Data Blog Menu

• Home
• About
• Posts
• Contact

Belief in supernatural evil and attitudes towards guns in the United States

An analysis of the correlation between belief in supernatural evil and pro-gun attitude in the United States

Posted by Christodoulos Santorinaios on July 30, 2022

Belief in Supernatural Evil and Guns

In this assignment we will replicate a study of belief in supernatural evil and attitudes towards guns in the United States. This assignment was done as a project for the Practical Data Science course taught by Mr. Panos Louridas at the Athens University of Economics and Business. The study to be replicated is:

Christopher G. Ellison, Benjamin Dowd-Arrow, Amy M. Burdette, Pablo E. Gonzalez, Margaret S. Kelley, Paul Froese, "Peace through superior firepower: Belief in supernatural evil and attitudes toward gun policy in the United States", Social Science Research, Volume 99, 2021, https://doi.org/10.1016/j.ssresearch.2021.102595.

---

The data we are going to use come from Wave Four of the Balor Religion Survey (BRS) as it is mentioned in chapter 1.4 of the paper.

To test these hypotheses, we use data from Wave Four of the Baylor Religion Survey (BRS), which was conducted in January of 2014. Briefly, the BRS is a national random sample of 1572 non-institutionalized respondents ages 18 and over who reside in the continental United States.

With a google search for the term Baylor Religion Survey we find the website https://www.baylor.edu/baylorreligionsurvey/ containing the data.

We click on the data tab and then at the DATA AT THEARDA.COM button

In the page that opens up, we choose the Baylor Religion Survey, Wave IV (2014) - Instructional Dataset as it is already prepared for easier use in the classroom.

On the new page we can click the Download tab and then download the excel (.xlsx) file to our computer by first agreeing to the Privacy Policy and hitting Continue.

So, now that we have the dataset, let's import some libraries and load it.

import pandas as pd
import matplotlib.pyplot as plt
%matplotlib inline
import numpy as np
import seaborn as sns

df = pd.read_excel("Baylor Religion Survey, Wave IV (2014) - Instructional Dataset.XLSX")

df.shape

(1572, 288)

df.sample(10)

	MOTHERLODE_ID	PROJECT_ID	RESPONDENT_ID	METHOD_ID	METHOD_TYPE	PROJECT_NUMBER	CREATED_ON	PRACTICE	RESPONDENT_DATE	RESPONDENT_LANGUAGE	...	LIBCONR	PARTYIDR	CHILDSR	HRSWORKD	EDUCR	I_GENDER	I_EDUC	I_MARITAL	I_RELIGION	I_ATTEND
267	128171109	33370	01_000285_00000012	162048140_01	SCAN	162048140	3/26/2014	OTHER	1/28/2014	en-US	...	3.0	2.0	NaN	0.0	1.0	1.0	1.0	1.0	1.0	6.0
464	128171905	33370	01_000482_00000020	162048140_01	SCAN	162048140	3/26/2014	OTHER	1/28/2014	en-US	...	1.0	1.0	3.0	3.0	3.0	2.0	3.0	2.0	5.0	6.0
1269	128172732	33370	01_001468_00000059	162048140_01	SCAN	162048140	3/26/2014	OTHER	2/25/2014	en-US	...	2.0	1.0	1.0	1.0	4.0	2.0	5.0	2.0	2.0	6.0
484	128171911	33370	01_000502_00000021	162048140_01	SCAN	162048140	3/26/2014	OTHER	1/28/2014	en-US	...	3.0	3.0	NaN	2.0	2.0	1.0	2.0	4.0	3.0	3.0
219	128171078	33370	01_000237_00000010	162048140_01	SCAN	162048140	3/26/2014	OTHER	1/28/2014	en-US	...	1.0	1.0	1.0	4.0	2.0	1.0	2.0	4.0	1.0	3.0
1173	128171675	33370	01_001305_00000053	162048140_01	SCAN	162048140	3/26/2014	OTHER	2/19/2014	en-US	...	2.0	2.0	1.0	4.0	3.0	1.0	4.0	2.0	4.0	4.0
841	128171493	33370	01_000885_00000036	162048140_01	SCAN	162048140	3/26/2014	OTHER	2/4/2014	en-US	...	2.0	3.0	NaN	3.0	NaN	2.0	NaN	2.0	1.0	6.0
1524	128172319	33370	01_001830_00000074	162048140_01	SCAN	162048140	3/26/2014	OTHER	3/21/2014	en-US	...	1.0	1.0	1.0	5.0	3.0	1.0	3.0	3.0	1.0	3.0
1245	128172716	33370	01_001411_00000057	162048140_01	SCAN	162048140	3/26/2014	OTHER	2/21/2014	en-US	...	1.0	2.0	NaN	0.0	NaN	2.0	NaN	2.0	4.0	3.0
286	128171126	33370	01_000304_00000013	162048140_01	SCAN	162048140	3/26/2014	OTHER	1/28/2014	en-US	...	1.0	1.0	2.0	NaN	3.0	2.0	3.0	2.0	4.0	3.0

10 rows × 288 columns

To understand the columns we refer to the Codebook of the dataset that can be found here.

