Human Trafficking, an Exploration of Survivors

Human trafficking can happen anywhere, and predominatly affects children and young adults. This is a difficult statistic to track, as getting help to those entraped by human trafficking is difficult, and keeping their information safe once they become surviors is imperative. The Counter-Trafficking Data Collaborative has created a international platform where individual organizations are able to upload their tracking data of survivors. As this only tracks those who have been able to escape trafficking, we can assume the number of individuales trafficking affects is much larger. Using this data however, we can get a glimps of who is affected, and how, leading to better strategies of breaking the cycle as a whole. Let's explore with knowlege learned from the course Data Analysis with Python: Zero to Pandas.

Downloading the Dataset

Let's import the packages needed to download the data and explore it.

#Importing the packages:

import pandas as pd
pd.options.mode.chained_assignment = None
import numpy as np
import matplotlib.pyplot as plt
%matplotlib inline
import seaborn as sns
import plotly.express as px
import os
!pip install --quiet pycountry_convert
from pycountry_convert import country_alpha2_to_country_name, country_name_to_country_alpha3

Let's begin by downloading the data, and listing the files within the dataset.

#Importing the uncleaned world data:
world_data = pd.read_csv('The Global Dataset 14 Apr 2020.csv', dtype={'typeOfSexConcatenated': str, 'RecruiterRelationship': str, 'majorityStatusAtExploit':str})

The dataset has now been read into the DataFrame world_data.

Data Preparation and Cleaning

Now that we have the data loaded, we need to clean it. We will want to parse down the number of rows, and make the values uniform, if they are not.

#Let's take a quick look at the data
world_data.head()

	By using this data you agree to the Terms of Use: https://www.ctdatacollaborative.org/terms-use	yearOfRegistration	Datasource	gender	ageBroad	majorityStatus	majorityStatusAtExploit	majorityEntry	citizenship	meansOfControlDebtBondage	...	typeOfSexPrivateSexualServices	typeOfSexConcatenated	isAbduction	RecruiterRelationship	CountryOfExploitation	recruiterRelationIntimatePartner	recruiterRelationFriend	recruiterRelationFamily	recruiterRelationOther	recruiterRelationUnknown
0	NaN	2002	Case Management	Female	18--20	Adult	-99	-99	CO	-99	...	-99	-99	-99	-99	-99	0	0	0	0	1
1	NaN	2002	Case Management	Female	18--20	Adult	-99	-99	CO	-99	...	-99	-99	-99	-99	-99	0	0	0	0	1
2	NaN	2002	Case Management	Female	18--20	Adult	-99	-99	CO	-99	...	-99	-99	-99	-99	-99	0	0	0	0	1
3	NaN	2002	Case Management	Female	18--20	Adult	-99	-99	CO	-99	...	-99	-99	-99	-99	-99	0	0	0	0	1
4	NaN	2002	Case Management	Female	18--20	Adult	-99	-99	CO	-99	...	-99	-99	-99	-99	-99	0	0	0	0	1

5 rows × 64 columns

world_data.shape

(48801, 64)

world_data.head()

	By using this data you agree to the Terms of Use: https://www.ctdatacollaborative.org/terms-use	yearOfRegistration	Datasource	gender	ageBroad	majorityStatus	majorityStatusAtExploit	majorityEntry	citizenship	meansOfControlDebtBondage	...	typeOfSexPrivateSexualServices	typeOfSexConcatenated	isAbduction	RecruiterRelationship	CountryOfExploitation	recruiterRelationIntimatePartner	recruiterRelationFriend	recruiterRelationFamily	recruiterRelationOther	recruiterRelationUnknown
0	NaN	2002	Case Management	Female	18--20	Adult	-99	-99	CO	-99	...	-99	-99	-99	-99	-99	0	0	0	0	1
1	NaN	2002	Case Management	Female	18--20	Adult	-99	-99	CO	-99	...	-99	-99	-99	-99	-99	0	0	0	0	1
2	NaN	2002	Case Management	Female	18--20	Adult	-99	-99	CO	-99	...	-99	-99	-99	-99	-99	0	0	0	0	1
3	NaN	2002	Case Management	Female	18--20	Adult	-99	-99	CO	-99	...	-99	-99	-99	-99	-99	0	0	0	0	1
4	NaN	2002	Case Management	Female	18--20	Adult	-99	-99	CO	-99	...	-99	-99	-99	-99	-99	0	0	0	0	1

5 rows × 64 columns

As we can see the data seems to lack null values. This is because the mull values are signified by the string -99. We also want to remove the fist column, as well as a few of the inner columns that we will not be exploring today.

