Investigating the Relationship Between Birth Month and Business Success
By Alex Freedman
Malcolm Gladwell in his famous book, Outliers, postulated that, for youth hockey players in Canada, being born earlier in a given year provided an enormous step up compared to those youths born later in the year. In order to make teaching children efficient, children must be grouped into arbitrary sets so as to create a limit on the number of children per set. Typically, children are grouped into grades with a cutoff point of January 1st. These grades are then used to determine extracurricular groupings including sporting teams, summer camps, etc. This leads to those children born in earlier months of a given year with distinct mental and physical growth advantages as well as significantly larger experiential banks over their peers born in later months. Gladwell found evidence that these advantages lead to higher rates of success in youth hockey for those youths born earlier in a year than their younger counterparts born in the middle and end of that same year.
I am setting out to continue developing Mr. Gladwell’s theory, by determining whether earlier birth months in a year for an individual can indicate future success in business. Unlike youth hockey, where much of early success can be due to sheer physical advantages that an older child may have over a slightly younger one, success in business should not be determined by physical growth advantages. Additionally, for the purposes of this analysis, where “success in business” will be defined by a dataset of prominent fortune 500 CEO’s, success has occurred far into adulthood, when any developmental head start would not be as critical of a factor. Essentially, the question is, do developmental head starts, relative to your arbitrarily established peer group, snowball into a lifetime of relative advantage, and therefore a predisposition for success in business?
In [2]:
import numpy as np #importing numpy
import pandas as pd #importing pandas
from bs4 import BeautifulSoup #importing Beautiful Soup
import requests
import html5lib #importing html5lib, as per Pandas read_html request
import re
The following data was gathered from http://www.statisticbrain.com/birth-month-statistics/ and averages out each month's number of births in the United States over the last 30 years. The csv was copied and was imported as a pandas dataframe. Later, we will discuss the results graphically.
In [3]:
path_to_bday_frequencies = '/Users/alexfreedman/Desktop/Stern/Data_Bootcamp/bday_frequencies 2.csv'
bday_densities = pd.read_csv(path_to_bday_frequencies)
bday_densities.sort_values('month_num')
Out[3]:
Using the help of my data scientist friend I was able to scrape a wikipedia table of prominent CEO's who will serve as my dataset for this project. The process underwent is as follows:
requests.get function to pull the html of the relevant Wikipedia TableBeautifulSoup object.inspect source feature, we can determine that we want the table html tag corresponding to the attribute class=wikitable sortable.pandas's handy read_html function to convert the html table into a DataFrame.
In [4]:
ceo = requests.get("http://en.wikipedia.org/wiki/List_of_chief_executive_officers") #use the
# requests package 'get' function to pull the html of the wikipedia page of CEO's
ceo_soup = BeautifulSoup(ceo.content)
ceo_table = ceo_soup.find('table', attrs={'class':'wikitable sortable'})
#wikitable sortable is the beginning of the table in the html (viewed with view source of chrome)
dfceo = pd.read_html(str(ceo_table),skiprows=0,encoding="utf-8")[0]
#the pandas read_html function reads the ceo_table html and turns it into a dataframe
dfceo.columns = dfceo.iloc[0] #make the first row the columns of the df
dfceo = dfceo.reindex(dfceo.index.drop(0)) #reindex the df to start at line 1
The next three functions will do the heavy lifting associated with:
regex. (Otherwise an error will be thrown.)
Each function uses a list comprehension to return an array. They will be used in a pandas
assigncall below to append the scraped data to the CEO dataframe created above.The original
BeautifulSoupcall is reused for these purposes.
