This study measured implicit bias in Hindu-Muslim relations in India using an Implicit Association Test (IAT). The IAT was administered to college students and measured reaction times when pairing Hindu or Muslim concepts with good or bad evaluations. Statistical analysis found most students showed a slight implicit preference for Hindus over Muslims. Characteristics like gender, religion, and political views were also analyzed for their relationship to levels of implicit bias. The study contributes to limited research on measuring implicit attitudes regarding Hindu-Muslim relations in India and how individual characteristics relate to degrees of implicit bias.
1. Measuring Implicit Bias in Religion
Analyzing the invisibly inherent bias in the Hindu-Muslim thought.
Aashay Verma and Daksh Baheti
April, 2019
Abstract
Given the current political climate and the reliance of political parties on religion-based
support bases in India, it is important to study the biases that historical and contemporary
events have on the thought process of the Indian population; especially the electorate. A
major component of this study is the analysis of implicit associations that might not be
visibly apparent but might play a crucial role in the formation of ideas and opinions at all
levels of interaction: individual, group, societal and national, among others. This paper
analyses one such Implicit Association Test (IAT) which was conducted among college
students from varying political, geographical, linguistic and religious backgrounds. The
paper notes that there is indeed an inherent implicit bias within the sample towards Hindus
when compared to Muslims, and it tries to draw correlations between certain individual
characteristics and the degree of implicit association using statistical methods and tools.
Keywords: Hindu, Muslim, Implicit Association Test, Bias.
1
2. Introduction and Motivation
Ghosh and Kumar (1991) say that “The Hindu-Muslim equation has perplexed social
scientists in modern India” and that they are the most salient intergroup relations within
the country. After India gained independence from British rule in 1947, in its constitution
it called itself a secular, democratic republic, in a way committing to “unity in diversity”,
the slogan that echoed the Indian identity across the globe.
Although the first contact between the people of the two religions dates back to more
than a thousand years when Muslim kings invaded India and formed their kingdoms, ten-
sions between the two communities have been rising over the recent years in various spheres
of interaction. Engineer (1999) says, “It is banal to state that even after years of indepen-
dence, the Hindu-Muslim problem is as distant from a solution as it ever was. If anything,
it has worsened.” This deterioration and the perplexity, not only amongst academics, but
also in the general public, form the core of the drive on which this paper is partly based.
Communal clashes have taken place all over the country but have become more common
and frequent in areas with a relatively larger proportion of Muslims, typically ranging from
20% to 40% (Ghosh and Kumar, 1991). The relationship between the two communities has
worsened over the past couple of decades, with the infamous demolition of the Babri Mosque
in 1992 by Hindu fanatics, the 2002 riots in Gujarat, and hundreds of other isolated cases
that got press coverage in the mainstream as well as local media. The saliency of religious
identity is often the differentiating factor between life and death during instances of violence,
especially across North India (Hansen, 1996). In spite of all this, political parties continue
to strategize their rhetoric to win Muslim (minority) votes.
All these factors make the study of this Hindu-Muslim connection important given the
historical trajectory, especially the current political climate of the country. It is no secret
that India operates on several identities, but what is most intriguing is the parallel of two
notions of India being realized at the same time - that of a secular India as defined by the
2
3. constitution, and that of India as a Hindu nation (Battaglia, 2017 and Singh, 2005). This
paper explores the notion of this divide between the Hindu and Muslim community in the
realm of implicit thought using an Implicit Association Test.
We test for the presence of implicit biases, which by definition are not a part of conscious
thought, through an IAT which provides us with an opportunity to look beneath the surface
of conscious thought given the religious context. This paper adds to the amalgamation of
the thin strokes of literature that currently exists on the interaction of the religious divide
in India with implicit thought measurement.
The next section provides a brief review of the existing literature and is followed by
the section on methodology and data. The section titled ‘Results and Inference’ reports
and discusses the outputs from the IAT and tries to contextualize certain results. The
penultimate section sheds some light on the limitations of this study. An appendix is
provided containing additional information about the study.
Literature Review
The Implicit Association Test
The general saying: People don’t speak what they think can be substantiated by Green-
wald et al. (2009) - who note that individuals possess mental associations that have impor-
tant roles in cognition and behaviour and that these associations exist largely outside the
awareness of the individual - in conjunction with Fazio and Olson (2003) - who highlight
that individuals may be unwilling to admit some mental association.
This can be perceived as the difference between the observed and the unobserved when
it comes to the thought processes that go behind the formulation of ideas, opinions and
conversations. Undoubtedly, the measurement of such ‘implicit’ biases can serve as a very
important tool not only in psychological research (Greenwald and Banaji, 2017 and Nosek et
3
4. al., 2007) but also in analysing the manifestations of varying attitudes (Greenwald, Klinger
and Liu, 1989), stereotypes (Bargh, 1994 and Devine, 1989), priming effects, context effects
and introspective access (Greenwald and Banaji, 1995). One of the most important uses
of measurement of the implicit association is to get around the ‘difficult’ problem of the
social-desirability bias (see Devine et al., 2002).
