Econometrics Project

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 18

Exploring the Influence of Religious Beliefs on Attitudes towards

Gender and Gender Roles: A Cross-Country Analysis of Economic


Development Levels
Eva Craig - 20334236

1. Hypothesis
An individual's religious beliefs have an effect on their attitudes towards gender roles & these
perceptions differ across different levels of economic development of a country.

2. Introduction
This econometrics paper analyses the impact of religious values on perceptions towards
misogyny using data from the World Values Survey's (WVS) 7th wave. I will investigate
whether the relationship between respondents' religious levels and their views on whether men
make better executives, whether men have more of a right to a job than women and whether
women making more money would cause problems in a marriage. The results indicate that
religious values could have an impact on whether an individual has misogynistic values. The
study also conducts a heterogeneity test to compare perceptions of gender roles between
developed and developing countries. The findings suggest contradictory evidence to support
the overall hypothesis. Altogether, this study contributes to the growing body of research on
the role of religion in shaping attitudes towards gender roles.

3. Theoretical Framework
The literature review for this econometrics paper investigates the impact of religious beliefs on
individuals' attitudes towards gender roles, as well as the potential differences in these
perceptions across various levels of economic development. A study conducted by Inglehart,
R., Norris, P., and Welzel, C., which argues that as societies become more economically
developed, they tend to experience a shift towards more liberal values, such as increased
support for gender equality (Inglehart et al.,2019). The series of essays, ‘Religion and Sexism’
argues that certain religions are the single most important shaper and enforcer of the image and
role of women in Western society. The author explores how many religions focus on an
exclusively male, patriarchal God, corresponding with feminine imagery of subservience and
roles of worship (Ruether,1998). An enriching contemporary study, by Etengoff&Lefevor,
examines how individual differences (e.g. conservatism), congregational (e.g.
doctrine/policies, limited contact), and cultural factors (e.g. power hierarchies), which are
associated with the development of a country, mutually contribute to the landscape of religion
and feminism. These individual differences highlight that perceptions about feminism and
religion likely differ across different levels of economic development of a country
(Etengoff&Lefevor, 2021). It is important to note that not all religious traditions promote these
values. Some religious traditions promote gender equality and challenge traditional gender
roles. The research suggests that religion can have a significant impact on attitudes towards
gender and gender roles, with individuals who hold more conservative religious beliefs often
exhibiting more traditional views on these issues. Additionally, the literature indicates that the
relationship between religion, gender, and economic development may vary significantly
across different cultural contexts.
4. Data and Descriptive Statistics
The WVS is an international research program that conducts a representative comparative
social survey globally every five years, using data from 77 countries and 129,000 respondents,
including developed and developing countries. The most recent 7th wave of the survey was
conducted from 2017-2020, with sample sizes ranging from N=1000 to N=5000, depending on
population size. The questionnaire is face-to-face and asks questions on various perceptions
about culture, attitudes and beliefs towards gender, family, and religion etc. The questionnaire
itself occurred at respondents place of residence. Respondents answered each question on a
scale from agree to disagree (WorldValuesSurvey,2022).

FIG 1: Example Question

FIG 2: Sample size for each country (WVS)

Methodology:
WVS is an association ensured a random sample of the full adult population. This was achieved
through a multi-stage territorial stratified sampling process, where the population was divided
into subgroups and final subjects were selected randomly from different strata in proportion to
their population(WorldValuesSurvey,2022). A Simple probability sampling was used for the
second stage, resulting in a comprehensive and representative sample. There are possible
limitations to this internal validity I explain below.

