Analyzing Views of 'The Office'

The Office is an American mockumentary sitcom television series that depicts the everyday work lives of office employees in the Scranton, Pennsylvania, branch of the fictional Dunder Mifflin Paper Company. It aired from March 24, 2005, to May 16, 2013, spanning a total of nine seasons and 201 episode.

The series was filmed in a single-camera setup without a studio audience or a laugh track to simulate the look of an actual documentary. The Office originally featured Steve Carell, Rainn Wilson, John Krasinski, Jenna Fischer, and B. J. Novak as the main cast and many other casts were also involved as guest stars during the filming of the show.

Today, we will analyze the variation of total views of each of the episode of the show along with its ratings. We will also identify the guest start who were involved in the most watched episode of the show. The information on the shows is available in the original dataset provided in Kaggle. The data set is available locally named as office_episodes.csv.

The data set contains the following information on a the episodes of the show.

office_episodes.csv

To begin the analysis, we first import the necessary libraries.

import pandas as pd
 import matplotlib.pyplot as plt

Since we'll be analyzing the dataset with a scatter plot, we will import the matplotlib library.

office_data = pd.read_csv('datasets/office_episodes.csv')

We read the dataset in a variable named office_data and use the read_csv function of the pandas library to read the dataset.

Let us now review the data contained in the dataset.

office_data.info()

which yields the following output.

<class 'pandas.core.frame.DataFrame'>
 RangeIndex: 188 entries, 0 to 187
 Data columns (total 14 columns):
  # Column Non-Null Count Dtype
 --- ------ -------------- -----
  0 episode_number 188 non-null int64
  1 season 188 non-null int64
  2 episode_title 188 non-null object
  3 description 188 non-null object
  4 ratings 188 non-null float64
  5 votes 188 non-null int64
  6 viewership_mil 188 non-null float64
  7 duration 188 non-null int64
  8 release_date 188 non-null object
  9 guest_stars 29 non-null object
  10 director 188 non-null object
  11 writers 188 non-null object
  12 has_guests 188 non-null bool
  13 scaled_ratings 188 non-null float64
 dtypes: bool(1), float64(3), int64(4), object(6)
 memory usage: 19.4+ KB

Here we see that the data is correctly placed with the data type correctly identified for almost all the data except for the release_date column which is identified as an object. We need to correct the data type for the column.

Also, we can see that the data contains 14 columns with 188 different data.

office_data['release_date'] = pd.to_datetime(office_data['release_date'])

On plotting the given data with episode_number on the horizontal axis and viewership_mil on the vertical axis, we get,

fig = plt.figure()
 plt.scatter(data=office_data, x='episode_number', y='viewership_mil')
 plt.show()

Currently, the plot is not very informative in itself. We'll be adding more details in the plot.

The first step is to differentiate the plot based on the ratings of the episodes.

The ratings column contains data which rate each episode through the scale of 1 to 10. So instead of using this scale, we differentiate the episodes through their scaled rating provided in the scaled_ratings columns.
 
 The scaled rating defined here is the normalized rating which is given as:

𝑋′ = (𝑋−𝑋𝑚𝑖𝑛)/(𝑋𝑚𝑎𝑥−𝑋𝑚𝑖𝑛)

where 𝑋𝑚𝑖𝑛 and 𝑋𝑚𝑎𝑥 represents the minimum and maximum value respectively.

 The normalized rating (Scaled rating) has value between 0 and 1 which allows more efficient way for representing them. Using the scaled values, we assign various colors for various ratings defined as:

To implement the above condition, we define a function color_r which takes in a value and assign the color to the value.

def color_r(scale):
  if scale < 0.25: return 'red'
  elif (scale>= 0.25) & (scale<0.50): return 'orange'
  elif (scale>=0.50) & (scale<0.75): return 'lightgreen'
  elif (scale>=0.75): return 'darkgreen'

We then define a new column color_rat which takes in the color parameter assigned from the function defined above. The function is applied via the apply function in the scaled_rating column.

office_data['color_rat'] = office_data['scaled_ratings'].apply(color_r)

Let' s apply color to the plots.

fig = plt.figure(figsize=(12,10))
 plt.scatter(data=office_data, x='episode_number', y='viewership_mil', c='color_rat')
 plt.show()

Now that we have applied different colors for different ratings with red representing the lowest rating while green representing the highest rating, we will now vary the size of the markers.

To differentiate the plot with the size parameters for episodes with appearance of guest stars, we define a list which contains bool value from the has_guests column.

size = list(office_data['has_guests'])

Based on the values in the list size, we assign size of 25 if the value is False and a value of 250 if the value is True.

for i, val in enumerate(size):
  if val == True: size[i] = 250
  else: size[i] = 25

We then create a column m_size with the size parameter in it and plot the graph with the size included.

office_data['m_size'] = size

fig = plt.figure(figsize=(12,10))
 plt.scatter(data=office_data, x='episode_number',y='viewership_mil', c='color_rat', s='m_size')
 plt.show()

Let's not change the marker style in addition to the marker size. In order to change the marker style, we need to pass the marker parameter to the scatter plot. But first, we need to separate the values from the office_data table.

For this we define two new tables: has_guest and no_guest and set the marker style to star (*) for episodes with guest start appearance.

has_guest = office_data[office_data['has_guests']==True]
 no_guest = office_data[office_data['has_guests']==False]

We then create separate plot but within the same figure as follows:

fig = plt.figure()
 plt.scatter(data=has_guest, x='episode_number',y='viewership_mil', c='color_rat', s='m_size', marker='*')
 plt.scatter(data=no_guest, x='episode_number',y='viewership_mil', c='color_rat', s='m_size', marker='o')
 plt.show()

This completes the plot and the information we need to show. Let's now finish the plot by adding axis titles, chart title and legend to the plot.

fig = plt.figure()
 plt.scatter(data=has_guest, x='episode_number', y='viewership_mil', c='color_rat', s='m_size', marker='*', label='Guest Start Appearance')
 plt.scatter(data=no_guest, x='episode_number', y='viewership_mil', c='color_rat', s='m_size', marker='o', label='No Star Appearance')
 plt.xlabel('Episode Number')
 plt.ylabel('Viewership (Millions)')
 plt.title('Popularity, Quality, and Guest Appearances on the Office')
 plt.legend()
 plt.show()

If we look at the plot above, we see that the total viewers start to decrease as the series extends i.e. as the episode number gets larger. Also the appearance of the guest stars are more frequent for the later episodes.
 
 Let's now see the guest start appearance for the most viewed episode of the series.

 For this we first set the max value of the viewership_mil column.

x = office_data['viewership_mil'].max()

We now iterate through the rows of the DataFrame and for the maximum value of viewership_mil we extract the names of the guest stars. Since there are many appearance in a single episode, we split the value using the str.split function. This gives a list with the names of the stars.

for i, val in office_data.iterrows():
  if val['viewership_mil'] == x:
  z=str(val['guest_stars'])
  y=z.split(',')

print(y)

This gives us the output:

['Cloris Leachman', ' Jack Black', ' Jessica Alba']

This post is based on the project 'Investigating Netflix Movies and Guest Stars in The Office' available in DataCamp.

The notebook for the complete is available in my GitHub repo.

Wikipedia has been referred for the creation of this post.