Questions

Q1: Belief in Supernatural Evil Metric

In order to measure the belief in supernatural evil, we will use the answers to three questions asked by the participants in the survey:

• Whether the respondent believes in the devil.

• Whether the respondent believes in hell.

• Whether the respondent believes in demons.

You will investigate how the answers to these three questions can be combined to a single metric. Justify your approach.

Based on the Codebook the answers to the above questions are in columns Q23A, Q23C and Q23G for belief in devil, hell and demons respectively. The response categories for each item were:

1 = Absolutely not
2 = Probably not
3 = Probably
4 = Absolutely

We will follow the paper's approach that is indeed logical. Responses to the three items are summed and averaged across the number of items answered to create the index we need.

df[['Q23A', 'Q23C', 'Q23G']].isna().sum()

Q23A    68
Q23C    73
Q23G    80
dtype: int64

df[['Q23A', 'Q23C', 'Q23G']]

	Q23A	Q23C	Q23G
0	4.0	4.0	4.0
1	4.0	4.0	4.0
2	3.0	2.0	2.0
3	4.0	4.0	4.0
4	4.0	4.0	4.0
...	...	...	...
1567	4.0	2.0	4.0
1568	3.0	3.0	3.0
1569	4.0	4.0	3.0
1570	3.0	3.0	3.0
1571	NaN	NaN	NaN

1572 rows × 3 columns

df[df['Q23A'].isna()][['Q23A', 'Q23C', 'Q23G']]

	Q23A	Q23C	Q23G
53	NaN	NaN	NaN
115	NaN	NaN	NaN
141	NaN	NaN	NaN
166	NaN	NaN	NaN
169	NaN	NaN	NaN
...	...	...	...
1538	NaN	NaN	4.0
1542	NaN	NaN	NaN
1556	NaN	NaN	4.0
1560	NaN	NaN	NaN
1571	NaN	NaN	NaN

68 rows × 3 columns

We have some NaN values in the above fields. We aren't going to fill them yet because we see that there are questions that are answered in the same line that there are NaN so we are going to take the mean across the rows. The mean will disregard the NaN in the rows that have at least one value. This will create values for our focal variable even if one or two values are missing from the line. Also, looking at the values I notice that most scores stay almost the same across these questions for the same individual so it's a good reasoning that the mean will in fact fill them with the mean of the columns that have values.

After creating the supernatural evil column we will have some NaN values that we will fill with the mean across columns.

df['Supernatural Evil'] = df[['Q23A', 'Q23C', 'Q23G']].mean(axis=1)

df['Supernatural Evil'].isna().sum()

55

We are left with only 55 NaN values. We will fill them with the mean.

df['Supernatural Evil'].mean()

3.089540760272468

The mean of our new variable is 3.09 rounded to 2 decimals as it is the case in the paper.

df['Supernatural Evil'].fillna(df['Supernatural Evil'].mean(), inplace=True)

Q2: Variables Selection

Apart from the belief in supernatural evil metric, you will use several other variables to control your estimates. The variables are (see Appendix B of the original publication):

• Dependent Variables

  • Ban on Semi-Auto Guns
  • Ban on High-Capacity Ammo Clips
  • Banning Civilian Handguns
  • Support for Concealed Carry Laws
  • More Armed Security at Schools
  • More Teachers/Faculty having Guns
  • More Gun Safety Programs
  • Expanded Mental Health Screening

• Independept Variables

  • Religious Variables
    • Attendance
    • Bible (Human Error, History and Fables)
    • Biblical Inerrancy
    • Biblical Literalism
  • Religious Affiliation
    • Conservative Protestant
    • Mainline Protestant
    • Black Protestant
    • Catholic
    • Other
    • No Affiliation
  • Political Ideology
  • Age
  • Sex
    • Female
    • Male
  • Race
    • White
    • Hispanic
    • African American/Black
    • Other
  • Education
    • Less Than High School
    • High School or Equivalent
    • Some College
    • College Degree
    • Post-graduate Degree
  • Household Income
  • Marital Status
    • Not Partnered/Single
    • Married/Cohabitating
  • Children
    • No kids under 18 in home
    • Kids under 18 in home
  • Area Demographics
    • Small Town/Rural
    • Urban Area
  • Region
    • South
    • Other Region

Derive descriptive statistics of your variables and encode them with dummy variables where needed.

Be very careful in your dummy variables encoding. In the end, you should use the variables as shown in Table 1 and Table 2 of the original publication.