FIG 3: Variable names


Data Descriptions
Column1
for Selected Variables
Column2 Column3 Column4
Variable Variable name Question Description Answer: Dummy Variable
y1 Mmorerighttojob Q33 When jobs are scarce, men should have more right to a job than women 1 if agree (sexist perceptions) 0 if disagree (equal perceptions)
y2 Wmoneyproblems Q35 If a woman earns more money than her husband, it's almost certain to cause problems 1 if agree (sexist perceptions) 0 if disagree (equal perceptions)
y3 Mbetterexec Q31 On the whole, men make better business executives than women do 1 if agree (sexist perceptions) 0 if disagree (equal perceptions)
q_170 Q170 The only acceptable religion is my religion. 1 if agree(religious) 0 if disagree (not religious)
q_171 Q171 Apart from weddings and funerals, about how often do you attend religious services these days? 1 if more than monthly (religious) 0 if less than monthly (not religio
q_173 Q173 Independently of whether you attend religious services or not, would you say you are…? 1 if religious person(religious) 0 if not religious person (not religious
q_164 Q164 How important is God in your life? 1 if important (religious) 0 if not important (not religious)
q_6 Q6 How important is Religion in your life? 1 if important (religious) 0 if not important (not religious)
deveopled If country is developed or developing 1 if developed 0 if developing
dev_q_170 Interaction term between developed variable and the only acceptable religion is my religion.
dev_q_171 Interaction term between developed variable and how often do you attend religious services
dev_q_173 Interaction term between developed variable and agree you are religious person
dev_q_164 Interaction term between developed variable and agree religion is important in your life
dev_q_6 Interaction term between developed variable and how important is Religion in your life?

5. Empirical Framework
i) Estimation Strategy
This research focuses on analysing the impact of religious values on sexist perceptions using
three dependent variables (Y1, Y2, and Y3). The study uses five questions related to religious
beliefs and scales the responses to 1 or 0, representing agreement or disagreement, respectively.
The research employs statistical inference, utilizing R2 values and P-values to measure the
significance of the independent variables.

The first relationship is examined using linear OLS robust regression, which accounts for
outliers that may affect the data. The second relationship is explored using a logistic regression
model, preferably to a probit model. Finally, a heterogeneity test is conducted to investigate
whether perceptions towards gender inequality differ across countries with varying economic
development levels.

Overall, this research provides insight into the relationship between religious values and sexist
perceptions and how it differs across countries. The study's use of various regression models
and statistical inference techniques strengthens its analysis and conclusions.

ii) Identification strategy


To establish a causal relationship between Y-variables and dependent variables , this study uses
a linear probability model. This study examines the relationship between perceptions about
gender and religion using dummy-variables to generate a control group. By comparing the
views about gender roles for individuals who are and are not religious, we can assess the causal
effect of religion. Additionally, there may be other factors that influence the relationship
between the variables that are not controlled for in the experiment. These factors can lead to
omitted-variable-bias, making it difficult to estimate the true causal effect of the variables of
interest.
To better model the relationship between the dependent and independent variables, this study
uses a logit model instead of the OLS linear regression model. This is because the linear
regression model assumes a linear relationship, which may not hold in all cases. By using a
logit model, we can assume a logistic distribution that may better match the survey data. This
allows for a more accurate estimation of the relationship between the variables of interest.

This study also conducts a heterogeneity test to evaluate how living in a developed country
versus a developing country affects a person's views on gender roles and religion. To account
for any missing relationships between living in a developed region and having religious values,
interaction-terms are included in the model.

6. Empirical Results
The coefficients in this regression are mostly negative which indicates that people who have
strong religious values is associated with disagreeing with: ‘when jobs are scarce men have
more right to a job than women’. Each variable is statistically significant, each with a P-value
of 0. The R2 is 0.148 which indicates 15% of the included variables offer explanatory power
for the dependant variable, the omitted variables are included in the error term. There is likely
omitted-variable-bias present in this regression and I tried to reduce this by including as many
variables included in the WVS questionnaire that are correlated with sexist perceptions and
religion. These regressions are robust meaning that it is less sensitive to outliers.

The coefficient -0.33 associated with q_170 (see Fig) demonstrates the expected change on
the sexism variable when a person agrees that their religion is the only acceptable religion (in
comparison to if they disagree), holding all other independent variables constant. This means
that the likelihood of being an individual agreeing that men have more right to a job than
women, is 33% lower for individuals who are view their religion as their only acceptable
religion compared to those who disagree, holding all other variables constant. This negative
coefficient indicates that being religious is associated with a lower likelihood of being sexist.