#Creating a subset of the orginal dataframe
world_data = world_data[['CountryOfExploitation', 'yearOfRegistration', 'Datasource', 
                         'gender', 'ageBroad', 'majorityStatus', 'majorityStatusAtExploit', 
                         'majorityEntry', 'citizenship', 'isForcedLabour', 'isSexualExploit', 
                         'isOtherExploit', 'isSexAndLabour','typeOfExploitConcatenated']]

#Replacing the values -99 to represent null values
world_data.replace('-99', np.nan, inplace=True)
world_data.replace(-99, np.nan, inplace=True)

Let's now see what the cleanded data looks like:

world_data.head()

	CountryOfExploitation	yearOfRegistration	Datasource	gender	ageBroad	majorityStatus	majorityStatusAtExploit	majorityEntry	citizenship	isForcedLabour	isSexualExploit	isOtherExploit	isSexAndLabour	typeOfExploitConcatenated
0	NaN	2002	Case Management	Female	18--20	Adult	NaN	NaN	CO	0.0	1.0	0.0	0.0	Sexual exploitation
1	NaN	2002	Case Management	Female	18--20	Adult	NaN	NaN	CO	0.0	1.0	0.0	0.0	Sexual exploitation
2	NaN	2002	Case Management	Female	18--20	Adult	NaN	NaN	CO	0.0	1.0	0.0	0.0	Sexual exploitation
3	NaN	2002	Case Management	Female	18--20	Adult	NaN	NaN	CO	0.0	1.0	0.0	0.0	Sexual exploitation
4	NaN	2002	Case Management	Female	18--20	Adult	NaN	NaN	CO	0.0	1.0	0.0	0.0	Sexual exploitation

world_data.shape

(48801, 14)

world_data.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 48801 entries, 0 to 48800
Data columns (total 14 columns):
 #   Column                     Non-Null Count  Dtype  
---  ------                     --------------  -----  
 0   CountryOfExploitation      38626 non-null  object 
 1   yearOfRegistration         48801 non-null  int64  
 2   Datasource                 48801 non-null  object 
 3   gender                     48801 non-null  object 
 4   ageBroad                   36439 non-null  object 
 5   majorityStatus             36439 non-null  object 
 6   majorityStatusAtExploit    9290 non-null   object 
 7   majorityEntry              6491 non-null   object 
 8   citizenship                48523 non-null  object 
 9   isForcedLabour             26102 non-null  float64
 10  isSexualExploit            23861 non-null  float64
 11  isOtherExploit             30938 non-null  float64
 12  isSexAndLabour             23456 non-null  float64
 13  typeOfExploitConcatenated  32627 non-null  object 
dtypes: float64(4), int64(1), object(9)
memory usage: 5.2+ MB

Now that the data has been cleaned, we can start exploring it.

Exploratory Analysis and Visualization

One of the best ways to start exploring data is to plot it, and visualize the relationships. That is what we are going to do below. Using matplotlib, seaborn, and plotly, let's dive into the data.

