Content uploaded by Steffen Zitzmann
Author content
All content in this area was uploaded by Steffen Zitzmann on Jun 19, 2021
Content may be subject to copyright.
https://doi.org/10.1177/01461672211024422
Personality and Social
Psychology Bulletin
1 –21
© 2021 by the Society for Personality
and Social Psychology, Inc
Article reuse guidelines:
sagepub.com/journals-permissions
DOI: 10.1177/01461672211024422
journals.sagepub.com/home/pspb
Empirical Research Paper
As reflected in many political and societal debates, manag-
ing diversity has become increasingly challenging for plural-
istic societies. Besides the many positive aspects that come
with diversity (e.g., cultural exchange and learning), there is
also the risk that, driven by categorizations of others into
“us” or “them,” diversity can lead to conflicts between
groups (intergroup conflict). This has led to calls for mutual
tolerance (Scanlon, 2003). However, many scholars (e.g.,
Simon, Eschert, et al., 2019; Verkuyten & Yogeeswaran,
2017; Vogt, 1997) have pointed out that tolerance has
scarcely been addressed systematically in social-psychologi-
cal theorizing and empirical research.
In social psychology, tolerating others has long been
thought to involve liking their beliefs, preferences, and prac-
tices, or regarding them as something good (see Simon,
Eschert, et al., 2019). However, in recent years, social psy-
chologists have adopted ideas from philosophy (e.g., from
Scanlon, 2003; see also Forst, 2013) that have gradually
shifted their understanding toward a theoretically more
sophisticated view of tolerance as the attitude that one accepts
the different ways of life practiced by outgroups (i.e., their
beliefs, preferences, and practices) despite one’s disapproval
of them. According to this emerging view, disapproval of oth-
ers (i.e., of their ways of life) is a definitional condition for
tolerance. In other words, it makes sense to say that one toler-
ates other people or things only when one first disapproves of
them (e.g., Gibson et al., 1992). Whereas tolerance is a popu-
lar and loaded word not only in everyday social and political
life but also in social scientific discourses, a great deal of
controversy surrounds the exact meaning of tolerance. This
situation testifies to the “difficulty of tolerance” (Scanlon,
2003) as an interpretive concept, which allows for rival
interpretations (Dworkin, 2013). In other words, tolerance is
1024422PSPXXX10.1177/01461672211024422Personality and Social Psychology BulletinZitzmann et al.
research-article2021
1University of Tübingen, Germany
2Kiel University, Germany
3University Medical Center Hamburg-Eppendorf, Germany
Corresponding Author:
Steffen Zitzmann, Hector Research Institute of Education Sciences and
Psychology, University of Tübingen, Europastraße 6, Tübingen 72072,
Germany.
Email: steffen.zitzmann@uni-tuebingen.de
Does Respect Foster Tolerance?
(Re)analyzing and Synthesizing
Data From a Large Research Project
Using Meta-Analytic Techniques
Steffen Zitzmann1, Lukas Loreth2,
Klaus Michael Reininger3, and Bernd Simon2
Abstract
Our own prior research has demonstrated that respect for disapproved others predicts and might foster tolerance toward
them. This means that without giving up their disapproval of others’ way of life, people can tolerate others when they
respect them as equals (outgroup respect–tolerance hypothesis). Still, there was considerable variation in the study features.
Moreover, the studies are part of a larger research project that affords many additional tests of our hypothesis. To achieve
integration along with a more robust understanding of the relation between respect and tolerance, we (re)analyzed all existing
data from this project, and we synthesized the results with the help of meta-analytic techniques. The average standardized
regression coefficient, which describes the relationship between respect and tolerance, was 0.25 (95% confidence interval
[CI] = [0.16, 0.34]). In addition to this overall confirmation of our hypothesis, the size of this coefficient varied with a
number of variables. It was larger for numerical majorities than for minorities, smaller for high-status than for low-status
groups, and larger for religious than for life-style groups. These findings should inspire further theory development and spur
growth in the social-psychological literature on tolerance.
Keywords
disapproval, respect, tolerance, meta-analysis
Received October 18, 2019; revision accepted May 18, 2021
2 Personality and Social Psychology Bulletin 00(0)
a contested concept, which allows and in fact calls for com-
peting conceptions. Note that we do not aspire to dictate or
prescribe a specific definition of tolerance. Rather, we offer a
particular conception of tolerance that sheds new light on the
processes and phenomena associated with tolerance and thus
helps to deepen our understanding of the value of tolerance
but also of the controversies that surround it.
On the basis of this understanding of tolerance, social
psychologists have begun to study the sources behind toler-
ance. One source is associated with the multilevel nature of
social categorization processes. Whereas groups often differ
in their ways of life, they share membership in higher level
groups such as a common society or nation-state (Turner
et al., 1987). Such a common higher level group membership
then operates as a social-psychological source of a mutual
recognition of equality (and corresponding entitlements)
across lower level group boundaries (Simon, 2020). Members
of different (lower level) groups can thus recognize each
other as equals at the higher level (e.g., as fellow citizens
with the same rights, duties, and liberties).
Moreover, in line with Honneth’s (1995) recognition the-
ory, which assumes that equality recognition underlies
respect, Simon and Grabow (2014) found that out of the
three recognition principles (need, equality, and achievement
recognition), equality recognition was indeed most predic-
tive of the experience of being respected (see also Simon
et al., 2015). Therefore, respect is hereafter also referred to as
respect for others as equals or equality-based respect (see,
for example, Eschert & Simon, 2019; Renger et al., 2017).
Notice that unlike alternative conceptions (e.g., Huo &
Molina, 2006), according to the definition employed by us
here, respecting others as equals is not at odds with disap-
proving of them (see Crane, 2017; Scanlon, 2003; see also
Simon, Eschert, et al., 2019; Verkuyten et al., 2019); that is,
one can respect others as equal citizens, while disapproving
of their ways of life.
The role of respect in tolerance has been highlighted by the
disapproval-respect model of tolerance (DRM; Simon &
Schaefer, 2016). The central hypothesis derived from this
model is the outgroup respect–tolerance hypothesis. According
to this hypothesis, people’s respect for outgroups they disap-
prove of influences their tolerance toward these groups. In line
with the hypothesis, Simon and Schaefer (2016) found that the
respect paid by Muslims in Germany to groups they disap-
prove of predicted the Muslims’ tolerance toward these groups.
More recently, using a longitudinal research design with time-
lagged measures of tolerance, Simon, Eschert, et al. (2019)
showed that Tea Party supporters’ respect for groups they dis-
approve of can foster the development of tolerance toward
these groups over time (Study 1). Moreover, experimental data
have suggested that respect even causally influences tolerance
(Simon, Eschert, et al., 2019; Study 2).
However, although empirical research has supported the
notion that tolerance and respect are related to one another,
the sizes reported for this relation (i.e., the standardized
regression coefficients) have varied considerably, possibly
due to differences between studies, which exist, for example,
in the sample or the design of the study. Therefore, the aim of
the present study was to disentangle these sources of varia-
tion and, thus, to provide a more comprehensive picture of
the relation between respect and tolerance. To this end, we
(re)analyzed the data that were collected for a large multi-
study research project directed by the last author and synthe-
sized the results with the help of meta-analytic techniques
while taking study variability into account. We investigated
the following substantive research questions and method-
ological questions. Because our overall aim was to test the
outgroup respect–tolerance hypothesis, we first asked
whether respect would predict tolerance on average (aggre-
gated across the results for this relation).
Second, we also examined how the results would vary in
accordance with the features of the studies or the data sets.
Exploring moderators considered in theoretical models is
one way to enrich theoretical reasoning and to contribute to
the further development of these models.
Therefore, we investigated moderators of theoretical
importance, which is novel to the research on the DRM but
quite common in social-psychological research in general.
For example, Tropp and Pettigrew (2005) found that the rela-
tionship between the variables that they studied differed
between members of a majority group and members of a
minority group, which the authors explained by a psycho-
logical asymmetry between these two groups. The DRM in
its current form does not predict that—or how—the respect–
tolerance link should vary as a function of people’s social
position in society at large, although such variables may
affect the level of tolerance expressed toward outgroups
(Simon et al., 2001). Nevertheless, given the role of these
variables in other social-psychological models and research,
one may also speculate about their moderating influence on
the respect–tolerance link. It stands to reason that, for exam-
ple, the members of a numerical majority or a high-status
group, which are particularly likely to regard themselves as
prototypical of and thus normative for society at large, may
be in greater need of a good reason to tolerate the “deviant”
others. Respect for them should be such a reason so that the
more respect members of majorities or high-status groups
actually grant them, the stronger their reason for tolerance
toward them. In other words, we speculated that tolerance
among members of majorities or high-status groups could be
more dependent on explicit reasoning or considerations con-
cerning their respect for others which should then be reflected
in a stronger respect–tolerance link relative to that for mem-
bers of minority or low-status groups. The same could be
true for religious groups. Relative to non-religious groups,
they may also be in greater need of a good reason to tolerate
those who fail to see “the Godly truths.” Taken together,
these group membership variables lend themselves as plau-
sible candidates to an empirical investigation of potential
moderator variables.
Zitzmann et al. 3
Moreover, we were interested in whether the methodol-
ogy employed in the different studies would affect the
respect–tolerance link. To give an example, to test the out-
group respect–tolerance hypothesis, Simon and Schaefer
(2016) employed a cross-sectional design, which—as with
all correlational studies—relied on the naturally occurring
variation between variables. Unlike this study, Simon,
Eschert, et al. (2019) employed a longitudinal design (Study
1). However, because longitudinal studies are still vulnerable
to unwanted confounding by unobserved third variables, the
authors additionally conducted an experiment (Study 2) that
allowed for more control by means of randomization. Thus,
we asked whether the design would explain variation in
results for the respect–tolerance link.
In statistical terms, one way to model this link is with a
linear model with tolerance as the dependent variable and a
measure of respect or a manipulation thereof as the predictor
variable in cross-sectional or experimental designs and, in
longitudinal designs, often also with an additional measure
of tolerance as the lagged criterion. At the same time, one
might wish to exclude reasonable alternative explanations by
adding one or more “third variables” as covariates to the
model. Even in the analysis of experimental data, it might be
helpful to include such covariates, particularly when ran-
domization is not perfect and confounding cannot be pre-
cluded. We thus asked whether choosing a model with
covariates would affect the results.