There is a positive coefficient for Q173. The coefficient (0.017) means the likelihood of having
sexist perceptions is 1.7% higher for individuals who are religious compared to those who are
not religious, ceteris paribus. This result does support my hypothesis as it highlights that being
sexist is associated with a higher likelihood of being religious.
We can see a similar result in for the dependant variable y2 which represents the respondent’s
answer to Q35: If a woman earns more money than her husband, it's almost certain to cause
problems. The only positive coefficient was for Q173 which is explained above. The
coefficient is 0.011 which means that the likelihood of having inequality gender perceptions is
1.1% higher for individuals who view themselves as religious compared to those who do not ,
ceteris paribus. This variable however is insignificant at the 10% level which may make it an
unreliable source of evidence.
This OLS linear-regression provided interesting results as nearly all of the coefficients are
positive and statistically significant which provides evidence for my hypothesis that that being
sexist is associated with a higher likelihood of being religious. The coefficient associated with
Q_170 conveys that the likelihood of being an individual agreeing that men make better
executives than women, is 25% higher for individuals who are view their religion as their only
acceptable religion compared to those who disagree, holding all other variables constant.

If one agrees their religion is the only acceptable religion then they are less likely to think
women are better or equal executives to men. Q171, 164 and 6 are different measures of
individuals religious values (see reference FIG) and demonstrate positive coefficients which
indicate that being religious is associated with a higher likelihood of being sexist, supporting
my hypothesis.

Summary
There is likely omitted-variable-bias which could lead to inaccurate estimates of the true
relationship between the variables. The first regression includes both negative and positive
coefficients, indicating evidence both supporting and contradicting the hypothesis that
religious values impact an individual's sexist values. The second regression also shows a
similar trend with slightly more evidence contradicting the literature review. In contrast, the
third regression presents stronger supporting evidence, as it mostly includes positive
coefficients supporting the hypothesis. These results bring us closer making a causal-statement
about the relationship between perceptions about religion and misogyny.

Logit Regression Model


I compared using a probit model and logistical model and found the logistical regression to be
a better match my for my data, as it demonstrated a higher R2 value.

Using a logit model is advantageous as it enables a non-linear relationship between the


independent variables and the dependent variable, which is not possible with linear OLS
regression. Linear OLS regression is not always appropriate for binary dependent variables
because it assumes a continuous, normally distributed dependent variable (AmiyaRanjanRout,
2023).

The Pseudo R2 is lower in each logistical regression than the OLS linear regression models
previously depicted. This indicates that the OLS linear-regression is better suited to the data.
The coefficient affiliated with q_170 is -1.39 which highlights a concern in the linear
probability model. The probability of being an individual agreeing that men have more right to
a job than women, is 139% lower for individuals who are view their religion as their only
acceptable religion compared to those who disagree, ceteris paribus, (assuming a logistical
distributions). This is over 4 times the value associated with the same relationship in the OLS
model. There is a large discrepancy because the coefficients in a logistic regression model
represent the change in the log odds of the dependent variable for a one-unit increase in the
independent variable, while the coefficients in an OLS regression model represent the change
in the dependent variable for a one-unit increase in the independent variable. This negative
coefficient indicates that being religious is associated with a lower likelihood of being sexist.

Summary
Again, there is likely omitted-variables which are correlated with the independent and
dependant variables which would cause my estimators to be biased. This table includes both
negative and positive coefficients, indicating evidence both supporting and contradicting the
hypothesis that religious values impact an individual's sexist values.

Heterogeneity Test
The heterogeneity test allows us to assess how living in a developed country versus a
developing region affects individuals' views on gender roles and religion. To account for any
missing relationships between living in a developed region and having religious values, we
include interaction-terms in the model. Specifically, we include an interaction-term between
the independent variable and a set of dummy variables that represent the differing countries'
economic development. The inclusion of this interaction-term captures the degree to which the
effect of sexism perceptions varies across different countries with different levels of
development. By measuring if these interaction-terms are statistically significant, we can
determine the presence of heterogeneity in the relationship between the variables. This
heterogeneity test is the final regression series in our study and provides fruitful insights into
the complex relationship between living in a developed region and having religious values.
I am focusing particularly on Y1 and Y3 as they are a better representation of the perception
of gender roles rather than household relationships.