Let's firstly create a mapping of the variables to the columns of the dataset based on the Codebook, so we have it for easy reference.

• Dependent Variables

  • Ban on Semi-Auto Guns: Q73A (Favor: 1, Oppose: 0)
  • Ban on High-Capacity Ammo Clips: Q73E (Favor: 1, Oppose: 0)
  • Banning Civilian Handguns: Q73G (Favor: 1, Oppose: 0)
  • Support for Concealed Carry Laws: Q73H (Favor: 1, Oppose: 0)
  • More Armed Security at Schools: Q73C (Favor: 1, Oppose: 0)
  • More Teachers/Faculty having Guns: Q73F (Favor: 1, Oppose: 0)
  • More Gun Safety Programs: Q73B (Favor: 1, Oppose: 0)
  • Expanded Mental Health Screening: Q73D (Favor: 1, Oppose: 0)

• Independept Variables

  • Religious Variables
    • Attendance: Q4 (0-8)
    • Bible (Human Error, History and Fables): Q17 options 3 and 4
    • Biblical Inerrancy: Q17 option 2
    • Biblical Literalism: Q17 option 1
  • Religious Affiliation
    • Conservative Protestant: RELTRAD option 1
    • Mainline Protestant: RELTRAD option 2
    • Black Protestant: RELTRAD option 3
    • Catholic: RELTRAD option 4
    • Other: RELTRAD options 5 and 6
    • No Affiliation: RELTRAD option 7
  • Political Ideology: Q31 (1-7)
  • Age: AGE
  • Sex
    • Female: Q77 option 2
    • Male: Q77 option 1
  • Race
    • White: Q88A option 1
    • Hispanic: Q89 options 2,3,4,5
    • African American/Black: Q88B option 1
    • Other: Q88C option 1, Q88D option 1, Q88E option 1, Q88F option 1
  • Education
    • Less Than High School: Q90 options 1 and 2
    • High School or Equivalent: Q90 option 3
    • Some College: Q90 options 4 and 5
    • College Degree: Q90 option 6
    • Post-graduate Degree =: Q90 option 7
  • Household Income: Q95
  • Marital Status: Q51A
    • Not Partnered/Single: Options 1,4,5,6
    • Married/Cohabitating: Options 2,3
  • Children Q92_NONE Q93_NONE CHILDSR
    • No kids under 18 in home
    • Kids under 18 in home
  • Area Demographics Q80
    • Small Town/Rural: Options 3,4
    • Urban Area: Options 1,2
  • Region
    • South: STATE
    • Other Region

Now, let's show how these will be as final columns in our dataset.

• Focal Variable
  • Supernatural Evil
    
    

• Religious Variables

  • Attendance: Q4 (0-8) We will leave attendance as is because it is a categorical ordinal variable
  • Biblical Inerrancy: 1 if Biblical Inerrancy
  • Biblical Literalism: 1 if Biblical Literalism We need 2 dummy variables to encode biblical inerrancy, biblical literalism and Bible (Human Error, History and Fables) as 0 and 0 will mean Bible (Human Error, History and Fables)
    
    
    • Religious Affiliation
    • Conservative Protestant: Column with 1 if Conservative Protestant
    • Mainline Protestant: Column with 1 if Mainline Protestant
    • Black Protestant: Column with 1 if Conservative Protestant
    • Catholic: Column with 1 if Conservative Protestant
    • Other: Column with 1 if Conservative Protestant We need 5 dummy variables to encode the above categories and No Affiliation. If everything is 0 it means No Affiliation
      
      

  • Political Ideology: Q31 (1-7) We will leave political ideology as is because it is a categorical ordinal variable
    
    

  • Age: AGE We will leave Age as is and replace Ages having 0 with the mean of Ages column ignoring zeros
    
    

  • Sex

    • Female: Column with 1 if Female We need 1 dummy for Gender, if 0 it means Male

  • Race

    • White: Q88A option 1
    • Hispanic: Column with 1 if Hispanic
    • Non-Hispanic Black: Column with 1 if African American/Black
    • Other: Column with 1 if Other We need 4 dummy variables for race, if all 0 then White

  • Education

    • Less Than High School: 1 if Less Than High School
    • Some College: 1 if Some College
    • College Degree: 1 if College
    • Post-graduate Degree: 1 if Post-graduate Degree We need 4 dummies for Education, if all 0 then High School or Equivalent

• Household Income: Q95 we leave it as it is
  
  
• Marital Status
  • Married/Cohabitating: 1 if Married/Cohabitating We need 1 dummy for marital status, if 0 then Not Partnered/Single
    
    
• Children
  • Kids under 18 in home: 1 if there are kids under 18 in home We need 1 dummy for this, if 0 then No kids under 18 in home
    
