Attrition

We begin by looking at the attrition variable and want to find out how many employees, of the 1,470 recorded observations, have left the company (i.e. where attrition equals “Yes”).

# sum number of "Yes" for attrition
sum(hr_cleaned$attrition == "Yes")

## [1] 237

#create bar plot for attrition
ggplot(hr_cleaned, 
       aes(attrition)) +
        geom_bar() +
        labs(title = "Only few employees leave", 
             subtitle = "Count of employees that have left and have not left the company",
             x = "Attrition?",
             y = "Number of employees") +
                    NULL

Summing all the observations that were “Yes” for attrition and looking at the bar plot below, we find that 237 employees left the company, whereas X of the observed employees still work for IBM.

This means that on average …

# divide the people that have left the company by all 
sum(hr_cleaned$attrition == "Yes") / count(hr_cleaned$attrition)

##  n_No 
## 0.192

… around 19 out of 100 people leave the company.

It is relatively difficult to judge an attrition rate of 19% with no further information about the industry of the company and what would be a normal attrition in that industry.

Oftentimes, employees leave a company because they are not satisfied with their job or their work-life balance. In order to make further conclusions about the attrition rate, we will evaluate employees’ happiness.

Employee Happiness

In the following, we will look at the distributions of job_satisfaction and work_life_balance.

#put job satisfaction levels in order
hr_cleaned$job_satisfaction <- factor(hr_cleaned$job_satisfaction, levels = c("Low", "Medium", "High", "Very High"))

#create job satisfaction plot
ggplot(hr_cleaned, 
       aes(job_satisfaction, 

#calculate percentage values per bar
           label = scales::percent(prop.table(stat(count))))) +
  
#create bar plot
            geom_bar() +
  
#add percentage values
            geom_text(stat = 'count',
            vjust = -0.5, # put labels above bars
            size = 5) + # size of label font

#add plot labels
            labs(title = "Most employees are happy with their jobs", 
                 subtitle = "Distribution of employees' satisfaction with their jobs",
                 x = "Job satisfaction",
                 y = "Number of employees") +
                    NULL

As can be seen, a majority of 61.29% of employees are highly or very highly satisfied with their job.

#put work-life balance levels in order
hr_cleaned$work_life_balance <- factor(hr_cleaned$work_life_balance, levels = c("Bad", "Good", "Better", "Best"))

#create work life balance satisfaction plot
ggplot(hr_cleaned, 
       aes(work_life_balance, 

#calculate percentage values per bar
           label = scales::percent(prop.table(stat(count))))) +
  
#create bar plot
            geom_bar() +
  
#add percentage values
            geom_text(stat = 'count',
            vjust = -0.5, #put labels above bars
            size = 5) + #size of label font

#add plot labels
            labs(title = "Most employees think their work-life-balance is great\n- but not perfect", 
                 subtitle = "Distribution of employees' satisfaction with their work-life-balance",
                 x = "Work-life-balance",
                 y = "Number of employees") +
                    NULL

As can be seen in this plot, a small minority of 5.4% of employees is not satisfied with their work-life balance. A majority of 60.7% of employees rates their work-life balance 3 out of 4, namely “better”. Nevertheless, the scale could be improved by introducing a “medium” option or balancing the positive and negative attributes as of now, there is one bad and three good options to describe one’s work-life balance.

Altogether, most employees seem to be satisfied with their job and work-life balance, indicating that the 19% attrition rate is either rather low compared to industry standards or there are other reasons than employees’ happiness causing attrition.

Education

Firstly, we will have a look at the relationship between employees’ monthly income and their respective highest level of education.

#put education levels in order
hr_cleaned$education <- factor(hr_cleaned$education, levels = c("Below College", "College", "Bachelor", "Master", "Doctor"))

#plot education vs monthly income boxplot
ggplot(hr_cleaned, 
       aes(x = education, y = monthly_income)) +
          geom_boxplot() +
          labs(title = "The higher the education, the higher the pay?",
               subtitle = "Relationship between monthly income and education",
               x = "Education level",
               y = "Monthly income ($)") +
     NULL

Generally, we can see that a higher level of education means a higher median monthly income. However, all education levels apart from Doctor show several outliers that earn between 15,000-20,000USD in contrast to relatively low medians around/below 5,000USD and 3rd quartile values of significantly lower than 10,000USD. No significant relationship can be observed.

For further investigation, we calculate and plot a bar chart of the mean (also called average) and median income by education level.