We can see that the R2 value for the following regressions is higher than the previous
regressions. This indicates that the new variables added (developed and the interaction
variables) create a model that better fits the data.
Developed variable
The coefficient associated with the developed variable is the average difference in Y1 between
developed and not developed respondents when a person is not religious (q_171, q_164, q_170,
q_6, q_173 = 0). This tells us that there is an average difference of 18% in sexist perceptions
(when a person has sexist perceptions versus them having more equal perceptions) in
developed countries when a person is not religious ceteris paribus.

Original variables
The coefficient of the original variables have all changed in this regression and which
highlights that the previous regressions had omitted-variable-bias. The coefficient for q_6
(dummy variable = 1 when person perceives religion as important in their life, 0 if otherwise)
is -0.0658 which tells us that the likelihood that someone has misogynistic values is 6.58%
lower in non-developed countries, ceteris paribus. The coefficient for q_171, q_173 and q_164
are insignificant at the 5% level which leads me to eliminate their explanatory value. The other
dependant variable which measures strong religious values is q_170 which is also negative
which indicates that religious people are less likely to be misogynistic in non-developed
countries.

Interaction-terms
The significant interaction-terms are: dev_q_164,dev_q_170 and dev_q_13. The coefficient
on dev_q_164 represents there is a 1.3% increase in the probability that someone is sexist due
to being religious (people who view God as important in their life) in a developed country,
versus being religious in a developing country, holding all other variables constant. This
interaction-term is extremely interesting as it supports my previous claim. The variables
dev_q_170 and dev_q_13 are also positive coefficients which indicates that religious values
have a causal-link to misogynistic values between developed and developing countries.
Developed variable
The coefficient associated with the developed variable is -0.0843 which explains the average
difference in sexist perceptions between developed and not developed respondent’s when they
don’t hold strong religious values. When a person is not religious (original variables=0), the
coefficient linked to the developed variable represents the mean variation in Y1 between
respondents who live in developed areas and those who do not. Specifically, the coefficient
value of -0.0843 indicates that the average difference in sexist perceptions between developed
and non-developed respondents is -8.43% when they do not strongly identify with religious
beliefs, ceteris paribus.

Original variables
The coefficients associated with the original variables are all positive and have changed from
the first regression series; this enforces that the earlier regressions suffered from omitted-
variable-bias as the coefficients were correlated with the variables added. The q_171 and q_173
variables are not significant (which differs from earlier regressions which indicates that their
earlier significance was likely due to the presence of the added terms). The coefficient for the
q_6 variable (respondent agrees that religion is important in their life) indicates that the average
likelihood that someone has sexist perceptions is 3.37% higher, when religious, than having
equal perceptions in non-developed countries, ceteris paribus. The coefficient for the q_164
variable, which represents the respondent's agreement that God is important in their life,
indicates that in non-developed countries, the average probability of someone having sexist
perceptions is 6.77%, when religious, higher than the probability of having equal perceptions,
holding other variables constant. In summary, these coefficients indicate a positive relationship
between having religious values and being having misogynistic values in non-developed
countries. This is contradictory to the previous heterogeneity test as it led us to believe the
opposite, this model has more significant variables which indicates higher illustrative power

Interaction-terms
The regression analysis identified three significant interaction-terms, namely dev_q_164,
dev_q_170, and dev_q_713. Notably, dev_q_164 showed a coefficient of -1.6%, which
represents the average difference in the probability of someone being displaying misogyny
(Y3) based on their belief in the importance of God in developed and developing countries.
The metric indicates that individuals who view God as important in their lives are 1.6% less
likely to be considered sexist than those who don't hold this belief, across both developed and
developing countries. This finding is particularly intriguing as it contradicts the initial
hypothesis.

On the other hand, dev_q_170 and dev_q_713 had positive coefficients, indicating that
religious values have a causal-link to misogynistic values in both developed and developing
countries. These variables provide strong support for the hypothesis and validate the theoretical
framework presented in the paper.

Summary
The results show that religious people are less likely to be misogynistic in non-developed
countries, but those who view God as important in their lives are more likely to be considered
sexist. In addition, the likelihood of someone having sexist perceptions is higher in non-
developed countries. These variables provide strong evidence for the hypothesis and validate
the theoretical framework presented in the paper. However, there may be omitted-variable-
bias.

