding cohort variable, setting years before 1930 and after 2000 to NA
cohort = ifelse(yrbrn < 1930 | yrbrn > 2000, NA, yrbrn),
# Recoding generational cohorts based on year of birth
gen = case_when(
yrbrn %in% 1900:1945 ~ "1",
yrbrn %in% 1946:1964 ~ "2",
yrbrn %in% 1965:1979 ~ "3",
yrbrn %in% 1980:1996 ~ "4",
TRUE ~ as.character(cohort) # Keeping other values as character if they do not fit the ranges
),
# Converting cohort to a factor with labels
gen = factor(gen,
levels = c("1", "2", "3", "4"),
labels = c("Interwar", "Baby Boomers", "Gen X", "Millennials"))
)
table(model_data$gen)##
## Interwar Baby Boomers Gen X Millennials
## 1026 2580 2214 1811model1 <- lm(auth ~ gen, data = model_data)modelsummary(
list(model1),
fmt = 1,
estimate = c( "{estimate} ({std.error}){stars}"),
statistic = NULL,
coef_rename = c("genBaby Boomers" = "Baby Boomers", "genGen X" = "Gen X", "genMillennials" = "Millennials"),
coef_omit = "Intercept")| (1) | |
|---|---|
| Baby Boomers | −1.1 (0.6)* |
| Gen X | −2.1 (0.6)*** |
| Millennials | −3.6 (0.6)*** |
| Num.Obs. | 7631 |
| R2 | 0.006 |
| R2 Adj. | 0.006 |
| AIC | 62934.0 |
| BIC | 62968.7 |
| Log.Lik. | −31462.009 |
| RMSE | 14.94 |
Coefficients Interpretation: Baby Boomers (-1.1, p<0.001): Compared to the Interwar generation, Baby Boomers exhibit scores that are, on average, 1.1 points lower on the authoritarian values scale (0-100). The one asterisks ( * ) denote statistical significance at the exact threshold of 0.05 of p-value. Gen X (-2.1, p<0.001): Generation X shows an even more pronounced decrease, with scores 2.1 points lower than the Interwar generation on the authoritarian values scale. The triple asterisks ( *** ) denote statistical significance at a very low p-value. (under our alpha threshold of 0.05) Millennials (-3.6, p<0.001): Millennials exhibit the most considerable decrease in authoritarian values, with their scores being 3.6 points lower than those of the Interwar generation. This marked difference is statistically significant, indicating that Millennials diverge most strongly from the Interwar generation in comparing their authoritarian values. Again, the triple asterisks ( *** ) denote statistical significance at a very low p-value. (under our alpha threshold of 0.05)
Adjusted R-Squared (Adj. R² = 0.006): the adjusted R-squared value of 0.006 means that the ‘gen’ variable (with our four coded generations) explain approximately 0.6% of the variance in the authoritarian values scale for Italy, after accounting for the number of predictors. While statistically significant, this suggests that the model explains only a small portion of the variability in authoritarian values, pointing to other factors outside generational differences that may play critical roles in shaping these attitudes.
Task 5
Now visualize, using the modelplot() function from the modelsummary package, coefficient estimates and 95% confidence intervals for the following three models:
Model 1 = populist attitudes scale as the outcome; left/right as the single predictor (omitting moderates as we did in the tutorial), and using data for Greece.
Model 2 = populist attitudes scale as the outcome; gender as the single predictor, and using data for Greece.
Model 3 = populist attitudes scale as the outcome; generations as the single predictor (using the four generational category recode we did in the tutorial), and using data for Greece. Interpret.
