“There comes a point in time where you can’t take it any more. It’s like, enough is enough” (former American Express employee Nick Williams after allegedly being fired for being white, see Flood and Lanum Reference Flood and Lanum2022). Angry White Man. “White male authors face another form of racism when it came to trying to break through as writers in TV, film, theater or publishing” (author James Paterson, see Baxter Reference Baxter2022). Angry White Man. “Middle-class white men are the most discriminated against in the television industry” (news presenter Jeremy Paxman, see Martin Reference Martin2008). Angry White Man. Nick, James, and Jeremy are not alone. Whether rich or poor, highly educated or not, many white men appear to be angry (Gest Reference Gest2016; Kimmel Reference Kimmel2017).
Survey results confirm this broad sensation of white male malaise. Data from the British Election Study in the United Kingdom and the General Social Survey (GSS) in the United States show that white men report lower level of happiness on average, and find life less worthwhile (Supplementary Tables D.1 and D.2). In the United States, for which we can track happiness from 1972 to 2022, white men’s relative unhappiness contrasts with the 1990s and 2000s and is worse than at any point in time (see Supplementary Figure D.1).
There is more. Recent polls have documented a divide between young men (albeit of all races) and young women on liberal attitudes (Burn-Murdoch Reference Burn-Murdoch2024) and on tolerance toward feminism (Skinner Reference Skinner2024). White men have been the core constituency of Donald Trump (Igielnik, Keeter, and Hartig Reference Igielnik, Keeter and Hartig2021; Pew Research Center 2018) and they seem much more likely than any other group to have voted Leave in the 2016 Brexit referendum (House of Commons Library 2016; Alabrese et al. Reference Alabrese, Becker, Fetzer and Novy2019).
How are we to understand white men’s anger? Anger, as social psychologists explain, arises when an individual feels they get less than they deserve (Carver and Harmon-Jones Reference Carver and Harmon-Jones2009), when they feel unfairly denied some achievement (Haidt Reference Haidt, Davidson, Scherer and Goldsmith2003), when they encounter a challenge against core norms that threatens their situation (Marcus et al. Reference Marcus, Valentino, Vasilopoulos and Foucault2019). Anger is not irrational, it is a response to the actions of others (Roseman Reference Roseman1984), it is a form of backlash against changes which harm an individual or a group.
There are many possible causes of white men’s anger. Automation and globalization are two of them, but they should affect low-skill workers rather than white men. Immigration is another one, but all natives could feel economic or cultural losses, not white men specifically. I turn to two alternative sources of white men’s anger. The first regards the provision of new information. Recent years have seen the rise of various movements highlighting the discrimination against women (e.g., the #MeToo; Hillstrom Reference Hillstrom2018) or the discrimination against African Americans (e.g., the Black Lives Matter; Taylor Reference Taylor2016), which reveal, by contrapositive, how easy life has been for white men. The second consists in policy changes equalizing chances of access to socially valuable position, such as affirmative action or quotas for women in company boards and politics.
I use a stylized formal framework, which relies on three key assumptions. First, individuals are characterized by their group identity, their social status, and their ability. The society is divided between a dominant group D (here, white men) and a disadvantaged group d. Social status corresponds to elite (upper class, college educated, wealthy) versus non-elite. The ability, in turn, affects the chances that one reaches a high status. Second, the system is meritocratic: individuals with higher ability are more likely to succeed socially. However, there is uncertainty about the easiness of joining the elite for members of each group. Lastly, individuals care about their social status and their social reputation, defined as their expectation of others’ perceptions of their ability. Social reputation in my model is closely connected to Gidron and Hall’s (Reference Gidron and Hall2017, S67) concept of “subjective social standing” (“the level of social respect or esteem people believe is accorded to them within the social order”).
I first show that the arrival of new information about the relative chances of joining the elite for each group is likely to polarize individuals along identity lines. Learning that individuals from the dominant group can easily join the elite diminishes the accomplishment of group D members who have succeeded socially, reducing their social reputation and payoff. Such information also exacerbates the failures of group D members who do not belong to the elite, also worsening their social reputation. The reverse holds for disadvantaged group members.
Reducing inequalities in the chances of accessing the elite while keeping the size of the elite constant has, in turn, two contrasting effects. It lowers the chances that individuals from group D obtain a high social status. It also increases the reputation of all group-D citizens: embellishing success, which is now harder to obtain, justifying failures, which are now more frequent. I explain how this dual impact can split group D members. Individuals with very high ability and very low ability support policies helping the other group. Even after the reform, a high ability individual has high chances of joining the elite, whereas a low ability individual is always unlikely of becoming an elite member. Both, however, see an improvement in their social reputation. Individuals with intermediary ability are the losers; the reputational gain is insufficient to compensate for their lower chance of social success.