The variables provide support for the hypothesis that religious beliefs have an impact on
attitudes towards gender and gender roles, but also make many contradictions. Nonetheless,
there is still likely omitted-variable-bias. These results bring us closer than ever to making a
causal statement about the relationship. It is difficult to examine if there is an upward or
downward bias as the coefficients don’t consistently over estimate or under estimate the more
accurate values in this model.

7. Discussion, possible extensions

To check for robustness, it might be possible to look at different measures of the Y variables
which measure if an individual has more sexist perceptions (Bellemare, 2020). They could
include more questions in the WVS survey which examine people’s perception of gender roles
(e.g. Do you think women are less capable to make rational decisions?). This would allow us
to examine more directly people’s views about gender roles.

Further treatment heterogeneity checks could be conducted. I would love to extend this study
across more subgroups such as (ethnicity, eductaion)(Bellemare,2020). This would allow us to
examine whether the different groups have significant impacts on the Y-variables.
Limitations
Due to likely omitted-variables, which are correlated with our Yvariables and religious
perception variables, we might not be able to make a casual statement, however this paper
offers evidence in the literature and in the regression models to paint us a better picture of the
relationship between religion and misogyny, across developed and developing countries.

There may also be internal validity issues within the survey as it was being conducted. An
example of this is if they conducted this survey to only people who own houses, then they are
not including the millions of homeless people around the world or people living in slums.
People who live in slums or are homeless are less likely to be educated (due to low-income)
which may cause them to have more sexist values. This would threaten the internal and external
validity of the data.

8. Summary, Concluding Remarks


Based on the regressions and analysis performed in this paper, there appears to be some support
for the hypothesis that religious beliefs impact an individual's attitudes towards gender roles,
and that this effect varies across countries with different levels of economic development.
However, the results are not conclusive and contain several limitations, including potential
omitted-variable-bias. While the regressions do support the claim to some extent, they also
present some contradictory findings. Nevertheless, this study represents a step in the right
direction towards better understanding the relationship between religious beliefs and attitudes
towards gender roles, and further research can build upon these findings to gain a more
comprehensive understanding of the topic. Overall, the study provides important insights into
the complex relationship between religion, culture, and gender equality.

9. References
- World Values Survey. (2022). WVS Database. WVS Database.
https://www.worldvaluessurvey.org/WVSContents.jsp
- Bellemare, M.F., 2020. How to Write Applied Papers in Economics [WWW
Document]. URL http://marcfbellemare.com/wordpress/wp-
content/uploads/2020/09/BellemareHowToPaperSeptember2020.pdf (accessed
4.13.23).
- AmiyaRanjanRout. (2023, January 10). Advantages And Disadvantages Of Logistic
Regression - GeeksforGeeks. GeeksforGeeks.
https://www.geeksforgeeks.org/advantages-and-disadvantages-of-logistic-regression/
- Inglehart, R., Norris, P., Welzel, C., 2019. Cultural Evolution, People’s Motivations
are Changing, and Reshaping the World. Cambridge.
- Ruether, R., 1998. Religion and Sexism. Wipf and Stock Publishers.
- Etengoff, C., & Lefevor, T. G. (2021). Sexual prejudice, sexism, and religion. Current
Opinion in Psychology, 40, 45–50. https://doi.org/10.1016/j.copsyc.2020.08.024
-
10. Appendix
Data source :
https://www.worldvaluessurvey.org/WVSContents.jsp
- Stratified sampling
- Cross sectional data
- Wave 7 :2017-2021
- 77 countries

DO FILE

import delimited using "/Users/eva/Downloads/Econometrics Project/ECON DATA.csv",


clear varn(1)

keep a_year b_country b_country_alpha q31 q33 q33_3 q35 q35_3 q6 q164 q170 q171 q174
q173

br

des

*renames*
rename q31 Mbetterexec
rename q33_3 Mmorerighttojob
rename q35_3 Wmoneyproblems

