---
title: " Graphical summary of categorical response"
author: "Statistics III: Multivariate Data and Regression"
date: "2025-10-03"
output: "beamer_presentation"
---

# Bar plot
- Bar plots visualize how a single variable is distributed by providing a count of how many times each value (discrete) of the variable is attained.

# Bar plot of counts 
```{r  bar plot}
#install.packages("CatDataAnalysis")
#library("CatDataAnalysis")
library(ggplot2)
bar1<-ggplot(diamonds, aes(x=cut, fill=cut )) + 
     geom_bar( ) +
     scale_fill_hue(c = 40) +theme(legend.position="none")
```

#
```{r bar1, echo=FALSE}
plot(bar1)
```


# Bar graph to show the proportions
```{r  bar plot with proportion}
bar2<-ggplot(data = diamonds) +
  geom_bar(mapping = aes(x = cut, y = after_stat(prop), 
  group = 1))
```
#
```{r echo=FALSE}
plot(bar2)
```


# Other alternatives are pie chart, donut chart and treemap.  
```{r pie chart}
dt<-table(diamonds$cut)
data<-as.data.frame(dt)
pie<-ggplot(data, aes(x="", y=Freq, fill=Var1)) +
     geom_bar(stat="identity", width=1, color="white") +
     coord_polar("y", start=0) +
     theme_void()
```

#
```{r echo=FALSE}
plot(pie)
```

# Grouped bar plot
- Plot frequencies for two categorical variables. 
- You should use grouped bar chart when making comparisons across different levels of a factor. 
- Use it when you want to look at how the second category variable changes within each level of the first and vice versa.
- So this is a situation where we have one categorical predictor and one categorical response.
- In the following we have used the type of device as predictor and gender as response. Which gender is using more laptops?
- One can look at the other way around depending on the scientific questions. Which device do men prefer?  

#
```{r  grouped bar plot}
# https://github.com/rsquaredacademy-education/
# online-courses/tree/master/categorical-data-in-r
data <- readRDS('analytics.rds')
data[1:3,1:4]
bar3<-ggplot(data) +
  geom_bar(aes(x = device, fill = gender), 
  position = "dodge") +
  xlab("Device") + ylab("Count")
```

#
```{r echo=FALSE}
plot(bar3)
```

# Stacked bar plots
- The bars are stacked on top of each other instead of placing them next to each other. Use stacked bar plots while looking at cumulative value.
```{r  stacked bar plot}
bar4<- ggplot(data) +
  geom_bar(aes(x = device, fill = gender)) +
  xlab("Device") + ylab("Count")
```

#
```{r echo=FALSE}
plot(bar4)
```


# Stacked proportional bar plot
- The height of all bars in this plot are the same. 
- The distribution of the second categorical variable is scaled to 1.
- The length of each bar is determined by its share in the category. 
- In the example, we see higher proportation of gender data is available for desktop users than for mobile and tablet users.
- We lose the information that overall there are very few tablet users.

#
```{r  stacked proportional bar plot}
bar5<- ggplot(data, aes(x = device, fill = gender)) + 
  geom_bar(position = "fill") +
  labs(y = "Proportion")
```

#
```{r echo=FALSE}
plot(bar5)
```

# Tile/Mosaic chart
- Useful for multiple predictors.
- In the Titanic example there are three predictors(gender, child/adult and class) and one response (saved yes/no). All categorical. This is a regression type data.
- In the HairEyeColor example, the role of predictor/response is unclear. This is more of a multivariate data setting (not regression). There are 3 categorical variables: gender, Hair Color and Eye Color.

#
```{r tile, fig.show='hide'}
if(!require('vcd')) {
  install.packages('vcd')
  library('vcd')
}
data("Titanic")
mosaicplot(Titanic,color=TRUE)
mosaicplot(HairEyeColor)
```

#
```{r tile plot, echo=FALSE}
mosaicplot(Titanic,color=TRUE)
```

#
```{r tile plot2, echo=FALSE}
mosaicplot(HairEyeColor)
```






# Parallel coordinates plot.
- Multiple quantitative responses. Categorical predictor.
- Fisher's Iris data. Predictor is species. Response are Sepal/Petal Length/Width.

```{r parallel coor, fig.show='hide'}
library(MASS)
ir <- rbind(iris3[,,1], iris3[,,2], iris3[,,3])
parc<-parcoord(ir, col = 1 + (0:149)%/%50)
```



#
```{r plot parallel, echo=FALSE}
parcoord(ir, col = 1 + (0:149)%/%50)
```