greece_data <- read.fst("greece_data.fst")
# Populist scale (Norris/Inglehart use low levels of trust in parliaments, politicians and parties as an indicator of populism)
# Note: Schafer does it 2 ways, we will focus now on the most straightforward
# Setting values greater than 10 to NA
greece_data$trstplt <- ifelse(greece_data$trstplt > 10, NA, greece_data$trstplt)
greece_data$trstprl <- ifelse(greece_data$trstprl > 10, NA, greece_data$trstprl)
greece_data$trstprt <- ifelse(greece_data$trstprt > 10, NA, greece_data$trstprt)
# Creating and rescaling the trust variable
greece_data$trust <- scales::rescale(greece_data$trstplt + greece_data$trstprl + greece_data$trstprt, na.rm = TRUE, to = c(0, 100))
# Here's how you would "flip it" because populist is conceived by N + I as 'anti-trust'
# As Schafer explains there are some issues with that conceptualization, but N + I is also widely applied
greece_data$populist <- scales::rescale(greece_data$trust, na.rm = TRUE, to=c(100,0))model_data <- greece_data %>%
mutate(
# Recoding gender
gndr = case_when(
gndr == 1 ~ "Male",
gndr == 2 ~ "Female",
gndr == 9 ~ NA_character_,
TRUE ~ as.character(gndr)
),
lrscale = case_when(
lrscale %in% 0:4 ~ "Left", # This time: Values 0 to 3 in 'lrscale' are categorized as "Left"
lrscale %in% 6:10 ~ "Right", # This time: Values 6 to 10 in 'lrscale' are categorized as "Right"
lrscale %in% c(77, 88, 99) ~ NA_character_, # Values 77, 88, 99 in 'lrscale' are set as NA
TRUE ~ NA_character_ # Any other values not explicitly matched above
),
# Recoding cohort variable, setting years before 1930 and after 2000 to NA
cohort = ifelse(yrbrn < 1930 | yrbrn > 2000, NA, yrbrn),
# Recoding generational cohorts based on year of birth
gen = case_when(
yrbrn %in% 1900:1945 ~ "1",
yrbrn %in% 1946:1964 ~ "2",
yrbrn %in% 1965:1979 ~ "3",
yrbrn %in% 1980:1996 ~ "4",
TRUE ~ as.character(cohort) # Keeping other values as character if they do not fit the ranges
),
# Converting cohort to a factor with labels
gen = factor(gen,
levels = c("1", "2", "3", "4"),
labels = c("Interwar", "Baby Boomers", "Gen X", "Millennials"))
)
table(model_data$gen)##
## Interwar Baby Boomers Gen X Millennials
## 2978 3484 3498 2330model1 <- lm(populist ~ lrscale, data = model_data)
model2 <- lm(populist ~ gndr, data = model_data)
model3 <- lm(populist ~ gen, data = model_data)models <- list(
"Model 1" = lm(populist ~ lrscale, data = model_data),
"Model 2" = lm(populist ~ gndr, data = model_data),
"Model 3" = lm(populist ~ gen, data = model_data))
modelsummary(models, fmt = 1,
estimate = c("{estimate} ({std.error}){stars}",
"{estimate} ({std.error}){stars}",
"{estimate} ({std.error}){stars}"),
statistic = NULL)| Model 1 | Model 2 | Model 3 | |
|---|---|---|---|
| (Intercept) | 72.0 (0.5)*** | 70.0 (0.3)*** | 65.8 (0.5)*** |
| lrscaleRight | −8.8 (0.6)*** | ||
| gndrMale | −1.0 (0.5)* | ||
| genBaby Boomers | 3.5 (0.7)*** | ||
| genGen X | 5.6 (0.7)*** | ||
| genMillennials | 5.8 (0.7)*** | ||
| Num.Obs. | 4936 | 9823 | 9562 |
| R2 | 0.037 | 0.000 | 0.009 |
| R2 Adj. | 0.036 | 0.000 | 0.009 |
| AIC | 44723.8 | 88873.2 | 86473.9 |
| BIC | 44743.3 | 88894.8 | 86509.7 |
| Log.Lik. | −22358.894 | −44433.590 | −43231.944 |
| RMSE | 22.44 | 22.30 | 22.25 |
modelplot(models, coef_omit = 'Interc')
Coefficients Interpretation: genBaby Boomers (3.5): Compared to the Interwar generation (reference category that is omitted), Baby Boomers exhibit scores that are, on average, 3.5 points higher on the authoritarian values scale (0-100). The triple asterisks ( *** ) denote statistical significance at a very low p-value (under our alpha threshold of 0.05) genGen X (5.6): Generation X shows an even more pronounced increase, with scores 5.3 points higher than the Interwar generation on the authoritarian values scale. Again, the triple asterisks ( *** ) denote statistical significance at a very low p-value (under our alpha threshold of 0.05) genMillennials (5.8): Millennials exhibit the most considerable increase in authoritarian values, with their scores being 5.8 points higher than those of the Interwar generation. This marked difference is statistically significant, indicating that Millennials diverge fewest strongly from the Interwar generation in comparing their authoritarian values.
Adjusted R-Squared (Adj. R² = 0.009): the adjusted R-squared value of 0.009 means that the ‘gen’ variable (with our four coded generations) explain approximately 0.9% of the variance in the authoritarian values scale for Greece, after accounting for the number of predictors (just one in this case). While statistically significant, this suggests that the model explains only a small portion of the variability in authoritarian values, pointing to other factors outside generational differences that may play critical roles in shaping these attitudes.
Interpretation: Male, relative to female, is not statistically significant. Also, the left/right model shows statistical significance, with staying in the “Right” group (relative to the reference category of “Left”) holding a score which is, on average, 8.8 points lower on the authoritarian value scale (of 0-100). Yet, the variance explained is 3.6%, which is the highest.