    
• Area Demographics
  • Urban Area: 1 if urban We need 1 dummy for this, if 0 then Rural
    
    
• Region
  • South: 1 if South We need 1 dummy for this, if 0 then Other Region

Note: For the Non-Hispanic Black there is an overlap between people being in the Black group and people in the Hispanic group but I am not removing anyone as there is no clear indicator how many Hispanics are Black.

bible_dict = {
    1: 'Biblical Literalism',
    2: 'Biblical Innerancy',
    3: 'Bible',
    4: 'Bible'
}

rel_aff_dict = {
    1: 'Conservative Protestant',
    2: 'Mainline Protestant',
    3: 'Black Protestant',
    4: 'Catholic',
    5: 'Other',
    6: 'Other',
    7: 'No Affiliation'
}

education_dict = {
    1: 'Less Than HS',
    2: 'Less Than High School',
    3: 'High School or Equivalent',
    4: 'Some College',
    5: 'Some College',
    6: 'College Graduate',
    7: 'Postgraduate'
}

marital_dict = {
    1: 'Not Partnered/Single',
    2: 'Married/Cohabitating',
    3: 'Married/Cohabitating',
    4: 'Not Partnered/Single',
    5: 'Not Partnered/Single',
    6: 'Not Partnered/Single'
}

place_dict = {
    1: 'Urban Area',
    2: 'Urban Area',
    3: 'Small Town/Rural',
    4: 'Small Town/Rural'
}

# https://stackoverflow.com/questions/26886653/pandas-create-new-column-based-on-values-from-other-columns-apply-a-function-o
def race(row):
    if row['Q88A'] == 1:
        return 'White'
    elif row['Q88B'] == 1:
        return 'Non-Hispanic Black'
    elif row['Q88C'] == 1 or row['Q88D'] == 1 or row['Q88E'] == 1 or row['Q88F'] == 1:
        return 'Other'
    elif row['Q89'] in [2, 3, 4, 5]:
        return 'Hispanic'
    
def children(row):
    if row['Q93_NONE'] == 1:
        return 'No kids under 18 in home'
    elif row['CHILDSR'] in [1, 2, 3]:
        return 'Kids under 18 in home'
    
def state(row):
    if row['STATE'] in ['MD', 'DE', 'VA', 'WV', 'KY', 'TN', 'NC', 'SC', 'FL', 'GA', 'AL', 'MS', 'LA', 'AK', 'TX', 'OK']:
        return 'South'
    else:
        return 'Other Region'

df['Gender'] = df['Q77'].apply(lambda x: 'Female' if x==2 else ('Male' if x==1 else x))
df['Attendance'] = df['Q4']
df['Bible Belief'] = df['Q17'].apply(lambda x: bible_dict[x] if x in bible_dict.keys() else x)
df['Religious Affiliation'] = df['RELTRAD'].map(rel_aff_dict)
df['Political Ideology'] = df['Q31']
df['Age'] = df['AGE']
df['Race'] = df.apply(lambda x: race(x), axis=1)
df['Education Level'] = df['Q90'].map(education_dict)
df['Household Income'] = df['Q95']
df['Marital Status'] = df['Q51A'].map(marital_dict)
df['Children'] = df.apply(lambda x: children(x), axis=1)
df['Place'] = df['Q80'].map(place_dict)
df['Region'] = df.apply(lambda x: state(x), axis=1)

df[['Gender', 'Attendance', 'Bible Belief', 'Religious Affiliation', 'Political Ideology', 'Race', 'Education Level', 'Marital Status', 'Children', 'Place', 'Region']].fillna()

for column in ['Gender', 'Attendance', 'Bible Belief', 'Religious Affiliation', 'Political Ideology', 'Race', 'Education Level', 'Marital Status', 'Children', 'Place', 'Region']:
    df[column].fillna(df[column].mode(), inplace=True)

for column in ['Age', 'Household Income']:
    df[column]. fillna(df[column].mean(), inplace=True)

df_dummies = pd.get_dummies(df[['Gender', 'Bible Belief', 'Religious Affiliation', 'Race', 'Education Level', 'Marital Status', 'Children', 'Place', 'Region']])

df2[['Supernatural Evil', 'Attendance', 'Biblical Inerrancy',
    'Biblical Literalism', 'Mainline Protestant', 'Black Protestant', 'Conservative Protestant',
    'Catholic', 'Other Religion', 'Political Ideology', 'Non-Hispanic Black', 'Hispanic', 'Other Race',
    'Female', 'Less Than HS', 'Some College', 'College Graduate', 'Postgraduate', 'Age', 'Income', 'Married',
    'Children', 'Urban', 'South']] = df.copy()[]

---

← Previous Post Next Post →

---

Copyright © Christodoulos Santorinaios 2022