But first, lets create a smaller dataframe to explore:

#Setting new plot defaults
import seaborn as sns
import matplotlib
import matplotlib.pyplot as plt
%matplotlib inline
sns.set_palette('Set3')
sns.set_style('darkgrid')
matplotlib.rcParams['font.size'] = 14
matplotlib.rcParams['figure.figsize'] = (9, 5)
matplotlib.rcParams['figure.facecolor'] = '#00000000'

#Selecting a new dataframe of only surviors who were exploited in the US.
us_data = world_data[world_data['CountryOfExploitation']=='US']

#Selecting a new dataframe of only US surviors who were recorded in 2017 and 2018.
us_17_18_data = us_data[us_data['yearOfRegistration'] >=2017]

Our new dataframe, reported US survivors of human trafficking for 2017 & 2018

#Checking a random sample to make sure the new dataframe is correctly filtered.
us_17_18_data.sample(10)

	CountryOfExploitation	yearOfRegistration	Datasource	gender	ageBroad	majorityStatus	majorityStatusAtExploit	majorityEntry	citizenship	isForcedLabour	isSexualExploit	isOtherExploit	isSexAndLabour	typeOfExploitConcatenated
48361	US	2018	Hotline	Female	9--17	Minor	Minor	NaN	00	NaN	NaN	0.0	NaN	NaN
40891	US	2017	Hotline	Female	27--29	Adult	NaN	NaN	US	0.0	1.0	0.0	0.0	Sexual exploitation
40123	US	2017	Hotline	Female	21--23	Adult	Minor	NaN	00	0.0	1.0	0.0	0.0	Sexual exploitation
48610	US	2018	Hotline	Male	0--8	Minor	Minor	Minor	00	0.0	1.0	0.0	0.0	Sexual exploitation
41418	US	2017	Hotline	Female	39--47	Adult	NaN	NaN	US	0.0	1.0	0.0	0.0	Sexual exploitation
46618	US	2018	Hotline	Female	30--38	Adult	NaN	NaN	00	0.0	1.0	0.0	0.0	Sexual exploitation
45355	US	2018	Hotline	Female	21--23	Adult	NaN	NaN	00	0.0	1.0	0.0	0.0	Sexual exploitation
45556	US	2018	Hotline	Female	21--23	Adult	NaN	NaN	00	0.0	1.0	0.0	0.0	Sexual exploitation
46396	US	2018	Hotline	Female	30--38	Adult	NaN	NaN	00	0.0	1.0	0.0	0.0	Sexual exploitation
47529	US	2018	Hotline	Female	9--17	Minor	Minor	NaN	00	0.0	1.0	0.0	0.0	Sexual exploitation

Plotting

#How many survivors were reported in 2017 vs. 2018? 
ax = sns.barplot(data=us_17_18_data, 
                 x='yearOfRegistration', 
                 y=us_17_18_data.index);

As we can see above, there are roughly the same number survivors reported in these 2 years, being just over 4,000.

#Exploring the number of male and female survivors
sns.catplot(x='gender', 
            kind='count', 
            palette="RdBu", 
            data=us_17_18_data);

Overwhelmingly, more female survivors were reported in 2017 and 2018, with over 7,000 cases in this 2 year span. What does this look like in percentages?

#Creating a plotly pie chart to break down gender prevalence:
gender = us_17_18_data.groupby(['gender', 'ageBroad']).size().reset_index()
gender.rename(columns = {0:'Number of Survivors'}, inplace=True)

fig = px.pie(gender.groupby('gender').sum().reset_index(), 
             values = 'Number of Survivors', 
             names = 'gender', 
             title = 'Gender of Human Trafficking Survivors', 
             color_discrete_sequence=px.colors.sequential.GnBu)
fig.show();

#At what age range did these victims become survivors?
count = us_17_18_data.ageBroad.value_counts()
plt.plot(count);

A majority of individuales who escape human trafficking are 9-20 years old. What does this look like when we also account for gender?

fig = px.bar(gender, x = 'ageBroad', y = 'Number of Survivors', color = 'gender', color_discrete_sequence=px.colors.qualitative.D3,
            category_orders = {'ageBroad': ['0--8', '9--17', '18--20', '21--23', '24--26', '27--29', '30--38', '39--47', '48+']})
fig.show()

It appears the majority of reported males are 17 and younger, with a few in the age range of 21-23.

Asking and Answering Questions

TODO - write some explanation here.