Implicit attitudes, according to Greenwald and Banaji (1995), are the ‘manifestations of
actions or judgements’ which are controlled by ‘automatically activated evaluation’ without
the causal awareness of the performer. The Implicit Association Test, given by Greenwald,
McGhee and Schwartz in 1998, is, therefore, a procedure to measure such underlying auto-
matic evaluation which is similar in intent to cognitive priming procedures for measuring
automatic effect and attitude. In recent years, this test has taken the form of a computer-
based exercise and is measured through latencies and errors in individual screens of the
test.
The IAT requires the test taker to sort stimulus from four concepts using two response
options (with each response option being assigned to two stimuli). It relies on the notion
that sorting should be easier when the two concepts that share a response are strongly
associated as compared to when they are weakly associated (Nosek, Greenwald and Banaji,
2007). The procedural specification of each IAT is the same, the only differences are in the
choice of the four concepts and their stimuli.
Hindu-Muslim relations
This subsection reviews the literature that has studied the divide between Hindus and
Muslims in India - both theoretically and empirically. On the point of the theoretical
origins of this divide, Misra (2000) argues that, in independent India, it arose out of the lax
approach to bettering the relations between the two communities by the Congress party.
The paper notes the two policies that were put in place by India’s ‘founding fathers’ to
4
5. enable seamless integration of the minorities (Muslims) with the ‘mainstream’, and then
explains how these well-intentioned measures had an exactly opposite effect in India’s case,
i.e., the minorities taking advantage of these provisions to continue their own ways of life,
safe in the knowledge that the law of the land had protections for them, much to the
displeasure of the majority Hindu population. As a measure to curb the ‘hegemony’ of the
Muslim minorities, the paper continues, the Hindu communities resorted to ‘extreme forms
of violence and vandalism’ such as the demolition of the Babri Mosque in 1992 (by a crowd
of 2,00,000 Hindus).
This chain of thought emphasizes the role of religion in political bureaucracy and vice
versa, henceforth forming a sizable chunk of the existing literature on the Hindu-Muslim
divide and its contextual importance in historic and contemporary India. Similar theoretical
arguments can be found in papers by Stephan and Stephan, 2000 and Ghosh and Kumar,
1991 among others.
Another branch of the study in the existing literature is of the inspection of particular
events in history that pertain to the Hindu-Muslim divide. We review two such events -
the demolition of the Babri Mosque and the Hindu-Muslim Riots of Gujarat - to give a
taste of the sort of literature that deals with data on such issues. The goal here is to see
the dynamics of the interaction between the two communities and how it affected their
interactions.
After the demolition of the Babri Mosque in 1992, a survey was conducted asking re-
spondents whether they approved of the events that unfolded at the site. It was found
that the Hindu respondents were more approving of the incident, less inclined to criminally
charge the perpetrators and ‘apportioned’ lesser blame to the central government (Chhibber
and Misra, 1993). However, since Hinduism itself is composed of a wide variety of castes,
and it is not prudent to generalize this finding to this diverse community, the paper found
that this anti-Islamic sentiment that existed among the Hindus was not ‘a widespread phe-
5
6. nomenon’, but was limited to a few groups holding particular beliefs. This led the authors
to conclude that “insofar as communal feelings are supported by the Hindu middle classes,
they contain a real threat to India’s polity”.
They also show that, even after the incident, the sentiment of the Hindu community,
albeit a small section of it, towards the Muslims was still adversarial and had possibly
changed for the worse, given how they felt that this incident was justifiable in a retaliatory
manner to give themselves a sense of security by giving a subtle reminder to the Muslims
of their minority status in India. This not only reinforces the theory that Misra (2000) put
forth, but also acts as its extension after the events of the Babri Mosque incident.
Unfortunately, even as the Babri Mosque conflict might be isolated in its geography, it
is far from being the only Hindu-Muslim standoff that India has borne witness to. Brass
(2002) says, “Events labelled ‘Hindu-Muslim riots’ have been recurring features in India for
three-quarters of a century or more...there are numerous cities and towns where riots have
become endemic.” He points to the infamous riots of Gujarat when over 100,000 Muslims
were forced to shift to the state’s refugee camps under ‘abysmal’ standards of living. These
riots occurred at such a scale that the author has been prompted to describe it as a ‘pogrom’
1 and not a riot.
