Getting the Data

In earlier, less techinical posts (post 1 and post 2), we used some basic preprocessing techniques to focus in and get a quick picture of our data.

emp <- emp[emp$CalendarYear == 2014,] #filtering to just 2014

pol <- emp[emp$Department == "Louisville Metro Police", ] #filter to just police 
summary(pol$AnnualRate) # quick check for crazy values or missing data. 
pol$Department <- factor(pol$Department) #resetting to dump empty Department levels

##    Min. 1st Qu.  Median    Mean 3rd Qu.    Max. 
##   12440   41890   50340   49920   55100  175100

Checking Number of Observations

Before we start throwing statistics all over the place, let’s first make sure we have a sufficient number of observations for the jobs we will look at. This won’t be an issue with a metropolitan police department, but you should always check this first. Always.

jobs <- data.frame(table(pol$JobTitle))# count number of people with a given job
jobs <- jobs[order(jobs$Freq, decreasing = TRUE),] # sort by frequency
head(jobs, 10) # take a look at the first 10 items in the list

##                                Var1 Freq
## 895  Police Officer                 1014
## 898  Police Sergeant                 172
## 893  Police Lieutenant                70
## 896  Police Recruit                   53
## 1116 Traffic Guard III                46
## 1114 Traffic Guard I                  39
## 1115 Traffic Guard II                 31
## 603  Information Process Technician   25
## 16   Administrative Clerk-Police      20
## 283  Crime Scene Technician I         15

Selecting the Top 3 Police Roles

sel_jobs <- jobs$Var1[c(1:3)] # selecting three levels of officers
pol <- pol[as.character(pol$JobTitle) %in% sel_jobs, ] #filtering to just those selected jobs 
pol$JobTitle <- factor(pol$JobTitle) # redefine the JobTitle factor levels

A Bit More Cleaning

Sometimes text elements come with extra white space that can create problems. Because R treats the “JobTitle” category as a factor here, we can use the “levels” command to puts quotes around the entire label. This will help us see any unexpected whitespace.

levels(pol$JobTitle)

## [1] "Police Lieutenant             " "Police Officer                "
## [3] "Police Sergeant               "

Let’s eliminate that whitespace by just renaming the factor levels without it.

levels(pol$JobTitle) <- c("Police Lieutenant", "Police Officer", "Police Sergeant")

Interpreting the Salary Data

A few things are clear just from the numbers. First, Lieutenants make substantially more than Officers or Sergeants. These differences are reflected in the means and medians as well as the maximum salaries within each of these positions.

There are also some differences when it comes to standard deviations though. The SD for the Sergeant salaries, for example, is less than half of that for Officers. This means that Sergeant salaries are generally similar to each other while those of Police Officers are more spread out.

In an HR setting it is helpful to look at the spread of salaries to see whether people with the same job title receive comparable salaries. There are always other elements to consider, but looking at the standard deviations is usually a great place to start.

Visualizing the Salary Data

Let’s see these distributions to get a feel for what’s really behind the numbers. We can use the histogram function from the lattice library to see a different AnnualRate histogram for each role.

We’ll also specify the break points (the bins) by using the using the sequence command. This let’s us use the same groupings and directly compare the distributions.

# install.packages("lattice") # run this if you don't have lattice installed
library(lattice)
pol$JobTitle = factor(pol$JobTitle,levels(pol$JobTitle)[c(1,3,2)]) #reordering the levels for better visual
histogram(x = ~AnnualRate| JobTitle, data = pol, layout = c(1,3), col = "purple",
          breaks = seq(from = 36000, to = 100000, by = 2000))

In the case of Police Officers, we see a substantial number of officers split off from the rest. That explains the larger SD. In HR, this is a strong signal to dig deeper into the extreme scores to understand the WHY behind the big differences.

The Sergeants give us something resembling a balanced, normal distribution, although you should be sure to note the more limited range of salaries for this group compared to the Officers.

Finally, for the Lieutenants, we see a sort of cutoff on the low end. This makes sense given that these individuals have a meaningfully higher rank and tend to have a lot more experience on the job as a group.

Note also the absence of overlap between the salaries of the three levels. Lieutenant salaries are substantially and visibly higher, again consistent with their experience and rank.