**Regressions***
reg Mbetterexec q6 q164 q170 q171 q174 q173, robust

*************elimiate 3*
reg Mmorerighttojob q6 q164 q170 q171 q174 q173 if Mmorerighttojob!=3

reg Wmoneyproblems q6 q164 q170 q171 q174 q173 if Wmoneyproblems!=3

*label variables dependant *


gen

label def labela 0"sexistperceptions" 1 "equalperceptions"


foreach x in Mmorerighttojob Wmoneyproblems{
label var `x' labela
}
gen y1 = 0 if Mmorerighttojob == 1 | Mmorerighttojob == 2
replace y1 = 1 if Mmorerighttojob == 2

gen y2 = 0 if Wmoneyproblems == 1 | Wmoneyproblems == 2


replace y2 = 1 if Wmoneyproblems == 2

gen y3 = 0 if Mbetterexec>0
replace y3 = 1 if Mbetterexec == 1 | Mbetterexec == 2

label def labelb 1"sexist perceptions" 2 "sexistperceptions" 3 "middle" 4 "equality" 5


"equality"
label var Mmorerighttojob Wmoneyproblems labelb

***code independant variables into dummys* religion variables****

*The only acceptable religion is my religion*


gen q_170 = 0 if q170>0
replace q_170 = 1 if q170 == 1 | q170 == 2

*how often do you attend religious services*


gen q_171 = 0 if q171>0
replace q_171 = 1 if q171 == 1 | q171 == 2 | q171 == 3

*are you a religious person*


gen q_173 = 0 if q173 == 2 | q173 == 3
replace q_173 = 1 if q173 == 1

*How important is God in your life?*


gen q_164 = 0 if q164>0
replace q_164 = 1 if q164 == 5 | q164 == 6 | q164 == 7 | q164 == 8 | q164 == 9 | q164 == 10

*how important religion is in your life?*


gen q_6 = 0 if q6>0
replace q_6 = 1 if q6 == 1 | q6 ==2

*regressions with coded dummys in independant variables*


***

reg y1 q_170 q_171 q_173 q_164 q_6, robust

outreg2 using OLSregressions1.doc

reg y2 q_170 q_171 q_173 q_164 q_6, robust

outreg2 using OLSregressions2.doc


reg y3 q_170 q_171 q_173 q_164 q_6, robust

outreg2 using OLSregressions3.doc

*****run robust regression*****

reg Mmorerighttojob q_170 q_171 q_173 q_164 q_6 if Mmorerighttojob!=3, robust

reg Wmoneyproblems q_170 q_171 q_173 q_164 q_6 if Wmoneyproblems!=3, robust

***
****probit model*??** asssume normal distribution*
probit y1 q_170 q_171 q_173 q_164 q_6
margins, dydx(*)
*odds ratio**

probit y2 q_170 q_171 q_173 q_164 q_6


margins, dydx(*)

probit y3 q_170 q_171 q_173 q_164 q_6


margins, dydx(*)

*assume logistic distribution*


logit y1 q_170 q_171 q_173 q_164 q_6
outreg2 using logy1.doc

logit y2 q_170 q_171 q_173 q_164 q_6

logit y3 q_170 q_171 q_173 q_164 q_6

*hetreogeneity test between developed and non-developed countries*

gen developed = 1 if b_country == 20 | b_country == 32 | b_country == 124 | b_country ==


158 | b_country == 196 | b_country == 203 | b_country == 276 | b_country == 300 |
b_country == 344 | b_country == 392 | b_country == 410 | b_country == 446 | b_country ==
528 | b_country == 554 | b_country == 702 | b_country == 703 | b_country == 826 |
b_country == 840 | b_country == 909
replace developed = 0 if mi(developed)

reg y1 developed, robust

reg y2 developed, robust

reg y3 developed, robust

reg developed q_170 q_171 q_173 q_164 q_6, robust


gen dev_q_170=developed*q_170
gen dev_q_171=developed*q_171
gen dev_q_6=developed*q_6
gen dev_q_173=developed*q_173
gen dev_q_164=developed*q_164

reg y1 developed q_6 dev_q_6 q_164 dev_q_164 q_170 dev_q_170 q_171 dev_q_171 q_173
dev_q_173, robust

outreg2 using y1.doc

reg y2 developed q_6 dev_q_6 q_164 dev_q_164 q_170 dev_q_170 q_171 dev_q_171 q_173
dev_q_173, robust

reg y3 developed q_6 dev_q_6 q_164 dev_q_164 q_170 dev_q_170 q_171 dev_q_171 q_173
dev_q_173, robust

outreg2 using y3.doc

You might also like