Even an incident like this - where so many lives were lost - did not affect the conser-
vative Hindu community in the country. The Vishwa Hindu Parishad, a prominent Hindu
nationalist body, described the Gujarat incident as ‘the first positive response of Hindus to
Muslim fundamentalism in 1000 years’ (Brass 2002). Brass (2002), in joint cohesion with
Chhibber and Misra (1993) concludes that if one were to take a closer look at the facts, it
would become apparent that these ‘endemic and widespread’ riots in India cannot really be
described in either of those ways, because such violent incidents take place in very particu-
lar locations around the country, and are the product of some specific relationships which
1
A pogrom is defined as an act of organized cruel behaviour or killing that is done to a large group of
people because of their race or religion.
6
7. became further devolved by the aftermath of these incidents.
The Interaction
Having looked at the literature that exists in both the realms - that of the Implicit
Association Test and Hindu Muslim relations separately - it becomes important to see their
intersection. The paper by Dunham et al. (2013) measures the development of implicit
and explicit attitudes towards caste and religion in minority status Muslim children and
majority status Hindu children. Their results indicate that, in the religion aspect, both
lower-status Muslim children and higher-status Hindu children show strong implicit in-group
preferences. They suggest that these in-group preferences act as a protective mechanism to
insulate children from the internalization of stigma.
Another paper by Barnhardt (2009) provides experimental evidence on the question
of religiously diverse neighbourhoods having effects on attitudes about other religions and
preferences for interreligious living. He conducts an explicit and implicit attitudes test
covering 1363 households and finds that there is a significant fall in the implicit bias among
Hindu children for Muslims when the interaction between the two increases. On the whole,
he observes the convergence of attitudes across religious groups.
However, given all the existing literature in the two fields (separate), the interaction
between the two spheres - that this paper aims to address - has remarkably received minimal
to no attention. Therefore, this paper contributes to the literature and sets itself as a
preliminary attempt to bring the above-mentioned interaction to light. This paper also sets
itself apart from other papers since it aims to paint a more general picture of the role and
manifestation of implicit bias in the context of the Hindu Muslim interaction.
7
8. Methodology and Data
The Implicit Association Test
This paper uses data collected through the administration of an Implicit Association Test
(IAT) among a few undergraduate students of Ashoka University. An IAT is usually taken by
a participant through computer software and it measures the strength of association of each
participant between ‘concepts’ (e.g., Hinduism and Islam for this study) and ‘evaluations’
(e.g., good or bad for this study). The test does this by recording the response times of each
participant on a number of screens where it pairs up a concept with an evaluation (e.g.,
Hinduism and Bad or Islam and Good). The idea is that a respondent with an implicit
bias for Hinduism will have a shorter response time for categorising Hinduism and Good in
the same corner as compared to Islam and Good, and vice versa. The test itself works on
the framework provided by Greenwald, McGhee and Schwartz (1998) and follows the same
methodology.
For the purposes of this paper, the IAT design had four main categories, with five stimuli
for each. The following table summarizes the categories and stimuli used.
Category Stimulus
Hindu Veda, Shiv, Hindu, Ram, Lakshman
Muslim Quran, Muslim, Hijab, Allah, Mohammed
Good Love, Peace, Joy, Happy, Success
Bad Hate, War, Evil, Pain, Failure
Survey
The experiment also included a demographic and opinion survey. Information such as
that on age (numeric entry), gender, faith, religiosity, opinions, major of study, political
preference (multiple choice) was collected (the appendix lists the questions presented). This
information was captured to draw compatative statics and conduct further analysis.
8
9. Analysis Framework
The paper, for the analysis of the data obtained, used comparative statics and a regres-
sion framework. The former is based on tabulations of data across variables obtained from
the survey and observing interesting patterns. The latter uses a linear regression framework
with the D score as the dependent variable and looks the the effect of various individual
characteristics on the degree of implicit bias. The full blown regression specification is as
follows:
(DScore)i = β0 + β1Agei + β2Genderi + β3Faithi + β4Religioni + β5Opinioni +
β6IATi + β7PPi + β7Majori + β8Newsi + φRegioni + i
Since many of these variables are categorical (including gender, faith, religion, opinion
etc.) in nature, the paper uses dummies to identify the individual impact of all such
categories on the dependent variable, along with region fixed effects. The construction
of these variables is explained in detail in the appendix. There were 40 participants but
owing to some technical issues, the data for 2 was unusable. Therefore, the total number of
datapoints were 38.
An additional variable called ’preference’ has been constructed and divided into 7 groups
for better visualization, namely: Strong preference for Muslims as compared to Hindus,
Moderate preference for Muslims as compared to Hindus, Slight preference for Muslims as
compared to Hindus, No preference, Slight preference for Hindus as compared to Muslims,
Moderate preference for Hindus as compared to Muslims and Strong preference for Hindus
as compared to Muslims.
Due to the paucity of datapoints, the full blown regression equation was not run. Instead,
the paper reports the regression results from the independent regression of some of the
explanatory variable on the dependent variable.
9
10. Comparative Statics and Discussion
Using multivariate tabulation to depict the relationship between the preferences, as
revealed by the Implicit Association Test and other individual characteristics, this section
provides some interesting patterns.