Overall, the model highlights that unfavorable information can generate a backlash by all individuals from the dominant group. Anger is less widespread in group D following policy changes. In both cases, (all or some) individuals from the dominant group react negatively because they lose socially. White men’s anger does not need to come from racist or sexist attitudes. It can be rooted in a sense of loss of their social standing.
Before turning to the model, I briefly connect my work to the most related formal literature. A long tradition considers how individuals use identity to form judgments about others (e.g., Phelps Reference Phelps1972). This has led individuals from disadvantaged groups to seek to erase their identity and assimilate into the dominant group (Eguia Reference Eguia2017; Fang Reference Fang2001). As a reaction, both members of the dominant group and those left behind in the disadvantaged group can “unite” to increase the cost of abandoning one’s original identity (Austen-Smith and Fryer Reference Austen-Smith and Fryer2005; Carvalho Reference Carvalho2013; Schnakenberg Reference Schnakenberg2013). For example, members of the dominant group can form stereotypes to sustain their social dominance (McGee Reference McGee2023). My article complements these important works with one twist. Even when the distribution of ability in all groups is known to be the same, differences in reputations can arise when individuals are uncertain about what it takes to join the elite. As such, my work is also connected to Ashworth, Berry, and Bueno de Mesquita (Reference Ashworth, Berry and de Mesquita2024), who study the sources of women’s underrepresentation in politics. Like in the present work, many of the theoretical results in Ashworth, Berry, and Bueno de Mesquita (Reference Ashworth, Berry and de Mesquita2024) rely on differences in reputation between men and women. Yet the causes of white men’s anger I highlight (information and policy changes) are completely distinct from the origins of women’s underrepresentation they study (voters’ discrimination and differential costs of running).
A FORMAL ILLUSTRATION OF THE ARGUMENT
Baseline Setup
Take a society with a mass of individuals. Individuals are characterized by their group identity, their social status, and their ability. A proportion
$ \alpha $
of citizens belong to the dominant group D. The rest (
$ 1-\alpha $
) belongs to the disadvantaged group d.
Regarding social status, a proportion e, commonly known, of the population belongs to the elite (
$ s=1 $
), with the rest being non-elite (
$ s=0 $
). The composition of the elite is unknown, but I suppose that the social status of an individual i would be observed in (unmodeled) social interactions. In contrast, ability, which I denote by
$ {\theta}^i $
, is an individual’s private information (type). It is common knowledge that each citizen’s ability is drawn independently and identically (i.i.d.) according to the cumulative distribution function (CDF)
$ F(\cdot ) $
, with associated probability density function (pdf) function
$ f(\cdot ) $
, over the interval
$ [\underset{\_}{\theta },\overline{\theta}] $
, with
$ \overline{\theta}>\underset{\_}{\theta } $
.
Ability matters to reach an elite social status. So does luck, which I capture by an unobserved random shock
$ {\epsilon}^i $
, distributed i.i.d. for each i according to the CDF
$ \Lambda (\cdot ) $
and pdf
$ \lambda (\cdot ) $
, over the interval
$ [-\overline{\epsilon},\overline{\epsilon}] $
. Individual i from group
$ g\in \{D,d\} $
belongs to the elite if the sum of their ability and luck is above a threshold
$ {E}_g\hskip-0.3em : $
$ {\theta}^i+{\epsilon}^i\ge {E}_g $
. Each citizen knows the way the system works. Individuals are, however, uncertain about the value of the relevant threshold for each group. The common knowledge and shared prior is that
$ {\overset{\sim }{E}}_D $
is distributed according to the CDF
$ \Gamma (\cdot ) $
, and pdf
$ \gamma (\cdot ) $
, over the interval
$ [{\underset{\_}{E}}_D,{\overline{E}}_D] $
with
$ 0<{\underset{\_}{E}}_D<{\overline{E}}_D<\overline{\theta} $
. The combination of
$ {E}_D $
with the elite size e fully determines the value of
$ {E}_d $
.
An individual i cares about their elite status
$ {s}^i\in \{0,1\} $
, with
$ {s}^i=1 $
denoting a member of the elite, and their social reputation. The latter consists of individual i’s expectation about other individuals’ perception of their ability given their social status. I denote it by
$ {\theta}_g^{*}({s}^i,{\theta}^i)\equiv {E}_{-i}^i(\overset{\sim }{\theta }|{s}^i,g,{\theta}^i) $
. A citizen i’s payoff is thus
$$ \begin{array}{rll}{U}^i({g}^i,{s}^i)={s}^i+{\theta}_g^{*}({s}^i,{\theta}^i).& & \end{array} $$
The game, in turn, proceeds as follows. Nature determines each individual’s ability
$ {\theta}^i $
and each citizen’s luck
$ {\epsilon}^i $
. Individuals in each group
$ g\in \{D,d\} $
with
$ {\theta}^i+{\epsilon}^i $
above the threshold
$ {E}_g $
become elite members. Individuals compute their social reputation (
$ {\theta}_g^{*}(\cdot ) $
) knowing their social status and ability. Payoffs are realized.