In [5]:
def find_link(s, prestr = 'https://wikipedia.org'): #grab each wiki link from wiki table rows
return [prestr + ceo_soup.find('a', text=elem)['href'] for elem in s]
def remove_footnote(s): #use regex to remove footnotes (from links, but can be used elsewhere)
return [re.sub('\[.*?\]','',elem) for elem in s]
def find_birthday(wl): #span with class=bday
#relevancy is determined from Chrome's View Source
bdays = [BeautifulSoup(requests.get(elem).content)
.find('span', attrs={'class':'bday'}) for elem in wl]
bday_stage = ["" if elem is None else elem.text for elem in bdays]
bday_format = pd.to_datetime(bday_stage, format='%Y-%m-%d', errors='coerce') #format
#the string birthday's into pandas datetime objects. errors='coerce' is required
#to not throw errors when trying to read empty ("") birthdays, and instead convert to NaN
return bday_format
Using the functions defined above, create new (final) df called dfceo_final. find_birthday() converts the string birthday elements into pandas datetime objects in birth_date. Once they are formatted as such, creating birth_year, birth_quarter, birth_month, and birth_day variables are straightforward.
In [6]:
dfceo_final = (dfceo.assign(wiki_link = lambda x: find_link(s=remove_footnote(x.Executive)))
.assign(birth_date = lambda x: find_birthday(x.wiki_link))
.assign(birth_year = lambda x: x.birth_date.dt.year,
birth_quarter = lambda x: x.birth_date.dt.quarter,
birth_month = lambda x: x.birth_date.dt.month,
birth_day = lambda x: x.birth_date.dt.day))
In [7]:
dfceo_final
Out[7]:
Per the advice of my data scientist friend, I used the ggplot package to plot the first two graphs instead of matplotlib. I reviewed the following websites to get to know ggplot:
http://docs.ggplot2.org/current/geom_density.html
http://docs.ggplot2.org/current/geom_histogram.html
For the last graph, I used the seaborn package to create a violin plot.
In [8]:
%matplotlib inline
from ggplot import *
import seaborn as sns
import matplotlib.pyplot as plt
In [9]:
dfceo_final.columns
Out[9]:
The below graph shows that there is a relatively uniform distribution of births throughout a given year. As shown in the graph, if anything, there would be a slight liklihood for births occuring in the third quarter of a given year. This data was included to show that there is not a natural propensity for first quarter births.
In [11]:
plot1_cols = ['month_num', 'freq']
df_plot1 = (bday_densities[plot1_cols]
.dropna())
plot1_breaks = (df_plot1
.month_num
.unique()
.astype(int)
.tolist())
ggplot(aes(x='month_num', y='freq'),
data=df_plot1) +\
geom_histogram(stat='identity',
fill="blue",
alpha=.8) +\
xlab('Birthday') + ylab('Count') + ggtitle('Histogram of USA Birthdays By Month')
Out[11]:
Below is a density plot for CEO birthdays by quarter using the scraped data from the wikipedia table. It is interesting to note that this dataset shows a disproportionate number of Q1 and Q2 births compared to the population plotted above. Further, Q3, which had the most births above, has the least births in the CEO dataset.
In [14]:
plot2_cols = ['birth_quarter']
df_plot2 = (dfceo_final[plot2_cols]
.dropna())
plot2_breaks = [1,2,3,4]
plot2_labels = ['Q1','Q2','Q3','Q4']
ggplot(aes(x='birth_quarter'),
data=df_plot2) +\
geom_density(colour="darkblue",
size=2,
fill="purple",
alpha=.2) +\
scale_x_continuous(breaks = plot2_breaks,
labels = plot2_labels) +\
xlab('Birthday') + ylab('Density') + ggtitle('Density Plot of CEO Birthday Quarters')
Out[14]:
The violin plot below shows the distribution of CEO birth years across the quarters in which those CEO's were born. Q4 CEO's appear to be slightly younger, on average. Q2 has outliers, both young and old, because Q2 births make up the majority of the dataset.
In [15]:
plot3_cols = ['birth_quarter', 'birth_year']
df_plot3 = (dfceo_final[plot3_cols]
.dropna())
ax = sns.violinplot(x="birth_quarter", y="birth_year", data=df_plot3)
ax.set(xlabel='Birth Quarter', ylabel='Birth Year')
ax.set(xticklabels=['Q1','Q2','Q3','Q4'])
Out[15]:
Conclusion
While the data set was a relatively small sample, it is interesting to see that when it is compared to the population set, Fortune 500 CEO's do appear to have a higher likelihood, albeit slightly, to be born in the first six months of a given year. Given this data, it could be theorized that early developmental advantages and the natural leadership associated with being older than one's peers, could indicate a future of business success.