Finally, constructs in social psychology cannot be mea-
sured without error. A person’s attitude or belief could be
assessed by asking him or her to rate the respective variable
either on a single item or across multiple items, which are
then aggregated into a scale score. From a measurement per-
spective, the scale score can be more reliable than a single
item. However, in general, neither the single item nor the
scale score can provide a perfectly reliable measure, and if
errors in the predictor variables are ignored, regression coef-
ficients will be biased (e.g., Fuller, 1987). To mitigate this
problem, one can adopt a latent-variable approach, which
models a variable’s true score and thus accounts for the mea-
surement error. In fact, the research considered in this article
has used different methods to assess respect. Therefore, in
addition to the analysis models with or without covariates,
our analytic approach covers three types of measures. We
raised the question of whether these types would yield results
of varying size.1 Specifically, we raised the question of
whether the use of the latent variable would yield larger sizes
of the relation between respect and tolerance than the use of
the scale score or a single item, and whether the use of the
scale score would yield larger sizes than the use of a single
item. Note that these are directed hypotheses.
The Present Work
Recent social-psychological research has indicated that
respect for others of whom one disapproves influences
one’s tolerance toward them. The aim of the present arti-
cle is twofold: First, despite a slow start, empirical work
on the respect–tolerance link has been accumulating, and
several studies have been published in the past few years
(e.g., Simon, Eschert, et al., 2019; Simon & Schaefer,
2016). The present work was aimed at summarizing this
emerging trend. Specifically, we tested whether, on aver-
age, respect influences tolerance, and we disentangled
sources of variation in the strength of the respect–toler-
ance link.
The added value of our study is noteworthy. Employing
meta-analytic techniques, we synthesized a comprehensive
corpus of research on the relation between respect and toler-
ance, and we believe this is an important milestone for future
research on tolerance in general. Moreover, we conducted
several moderator analyses, some of which were theoreti-
cally motivated, whereas others were more exploratory in
nature. We believe that studying moderators and finding evi-
dence for moderating effects can inform future study designs
in important ways, contribute to the development of inter-
ventions for promoting tolerance, and promote scientifically
informed debates in society at large on the important topic of
mutual tolerance.
In line with the DRM, we expected that the average size
of the respect–tolerance link would be positive and substan-
tial (main research question). Theoretical and methodologi-
cal thinking (see above) led us to investigate a number of
possible substantive and methodological moderators, which
are summarized in Table 1 (see also the “Coding” section).
For exploratory reasons, we also included a number of other
potential moderators. These analyses were exploratory
because theoretical assumptions about how these variables
would influence the relation between respect and tolerance
did not exist. Therefore, we did not state more specific
research questions for these variables.
Method
Data Retrieval and Preprocessing
The data came from a large research project directed by the
last author concerning life in pluralistic societies. To gain
access to the data from this project, we asked contributors to
the research project to submit the data sets from their already
published (Krys et al., 2020; Paffrath & Simon, 2020; Schaefer
& Simon, 2019; Simon, Eschert, et al., 2019; Simon et al.,
2016, 2019; Simon & Schaefer, 2016, 2018) or not yet pub-
lished articles (Schaefer et al., 2021) that included the relevant
measurements for integration into our analysis. Besides mea-
sures of tolerance and approval, the data sets had to contain at
least one measure of respect. We received a total of 11 distinct,
large data sets in response to the request, and we split these
into smaller subsets. An overview is provided in Table 2,
which shows the relevant features of the requested data (sub)
sets. For more details, see the Supplemental Material.
4 Personality and Social Psychology Bulletin 00(0)
According to the DRM, for tolerance toward a particular
group to exist, this group has to be an outgroup that is met
with disapproval. Because of this, participants had to be
excluded if they were members of that particular group or if
they did not disapprove of that group. To this end, for each
participant, we checked for whether the rating of his or her
approval on a 7-point bipolar scale ranging from −3 to 3
(with 0 as midpoint) was equal to or greater than zero. Note
that the approval measure assessed approval of what the out-
group stands for (i.e., the members’ beliefs, preferences, and
practices). Thereby, it tapped the psychologically or subjec-
tively essential, defining feature of the outgroup as a collec-
tive rather than some stereotypical characteristic. It also did
not assess any kind of interpersonal disliking of individual
outgroup members.
Coding
The second and last authors coded each data set along mul-
tiple dimensions: membership in a numerical minority or
majority, social status, type of group, design, analysis
model, respect measure, country of residence, sample,
mean disapproval, mean age, percentage of women, and
number of participants. Interrater agreement was perfect
except for the outgroup’s social status and its type of
group, where Rater 2 argued that for the outgroup of
Table 1. Overview of Moderators.
Type of moderator Moderator Categories
Substantive Numerical minority or majority (sample) Minority
Majority
Numerical minority or majority (outgroup) Minority
Majority
Social status (sample) Low
High
Social status (outgroup) Low
High
Type of group (sample) Political
Religious
Life-style
Type of group (outgroup) Political
Religious
Life-style
Ethnic
Methodological Design Cross-sectional
Longitudinal
Experimental
Analysis model No covariates
Covariates included
Respect measure Single item
Scale score
Latent variable
Additional Country of residenceaGermany
Poland
The United States
Brazil
SampleaTea Party supporters
Catholics
Protestants
Muslims
Alevis
LGBTs
University students
Mean agea—
Percentage of womena—
Number of participantsa—
Note. LGBTs = lesbian, gay, bisexual, and transgender individuals.
aModerators included for exploratory reasons.
5
Table 2. Selected Features of the Requested Data Sets.
Study ID Design
Sample characteristics
Outgroup
Standardized regression coefficient
Sample
Country of
residence Mean age
Percentage
of women Est. SEs L95%CI U95%CI
#01 (data used by
Schaefer etal., 2021;
Simon & Schaefer,
2016)
Cross-sectional
(T1)
Muslims Germany 30.39 49.1 Atheists 0.57 0.03 0.51 0.64
31.17 35.8 Christians 0.61 0.07 0.47 0.76
29.97 40.6 Feminists 0.58 0.04 0.50 0.66
29.91 42.8 Homosexuals 0.50 0.04 0.42 0.57
30.30 38.0 Jews 0.69 0.05 0.58 0.79
Cross-sectional
(T2)
Muslims Germany 30.92 48.9 Atheists 0.65 0.02 0.61 0.70
30.85 48.2 Christians 0.28 0.03 0.22 0.34
30.87 47.7 Feminists 0.39 0.03 0.33 0.45
30.80 48.1 Homosexuals 0.58 0.02 0.53 0.62
30.81 48.5 Jews 0.71 0.02 0.67 0.76
Longitudinal
(T1–T2)
Muslims Germany 30.47 49.6 Atheists 0.37 0.04 0.28 0.46
31.24 33.1 Christians 0.18 0.12 −0.06 0.43
30.08 39.6 Feminists 0.60 0.05 0.51 0.69
29.93 42.8 Homosexuals 0.12 0.04 0.04 0.21
30.50 36.4 Jews −0.05 0.11 −0.27 0.16
#02 (data used by Krys
etal., 2020; Schaefer
etal., 2021; Simon
etal., 2016)
Cross-sectional
(T1)
Protestants Brazil 40.36 31.8 Afro-Brazilians 0.30 0.07 0.16 0.43
40.20 30.4 Atheists 0.36 0.05 0.26 0.47
41.49 28.6 Catholics 0.29 0.10 0.10 0.48
39.95 29.9 Feminists 0.17 0.06 0.05 0.29
40.03 30.1 Homosexuals 0.25 0.06 0.13 0.36
Cross-sectional
(T2)
Protestants Brazil 39.63 28.6 Afro-Brazilians 0.30 0.05 0.20 0.40
39.64 29.2 Atheists 0.30 0.05 0.19 0.40
39.65 28.2 Catholics 0.10 0.05 −0.01 0.21
39.78 28.9 Homosexuals −0.07 0.06 −0.19 0.04
Longitudinal
(T1–T2)
Protestants Brazil 40.36 31.8 Afro-Brazilians 0.06 0.05 −0.04 0.16
40.20 30.4 Atheists 0.07 0.05 −0.02 0.17
41.49 28.6 Catholics 0.66 0.04 0.59 0.73
40.03 30.1 Homosexuals 0.01 0.05 −0.07 0.10
#03 (data used by Krys
etal., 2020; Schaefer
etal., 2021)
Cross-sectional
(T1)
Protestants Germany 49.55 46.6 Atheists 0.28 0.03 0.21 0.35
47.40 48.7 Catholics 0.31 0.07 0.18 0.44
48.55 42.4 Homosexuals 0.32 0.04 0.24 0.39
48.55 42.4 Muslims 0.27 0.04 0.20 0.34
Cross-sectional
(T2)
Protestants Germany 50.55 47.5 Atheists 0.37 0.03 0.32 0.43
50.71 47.3 Catholics 0.36 0.03 0.30 0.42
50.56 46.3 Homosexuals 0.36 0.03 0.30 0.42
50.61 46.6 Muslims 0.36 0.03 0.30 0.42
(continued)
6
Study ID Design
Sample characteristics
Outgroup
Standardized regression coefficient
Sample
Country of
residence Mean age
Percentage
of women Est. SEs L95%CI U95%CI
Longitudinal
(T1–T2)
Protestants Germany 49.14 46.4 Atheists 0.14 0.03 0.07 0.20
46.76 46.2 Catholics 0.26 0.06 0.13 0.38
48.36 42.1 Homosexuals 0.13 0.04 0.05 0.21