The overall sample has the following frequency distribution in relation to the preference
categories:
Category Absolute Number Percent
Strong Preference for Muslims 1 3
Moderate Preference for Muslims 6 16
Slight Preference for Muslims 2 5
No Preference 10 26
Slight Preference for Hindus 5 13
Moderate Preference for Hindus 7 18
Strong Preference for Hindus 7 18
Total 38 100
Table 1: Tabulation by Groups (rounded off to the nearest whole percent)
In varying degrees, 24% of the sample had a preference for Musilms over Hindus, a stark
26% had no preference, whereas around 50% of the sample showed a preference towards
Hindus as compared to Muslims. The sample consisted of 19 males and 19 females. However,
the religious distribution was not so balanced with 17 people identifying as Hindus, 12 as
Agnostic, 4 as Christian, 1 as Mulsim and 4 as None of the above.
Groups by Gender
As can be seen in the table, the proportion of participants that identified as female
and showed no preference for either religion is more than double of the participants that
identified as male in the same catogory. No participant who identified as a male exibited
a strog preference for Muslims as compared to Hindus whereas 5.3% of the participants
that identified as females belonged in the same category of preference. The proportions of
10
11. Gender
Groups Female Male Total
Strong Preference for Muslims 5.26 0.00 2.63
Moderate Preference for Muslims 10.53 21.05 15.79
Slight Preference for Muslims 5.26 5.26 5.26
No Preference 36.84 15.79 26.32
Slight Preference for Hindus 10.53 15.79 13.16
Moderate Preference for Hindus 15.79 21.05 18.42
Strong Preference for Hindus 15.79 21.05 18.42
Total 100.00 100.00 100.00
Table 2: Preference Groups by Gender (%)
participants that showed Slight preference for Muslims as compared to Hindus were equal
across tose who identified as a female or a male. Finally, more participants who identified
as male showed Strong preference for Hindus as compared to Muslims when compared to
those who identified as females in the same category.
Gropus by Opinion on Discrimination against Muslims in India
Participants were provided with the following prompt: Different social groups may face
discrimination in employment, education, housing, access to healthcare. and asked to rep-
sond to the following statement: Muslims are discriminated against in India. Do you agree?
The options were Strongly agree, slightly agree, slightly disagree and strongly disagree.
The neutral option was intentionally removed to make sure that the participants were forced
to think more and mark an option that truly representive of their beliefs, i.e., to get around
the bias.
11
12. Muslims are discriminated against in India
Groups Strongly Agree Slightly Agree Strongly Disagree
Strong Preference for Muslims 1 0 0
Moderate Preference for Muslims 4 2 0
Slight Preference for Muslims 2 0 0
No Preference 7 3 0
Slight Preference for Hindus 4 1 0
Moderate Preference for Hindus 5 1 1
Strong Preference for Hindus 6 1 0
Table 3: Groups by opinion on discrimination against muslims in India
As the table reports, no participant reported slight disagreement towards the given
statement. It is interesting to note that the participant who strongly disagreed with the
given statement had a moderate preference for Hindus as compared to Muslims. Of the
people who strongly agreed with the statement, 7 exibited no preference whereas 15 exibited
varying degrees of preference for Hindus over Muslims.
Regression Analysis and Discussion
The following tables (sections) present the results from the regression of the dependent
variable using the most intuitive independent varibale(s). In this case, as mentioned earlier,
the ceteris peribus interpretation does not hold for any result since no other factors are
controlled for while running these regressions.
12
13. D Score and Gender
Dependent Variable: D Score
VARIABLES D-Score
Gender = 2, Male 0.06
(0.13)
0.66
Constant, Female 0.25
(0.09)
0.01
Note: Standard errors in parentheses, p values below standard errors.
Results Table: D Score and Gender
As can be seen in the table above, when we don’t hold anything constant, the D Score for
the base category (those identifying as female), on average, is 0.25, indicating a preference
for Hindus over Muslims. This result is statistically significant at the 5% level of significance.
To interpret the result for those who identify as male, we need to add their regression
coefficients to the D Score of the base category, which is 0.06+0.25 = 0.31, which means
that on an average, those who identified as males had a higher implicit bias towards Hindus
than those who identified as females. This result was however not statistically significant
at any of the conventional levels of significance.
13
14. D Score and Faith
Dependent Variable: D Score
VARIABLES D-Score
Faith = 2, Christianity 0.63
(0.42)
0.14
Faith = 3, Hinduism 0.93
(0.38)
0.02
Faith = 4, Agnostic 0.84
(0.39)
0.04
Faith = 5, None of the above 0.88
(0.42)
0.04
Constant, Islam -0.57
(0.37)
0.14
Note: Standard errors in parentheses, p values below standard errors.
