gss-colrac.Rmd

---
title: "Predicting Attitudes Towards Racist College Professors"
author: "Gustavo Arruda"
date: "`r lubridate::today()`"
output: github_document
---

```{r setup, include=FALSE} 
knitr::opts_chunk$set(warning = FALSE, message = FALSE) 
```

## Load necessary libraries

```{r packages}
library(tidyverse)
library(rcfss)
library(randomForest)
library(knitr)
library(caret)
library(partykit)

theme_set(theme_minimal())

set.seed(1234)
```

```{r}
gss_colrac <- gss_colrac

gss_colrac_transformed <- gss_colrac %>%
  mutate(colrac = if_else(colrac == FALSE, "NO",
                           if_else(colrac == TRUE, "YES", NA_character_))) %>%
  mutate_if(is.character, as.factor)

train_control <- trainControl(method = "oob")

# Train the model
random_forest_model <- train(colrac ~., data = gss_colrac_transformed,
                    method = "rf", 
                    ntree = 200,
                    trControl = train_control,
                    na.action = na.omit
                    )

random_forest_model$finalModel
varImpPlot(random_forest_model$finalModel)

colrac_selected <- ctree(colrac ~ tolerance + age + egalit_scale + wordsum + authoritarianism, data = gss_colrac_transformed)

plot(colrac_selected,
     ip_args = list(
       pval = TRUE,
       id = FALSE),
     tp_args = list(
       id = FALSE)
)

```

I choose a random forest algorithm to build this model. Such algorithm iteratively select the best variables to construct a forest tree model, which does a good work to elicit relationships between categorical variables in a data set. The random_forest_model$finalModel chart shows that 'tolerance', 'age', 'egalit_scale', 'wordsum' and 'authoritarianism' are the most important predicting variables.


## Session info

```{r}
devtools::session_info()
```