48.29 42.1 Muslims 0.05 0.04 −0.03 0.13
#04 (data used by
Reininger etal., 2020;
Schaefer etal., 2021;
Simon, Eschert, etal.,
2019; Simon, Reininger,
etal., 2019)
Cross-sectional
(T1)
Tea Party
supporters
The United
States
44.51 46.0 Conservatives 0.65 0.11 0.43 0.87
53.17 40.9 Homosexuals 0.50 0.04 0.42 0.58
54.31 40.1 Liberals 0.42 0.04 0.35 0.49
53.40 43.7 Muslims 0.55 0.03 0.49 0.61
Cross-sectional
(T2)
Tea Party
supporters
The United
States
49.44 49.6 Conservatives 0.13 0.04 0.05 0.21
51.24 47.2 Homosexuals 0.51 0.03 0.45 0.57
52.33 44.9 Liberals 0.50 0.03 0.44 0.57
51.69 46.6 Muslims 0.59 0.03 0.53 0.65
Cross-sectional
(T3)
Tea Party
supporters
The United
States
50.53 47.2 Conservatives 0.06 0.04 −0.01 0.14
51.44 46.0 Homosexuals 0.53 0.03 0.47 0.58
52.14 45.0 Liberals 0.45 0.03 0.39 0.50
51.84 45.8 Muslims 0.54 0.03 0.48 0.60
Longitudinal
(T1–T2)
Tea Party
supporters
The United
States
52.81 42.0 Homosexuals 0.22 0.04 0.14 0.31
54.47 40.5 Liberals 0.30 0.04 0.23 0.37
53.18 44.6 Muslims 0.22 0.04 0.14 0.30
Longitudinal
(T1–T3)
Tea Party
supporters
The United
States
53.09 41.1 Homosexuals 0.36 0.04 0.28 0.44
54.41 39.9 Liberals 0.25 0.04 0.18 0.33
53.40 43.9 Muslims 0.18 0.04 0.10 0.27
Longitudinal
(T2–T3)
Tea Party
supporters
The United
States
51.03 47.2 Homosexuals 0.28 0.04 0.21 0.34
52.30 45.2 Liberals 0.17 0.03 0.11 0.24
51.63 46.4 Muslims 0.14 0.04 0.07 0.22
#05 (data used by
Schaefer etal., 2021)
Cross-sectional
(T1)
Alevis Germany 28.54 52.7 Atheists 0.45 0.08 0.29 0.62
29.81 35.4 Feminists 0.60 0.09 0.42 0.78
32.13 30.8 Homosexuals 0.58 0.08 0.43 0.73
31.93 46.4 Jews 0.63 0.09 0.45 0.82
32.51 42.9 Sunnis 0.41 0.07 0.28 0.54
Cross-sectional
(T2)
Alevis Germany 30.60 49.9 Atheists −0.07 0.05 −0.17 0.03
30.61 49.0 Christians 0.00 0.05 −0.10 0.09
30.55 49.2 Feminists 0.04 0.05 −0.05 0.14
30.66 49.4 Jews −0.03 0.05 −0.12 0.07
30.47 48.9 Sunnis 0.19 0.05 0.09 0.28
Table 2. (continued)
(continued)
7
Study ID Design
Sample characteristics
Outgroup
Standardized regression coefficient
Sample
Country of
residence Mean age
Percentage
of women Est. SEs L95%CI U95%CI
Longitudinal
(T1–T2)
Alevis Germany 28.68 51.9 Atheists 0.23 0.08 0.08 0.38
32.90 41.6 Sunnis 0.29 0.06 0.17 0.40
#06 (data used by
Paffrath & Simon, 2020;
Schaefer etal., 2021)
Cross-sectional
(T1)
Catholics Poland 40.81 48.1 Atheists 0.55 0.04 0.46 0.63
39.20 36.7 Feminists 0.52 0.05 0.44 0.61
40.75 36.2 Homosexuals 0.58 0.04 0.50 0.66
39.25 48.0 Muslims 0.50 0.04 0.41 0.58
38.86 43.6 Russian Orthodox people 0.73 0.07 0.60 0.86
Cross-sectional
(T2)
Catholics Poland 40.33 50.7 Atheists 0.54 0.03 0.48 0.60
39.97 48.3 Feminists 0.58 0.03 0.52 0.64
40.11 49.3 Homosexuals 0.70 0.03 0.64 0.75
40.28 50.8 Muslims 0.40 0.03 0.34 0.46
39.30 51.6 Russian Orthodox people 0.64 0.03 0.58 0.70
Longitudinal
(T1–T2)
Catholics Poland 40.33 46.1 Atheists 0.14 0.05 0.03 0.24
38.38 34.8 Feminists 0.19 0.06 0.08 0.30
40.42 36.3 Homosexuals 0.28 0.06 0.16 0.39
38.75 47.5 Muslims 0.16 0.05 0.06 0.26
37.23 48.9 Russian Orthodox people −0.18 0.08 −0.34 −0.02
#07 (data used by
Schaefer etal., 2021)
Cross-sectional
(T1)
LGBTs The United
States
38.97 42.3 African Americans 0.58 0.10 0.38 0.78
35.00 56.1 Asian Americans 0.45 0.12 0.21 0.68
38.02 51.2 Latinos/Hispanics 0.57 0.11 0.35 0.80
33.11 61.0 Tea Party members 0.44 0.04 0.35 0.52
32.71 59.1 Religious people 0.45 0.06 0.34 0.56
Cross-sectional
(T2)
LGBTs The United
States
33.33 60.7 African Americans 0.15 0.05 0.06 0.24
33.68 60.6 Latinos/Hispanics 0.13 0.04 0.05 0.22
33.05 59.9 Tea Party members 0.62 0.04 0.55 0.70
32.90 59.7 Religious people 0.39 0.04 0.30 0.47
Longitudinal
(T1–T2)
LGBTs The United
States
39.02 38.1 African Americans −0.53 0.10 −0.72 −0.34
32.99 61.1 Tea Party members 0.32 0.04 0.23 0.41
32.55 58.0 Religious people 0.23 0.05 0.12 0.33
#08 (data used by
Reininger etal., 2020;
Schaefer etal., 2021;
Simon, Reininger, etal.,
2019)
Cross-sectional
(T1)
LGBTs Germany 40.73 47.5 AfD 0.41 0.03 0.34 0.47
39.62 44.3 Catholics 0.35 0.05 0.26 0.45
41.16 21.0 Feminists 0.17 0.13 −0.08 0.41
40.77 39.3 Muslims 0.42 0.05 0.32 0.52
39.87 41.0 Protestants 0.34 0.07 0.21 0.48
39.87 42.4 Refugees 0.39 0.12 0.15 0.62
Table 2. (continued)
(continued)
8
Study ID Design
Sample characteristics
Outgroup
Standardized regression coefficient
Sample
Country of
residence Mean age
Percentage
of women Est. SEs L95%CI U95%CI
Cross-sectional
(T2)
LGBTs Germany 40.56 47.4 AfD 0.45 0.03 0.39 0.51
39.98 48.3 Catholics 0.20 0.04 0.11 0.29
40.28 47.8 Feminists 0.74 0.03 0.68 0.81
40.33 47.1 Muslims 0.36 0.04 0.29 0.44
39.26 47.2 Protestants 0.12 0.05 0.03 0.21
40.20 49.0 Refugees 0.19 0.04 0.12 0.27
Longitudinal
(T1–T2)
LGBTs Germany 40.72 47.7 AfD 0.24 0.03 0.17 0.31
39.41 43.7 Catholics −0.11 0.05 −0.21 −0.02
41.14 21.8 Feminists −0.09 0.18 −0.44 0.26
40.70 38.4 Muslims 0.08 0.06 −0.03 0.19
39.82 40.4 Protestants −0.04 0.06 −0.16 0.09
39.97 42.5 Refugees −2.15 0.23 −2.60 −1.69
#09 (unpublished data) Cross-sectional
(T1)
Catholics Poland 28.86 65.6 Atheists 0.51 0.05 0.42 0.60
27.47 57.4 Feminists 0.43 0.05 0.33 0.54
29.29 68.5 Homosexuals 0.46 0.05 0.37 0.56
28.45 62.5 Muslims 0.57 0.05 0.48 0.67
28.13 61.5 Russian Orthodox people 0.57 0.08 0.40 0.73
#10 (data used by Simon,
Eschert, etal., 2019)
Experimental Students Germany 21.10 70.6 Neues Zentruma0.39 0.11 0.16 0.61
#11 (unpublished data) Experimental Students Germany 27.98 50.0 Neues Zentruma0.59 0.10 0.39 0.79
Note. In addition to the features of the requested and preprocessed data (sub)sets, the table shows the estimates (Est.), the standard errors (SEs), and the lower (L95%CI) and upper (U95%CI) bounds
of the 95% confidence intervals for the standardized regression coefficients that describe the relations between tolerance (dependent variable) and respect (predictor). LGBTs = lesbian, gay, bisexual,
and transgender individuals.
aNeues Zentrum (English: New Center) is a fictitious German student group used in the cover story for the experiments. It was stated that the group is close to the Alternative für Deutschland (AfD,
English: Alternative for Germany), which is an existing right-wing party in Germany. For more details about the student group Neues Zentrum, see Simon, Eschert, etal. (2019).
Table 2. (continued)
Zitzmann et al. 9
atheists (disapproved of by Polish Catholics, for example),
it was hard to assign one of the predefined codes.
Disagreement was resolved via discussion, which finally
led to the assignment of missing values.
Membership in a numerical minority or majority. We coded
each sample and each outgroup as a numerical minority or a
numerical majority.
Social status. Each sample and each outgroup was coded as a
low or high social status group.
Type of group. Moreover, both the samples and the out-
groups were coded according to whether they were a
political group, a religious group, a life-style group, or an
ethnic group.
Design. Each study was coded as cross-sectional, longitudi-
nal, or experimental.
Analysis model. We coded each analysis model according to
whether it included one or more covariates in addition to the
measures of respect and tolerance, and thus controlled for
third variables.
Respect measure. Each measure of respect was coded as a
single item, a scale score, or a latent variable. We assigned
a missing value when respect was manipulated in an
experiment rather than measured and the respect variable
was a dummy. Notice, however, that in some experiments,
after manipulating respect, it was also measured to check
for whether its manipulation had been successful. In this
case, we assigned a code to this (additional) respect vari-
able depending on the respect measure that was used.
Country of residence. We coded each sample as stemming
from Germany, Poland, the United States, or Brazil.
Sample. The larger research project included samples of Tea
Party supporters; Catholics; Protestants; Muslims; Alevis;
lesbian, gay, bisexual, and transgender (LGBT) individuals;
or university students, which we coded accordingly.
Mean age. We computed the mean age for each sample by
aggregating across the participants’ age.
Percentage of women. For each sample, we computed the per-
centage of women in the sample as a proxy for the gender
distribution.
Number of participants. This was the sample size available
for analysis (for an explanation of why, in some samples, we
computed more than one mean age, percentage of women, or
number of participants depending on the specific outgroup,
see the “Analytic Approach” section).
These variables were used to identify possible modera-
tors. Our analytic approach proceeded in multiple steps, each
of which will be described in detail in the next section.
Analytic Approach
The analytic approach we employed consisted of two basic
steps. First, for each data set, we performed correlation anal-
yses. The resulting correlation matrices were then used as the
input of regression analyses to assess respect–tolerance links
with standardized regression coefficients. In the second step,
the respect–tolerance links were subjected to a meta-analysis
(to assess the overall link) or a meta-regression analysis. In
the meta-regression analysis, we used a moderator to explain
the differences in the respect–tolerance link across data sets.