#defining the mean per education level
mean_ed <- hr_cleaned %>%
  
  #group observations by their level of education
            group_by(education) %>%
  
  #create mean variable
            summarize(mean_inc = mean(monthly_income)) %>% 
  
  #arrange mean variable from highest to lowest
            arrange(desc(mean_inc))

#check mean per education level 
mean_ed

## # A tibble: 5 x 2
##   education     mean_inc
##   <fct>            <dbl>
## 1 Doctor           8278.
## 2 Master           6832.
## 3 Bachelor         6517.
## 4 College          6227.
## 5 Below College    5641.

#create plot for mean income by education level
ggplot(mean_ed,
       aes(x=education,
           y=mean_inc)) +
        geom_col() +
        theme(axis.text.x = element_text(angle = 90)) + #turn x axis label text 90 degrees
        labs(title = "The higher the education, the higher the (average!) pay",
             subtitle = "Average income by education Level",
             x="Education level",
             y="Average monthly income ($)")

#defining the median per education level
median_ed <- hr_cleaned %>%
  
  #group observations by their level of education
            group_by(education) %>%
  
  #create median variable
            summarize(median_inc = median(monthly_income)) %>% 
  
  #arrange median variable from highest to lowest
            arrange(desc(median_inc))

#check median per education level 
median_ed

## # A tibble: 5 x 2
##   education     median_inc
##   <fct>              <dbl>
## 1 Doctor             6203 
## 2 Master             5342.
## 3 College            4892.
## 4 Bachelor           4762 
## 5 Below College      3849

#create plot for median income by education level
ggplot(median_ed,
       aes(x=education,
           y=median_inc)) +
          geom_col() +
          theme(axis.text.x = element_text(angle = 90)) + #turn x axis label text 90 degrees
          labs(title = "The higher the education, the higher the median pay? - Not necessarily!",
               subtitle= "Median income by education level",
               x="Education level",
               y="Median monthly income ($)")

The median monthly income shows no direct positive relationship with the level of education as the median monthly income of employees with a bachelor is lower than that of people with a college degree. Nevertheless, the average (mean) monthly income shows a positive relationship with the education level as people with the lowest education level (Below College) earn the lowest monthly income on average, employees with the second lowest education level (College) earn the second lowest monthly income and so on.

Finally, we plot the distribution of monthly income by education level. The colouring also indicates the gender of the respective monthly income.

#plot distribution of monthly income
ggplot(hr_cleaned, 
       aes(x = monthly_income )) + 
            geom_histogram(bins = 50) + #set size of histogram bins
  
  #facet by education level
            facet_wrap(~education) + 
  
            labs(title = "Little for many, much for few - regardless of education",
                 subtitle = "Distribution of monthly income by education level",
                 x = "Monthly income ($)",
                 y = "Number of employees") +
     NULL

Here, again, we can only observe that there are a generally more low-salary jobs than high-salary jobs. The majority of all education levels except Doctor earn below 10,000USD but there is no clear indication as to which level of education earns the least. While there are only a few employees with a Doctor, their level of monthly income varies and shows no directly connected level of income.

Altogether, these plots may show that while it is generally possible for some people to earn a comparatively higher salary with a higher level of education, there is no strict rule and, thus, also no significant positive relationship between employees’ monthly income and highest level of education. We may find other factors such as gender or job role to be a more significant influence.

Gender

Next, we will investigate if there is a relationship between monthly income and gender.

#create boxplot of monthly income vs gender
ggplot(hr_cleaned, 
       aes(x = gender, y = monthly_income)) +
        geom_boxplot(alpha=.2) +
        labs(title = "No gender pay gap",
             subtitle = "Relationship between monthly income and gender",
             x = "Gender",
             y = "Monthly income ($)") +
     NULL

We see, again, that most employees generally are in the lower monthly income range (i.e., < 10,000USD) and the median employee salary is around 5,000USD - this is true for both men and women. We don’t see significant differences between male and female employees. In fact, the median salary of women is actually slightly higher than men’s within this group of employees.

While most women are in this lower range, there are a few women earning up to 20,000USD. Even though there are also fewer men earning within the higher income range than within the lower, there are overall more men earning such relatively high monthly incomes (more outliers visible).

Nonetheless, this graph does not take into consideration the number of female/male employees. Let’s have a look at that.

#Number of female employees
sum(hr_cleaned$gender == "Female")

## [1] 588

#Number of male employees
sum(hr_cleaned$gender == "Male")

## [1] 882

#Female in % of total
sum(hr_cleaned$gender == "Female")/(sum(hr_cleaned$gender == "Female") + sum(hr_cleaned$gender == "Male"))

## [1] 0.4

We find that there are fewer women in the company, namely 40% of the employees are female.

Let’s take a closer look at the average income per gender.