Before proceeding to the analysis, I impose a few restrictions on the model primitives. First, all pdfs (f,
$ \lambda $
,
$ \gamma $
) are continuous. Second,
$ \lambda (\cdot ) $
is symmetric around
$ 0 $
,
$ {\lambda}^{\prime }(\epsilon ) $
is continuous, and
$ \frac{\lambda^{\prime }(\epsilon )}{\lambda (\epsilon )} $
is decreasing with
$ \epsilon $
(the uniform and the normal distributions satisfy these properties). Third, for each level of ability
$ {\theta}^i $
, the full range of luck shocks is realized. Fourth, all individuals remain uncertain about the value of the threshold
$ {E}_D $
after observing their social status and ability. In terms of notation, I distinguish between random variables, denoted by
$ \overset{\sim }{\cdot } $
, and realization of a random variable, without tilde.
Preliminary Observations
The only quantity of interest is the social reputation of an individual. Absent additional assumptions, we cannot compare reputations across groups. The following lemma describes some properties of social reputations within each group (Supplementary Appendix A.2 contains the proofs for the baseline setup).
Lemma 1. Elite members have higher expected reputation than non-elite members: for all
$ {\theta}^i\in [\underset{\_}{\theta },\overline{\theta}] $
,
$ {\theta}_g^{*}(1,{\theta}^i)>{\theta}_g^{*}(0,{\theta}^i) $
for all
$ g\in \{d,D\} $
.
An individual’s social reputation increases with their own ability: for all
$ {\theta}_h^i,{\theta}_l^i\in {[\underset{\_}{\theta },\overline{\theta}]}^2 $
satisfying
$ {\theta}_h^i>\hskip-0.3em {\theta}_l^i,\hskip0.3em {\theta}_g^{*}({s}^i,{\theta}_h^i)>{\theta}_g^{*}({s}^i,{\theta}_l^i) $
for all
$ g\in \{d,D\} $
and
$ {s}^i\in \{0,1\} $
.
The first point is relatively straightforward. Given the meritocratic nature of the society, abler individuals have greater chances of joining the elite. Hence, individuals from the elite have higher reputation than non-elite members.
The second point is slightly subtler. It comes from individuals learning about the value of their group threshold from their social status and their ability. Take a successful individual (
$ {s}^i=1 $
). If that individual has a low ability, they understand that the threshold to join the elite is likely to be low (otherwise, it is unlikely they would have made it). Hence, they expect that social success comes with a small boost in reputation, their
$ {\theta}_g^{*}(1,{\theta}^i) $
is relatively low. In contrast, an individual with a high
$ {\theta}^i $
does not have the same consideration. They can make it to the elite with high probability whether the threshold is low or high. Hence, they expect others to hold them in high esteem, their
$ {\theta}_g^{*}(1,{\theta}^i) $
is relatively high.
The Effect of New Information
To look at the effect of information, I assume that all individuals receive a public signal z distributed over the interval
$ [\underset{\_}{z},\overline{z}] $
with CDF and associated pdf conditional on
$ {E}_D\hskip0.3em Z(\cdot |{E}_D) $
and
$ \zeta (\cdot |{E}_D) $
. Following Milgrom (Reference Milgrom1981), I assume that the conditional distributions satisfy the strict monotone likelihood ratio property:
$ \frac{\zeta (z|{E}_D^h)}{\zeta ({z}^{\prime }|{E}_D^h)}>\frac{\zeta (z|{E}_D^l)}{\zeta ({z}^{\prime }|{E}_D^l)} $
for all
$ z>{z}^{\prime } $
,
$ {E}_D^h>{E}_D^l $
. In words, a high threshold yields relatively more high than low signals than a low threshold.
The consequences of new information is summarized in the next proposition. To state it, I denote
$ {\theta}_g^{*}({s}^i,{\theta}^i|z) $
the social reputation of individual i in group g and social status
$ {s}^i $
and ability
$ {\theta}^i $
after public signal
$ z\in [\underset{\_}{z},\overline{z}] $
(recall
$ {\theta}_g^{*}({s}^i,{\theta}^i) $
is the pre-signal reputation).