Step 1: Assessing respect–tolerance links. Many data sets con-
tained additional, possibly confounding third variables,
including the strength of participants’ approval of the out-
group as well as the sociological variables—age, gender,
education, and income. Along with the respect and tolerance
variables, these third variables were subjected to correlation
analyses, which were computed in R (R Development Core
Team, 2016). In these analyses, we computed bivariate cor-
relations between the variables, which built the basis for fur-
ther analyses. Among these, the correlations with respect
differed in the extent to which the measurement error in the
respect variable was accounted for. To fully account for this
error, we adopted a latent-variable approach and used the R
package lavaan (Rosseel, 2012) to model respect by means
of an unconstrained measurement model with freely esti-
mated loadings and error variances. Alternatives to the use of
a latent-variable approach that are often less reliable are the
scale score and the single item, which we also used. Note that
the question of whether the choice of, for example, a latent
variable instead of a scale score would affect the respect–
tolerance link was addressed by performing an analysis of
moderating effects in Step 2 of our analytic approach (see
more below).
To deal with missing values in the variables, we applied
full information maximum likelihood (FIML), which allowed
us to use all the information available in the data. This
approach handles missing values in variables on the basis of
the assumption that the missing values depend on other vari-
ables in the data set (missing-at-random assumption; Rubin,
1987). Note also that FIML is recommended by many schol-
ars (e.g., Allison, 2003) and is often applied in psychological
research.
Because the correlation analyses resulted in more than
one correlation matrix per data set, we adopted a shifting-
unit-of-analysis approach (Cooper, 1998). In the first step of
this approach, we aggregated correlation matrices that came
from the same sample and did not differ with respect to the
moderator of interest (see Lipsey & Wilson, 2001). To this
end, we used the weighted average (weighted by the number
10 Personality and Social Psychology Bulletin 00(0)
of participants; Hunter & Schmidt, 2004). To illustrate, sup-
pose a data set in which multiple correlation analyses that
differed with respect to the outgroup were conducted. To pre-
pare the input for the estimation of the DRM, we aggregated
the correlation matrices across these different outgroups.
However, when our aim was to investigate whether the out-
group would act as a moderator in the DRM model, we did
not aggregate these correlation matrices but subjected them
to further analyses separately. This approach minimizes the
dependence that is due to the fact that multiple correlation
matrices stem from the same sample while preserving all the
information relevant for conducting the meta-regression
analyses (see O’Mara et al., 2006).
After we aggregated the correlation matrices, we esti-
mated the DRM for each aggregated correlation matrix by
fitting multiple regression models with the OpenMx soft-
ware (Boker et al., 2017). In the basic cross-sectional model,
tolerance was regressed only on respect:
TOLRESP
ii
i
=+
rβ1, (1)
where TOLi is the ith person’s tolerance toward an outgroup,
RESPi is the person’s respect for this group, and ri are residu-
als. β1 describes the relation between respect and tolerance,
and was thus the parameter that was of primary interest in
this model. In the longitudinal case, tolerance at time point t
was predicted by respect at an earlier time point t – 1, while
controlling for tolerance at that earlier time point (lagged cri-
terion). The model reads:
TOLRESPTOL
t,it,i t,
ii
=++rββ
12
−−11
. (2)
In accordance with the procedure used in the published
studies, the models could include one or more additional
covariates; for example, the participants’ age, gender, educa-
tion, income, or even approval of the outgroup, which
allowed us to preclude a potential confounding by these vari-
ables. The question of whether β1 would differ as a function
of covariates was addressed in Step 2.
Where the model was fit to an aggregated correlation
matrix, we used the average sample size N (aggregated over
different numbers of participants, which could vary from
outgroup to outgroup due to the DRM’s requirement that par-
ticipants must not be members of the group that is being met
with disapproval themselves). However, fitting the model to
an aggregated correlation matrix was not problem-free.
Therefore, before we estimated the model, we first checked
whether the correlation matrix was non-positive definite.
When this was the case, we applied Yuan and Chan’s (2008)
ridge technique to the matrix and passed the “cured” instead
of the degenerate matrix to OpenMx. The ridge technique
adds a small value a to the main diagonal of the singular
matrix R: R1 = R + aI, where I is the identity matrix (for
another application of this technique in a meta-analysis, see
Möller et al., 2020). From each fitted DRM, we extracted the
standardized regression coefficient (i.e., the relation between
respect and tolerance) and its standard error.
Step 2: Explaining differences in the respect–tolerance link. To
pool the respect–tolerance links, we fit a random-effects
model using metaSEM (Cheung, 2015b). Applying the com-
mon notation for hierarchical models (e.g., Raudenbush &
Bryk, 2002), the random-effects model reads at the first level:
Level1
:,te
iii
= θ+ (3)
where ti is the point estimate as derived from the ith correla-
tion matrix, θi is the respective true parameter, and ei is the
deviation of the point estimate from the true parameter. At
the second level, the following relation holds:
Level2
:,
θθ
ii
u= + (4)
where θ is the average population parameter, and ui is the
deviation of θi from θ. Unlike the fixed-effects model, this
model takes into account the variability across data sets (e.g.,
Overton, 1998). Notice that the meta-analytic approach of
pooling regression coefficients (sometimes also referred to
as the parameter-based approach because one or more model
parameters are meta-analyzed; see Becker & Wu, 2007;
Gasparrini et al., 2012; see also Cheung, 2015a) was chosen
here instead of pooling correlations so that comparisons
could be made with past tolerance studies, which used the
standardized regression coefficient to describe the respect–
tolerance link. Moreover, it allowed us to investigate the
influence of categorical and, most importantly, continuous
moderators (e.g., number of participants) on this link using
meta-regression analysis.
To this end, we estimated mixed-effects models for each
of the moderators. When a discrete moderator with m catego-
ries was investigated, we first created m – 1 dummy vari-
ables, CC Cm12 1
,,,−, which were then added as predictors
to the second level of the model:
Level2 11 22 11
:,
,, ,
θθγγ γ
iiimmi i
CC
Cu
=+ ++
++
−−
(5)
where γk
km
(,,, )
=−
12 1 is the difference between the cat-
egory that is coded by Ck and the reference category. For
continuous moderators, we first standardized the moderator
and then added it as a predictor:
Level2
:,
θθγ
ii
i
Mu
=+ + (6)
where γ is the slope of the moderator Mi. In addition, meta-
regression analysis also allowed us to add additional vari-
ables as covariates to the model to control for unwanted
confounding by these variables. Recall that some of the mod-
erators and covariates had missing values (see the “Coding”
section for an explanation for why the outgroup’s social sta-
tus had missing values, for example). Unfortunately,
Zitzmann et al. 11
estimating the mixed-effects models allowed us to use only
complete cases in the explanatory variables, which is why
we used listwise deletion in this case.
Results
To evaluate our main research question, which asked whether
respect would predict tolerance on average, we applied the
random-effects model as described above. The pooled stan-
dardized regression coefficient was 0.25 (p < .001), and its
95% confidence interval ranged from 0.16 to 0.34, thereby
clearly supporting the outgroup respect–tolerance hypothesis
that people’s respect influences their tolerance (see Table 3
for further details about the averaged results).
We tested the effects of the moderators by computing
meta-regression analyses. We proceeded by first analyzing
the moderator of interest separately without any further
moderators. However, for some moderators, because this
analysis was expected to be confounded by other variables,
we performed another analysis in which we controlled for
these potential confounding variables (e.g., design). The
resulting adjusted effects of the moderators were free from
any influence from the confounders. Notice also that
although we included only the variables that were consid-
ered most critical, one might argue that other variables could
also act as confounders (e.g., number of participants).
However, we did not control for them for three reasons. First
and foremost, whereas it is easy to justify why we controlled
for design in almost every analysis, it is less obvious to
explain how other variables could confound the analyses.
For example, it would be unlikely for the number of partici-
pants to affect the effects of the moderators. Moreover, the
variables analysis model and respect measure could not
affect the effect sizes of the moderators because they were
virtually uncorrelated (i.e., orthogonal) with all other vari-
ables. Third, there were missing values in the outgroup’s
social status and type of group variables. Because we used
only complete cases in the analyses, including these vari-
ables would have led to the (listwise) deletion of cases, and
we wanted to avoid doing this so that we could preserve as
much of the information as possible (see Zhou et al., 2019
for a similar argument for why variables with missing val-
ues were not entered as covariates in their analyses).
To test the effect that each moderator of interest had, we
compared two nested models (one unconstrained model
with the assumed effect for the moderator of interest vs.
one constrained model without it) using the likelihood
ratio test (LRT), which is also known as the chi-square dif-
ference test. To specify the constrained model, we fixed
the moderator’s slopes (γγ γ
12 1
,,,m− for categorical mod-
erators, γ for continuous moderators; see “Analytic
Approach” section) to zero. For categorical moderators
with more than two categories, we conducted additional
pairwise comparisons in addition to this omnibus test of
moderating effects to identify significant differences
between the categories (see Tables 4 and 5). For all statisti-
cal tests, α was set to the nominal 5% level.
In reporting the results, we begin with the results for the
research questions. In the next step, we present the results of
the exploratory analyses.
Membership in a Numerical Minority or Majority
When the two models (unconstrained vs. constrained) were
compared in the first meta-regression analysis (without any
additional moderators), the LRT with one degree of freedom
yielded a value of 7.01, which was significant (p < .01) and
thus clearly supported the speculation that the sample’s
membership in a numerical minority or majority would affect
the respect–tolerance link. This effect was considered large
because its size of f 2 = 1.20 exceeded the critical value of
0.35 for large effects. We used Cohen’s f 2 here because Selya
et al. (2012; see also Lorah, 2018) suggested that it would
quantify the effects in hierarchical models, and common
guidelines for its interpretation exist (see Cohen, 1988).
Moreover, as can be seen in Table 4, the difference in the
respect–tolerance link between the two categories “numeri-
cal minority” and “numerical majority” was −0.22 and sig-
nificant (p < .01), suggesting that participants belonging to a
numerical majority exhibited significantly stronger relations
between respect and tolerance than those belonging to a
numerical minority did. However, it was reasonable to
assume that this analysis was to some extent confounded by
other moderators, most likely the research design.
Moreover, we assumed that the sample’s social status and
its type could also act as confounders. Therefore, in the sec-
ond analysis, we controlled for these variables to check for
whether a moderating effect of the sample’s membership in a
numerical minority or majority would still occur when con-
founding by these variables was precluded. However, the
finding proved to be robust as indicated by a value of 15.46
of the LRT statistic with one degree of freedom, which was
significant (p < .001). Also, the comparison confirmed that
numerical majorities showed a significantly stronger link
than numerical minorities did (p < .001; see Table 5).