Results Table: D Score and Faith
Not holding anything constant, the average D Score of a participant who indicated Is-
lam as their faith (base category) is -0.57, indicating that they had an implicit bias towards
Muslims as opposed to Hindus. As before, the D Scores of the participants who indicated
their faith as something other than Islam can be obtained by adding their regression coeffi-
cients to that of the base category. Therefore, the average D Score for those who indicated
Hinduism was 0.93-0.57=0.36, which gives us an indication that those who indicated Hin-
duism as their faith exhibited an implicit bias towards Hindus as opposed to Muslims on an
average. This D Score was succeeded by those who indicated “None of the Above”, meaning
that they didn’t associate with any religion. The results for those who indicated Hinduism
and None of the Above as their faiths are significant at the 5% level of significance.
14
15. D Score and Major
Dependent Variable: D Score
VARIABLES D-Score
Major at Ashoka = 2, Computer Science and Entrepreneurship 0.64
(0.42)
0.14
Major at Ashoka = 3, Computer Science and Mathematics 0.76
(0.42)
0.09
Major at Ashoka = 4, Economics 0.07
(0.25)
0.79
Major at Ashoka = 5, Economics and Finance 0.41
(0.23)
0.09
Major at Ashoka = 6, English 0.11
(0.25)
0.66
Major at Ashoka = 7, English and Journalism 0.05
(0.42)
0.91
Major at Ashoka = 10, History and IR 0.52
(0.42)
0.23
Major at Ashoka = 11, Mathematics 0.44
(0.42)
0.31
Major at Ashoka = 12, Philosophy 0.14
(0.42)
0.75
Major at Ashoka = 13, Political Science -0.07
(0.25)
0.79
Major at Ashoka = 16, Psychology 0.34
(0.27)
0.22
Constant, Computer Science 0.07
(0.19)
0.70
Note: Standard errors in parentheses, p values below standard errors.
Results Table: D Score and Major
15
16. The above table presents the regression result of D Score of participants on the major
variable. Many interesting observations can be made from the table. When nothing else is
held constant, those participants majoring in Computer Science (base category) had a D
Score of 0.07, indicating a slight implicit bias towards Hindus over Muslims on an average.
To calculate the D Score of students of other majors, the respective regression coefficients of
the majors must be added to the D Score of the base category. This helps in making direct
comparisons among the majors. One thing worth noting is that those who were majoring in
Political Science had a D Score of -0.07+0.07 =0, indicating no implicit bias towards either
Hinduism or Islam. This result can be potentially attributed to the curriculum design of the
Political Science programme that exposes the students to multiple points of view through a
wide variety of courses (specifically on topics related to religion) in order for them to refine
their outlook towards the world.
On the other hand, those majoring in Computer Science and Mathematics had the
highest average D Score of all, 0.76+0.07 = 0.83, indicating that they had the highest
implicit bias for Hindus over Muslims on an average. Those majoring in Economics and
Economics and Finance had average D Scores of 0.14 and 0.48 respectively, indicating that,
on an average, the latter had more of an implicit bias for Hindus over Muslims than the
former. Of all the results, only those pertaining to the Computer Science, Mathematics and
Economics and Finance majors were statistically significant at the 10% level, and all others
were not significant at any conventional level of significance.
16
17. D Score and News following
Dependent Variable: D Score
VARIABLES D-Score
News Following = 2, Skim through occasionally -0.16
(0.14)
0.27
News Following = 3, Rarely find time -0.20
(0.21)
0.36
Constant, Follow the news closely 0.39
(0.12)
0.00
Note: Standard errors in parentheses, p values below standard errors.
Results Table: D Score and News Following
The participants were asked to report on how often they follow the news. The options
were: follow closely, skim through news and rarely follow news. The table presents the
regression result of this categorical variable on D-Scores.
The base category is of the respondents who indicated that they follow the news closely.
They have a D Score of 0.39, which is statistically significant at the 1% level, indicating
that they have an implicit bias for Hindus over Muslims. The other categories are of those
who say they rarely find time to follow the news (D Score = 0.19) and those who say they
skim through the news occasionally (D Score = 0.23). The latter two results lack statistical
significance at all conventional levels. A possible explanation for this trend in implicit biases
could be the surge in the anti-Islam and Islamophobic reporting that has been observed in
international media, which could possibly have influenced those who follow the news closely
to have more of an implicit bias against Muslims as compared to Hindus.
17
18. Limitations
Many scholars and psychologists have documented the flaws with the fundamental mech-
anism of the Implicit Association Test. Some of the major contentions are that of reliability
and validity of the test, the lack of its correlation with other established measures of preju-
dice and the high rates of false positives and false negatives associated with the test (Nagai,
2017).
Secondly, a case can always be made for slippage in the experimental design and the
execution of the experiment. Certain factors such as the number of fellow participants,
interaction with another participant before experiment, non-intentional priming etc. can
always lead to an implicit bias in the test takers and hence skew the data.
