10 Exercises

Code

# load libraries
library(tidyverse)
library(kableExtra)
library(ggplot2)
library(ggbeeswarm)
library(gridExtra)

Exercise 10.1 (Summarize diabetes data) Use below code to load diabetes data set and calculate BMI and add categorical variable obese (Yes) if \(BMI \ge 30\) and No otherwise. Summarize variables: obese, age and gender reporting mean and sample standard deviation for numerical variables and counts and percentage per group for categorical variables.

library(faraway)
library(tidyverse)

inch2m <- 2.54/100
pound2kg <- 0.45
data_diabetes <- diabetes %>%
  mutate(height  = height * inch2m, height = round(height, 2)) %>% 
  mutate(waist = waist * inch2m) %>%  
  mutate(weight = weight * pound2kg, weight = round(weight, 2)) %>%
  mutate(BMI = weight / height^2, BMI = round(BMI, 2)) %>% 
  mutate(obese= cut(BMI, breaks = c(0, 29.9, 100), labels = c("No", "Yes")))

Exercise 10.2 (Plot diabetes data) Try various plots to visualize diabetes variables: BMI and gender. Start by making a histogram and density plot for BMI and box plot of BMI stratified by gender status. Can you think of any other plots that may be useful here to visualize the relationships between BMI and gender?

Solutions: Descriptive statistics

Solution. Exercise 10.1

Age is a numerical variable and we can calculate mean and sample standard deviation for example as below:

data_diabetes %>%
  select(age) %>%
  summarize(age_mean = mean(age, na.rm = T), 
            age_sd = sd(age, na.rm = T)) %>%
  print()

  age_mean   age_sd
1 46.85112 16.31233

Gender and obesity status are categorical variables and we can calculate counts and percentages per groups as below:

summary_gender <- data_diabetes %>%
  select(gender) %>%
  group_by(gender) %>%
  summarize(n = n()) %>%
  mutate(percent = n * 100 / nrow(data_diabetes)) %>%
  print()

# A tibble: 2 × 3
  gender     n percent
  <fct>  <int>   <dbl>
1 male     169    41.9
2 female   234    58.1

summary_obese <- data_diabetes %>%
  select(obese) %>%
  group_by(obese) %>%
  summarize(n = n()) %>%
  mutate(percent = n * 100 / nrow(data_diabetes)) %>%
  print()

# A tibble: 3 × 3
  obese     n percent
  <fct> <int>   <dbl>
1 No      253   62.8 
2 Yes     144   35.7 
3 <NA>      6    1.49

Alternatively, we can use one of the many R data summaries packages, for instance arsenal to summarize obesity status by age and gender.

library(arsenal)

tab1 <- tableby(obese ~ gender + age, data=data_diabetes)
summary(tab1)

	No (N=253)	Yes (N=144)	Total (N=397)	p value
gender				< 0.001
male	128 (50.6%)	40 (27.8%)	168 (42.3%)
female	125 (49.4%)	104 (72.2%)	229 (57.7%)
age				0.734
Mean (SD)	47.103 (16.745)	46.521 (15.831)	46.892 (16.402)
Range	19.000 - 91.000	20.000 - 92.000	19.000 - 92.000

Another popular package is gtsummary that calculates descriptive statistics for continuous, categorical, and dichotomous variables in R, and presents the results in customizable summary table ready for publication.

library(gtsummary)

data_diabetes %>%
  select(age, gender, obese) %>%
  tbl_summary(by = obese, 
              statistic = list(all_continuous() ~ "{mean} ({sd})"))

Characteristic	No, N = 253¹	Yes, N = 144¹
age	47 (17)	47 (16)
gender
male	128 (51%)	40 (28%)
female	125 (49%)	104 (72%)
¹ Mean (SD); n (%)

Solution. Exercise 10.2

font.size <- 12
col.blue.light <- "#a6cee3"
col.blue.dark <- "#1f78b4"
my.ggtheme <- theme(axis.title = element_text(size = font.size), 
        axis.text = element_text(size = font.size), 
        legend.text = element_text(size = font.size), 
        legend.title = element_blank(), 
        legend.position = "top", 
        axis.title.y = element_text(angle = 0)) + theme_bw()
  
plt_hist <- data_diabetes %>%
  ggplot(aes(x = BMI)) +
  geom_histogram() + 
  my.ggtheme

plt_density <- data_diabetes %>%
  ggplot(aes(x = BMI)) +
  geom_density() + 
  my.ggtheme

plt_boxplot <- data_diabetes %>%
  ggplot(aes(x = gender, y = BMI)) + 
  geom_boxplot() + 
  my.ggtheme
  

plt_hist

plt_density

plt_boxplot

In addition, we could for instance try beeswarm plot and/or histogram stratified by gender. Or we can try also overlaying box plots over the jitter plot either for all BMI variables or separately for males and females. Sometimes, it may be also a good idea to plot summary statistics, e.g. a barplot at a height of means and error bars representing standard deviation, error bars or confidence intervals. See this post for inspiration if you’d like to try plotting the summary statistics instead http://www.cookbook-r.com/Graphs/Plotting_means_and_error_bars_(ggplot2)/