However, we did not find any significant effect of mem-
bership in a numerical minority or majority for the outgroup.
This finding held in the analysis that did not include the con-
founding variables, design, social status, or type of outgroup,
Δχ2(1) = 0.02, and when these confounders were included as
covariates, Δχ2(1) = 0.46.
Social Status
To test for a moderating effect of the sample’s social status
in the first analysis (without any other moderators), we
used an LRT with one degree of freedom. The LRT statistic
was 0.88 and failed to reach significance. In the second
analysis, we added three moderators as covariates to con-
trol for unwanted confounding: design, the sample’s
12
Table 3. Average Results for the Relation Between Respect and Tolerance.
Moderator Categories
Overall No. of
observations used
No. of data
sets used
Average No. of
observations used
Standardized regression coefficient
Est. SEs L95%CI U95%CI
No moderator 3,193 11 290 0.25*** 0.05 0.16 0.34
Substantive moderators
Numerical minority or
majority (sample)
Minority 2,455 7 351 0.18*** 0.04 0.10 0.26
Majority 738 4 184 0.40*** 0.06 0.28 0.52
Numerical minority or
majority (outgroup)
Minority 3,366 11 306 0.25*** 0.05 0.14 0.36
Majority 1,735 6 289 0.26*** 0.07 0.12 0.41
Social status (sample) Low 1,151 4 288 0.20** 0.07 0.06 0.34
High 2,042 7 292 0.29*** 0.06 0.18 0.39
Social status (outgroup) Low 3,050 11 277 0.21** 0.07 0.08 0.34
High 2,133 7 305 0.22** 0.08 0.06 0.38
Type of group (sample) Political 562 1 562 0.25†0.14 −0.02 0.53
Religious 1,921 6 320 0.23*** 0.06 0.11 0.34
Life-style 710 4 178 0.30*** 0.08 0.15 0.46
Type of group (outgroup) Political 3,098 10 310 0.36*** 0.05 0.26 0.46
Religious 2,714 9 302 0.19*** 0.05 0.08 0.29
Life-style 2,708 7 387 0.26*** 0.06 0.15 0.38
Ethnic 185 2 93 −0.34** 0.13 −0.58 −0.09
Methodological moderators
Design Cross-sectional 3,119 9 347 0.44*** 0.03 0.38 0.50
Longitudinal 2,769 8 346 0.18*** 0.03 0.12 0.24
Experimental 111 2 56 0.50*** 0.09 0.31 0.68
Analysis model No covariates 4,113 11 374 0.27*** 0.05 0.18 0.37
Covariates included 3,141 11 286 0.25*** 0.05 0.16 0.35
Respect measure Single item 3,203 11 291 0.23*** 0.06 0.12 0.34
Scale score 3,170 11 288 0.28*** 0.06 0.17 0.39
Latent variable 3,206 11 291 0.37*** 0.06 0.26 0.48
Additional moderators
Country of residence Germany 1,506 6 251 0.25*** 0.06 0.14 0.36
Poland 627 2 313 0.34*** 0.09 0.17 0.52
The United States 827 2 413 0.25** 0.09 0.07 0.42
Brazil 233 1 233 0.09 0.13 −0.16 0.33
Sample Tea Party supporters 562 1 562 0.25** 0.08 0.10 0.41
Catholics 627 2 313 0.35*** 0.06 0.23 0.47
Protestants 743 2 371 0.11†0.06 0.00 0.22
Muslims 410 1 410 0.28** 0.09 0.11 0.45
Alevis 141 1 141 0.18†0.11 −0.03 0.39
LGBTs 599 2 300 0.16** 0.06 0.04 0.29
University students 111 2 56 0.50*** 0.09 0.32 0.67
Note. In addition to the estimates (Est.), the table shows the standard errors (SEs) and the lower (L95%CI) and upper (U95%CI) bounds of the 95% confidence intervals for the average standardized
regression coefficients (aggregated across data sets) of the moderator variables. LGBTs = lesbian, gay, bisexual, and transgender individuals.
†p < 1. **p < .01. ***p < .001 (all ps two-sided).
Zitzmann et al. 13
Table 4. Pairwise Comparisons (Uncontrolled).
Moderator Comparisons
Difference
Est. SEs L95%CI U95%CI
Substantive moderators
Numerical minority or majority (sample) Minority – Majority −0.22** 0.07 −0.08 −0.36
Numerical minority or majority (outgroup) Minority – Majority −0.01 0.09 0.16 −0.19
Social status (sample) Low − High −0.09 0.09 0.09 −0.27
Social status (outgroup) Low − High −0.01 0.11 0.20 −0.22
Type of group (sample) Political – Religious 0.03 0.15 0.33 −0.27
– Life-style −0.05 0.16 0.27 −0.36
Religious – Life-style −0.08 0.10 0.12 −0.27
Type of group (outgroup) Political – Religious 0.17* 0.07 0.32 0.03
– Life-style 0.10 0.08 0.25 −0.05
– Ethnic 0.70*** 0.14 0.96 0.43
Religious – Life-style −0.07 0.08 0.08 −0.23
– Ethnic 0.52*** 0.14 0.79 0.26
Life-style – Ethnic 0.60*** 0.14 0.87 0.33
Methodological moderators
Design Cross-sectional – Longitudinal 0.26*** 0.04 0.35 0.18
– Experimental −0.05 0.10 0.14 −0.24
Longitudinal – Experimental −0.32** 0.10 −0.13 −0.51
Analysis model No covariates – Covariates included 0.02 0.07 0.15 −0.11
Respect measure Single item – Scale score −0.05 0.08 0.11 −0.21
– Latent variable −0.13†0.08 0.02 −0.29
Scale score – Latent variable −0.08 0.08 0.07 −0.24
Additional moderators
Country of residence Germany − Poland −0.10 0.11 0.11 −0.30
– The United States 0.00 0.11 0.21 −0.21
– Brazil 0.16 0.14 0.43 −0.11
Poland – The United States 0.10 0.13 0.34 −0.15
– Brazil 0.26†0.16 0.56 −0.05
The United States – Brazil 0.16 0.15 0.46 −0.14
Sample Tea Party
supporters
– Catholics −0.10 0.10 0.10 −0.29
– Protestants 0.15 0.10 0.33 −0.04
– Muslims −0.02 0.12 0.20 −0.25
– Alevis 0.08 0.13 0.33 −0.18
– LGBTs 0.09 0.10 0.28 −0.10
–University students −0.24* 0.12 −0.01 −0.47
Catholics – Protestants 0.24** 0.08 0.40 0.08
–Muslims 0.07 0.11 0.28 −0.14
– Alevis 0.17 0.12 0.41 −0.07
− LGBTs 0.18* 0.09 0.35 0.01
– University students −0.15 0.11 0.07 −0.36
Protestants – Muslims −0.17 0.10 0.03 −0.37
– Alevis −0.07 0.12 0.17 −0.31
− LGBTs −0.06 0.08 0.11 −0.22
– University students −0.39*** 0.11 −0.18 −0.60
Muslims – Alevis 0.10 0.14 0.37 −0.17
– LGBTs 0.11 0.11 0.32 −0.10
– University students −0.22†0.12 0.03 −0.46
Alevis – LGBTs 0.01 0.12 0.25 −0.23
– University students −0.32* 0.14 −0.05 −0.59
LGBTs – University students −0.33** 0.11 −0.12 −0.55
Note. In addition to the estimates (Est.), the table shows the standard errors (SEs) and the lower (L95%CI) and upper (U95%CI) bounds of the 95%
confidence intervals for the mean differences (aggregated across data sets) between the standardized regression coefficients of the moderators’
categories. LGBTs = lesbian, gay, bisexual, and transgender individuals.
†p < 1. *p < .05. **p < .01. ***p < .001 (all ps two-sided).
14
Table 5. Pairwise Comparisons Controlled for Possibly Confounding Variables.
Moderator Comparisons
Difference
ConfoundersEst. SEs L95%CI U95%CI
Substantive moderators
Numerical minority or
majority (sample)
Minority – Majority −0.18*** −0.04 −0.11 −0.26 Design, social status (sample), type of group
(sample)
Numerical minority or
majority (outgroup)
Minority – Majority −0.08 −0.12 0.15 −0.31 Design, social status (outgroup), cultural
background (outgroup)
Social status (sample) Low – High 0.19*** −0.04 0.28 0.11 Design, numerical minority or majority
(sample), type of group (sample)
Social status
(outgroup)
Low – High 0.07 −0.11 0.28 −0.14 Design, numerical minority or majority
(outgroup), type of group (outgroup)
Type of group
(sample)
Political – Religious 0.16*** −0.04 0.24 0.09 Design, numerical minority or majority
(sample), social status (sample)
– Life-style 0.28*** −0.06 0.40 0.16 —”—
Religious – Life-style 0.12* −0.05 0.21 0.02 —”—
Type of group
(outgroup)
Political – Religious 0.08 −0.07 0.22 −0.05 Design, numerical minority or majority
(outgroup), social status (outgroup)
– Life-style 0.06 −0.08 0.22 −0.11 —”—
– Ethnic 0.67*** −0.13 0.93 0.42 —”—
Religious – Life-style −0.03 −0.07 0.12 −0.17 —”—
– Ethnic 0.59*** −0.13 0.84 0.34 —”—
Life-style – Ethnic 0.62*** −0.13 0.87 0.36 —”—
Methodological moderators
Design Cross-sectional – Longitudinal 0.25*** −0.03 0.30 0.20 Numerical minority or majority (sample),
social status (sample), type of group (sample)
– Experimental −0.10 –0.09 0.08 −0.29 —”—
Longitudinal – Experimental −0.35*** –0.10 −0.16 −0.54 —”—
Additional moderators
Country of residence Germany − Poland −0.09†−0.05 0.01 −0.18 Design
– The United
States
−0.06 −0.04 0.03 −0.15 “
– Brazil 0.12†−0.06 0.23 0.00 “
Poland – The United
States
0.03 −0.06 0.13 −0.08 “
– Brazil 0.20** −0.07 0.34 0.07 “
The United
States
– Brazil 0.18** −0.06 0.30 0.05 “
Note. In addition to the estimates (Est.), the table shows the standard errors (SEs) and the lower (L95%CI) and upper (U95%CI) bounds of the 95% confidence intervals for the unconfounded mean
differences (adjusted by confounders) between the standardized regression coefficients of the moderators’ categories.