The major limitation of this paper however comes from the fact that it operates its entire
analysis on merely 38 data points. Furthermore, since the exclusion criteria that was used for
the study (only the Ashokan Undergraduate Body) was strict, no generalizable arguments
can be made about either the Ashokan population or the general population at large. The
lack of more data points also restricted the reach of this paper from multivariate regression
to only univariate regression which doesn’t allow for the ceteris paribus interpretation of
results.
18
19. References
Bargh, J. A. ”The four horsemen of automaticity: Awareness, intention, efficiency, and
control in social cognition”. Handbook of Social Cognition, 1994, pp. 1-40.
Battaglia, M. and Lebedinski, L. ”The curse of low expectations”. Economics of Tran-
sition and Institutional Change, vol. 25, 2017, pp. 681-721.
Barnhardt, Sharon. ”Near and Dear? Evaluating the Impact of Neighbor Diversity on
Inter-Religious Attitudes”. Working Paper, 2009.
Chhibber, Pradeep K., and Subhash Misra. “Hindus and the Babri Masjid: The Sec-
tional Basis of Communal Attitudes.” Asian Survey, vol. 33, no. 7, 1993, pp. 665–672.
Devine, P. G. ”Stereotypes and prejudice: Their automatic and controlled components.”
Journal of Personality and Social Psychology, vol. 56, no. 1, 1989, pp. 5-18.
Devine, P. G., Plant, E. A., Amodio, D. M., Harmon-Jones, E., and Vance, S. L. ”The
regulation of explicit and implicit race bias: The role of motivations to respond without
prejudice.” Journal of Personality and Social Psychology, vol. 82, no. 5, 2002, pp. 835-848.
Dunham, Yarrow, Srinivasan, Mahesh, Dotsch, Ron and Barner, David. ”Religion in-
sulates ingroup evaluations: the development of intergroup attitudes in India”. Develope-
mental Science, 2013, pp. 1-9.
Engineer, Asghar Ali. “Resolving Hindu-Muslim Problem: An Approach.” Economic
and Political Weekly, vol. 34, no. 7, 1999,pp. 396-400.
Ghosh, Emmanuel S. K., and Rashmi Kumar. “Hindu-Muslim Intergroup Relations in
India: Applying Socio-Psychological Perspectives.” Psychology and Developing Societies,
vol. 3, no. 1, 1991, pp. 93–112.
Ghosh, Emmanuel S. K., and Rashmi Kumar. “Hindu-Muslim Intergroup Relations in
India: Applying Socio-Psychological Perspectives.” Psychology and Developing Societies,
vol. 3, no. 1, 1991, pp. 93–112.
Greenwald, A. G., McGhee, D. E., and Schwartz, J. L. K. ”Measuring individual dif-
19
20. ferences in implicit cognition: The implicit association test.” Journal of Personality and
Social Psychology, vol. 74, no. 6, 1998, pp. 1464-1480.
Greenwald, A. G., Poehlman, T. A., Uhlmann, E. L., and Banaji, M. R. “Understanding
and using the Implicit Association Test: III. Meta-analysis of predictive validity.” Journal
of Personality and Social Psychology, vol. 97, no. 1, 2009, pp. 17-41.
Greenwald, Anthony G. and Banaji, Mahzarin R. ”Implicit Social Congition: Attitudes,
Self-Esteem and Stereotypes.” Psycological Review, vol. 102, no. 1, 1995, pp. 4-27.
Greenwald, Anthony G. and Banaji, Mahzarin R. ”The Implicit Revolution: Reconceiv-
ing the Relation Between Conscious and Unconscious.” American Psychologist, vol. 72, no.
9, 2017, pp. 861-871.
Greenwald, Anthony G., Klinger, Mark R., and Liu, Thomas J. ”Unconscious processing
of dichoptically masked words”. Memory & Cognition, vol. 17, no. 1, 1989, pp. 35-47.
Hansen, Thomas Blom. “Recuperating Masculinity: Hindu Nationalism, Violence and
the Exorcism of the Muslim ‘Other.’” Critique of Anthropology, vol. 16, no. 2, 1996, pp.
137-172.
Misra, Amalendu. “Hindu Nationalism and Muslim Minority Rights in India.” Interna-
tional Journal on Minority and Group Rights, vol. 7, no. 1, 2000, pp. 1–18.
Nosek, B. A., Greenwald, A. G., and Banaji, Mahzarin R. ”The Implicit Association
Test at age 7: A methodological and conceptual review”. Psychology Press. 2007, pp.
265–292.
Olson, Michael A., and Russell H. Fazio. Relations between Implicit Measures of reju-
dice: What Are We Measuring?” Psychological Science, vol. 14, no. 6, 2003, pp. 36–639.
Singh, Pritam. “Hindu Bias in India’s ’Secular’ Constitution: Probing Flaws in the
Instruments of Governance.” Third World Quarterly, vol. 26, no. 6, 2005, pp. 909–926.