#Separate female and male datasets
hr_cleaned_f <- hr_cleaned %>% 
                  filter(hr_cleaned$gender == "Female")

hr_cleaned_m <- hr_cleaned %>% 
                  filter(hr_cleaned$gender == "Male")

#total monthly income over all female / male employees
#sum(hr_cleaned_f$monthly_income)
#sum(hr_cleaned_m$monthly_income)

#average incomes for females
sum(hr_cleaned_f$monthly_income)/sum(hr_cleaned$gender == "Female")

## [1] 6687

#average income for males
sum(hr_cleaned_m$monthly_income)/sum(hr_cleaned$gender == "Male")

## [1] 6381

As we can see from the plot and the calculations, the average monthly income over all females in the company is 6,686USD whereas men earn 301USD less on average, namely 6,381USD. Even though there is a greater number of men that earn a relative high monthly income (than women), there is also a significantly higher amount of men that earn a lower-range monthly income than women. Since monthly income of men and women varies largely between all income classes and there are more men employed in the company, their amounts of monthly income are also wider spread than those of the fewer female employees. As there are more male employees and generally more low-salary employees, men’s average income is lower than that of the female employees.

While the mean follows the same trend as the median in terms here, the median salary is lower for both men and women, indicating that the mean is subject to outliers (employees with very, very high income).

Again, we find no significant relationship between an employee’s gender and their monthly income.

Job Role

Since we found that neither education nor gender is a definitive indicator for monthly income, we will look at the relationship between an employee’s job role and their monthly income next.

#put job role levels in order according to highest median monthly income
hr_cleaned$job_role <- factor(hr_cleaned$job_role, levels = c("Sales Representative", "Laboratory Technician", "Research Scientist","Human Resources", "Sales Executive", "Manufacturing Director", "Healthcare Representative",  "Research Director", "Manager"))

#create boxplot of monthly income vs job role
ggplot(hr_cleaned, 
       aes(x = monthly_income, y = job_role)) +
        geom_boxplot() +
        labs(title = "Little for many, much for managers and directors",
             subtitle = "Relationship between monthly income and job role",
             x = "Job role",
             y = "Monthly income ($)") +
      NULL

With this plot, we can clearly see that job roles can be a meaningful indicator of monthly income. When sorting the job roles by highest to lowest average monthly income, it becomes clear that high-seniority positions such as managers, directors and executives earn a higher level income than low-seniority jobs as a laboratory technician or sales representative.

This result is is not surprising as a commonly approved approach to determining an employee’s salary is to take their job role rather than their level of education or gender into consideration. These influencing factors, of course, also play a role but are nowhere near as decisive as what kind of role an employee takes on and how much responsibility he/she carries.

Age

As we have already seen some indications for a relationship between age and job_role in the beginning, we will, lastly, investigate the relationship between age and monthly income.

#create scatterplot of monthly income vs age
ggplot(hr_cleaned, 
       aes(x = age, y = monthly_income)) +
        geom_point() +
  
  #add trend/regression line
        geom_smooth(method = lm) + 
  
        labs(title = "Little for young, much for elder",
             subtitle = "Relationship between monthly income and age",
             x = "Age",
             y = "Monthly income ($)") +
     NULL

Although we can see that there are people of all ages earning a monthly income of below 5,000USD, it is clear that there are almost no people (just one person) younger than 30 that earn more than 10,000USD and only few (namely five) people between 30 and 40 that earn more than 15,000USD. Only employees above 40 years of age earn within the higher monthly income range of 15,000-20,000USD. This analysis clearly shows a positive relationship between age and monthly income, meaning the older an employee, the more likely he/she earns relatively more money.

As a higher age is commonly associated with a higher (more senior) job role and consequently also a higher salary, we facet the prior plot per job role for further analysis.

#put job role levels in order according to highest median monthly income
hr_cleaned$job_role <- factor(hr_cleaned$job_role, levels = c("Manager", "Research Director", "Healthcare Representative", "Manufacturing Director", "Sales Executive", "Human Resources","Research Scientist", "Laboratory Technician", "Sales Representative"))

#create scatterplot of monthly income vs age
ggplot(hr_cleaned, 
       aes(x = age, y = monthly_income)) +
        geom_point() +
  
  #add trend/regression line
        geom_smooth(method = lm) + 
  
  #facet by job role
        facet_wrap(~job_role) + 
  
        labs(title = "Little for the young many, much for the elder managers and directors",
             subtitle = "Relationship between monthly income and age by job role",
             x = "Age",
             y = "Monthly income ($)") +
     NULL

This analysis prints a cohesive picture with our previous analysis. There are no to only a few employees below 30 in high-seniority positions such as Manager and Reseach Director which both record the highest levels of monthly income. There are people of almost all ages in the other job roles. While people in HR are relatively middle-aged with a few employees at the lower and upper age range, most people working as Sales Representatives are relatively young with a few older people and only two employees above 50.