†p < 1. *p < .05. **p < .01. ***p < .001 (all ps two-sided).
—”— and “ indicate that the same confounders were controlled for as in the previous comparison.
Zitzmann et al. 15
membership in a numerical minority or majority, and its
type. The LRT with one degree of freedom yielded a much
larger value of 16.42, which was significant (p < .001).
Contrary to our speculation, the difference in the respect–
tolerance link between the two categories “low” and “high”
social status was significant and positive (0.19, p < .001),
which indicated that participants belonging to a low-status
group showed a significantly stronger link than those
belonging to a high-status group.
We did not find any significant effect of the outgroup’s
social status in the first analysis, Δχ2(1) = 0.01, nor did we
find one in the second analysis, Δχ2(1) = 0.38, when we
included design, membership in a numerical minority or
majority, and type of outgroup as covariates.
Type of Group
Model comparison with two degrees of freedom yielded a
value of 0.61 in the first analysis (without any other mod-
erators), which was not significant. However, when possi-
ble confounders (design, membership in a numerical
minority or majority, and the social status of the sample)
were included in the second analysis, the LRT statistic with
two degrees of freedom was 16.35 and thus supported the
significant impact of the type of sample on the respect–
tolerance link (p < .001). This effect was large in terms of
the suggested cutoffs (i.e., f 2 > 0.35). We found significant
differences between the category “political” group and the
other two types of groups. The political samples showed
stronger relations between respect and tolerance than the
religious and the life-style samples (ps < .001), and the
religious samples showed stronger relations than the life-
style samples (p < .05) (see Table 5).
A significant and large effect of the type of outgroup was
observed in the first analysis, Δχ2(2) = 3.09, p < .001, f 2 =
1.16, with the political outgroups exhibiting a stronger
respect–tolerance link than the religious (ps < .05) or ethnic
(p < .001) outgroups (see Table 4); the religious outgroups
exhibiting a stronger link than the ethnic outgroups (p <
.001); and the life-style outgroups exhibiting a stronger link
than the ethnic outgroups (p < .001). However, although the
moderating effect remained significant and large in the sec-
ond analysis in which we controlled for design, the out-
group’s membership in a numerical minority or majority, and
the outgroup’s social status, only the comparisons with the
ethnic outgroups were significant (all ps < .001).
Design
In the first analysis (without any other moderators), the result
of the LRT with two degrees of freedom for the moderating
effect of design was 21.42, which indicated a significant
effect (p < .001), thus supporting the expectation that the
design would affect the respect–tolerance link. With a size of
f 2 = 3.24, this effect was considered large, and pairwise
comparisons indicated that the respect–tolerance link was
significantly stronger for cross-sectional than for longitudi-
nal designs (p < .001) and significantly stronger for experi-
mental than for longitudinal designs (p < .01).
To test the robustness of these results, we controlled for
membership in a numerical minority or majority, social sta-
tus, and the type of sample in the second analysis. The LRT
with two degrees of freedom yielded a value of 35.81, indi-
cating significance (p < .001). Moreover, of the pairwise
comparisons, the difference between the cross-sectional and
longitudinal designs (Δ = 0.25) and the difference between
the experimental and longitudinal designs (Δ = 0.35) were
both significant (both ps < .001), with the cross-sectional
designs showing stronger relations between respect and tol-
erance than the longitudinal designs, and the experimental
designs showing stronger relations than the longitudinal
designs.
Analysis Model
To test for whether the analysis model affected the respect–
tolerance link, we conducted an LRT with one degree of
freedom. Recall that we distinguished between models with
one or more covariates (e.g., where approval of the out-
group was controlled for) and models without covariates in
which tolerance was predicted only by respect. The test
yielded a value of 0.09, which was not significant. Notice
also that the analysis model was among the variables that
could not be confounders for statistical reasons (because
they were virtually orthogonal to all other variables). This
also means that if the analysis model served as the modera-
tor in the moderator analysis, adding other variables as
covariates would not alter the size of the moderating effect.
Therefore, we did not conduct another analysis in which we
controlled for these variables.
Respect Measure
We conducted another analysis in which we tested the role of
the respect measure as a possible moderator but did not find
any significant effect, Δχ2(2) = 2.71. However, because we
had specific directed hypotheses on the differences between
the measures with regard to the size of the respect–tolerance
link, we considered the pairwise comparisons to be the criti-
cal tests. Table 4 shows that despite the absence of evidence
of an overall effect of the respect measure, we found support
for our hypothesis that the use of the latent variable would
yield larger sizes of this link than the use of a single item
would. Note that because the hypothesis was directed, we
considered marginal significance (i.e., p < .1, two-sided) to
be indicative of a reliable difference in this case. However,
although the latent variables tended to yield a stronger link
than the scale scores, this difference was not reliable as was
the difference between the scale scores and the single items
(ps > .1, two-sided). Note also that for the same reason as for
16 Personality and Social Psychology Bulletin 00(0)
when the analysis model was the moderator, we did not con-
duct a second analysis to control for possible confounders.
In an additional analysis, we investigated whether the five
respect items in the single-item category were differentially
related to tolerance and which of the five respect items in the
single-item category explained the most variance. The LRT
statistic with four degrees of freedom was 16.11 and signifi-
cant (p < .01). The size was f 2 = 1.10 and was thus rather
large. However, the pairwise comparisons suggested that sig-
nificant differences (all ps < .05) only existed between items
that were used in the correlational studies and items that
were exclusively used in the experiments (Studies 10 and
11), making it difficult to attribute these differences solely to
the items.
Next, we report the results for the additional moderators,
for which we did not have any hypotheses and which were
thus purely exploratory in nature.
Country of Residence
The LRT for the moderating influence of country of resi-
dence had three degrees of freedom and yielded a value of
2.35 in the first analysis (without any other moderators),
which was not significant. When we controlled for design in
the second analysis, the LRT statistic with three degrees of
freedom was 8.54, which indicated significance (p < .05).
Applying the common cutoffs, the effect could be considered
large (f 2 = 1.15).
As revealed by pairwise comparisons, participants from
Poland and the United States provided stronger respect–tol-
erance links than those from Brazil (ps < .01). Moreover, we
found a marginally significant difference between Germany
and Poland (p < .1), where the link was stronger in Poland
than in Germany. We also found a difference between
Germany and Brazil, where the link was stronger in Germany
than in Brazil.
Sample
We conducted a model comparison with six degrees of
freedom. The associated LRT statistic was 11.63, which
failed to reach significance, indicating that at most, there
was a tendency for the samples to vary with regard to the
respect–tolerance link. Additional pairwise comparisons
showed that university students provided significantly
stronger respect–tolerance relations than Tea Party sup-
porters (p < .05), Protestants (p < .001), Muslims (p <
.1), Alevis (p < .05), and LGBTs (p < .01). Moreover, we
found that the link was significantly stronger for Catholics
than for Protestants (ps < .01) and LGBTs (p < .05).
However, it should be noted that because the moderating
effect of the sample was mainly driven by the differences
between the students and all other samples and this group
participated exclusively in the experiments, the result
should be interpreted with caution.
Mean Age
A model comparison with one degree of freedom indicated
no significant effect of the mean age of the sample on the
respect–tolerance link. This finding held in the first analysis
without any other moderators, Δχ2(1) = 1.46, and the second
analysis with design as the covariate, Δχ2(1) = 0.26.
Percentage of Women
The LRT statistic of the test of the moderating influence of
the gender distribution had one degree of freedom. The value
was 5.23 and significant in the first analysis (without any
other moderators; p < .05). The size of the effect was f 2 =
0.05 and thus above the cutoff of 0.02 for small effects.
However, when we controlled for design in the second analy-
sis, a significant effect was no longer observed.
Number of Participants
We found a significant moderating influence of the number
of participants in the first analysis, without any other mod-
erators, Δχ2(1) = 5.06, p < .05. However, because of its size
of f 2 = 0.05, this effect was small in terms of the suggested
cutoffs. When we controlled for design in the second analy-
sis, a marginally significant effect was still observed, with
the larger samples showing a stronger respect–tolerance link.
Summary
The most important findings can be summarized as follows.
We found clear support for our main hypothesis that the rela-
tion between respect and tolerance would exist on average
and be substantial. Moreover, we also found evidence for
some moderating influences. First and foremost, the sam-
ple’s characteristics played a moderating role in the relation
between respect and tolerance. More specifically, respect
was more predictive of tolerance among members of numeri-
cal majorities than minorities, and—unexpectedly—it was
more predictive among members of low-status groups than
high-status groups. Moreover, we found that members of
religious groups exhibited a stronger respect–tolerance link
than members of life-style groups did and that members of
political groups exhibited an even stronger link than mem-
bers of religious or life-style groups did.
Second, we found some evidence that the methodology
played a moderating role. Cross-sectional studies and experi-
ments showed stronger relations between respect and toler-
ance than longitudinal studies, which was not very surprising.
For example, in experimental designs, respect is actively
manipulated and is thus under the control of the experimenter
as is the extent to which confounding variables are allowed
to interfere. But in longitudinal designs, the effect depends
on naturally occurring processes (i.e., processes that prompt
changes in the level of respect). Moreover, we found that the
Zitzmann et al. 17
respect–tolerance link was rather robust against variations in
the model employed for data analysis and in the type of mea-
sure used to assess respect. We did not find evidence that
adding one or more covariates to the DRM affected the
respect–tolerance link, suggesting that alternative explana-
tions in terms of these variables did not apply and that the
DRM in its simplest form (i.e., including only respect and
tolerance variables) need not be refined substantially. Also,
the respect–tolerance link was stronger for the latent variable
than for the single item, thereby pointing to the usefulness of
the latent-variable approach when respect is assessed with
multiple items. However, it should be noted that the results
did not suggest that the latent-variable approach was supe-
rior to the scale score, which encourages the view that the
scale score could be an attractive alternative to the latent-
variable approach in social-psychological research, particu-
larly in situations in which the adoption of a latent-variable
approach is troublesome (e.g., in small sample contexts in
which convergence issues are likely to occur).
Discussion
Motivated by the topic’s high relevance to society, we evalu-
ated whether respect for others as equals fosters tolerance
toward others who are met with disapproval. To this end, we
determined the pooled strength of this relation across a com-
prehensive set of empirical studies. Overall, respect was a
significant, substantial predictor of tolerance. Moreover, the
relation was influenced by a number of moderators in ways
that can be informative for future research. We discuss this
further below, where we also discuss the specific meaning of
the observed moderations as well as their relevance to the
model and the ways in which they may contribute to the
model’s further development.