Stephan, W. G., and Stephan, C. W. ”An integrated threat theory of prejudice. Applied
Social Psychology, 2000, pp. 23-45
20
21. Appendix
Demographic and Opinion Survey
A demographic and opinion survey was conducted following the IAT test (within the
software). The purpose of the survey was to capture individual specific characterstics that
would be then used to draw descriptive and comparative statics.
The detailed survey, along with response options, is as follows:
1. What is your age? (numeric entry)
2. What is your gender?
a) Female
b) Male
c) Non-Binary
3. What is your faith?
a) Islam
b) Christianity
c) Hinduism
d) Agnostic
e) None of the above
4. How religious do you consider yourself to be?
a) Not at all
b) Somewhat religious
c) Very religious
5. Different social groups may face discrimination in employment, education, housing,
access to healthcare.
Muslims are discriminated against in India.” Do you agree?
21
22. a) Strongly Agree
b) Slightly Agree
c) Slightly Disagree
d) Strongly Disagree
6. How warm or cold do you feel towards Islam?
a) Extremely Warm
b) Very Warm
c) Somewhat Warm
d) Slightly Warm
e) Slightly Cold
f) Somewhat Cold
g) Very Cold
h) Extremely Cold
7. Which statement best describes you?
a) I strongly prefer Islam to Hinduism
b) I moderately prefer Islam to Hinduism
c) I slightly prefer Islam to Hinduism
d) I slightly prefer Hinduism to Islam
e) I moderately prefer Hinduism to Islam
f) I strongly prefer Hinduism to Islam
8. How many Implicit Association Tests (IATs) have you previously performed?
a) 0
b) 1
c) 2
d) 3-5
e) 6+
22
23. 9. What is your political preference?
a) I tilt towards the right, I am likely to vote for BJP
b) I tilt towards the left, I am likely to vote for Congress
c) I am likely to vote for a third party
d) I am likely to vote for no one.
10. What is your major at Ashoka University?
a) Computer Science
b) Computer Science and Entrepreneurship
c) Computer Science and Mathematics
d) Economics
e) Economics and Finance
f) English
g) English and Journalism
h) English and Creative Writing
i) History
j) History and International Relations
k) Mathematics
l) Philosophy
m) Political Science
n) Politics and Society
o) Politics, Philosophy and Economics
p) Psychology
q) Sociology and Anthropology
11. Which part of India do you consider yourself to be from?
a) North (J&K, Himachal, Punjab, Chandigarh, Uttarakhand, Haryana, Delhi, Ra-
jasthan, UP)
23
24. b) North East (Arunachal, Nagaland, Manipur, Mizoram, Tripura, Assam, Meghalaya,
Sikkim)
c) East (Bihar, Jharkhand, West Bengal, Odisha)
d) Central (Madhya Pradesh, Chhattisgarh)
e) West (Gujarat, Maharashtra, Goa)
f) South (Telangana, Andhra, Karnataka, Tamil Nadu, Kerala)
g) Islander (Lakshadweep, Andaman)
12. Which of the following statements describes you best?
a) I follow the news closely: I read it every day.
b) I skim through the news occasionally.
c) I rarely find time to update myself with current affairs.
13. Which of the following brackets does your annual family income fall under?
a) INR 0 - 5,00,000
b) INR 5,00,000 - 10,00,000
c) INR 10,00,000 - 25,00,000
d) INR 25,00,000 - 50,00,000
e) More than INR 50,00,000
f) I’d rather not say
14. What is your mother tongue? (text response)
Construction of Variables
The D Score variable that has been used is the output of the Implicit Association Test.
No changes have been made to the variable.
Age is a numeric variable that takes values between 19 and 21 years. Since not much
variation is observed in this variable, it has been omitted from the main body of the paper.
24
25. Gender is a categorical variable that takes value 1 for Female, 2 for Male and 3 for Non-
Binary. However, there are no respondents in the 3rd category, but to retain the structure
of the origional data, the gender variable has not been converted to a dummy.
The variable Faith is a categorical variable that takes the value 1 for Islam, 2 for Chris-
tianity, 3 for Hinduism, 4 for Agnostic and 5 for None of the above.
The variable Religiosity is a categorical variable that takes the value 1 for Not at all, 2
for Somewhat religious and 3 for very religious.
The variable Opinion 1 is a categorical variable that takes the value 1 for Strongly agree,
2 for Slightly agree and 3 for Strongly disagree.
The variable Opinion 2 is a categorical variable that takes the value 1 for Extremely
warm, 2 for Very warm, 3 for Somewhat warm, 4 for Slightly warm, 5 for Slightly cold, 6
for Somewhat warm and 7 for Very cold.
The variable Preference is a categorical variable that takes value 1 for Moderately prefer
Islam, 2 for Slightly prefer Islam, 3 for Slightly prefer Hinduism, 4 for Moderately prefer
Hinduism and 5 for Strongly prefer Hinduism.