Instructions (delete this cell)

• Ask at least 5 interesting questions about your dataset
• Answer the questions either by computing the results using Numpy/Pandas or by plotting graphs using Matplotlib/Seaborn
• Create new columns, merge multiple dataset and perform grouping/aggregation wherever necessary
• Wherever you're using a library function from Pandas/Numpy/Matplotlib etc. explain briefly what it does

Q1: Are survivors more likely to be adults or minors?

#Below, we are going to make a stacked bar chart using plotly.
us_17_18_data['Survivors'] = len(us_17_18_data) #new column made

data_bar_mg = pd.DataFrame(us_17_18_data.groupby(['gender', 'majorityStatus'])['majorityStatus'].agg(Survivors='count')).reset_index() #creating a dataframe aggregating age groups
fig = px.bar(data_bar_mg, x="majorityStatus", y="Survivors", color="gender", 
            title="Human Trafficking Demographics at Time of Survivors Status",
            labels={'majorityStatus':'Age'},
              color_discrete_sequence= px.colors.sequential.Plasma_r) #this is the plot
fig.update_traces(texttemplate='%{value}', textposition='outside') #adding values to the plot
fig.update_layout(hovermode='x') #adding info when hovering over the plot
fig.show();

Based on this data, female survivors are more likely to adults, while males are more likely to become survivors while still minors.

Q2: At what age did these survivors first enter human trafficking?

#Using a similar plotly graph to the graph above, exploring age at first exploit.
exploit = us_17_18_data.groupby(['gender', 'majorityStatusAtExploit']).size().reset_index()
exploit.rename(columns = {0:'Number of Survivors'}, inplace=True)

fig = px.bar(exploit, x = 'majorityStatusAtExploit', 
             y = 'Number of Survivors', 
             color = 'gender',
             title="Human Trafficking Demographics at Time of First Exploit",
             labels={'majorityStatusAtExploit':'Age'},
            color_discrete_sequence= px.colors.sequential.Plasma_r,
            category_orders = {'majorityStatusAtExploit': ['Minor', 'Adult']})
fig.update_traces(texttemplate='%{value}', textposition='outside')
fig.show()

This chart contrasts with our first. First exploitation of individuals tends to be while they are minors, but it takes years for them to eventually escape human trafficking.

Q3: What type of exploitation has been occuring to these survivors?

#Using seaborn catplot to explore categories of exploit
g = sns.catplot(x='typeOfExploitConcatenated', 
            kind='count', 
            height=10, 
            palette="Set3", 
            data=us_17_18_data);
g.set_axis_labels("", "Number of Survivors").set_xticklabels(["Sexual Exploit", "Labor Exploit", "Both"]).despine(left=True);  

From the data we can see the majority of survivors reported sexual exploitation, and very few reported labor exploitation.

Q4: A broader view: How does the US statistics compare to the rest of the world based on all the data, from 2002-2019.

#Let's draw a map of where the survivors are being reported.
#First we have to get the countries into a plottable format using Python:

def get_alpha3(col):
    try:
        iso_3 =  country_name_to_country_alpha3(col)
    except:
        iso_3 = np.nan
    return iso_3

def get_name(col):
    try:
        name =  country_alpha2_to_country_name(col)
    except:
        name = np.nan
    return name

#Now let's group by country of exploitation:
world_data['CountryOfExploitation'] = world_data['CountryOfExploitation'].apply(lambda x: get_name(x)) #renaming the country of exploitation using function above
world_data['alpha_3'] = world_data['CountryOfExploitation'].apply(lambda x: get_alpha3(x)) #renaming the country to a 3 letter abbrevation, and making new column
exploitation_map = pd.DataFrame(world_data.groupby(['CountryOfExploitation', 'alpha_3'])['alpha_3'].agg(Survivors='count')).reset_index() #creating a dataframe with the new columns and renamed countries so plotly can read it

#Now for the plotly map graph:
fig = px.choropleth(exploitation_map, locations='alpha_3',
                    color='Survivors',
                    hover_name='CountryOfExploitation',
                    color_continuous_scale='Viridis_r')
fig.update_layout(title_text="Human Trafficking Surviors Based on Reported Country of Exploitation")
fig.show()