Methodologically, we demonstrated the use of meta-ana-
lytic techniques, which we employed to synthesize a com-
prehensive collection of existing data sets. We did this in a
manner similar to systematic literature reviews. However,
our work differed from most of these reviews. Systematic
reviews often involve a large number of different conceptu-
alizations of the constructs, which can blur the relations
between these and other constructs. By using the data from a
large multistudy research project, we ensured that the con-
ceptualizations (e.g., where respect was defined as viewing
others as equals) were consistent across the different studies
that were included in our analysis. This provided us with a
more robust test of the outgroup respect–tolerance hypothe-
sis. We found that the relation between respect and tolerance
was, on average, positive and substantial, thereby supporting
our hypothesis. Thus, we view the use of data from a larger
project as an advantage of the present work because interpre-
tational ambiguities due to different conceptualizations of
the constructs did not emerge.
Finally, to guarantee transparency, we attempted to be as
precise as possible with respect to the exact procedures we
employed, and we were clear about which of the analyses
were conducted to test hypotheses and which of them were
exploratory. Besides these strengths of the present work, the
following limitations should also be mentioned.
Limitations
Most of the data used in the present study were borrowed
from correlational studies, which were still vulnerable to
unwanted confounding by variables that were not measured
in the respective studies. Experiments, which are designed to
control for third variables by means of randomization, have
to date been rare so far in research on the respect–tolerance
link. Therefore, some caution is in order when interpreting
our main finding because, strictly speaking, it does not pres-
ent conclusive evidence of the notion that respect is not just
a correlate of tolerance but is rather a cause.
Although our analytic strategy met the requirements of
best practice, some of the methodological decisions may be
controversial and should be discussed further. First, the
parameter-based approach used in the present work has been
criticized in the past (e.g., Cheung, 2015a). However, the
main criticism of it does not apply here because we did not
face the problem of incomplete information when fitting the
analysis models. Each data set included one measure of
respect in addition to tolerance, which allowed us to estimate
the correlation between these variables and thus also the
DRM in its simplest form (i.e., the model with tolerance as
the dependent variable and respect as the predictor).
Second, to address the dependencies across the multiple
effect sizes, we applied the shifting-units-of-analysis
approach (Cooper, 1998), which is certainly not without
weaknesses (e.g., several steps need to be performed in this
analysis instead of just one step). On the contrary, it can be
very useful in practice (see Lipsey & Wilson, 2001; see also
O’Mara et al., 2006, for an example application of this
approach). It should also be noted that there are alternative
approaches such as three-level meta-analyses (Cheung,
2014). Although each approach accounts for the dependen-
cies across the effect sizes, they differ with regard to their
assumptions (see Cheung, 2019, for a discussion of these
approaches). Future research could apply both approaches
(or a combination thereof) to check for whether the findings
are robust (i.e., they do not change) against the use of differ-
ent approaches.
Third, missing values were almost completely absent
from the moderation analyses except for some moderators,
where the ratio of missing values was still extremely rare. In
the meta-regression analyses involving these variables, we
used only complete cases (i.e., listwise deletion).
Alternatively, we could have applied FIML. Not only does
FIML handle missing values in dependent variables (e.g.,
missing effect sizes), but it also allows for the handling of
missing values in moderators. However, to cope with miss-
ing values in moderators, additional assumptions must be
18 Personality and Social Psychology Bulletin 00(0)
made. For example, missing values must be assumed to be
missing at random and—to make this assumption plausi-
ble—additional variables on which the missing values
depend must be included (e.g., further study features).
However, recall that the missing values were partly due to
the raters’ decision not to assign any predefined code (see the
“Coding” section). Thus, it is questionable whether the cor-
rect use of FIML would have been feasible in the present
study because it was hard to see which variables the missing
values depended on.
Moreover, in our case, in which the effect sizes were fully
observed and the missing values in the moderators were
extremely rare, we did not consider FIML to be noticeably
superior to listwise deletion. Note, however, that we
employed FIML successfully in the preprocessing of the
data, where the missing values in the variables were more of
an issue.
Fourth, we focused on the main effects of the moderators.
However, it would be interesting to also investigate interac-
tions between them. For example, future research could
study the interplay between the sample’s and the outgroup’s
characteristics. We believe that this interplay can be best
studied with the help of experiments, which allow research-
ers to vary these characteristics in a systematic way.
Directions for Theory Development and Future
Empirical Research
The mechanism that the DRM is based on is straightforward.
The assumption is that outgroups can be recognized and
respected as equals, which then facilitates tolerance toward
them, even from others who disapprove of their ways of life.
The distinct contribution of the present work is the compre-
hensive test of the robustness of the respect–tolerance link
and our endeavor to uncover possible moderators that can
inform and guide the further development of the DRM but
also future research on tolerance more generally. Of the
many novel empirical findings we reported, we wish to high-
light the ones that appear particularly promising to us with
regard to their potential to contribute to theory development
and corresponding empirical research.
First, we obtained empirical support for our specula-
tion that members of numerical majority groups show
stronger respect–tolerance links than members of numer-
ical minority groups do. We had intuited that members of
majority groups would see themselves as relatively more
prototypical representatives of society with a greater
need of a good reason to tolerate the outgroup. Respect
for outgroup members should be such a reason. Hence,
the more respect majorities or high-status groups actually
grant outgroup members, the stronger their reason for tol-
erance. This account may also hold for our observation
that members of religious groups evinced a stronger
respect–tolerance link than members of (presumably less
obliging) life-style groups.
Another—not necessarily competing but possibly comple-
mentary—account comes into view when we turn to the par-
ticularly strong link observed for members of political groups
compared with both religious and life-style groups. The polit-
ical sphere—at least in modern democratic societies that
emphasize mutual respect for equal fellow citizens—should
be a particularly conducive medium for the operation of the
respect–tolerance link. From a social-psychological perspec-
tive, the cognitive representation of society as the common
(higher level) social and political entity comprising members
of many different (lower level) groups may well be the deci-
sive factor here. Heterogeneity is a characteristic of modern-
day societies (Durkheim, 1997), and the cognitive
representation of a society as a heterogeneous superordinate
ingroup should imbue the pivotal concept of respect for oth-
ers as equals with a particular meaning. That is, it should shift
its meaning from a narrow understanding of respect for others
as “identical” equals as found in homogeneous groups to the
wider understanding of respect for others as “different”
equals—”different” owing to their membership in different
(lower level) groups, “equal” owing to their membership in
the same heterogeneous society (Simon, 2020). Whereas the
narrow understanding of respect for others as identical equals
would be burdened by expectations of conformity, which in
turn strain the respect–tolerance link, the wider understanding
of respect for others as different equals should, on the con-
trary, be quite compatible with self–other differences so that
respect can easily and strongly be linked up with tolerance.
The cognitive-representation account can also contribute to
the explanation of the stronger respect–tolerance link
observed among majority members that we already discussed
above. Relative to minority members, majority members tend
to construe their ingroup as a more heterogeneous group
(Simon, 1992). If this tendency is combined with the ten-
dency to project the image of one’s ingroup onto society as a
whole (Wenzel et al., 2007), the foundation would be laid for
a strong respect–tolerance link. Moreover, our observation
that members of low-status groups produced a stronger
respect–tolerance link than members of high-status groups
could be another case in point here. It stands to reason that
members of low-status groups are particularly motivated to
see society as a heterogeneous entity with heterogeneity sig-
naling openness and encouraging social mobility aspirations.
In any case, if confirmed in future research, the cognitive-
representation account could turn out to be a particularly
powerful explanation because it integrates the specification of
a moderator variable (homogeneous vs. heterogeneous cogni-
tive representation of society) with the further specification of
the mechanism driving the respect–tolerance link (respect for
others as different rather than identical equals).
At this point, we are far from claiming that we are already
able to provide a conclusive and integrated theoretical expla-
nation for our numerous empirical observations. Still, these
observations challenged us to engage in further theoretical
elaboration, and we see additional value in the fact that they
Zitzmann et al. 19
pointed us (and hopefully other researchers as well) toward
important open questions concerning the role of respect in
tolerance. Among these are questions about why ethnic out-
groups seem to be particularly hard cases with regard to the
emergence of a strong respect–tolerance link and what
underlies the differences observed along the national and/or
cultural (i.e., country-of-residence) dimensions. Of course, a
reliable answer, especially to the latter question, will require
future research to employ representative sampling that goes
beyond the scope of the research presented in this article.
Finally, despite the complexity of the results and some
open questions, there should be no doubt that the present
work also has a clear and positive message with important
practical implications. People are indeed capable of develop-
ing tolerance toward outgroups, without necessarily having
to give up their disapproval of these groups. Of course, much
depends on whether we are willing to respect each other as
equals and therefore depends on the appropriate social and
political arrangements that will help people give mutual
respect to each other as equals. But once people experience
such respect, a positive reciprocity mechanism will likely be
set in motion with the potential to help develop tolerant,
peaceful, and possibly also cooperative social relationships
across group boundaries (see Reininger et al., 2020; Schaefer
& Simon, 2019; Simon & Schaefer, 2018).
Conclusion
To conclude, the outgroup respect–tolerance hypothesis that
we scrutinized in the present work was derived from the
DRM, which offers one step toward a broader social-psycho-
logical theory of intergroup conflict (Simon, 2020). An
important challenge for future research is to further develop
the model and its connections to other aspects of intergroup
relations (e.g., politicization and polarization). We hope that
the work presented in this article will serve as a source of
inspiration for such research and further theory building as
well as an instructive illustration of how (re)analyzing and
synthesizing results from multiple data sets coming from a
larger research project can contribute to such an important
endeavor.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect
to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support
for the research, authorship, and/or publication of this article: This
work was supported by Grant SI 428/20-1 from the German
Research Foundation (DFG) awarded to Bernd Simon.
ORCID iD
Steffen Zitzmann https://orcid.org/0000-0002-7595-4736
Supplemental Material
Supplemental material for this article is available online.
Note
1. Multiple tolerance measures were not included in the present
study because in all but two of the data sets used here, tolerance
was assessed with only a single item.