The variable Ploitical Preference is a categorical variable that takes the value 1 for
Rightist (BJP), 2 for Leftist (Congress), 3 for 3rd Party and 4 for Nobody.
The variable Major is a categorical variable that takes the value 1 for Computer Sci-
ence, 2 for Computer Science and Entrepreneurship, 4 for Economics, 5 for Economics and
Finance, 6 for English, 7 for English and Journalism, 10 for History and IR, 11 for Math-
ematics, 12 for Philosophy, 13 for Political Science, 16 for Psychology. Majors with values
3,8,9,14,15 have been dropped since they have no observations in the dataset. However, to
preserve the structure of the data, the majors have not been renumbered.
The variable Geographic is a categorical varibale which takes a value 1 for North, 2 for
East, 3 for West and 4 for South. The number 5 (islanders) have not been mentioned since
it had no observations.
25
26. The variable News is a categorical variable that takes the value 1 for Follow closely, 2
for Skim through occasionally and 3 for Rarely find time.
The variable Group is a categorical variable that has been created using intervals. It
ranges from Strong preference towards Muslims as compared to Hindus to Strong preference
towards Hindus as compared to Muslims. The exact categories can be found in the main
body of the paper.
The variable IAT is a categorical variable that takes the value 0 for 0 IATs, 1 for 1 IATs,
2 for 2 IATs, 3 for 3-5 IATs and 4 for 6+ IATs.
Experimental Setting
The entire test was conducted in a controlled environment on the 1st April, 2019 from
1900 hours to 2100 hours in Academic Block 02 (AC 02), LR 005, Ashoka University,
Sonipat, Haryana.
A sign up sheet was personally circulated (by the experimenters) to approximately 60
participants out of which 40 participated in the experiment. All participants were asked to
bring thier laptop devices with them. The slots were divided into 10 minutes each over the
duration of 2 hours. Since the data for 2 test takers was corrupted, the final sample size
was 38.
Instructions to Participants
There were no verbal instructions given to any participant. As they entered the exper-
iment room, they were directed towards a seat on which a consent form was kept. The
following written instructions were projected on the whiteboard throughout the course of
the experiment:
Please sit wherever you find a consent form and fill it.
Check your inbox for an email from Aashay. Click on the link and follow the instructions.
26
27. The test is anonymous.
The test ends when you’re redirected to a website containing some information.
Thank you so much for your time!
Consent Form
The consent form was collected as soon as the participant indicated that they had
completed the study. They were then, outside the experiment room, debriefed. All consent
forms have been attached as a separate file with this paper.
The name of any respondent was not captured during any part of the study to maintain
confidentiality. Only signatures of the participants were captured on the consent form.
Hence, no one-to-one idetification of any participant is possible in the dataset.
Here is a sample consent form:
Subject Eligibility
Only the undergraduate student body at Ashoka university was used, as test takers,
for this study. The test takes were from different batches and different majors across the
undergraduate batch. No other exclusion criteria was used for this study in terms of sample
selection.
27
28. Technical Details
IAT Software, by Michael, was used for conducting the Implicit Association Test. The
details of the software can be found at https://www.iatsoftware.net. The authors of
this paper were in constant touch with the creator of the software for technical support.
The .iat file which contains the code for the formation of the IAT has been attached for
reference.
The raw data file containing raw D-Scores and other data captured from the IAT software
was downloaded in .xlsx format and was accessed using Microsoft Excel which is a part of
the Microsoft Office Professional Plus 2016 package. The raw data file has been attached
for reference.
Activities such as data cleaning, labelling, grouping, analysis, graphing and tabulations
were performed using Stata 14.2 which is a software by StataCorp LLC. The log file is
embedded at the end of this document and the do file has been attached for reference. The
latter is a systematic guide to the activities that were performed on the data and can be
used to replicate the results in this paper.
The documentation and final output (this docuument) was produced using the LateX
package provided by Visual Studio Code.
Experiment Photographs
The following photographs were captured during the experiment at random times. The
date and time stamp feature was applied to ensure the authenticity of the images.
28
32. Presentation Feedback
Two points were brought up during the presentation of this paper in the Economics of
Discrimination Lecture(s). The first question was raised by Professor Deshpande and had
to do with estimation. In particular, she asked why other explanatory variables were not
included in the estimations to which the presenters responded. The reason for the above
question had already been incorporated in the first draft.
A second query was raised and had to do with the distribution of the sample with respect
to religion. The information has been included in the ’Comparative Statics and Discussion’
section.
No further feedback was recieved.
Author Information
Aashay Verma is a third year undergraduate at Ashoka University pursuing a degree in
Economics and Finance. He can be reached at this email address.
Daksh Baheti is a third year undergraduate at Ashoka University pursuing a degree in
Economics and Finance. He can be reached at this email address.
32