#What are the numbers?
exploitation_map[['CountryOfExploitation', 'Survivors']].set_index('CountryOfExploitation').sort_values(by='Survivors', ascending=False).head(10) #creating a quick series table

	Survivors
CountryOfExploitation
United States	12512
Ukraine	5399
Moldova, Republic of	4504
Russian Federation	2738
Philippines	1988
Indonesia	1777
Cambodia	1000
Malaysia	930
Ghana	544
United Arab Emirates	504

As we can see, the US has a very large number of reported survivors. Why is this, and are the demographics similar to the rest of the world?

Q5: Global exploitation trends, what are they?

#Let's look at this as a dataframe:
most_common_exploit = world_data.typeOfExploitConcatenated.value_counts(ascending=False).head(5) #taking only the top 5
most_common_exploit

Sexual exploitation              15989
Forced labour                     8969
Other                             7063
Slavery and similar practices      359
Forced marriage                    168
Name: typeOfExploitConcatenated, dtype: int64

#Seaborn bar graph of the most prevalent forms of exploitation:
g = sns.barplot(x=most_common_exploit.index, y=most_common_exploit.values)
labels=['Sexual Exploit', 'Forced Labor', 'Other', 'Slavery or Simiar', 'Forced Marriage'] #renaming the columns for ease of reading
g.set_xticklabels(labels=labels, rotation=80);

Globally, the most common form of human trafficking is sexual exploit, followed by forced labor. Unlike the US however, forced labor and other types of trafficking are between 40-50% as prevalent as sexual exploit, dramatically higher than what the United States sees.

Q6: Are males a miniorty of survivors globally?

#Using plotly to view male vs. female survivors in a pie chart
gender = world_data.groupby(['gender', 'ageBroad']).size().reset_index() #first creating a subset dataframe
gender.rename(columns = {0:'Number of Survivors'}, inplace=True) #renaming the columns

fig = px.pie(gender.groupby('gender').sum().reset_index(), 
             values = 'Number of Survivors', 
             names = 'gender', 
             title = 'Gender of Human Trafficking Survivors', 
             color_discrete_sequence=px.colors.sequential.RdBu) #graph created
fig.show();

Males survivors are still less prevalent than females, but it's closer to a quarter world wide, compared to the 3.7% in the US.

Inferences and Conclusion

As we can see, thousands of people have been reported surviors of human trafficking around the world. In this exploration, we just scratched the surface of who this affects and where they are affected.

Because this dataset relies on individual organizations to upload the contacts they made with survivors, we don't know if this is a full picture of what human trafficking is in the world. This basic comparison showed that in the past 2 years, a majority of US trafficking survivors have been young adult females who where sexually exploited. This is not mirrored in the larger global view of a 17 year period. The global comparison showed us about 75% of the survivors were female, and the most prevalent forms of exploit were split between sexual exploit, forced labor, and "other".

Looking at the global data we can also see that the US is leading in human trafficking survivors. But what does this mean? Does the US have more reported human trafficking survivors because it has the best organizations to help rescue individuals from trafficking? Better record systems in place to track survivors? More prevalence of human trafficking overall?

This is not a question easily answered with the data we have here. Very few organizations record and release human trafficking survivor data as most intrudes on the privacy of survivors. The resource I used here, Counter-Trafficking Data Collaborative, is still very new, beginning in 2017, and it will take time for more data to be collected and reported to the public, so that a better understanding of human trafficking can be developed.

With more awareness, we can help to fight human trafficking.

References and Future Work

As more data becomes avalible, more thorough analysis can be made on human trafficking data, as well as more robust strategies to combat it.

Source: Counter-Trafficking Data Collaborative (CTDC), [October, 2020]

Useful websites:

• https://stackoverflow.com/
• https://pandas.pydata.org/
• https://seaborn.pydata.org/introduction.html
• https://plotly.com/python/
• https://matplotlib.org/3.3.2/contents.html