References
Allison, P. D. (2003). Missing data techniques for structural equa-
tion modeling. Journal of Abnormal Psychology, 112, 545–
557. https://doi.org/10.1037/0021-843X.112.4.545
Becker, B. J., & Wu, M.-J. (2007). The synthesis of regression
slopes in meta-analysis. Statistical Science, 22, 414–429.
https://doi.org/10.1214/07-STS24
Boker, S. M., Neale, M. C., Maes, H. H., Wilde, M. J., Spiegel,
M., Brick, T. R., . . . Kirkpatrick, R. M. (2017). OpenMx 2.8.3
user guide [Computer software manual]. https://openmx.ssri.
psu.edu/
Cheung, M. W.-L. (2014). Modeling dependent effect sizes with
three-level meta-analyses: A structural equation modeling
approach. Psychological Methods, 19, 211–229. https://doi.
org/10.1037/a0032968
Cheung, M. W.-L. (2015a). Meta-analysis: A structural equation
modeling approach. Wiley.
Cheung, M. W.-L. (2015b). metaSEM: An R package for meta-
analysis using structural equation modeling. Frontiers in
Psychology, 5, 1–7. https://doi.org/10.3389/fpsyg.2014.01521
Cheung, M. W.-L. (2019). A guide to conducting a meta-analysis
with non-independent effect sizes. Neuropsychology Review,
29, 387–396. https://doi.org/10.1007/s11065-019-09415-6
Cohen, J. E. (1988). Statistical power analysis for the behavioral
sciences (2nd ed.). Wiley.
Cooper, H. M. (1998). Integrating research: A guide for literature
reviews (3rd ed.). SAGE.
Crane, T. (2017). The meaning of belief: Religion from an atheist’s
point of view. Harvard University Press.
Durkheim, E. (1997). The division of labor in society. The Free
Press.
Dworkin, R. (2013). Justice for hedgehogs. Harvard University
Press.
Eschert, S., & Simon, B. (2019). Respect and political disagree-
ment: Can intergroup respect reduce the biased evaluation
of outgroup arguments? PLOS ONE, 14, Article e0211556.
https://doi.org/10.1371/journal.pone.0211556
Forst, R. (2013). Toleration in conflict: Past and present. Cambridge
University Press.
Fuller, W. A. (1987). Measurement error models. Wiley.
Gasparrini, A., Armstrong, B., & Kenward, M. G. (2012).
Multivariate meta-analysis for non-linear and other multi-
parameter associations. Statistics in Medicine, 31, 414–429.
https://doi.org/10.1002/sim.5471
Gibson, J. L., Duch, R. M., & Tedin, K. L. (1992). Democratic val-
ues and the transformation of the Soviet Union. The Journal of
Politics, 54, 329–371. https://doi.org/10.2307/2132030
Honneth, A. (1995). The struggle for recognition: The moral gram-
mar of social conflicts. Polity Press.
20 Personality and Social Psychology Bulletin 00(0)
Hunter, J. E., & Schmidt, F. L. (2004). Methods of meta-analysis:
Correcting error and bias in research findings. SAGE.
Huo, Y. J., & Molina, L. E. (2006). Is pluralism a viable model of
diversity? The benefits and limits of subgroup respect. Group
Processes & Intergroup Relations, 9, 359–376. https://doi.
org/10.1177/1368430206064639
Krys, S., Simon, B., Reininger, K. M., Brause, A.-C., & Galeão
Silva, L.-G. (2020). The politicization of religious com-
munities: Effects on intergroup differentiation. Revista de
Psicología Social, 35, 116–147. https://doi.org/10.1080/0213
4748.2019.1682290
Lipsey, M. W., & Wilson, D. B. (2001). Practical meta-analysis.
SAGE.
Lorah, J. (2018). Effect size measures for multilevel models:
Definition, interpretation, and TIMSS example. Large-Scale
Assessments in Education, 6, 1–11. https://doi.org/10.1186/
s40536-018-0061-2
Möller, J., Zitzmann, S., Helm, F., Machts, N., & Wolff, F. (2020).
A meta-analysis of relations between achievement and aca-
demic self-concept. Review of Educational Research, 90, 376–
419. https://doi.org/10.3102/0034654320919354
O’Mara, A. J., Marsh, H. W., Craven, R. G., & Debus, R.
(2006). Do self-concept interventions make a difference?
A synergetic blend of construct validation and meta-anal-
ysis. Educational Psychologist, 41, 181–206. https://doi.
org/10.1207/s15326985ep4103_4
Overton, R. C. (1998). A comparison of fixed-effects and mixed
(random-effects) models for meta-analysis tests of moderator
variable effects. Psychological Methods, 3, 354–379. https://
doi.org/10.1037/1082-989X.3.3.354
Paffrath, J., & Simon, B. (2020). The significance of the superor-
dinate: Linking (dis-)embedded identity to non-normative ends
and means. Journal for Deradicalization, 24, 161–190.
Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear mod-
els: Applications and data analysis methods (2nd ed.). SAGE.
R Development Core Team. (2016). R: A language and environ-
ment for statistical computing [Computer software manual].
http://www.R-project.org
Reininger, K. M., Schaefer, C. D., Zitzmann, S., & Simon, B. (2020).
Dynamics of respect: Evidence from two different national and
political contexts. Journal of Social and Political Psychology,
8, 542–559. https://doi.org/10.5964/jspp.v8i2.1199
Renger, D., Renger, S., Miché, M., & Simon, B. (2017). A social
recognition approach to autonomy: The role of equality-based
respect. Personality and Social Psychology Bulletin, 43, 479–
492. https://doi.org/10.1177/0146167216688212
Rosseel, Y. (2012). lavaan: An r package for structural equation
modeling. Journal of Statistical Software, 48, 1–36. https://doi.
org/10.18637/jss.v048.i02
Rubin, D. B. (1987). Multiple imputation for nonresponse in sur-
veys. Wiley.
Scanlon, T. M. (2003). The difficulty of tolerance: Essays in politi-
cal philosophy. Cambridge University Press.
Schaefer, C. D., & Simon, B. (2019). Muslims’ experiences of
equality recognition in Germany: Effects on identification with
the larger society and on equality recognition of other societal
subgroups. Cultural Diversity & Ethnic Minority Psychology,
26, 32–41. https://doi.org/10.1037/cdp0000270
Schaefer, C. D., Zitzmann, S., Loreth, L., Paffrath, J., Grabow, H.,
Loewy, M., & Simon, B. (2021). The meaning of respect under
varying context conditions. Manuscript submitted for publication.
Selya, A. S., Rose, J. S., Dierker, L. C., Hedeker, D., & Mermelstein,
R. J. (2012). A practical guide to calculating Cohen’s f 2, a
measure of local effect size, from PROC MIXED. Frontiers in
Psychology, 3, 1–6. https://doi.org/10.3389/fpsyg.2012.00111
Simon, B. (1992). The perception of ingroup and outgroup
homogeneity: Reintroducing the intergroup context.
European Review of Social Psychology, 3, 1–30. https://doi.
org/10.1080/14792779243000005
Simon, B. (2020). A new perspective on intergroup conflict:
The social psychology of politicized struggles for recog-
nition. Theory & Psychology, 30, 147–163. https://doi.
org/10.1177/0959354319887227
Simon, B., Aufderheide, B., & Kampmeier, C. (2001). The social
psychology of minority-majority relations. In R. Brown & S.
Gaertner (Eds.), Blackwell handbook of social psychology:
Intergroup processes (pp. 303–323). Blackwell.
Simon, B., Brause, A.-C., & Galeão Silva, L.-G. (2016).
Contextualizing the meaning of respect: The case of
Evangelicals in Brazil. Revista de Psicología Social, 31, 399–
422. https://doi.org/10.1080/02134748.2016.1190130
Simon, B., Eschert, S., Schaefer, C. D., Reininger, K. M.,
Zitzmann, S., & Smith, H. J. (2019). Disapproved, but toler-
ated: The role of respect in outgroup tolerance. Personality
and Social Psychology Bulletin, 45, 406–415. https://doi.
org/10.1177/0146167218787810
Simon, B., & Grabow, H. (2014). To be respected and to respect:
The challenge of mutual respect in intergroup relations.
British Journal of Social Psychology, 53, 39–53. https://doi.
org/10.1111/bjso.12019
Simon, B., Grabow, H., & Böhme, N. (2015). On the meaning of
respect for sexual minorities: The case of gays and lesbians.
Psychology & Sexuality, 6, 297–310. https://doi.org/10.1080/1
9419899.2014.987683
Simon, B., Reininger, K. M., Schaefer, C. D., Zitzmann, S., &
Krys, S. (2019). Politicization as an antecedent of polarization:
Evidence from two different political and national contexts.
British Journal of Social Psychology, 58, 769–785. https://doi.
org/10.1111/bjso.12307
Simon, B., & Schaefer, C. D. (2016). Tolerance as a function of
disapproval and respect: The case of Muslims. British Journal
of Social Psychology, 55, 375–383. https://doi.org/10.1111/
bjso.12137
Simon, B., & Schaefer, C. D. (2018). Muslims’ tolerance towards
outgroups: Longitudinal evidence for the role of respect.
British Journal of Social Psychology, 57, 240–249. https://doi.
org/10.1111/bjso.12213
Tropp, L. R., & Pettigrew, T. F. (2005). Relationships between
intergroup contact and prejudice among minority and majority
status groups. Psychological Science, 16, 951–957. https://doi.
org/10.1111/j.1467-9280.2005.01643.x
Turner, J. C., Hogg, M. A., Oakes, P. J., Reicher, S. D., & Wetherell,
M. (1987). Rediscovering the social group: A self-categoriza-
tion theory. Blackwell.
Verkuyten, M., & Yogeeswaran, K. (2017). The social psychology
of intergroup toleration: A roadmap for theory and research.
Zitzmann et al. 21
Personality and Social Psychology Review, 21, 72–96. https://
doi.org/10.1177/1088868316640974
Verkuyten, M., Yogeeswaran, K., & Adelman, L. (2019).
Intergroup toleration and its implications for culturally diverse
societies. Social Issues and Policy Review, 13, 5–35. https://
doi.org/10.1111/sipr.12051
Vogt, W. P. (1997). Tolerance and education: Learning to live with
diversity and difference. SAGE.
Wenzel, M., Mummendey, A., & Waldzus, S. (2007). Superordinate
identities and intergroup conflict: The ingroup projection
model. European Review of Social Psychology, 18, 331–372.
https://doi.org/10.1080/10463280701728302
Yuan, K.-H., & Chan, W. (2008). Structural equation modeling with
near singular covariance matrices. Computational Statistics
and Data Analysis, 52, 4842–4858. https://doi.org/10.1016/j.
csda.2008.03.030
Zhou, S., Page-Gould, E., Aron, A., Moyer, A., & Hewstone, M.
(2019). The extended contact hypothesis: A meta-analysis on 20
years of research. Personality and Social Psychology Review,
23, 132–160. https://doi.org/10.1177/1088868318762647