Proposition 1. For all
$ g\in \{D,d\} $
, all
$ {\theta}^i\in [\overline{\theta},\underset{\_}{\theta }] $
, and all
$ {s}^i\in \{0,1\} $
, there exists a unique
$ {z}^0({s}^i,{\theta}^i,g)\in (\underset{\_}{z},\overline{z}) $
such that
-
•
$ {\theta}_g^{*}({s}^i,{\theta}^i|{z}^0({s}^i,{\theta}^i,g))={\theta}_g^{*}({s}^i,{\theta}^i) $
. -
• For all
$ z>(<){z}^0({s}^i,{\theta}^i,D) $
,
$ {\theta}_D^{*}({s}^i,{\theta}^i|z)>(<){\theta}_D^{*}({s}^i,{\theta}^i) $
.For all
$ z>(<){z}^0({s}^i,{\theta}^i,d) $
,
$ {\theta}_d^{*}({s}^i,{\theta}^i|z)<(>){\theta}_d^{*}({s}^i,{\theta}^i) $
.
If there exists an uninformative signal
$ {z}^u $
such that
$ \zeta ({z}^u|{E}_D)=\zeta ({z}^u|{E}_D^{\prime }) $
for all
$ {E}_D\ne {E}_D^{\prime } $
, then
$ {z}^0({s}^i,{\theta}^i,g)={z}^u $
.
A low signal reveals that the system is likely to be biased in favor of the dominant group. Individuals then realize that the bar
$ {E}_D $
is low for group-D members. Group-D individuals from the elite see their successes diminished; non-elite individuals from group-D see their failures exacerbated, both suffer a loss in social reputation.
For the disadvantaged group, the effect is exactly reversed given the fixed size of the elite. A low threshold for group D indicates a high threshold for their group. The successes of group-d elite members are embellished and the failures of non-elite individuals are easily excused. Both elite and non-elite group-d individuals experience an increase in their social reputation.
In Supplementary Appendix B.1, I show that the insights from Proposition 1 are robust to various changes to the information structure as long as individuals do not perfectly learn how the system works. An especially interesting case consists of uncertainty about the distributions of abilities in the two groups, their deservedness, instead of the value of the threshold. This changes the interpretation of some piece of information. For example, publicizing that the elite has a large proportion of group-D members is bad news if
$ {E}_D $
is unknown, as it indicates a low threshold, but good news if the distributions of ability are unknown, as it reveals group-D is more deserving. I briefly return to this below. For now, I note that information provision tends to unify members of the same group and polarize them with individuals from the other group.
Changing the Entry Conditions into the Elite
Instead of information (letting
$ {E}_D $
and
$ {E}_d $
be known to simplify computations), the dominant group may lose from policies meant to help the disadvantaged group. Fixing the size of the elite, I model such policies as
-
• increasing the threshold for the dominant group by
$ \Delta >0 $
, -
• decreasing the threshold for the disadvantaged group by
$ \delta (\Delta )>0 $
.
To study the effect of such policy, it is helpful to denote
$ {W}_D({\theta}^i,\Delta ) $
and
$ {W}_d({\theta}^i,\delta ) $
the expected utility of an individual i with ability
$ {\theta}^i $
(prior to their social status being determined) when the threshold increases by
$ \Delta $
and decreases by
$ \delta $
for groups D and d, respectively. Proposition 2 summarizes the effect of small changes to the thresholds.
Proposition 2. There exist
$ {\theta}_g^l,{\theta}_g^h $
,
$ {\theta}_g^l<{\theta}_g^h $
, unique if
$ {\theta}_g^j\in [\underset{\_}{\theta },\overline{\theta}] $
(
$ j\in \{l,h\} $
), such that:
-
• In group D, for all individuals with
$ {\theta}^i\in [{\theta}_D^l,{\theta}_D^h], $
$ \left.\frac{\partial {W}_D({\theta}^i,\Delta )}{\partial \Delta}\right|{}_{\Delta =0}\le 0 $
, for all individuals with
$ {\theta}^i\hskip0.3em \notin \hskip0.3em [{\theta}_D^l,{\theta}_D^h] $
,
$ \left.\frac{\partial {W}_D({\theta}^i,\Delta )}{\partial \Delta}\right|{}_{\Delta =0}>0 $
. -
• In group d, for all individuals with
$ {\theta}^i\in [{\theta}_d^l,{\theta}_d^h], $
$ \left.\frac{\partial {W}_d({\theta}^i,\delta )}{\partial \delta}\right|{}_{\delta =0}\ge 0 $
, for all individuals with
$ {\theta}^i\hskip0.3em \notin \hskip0.3em [{\theta}_d^l,{\theta}_d^h] $
,
$ \left.\frac{\partial {W}_d({\theta}^i,\delta )}{\partial \delta}\right|{}_{\delta =0}<0 $
.
Consider individuals from group D. Changing the thresholds has a direct and an indirect effect. Such change directly reduces the chances that individuals from group D join the elite. Indirectly, by moving the bar upward, the policy increases the reputation of all individuals from the dominant group.
Individuals with low ability have little chances to join the elite. They may care little about the direct effect of the policy change and mostly benefit from the reputational gain. Individuals with very high ability always have good odds to become elite members pre- or post-reform, so they may also mostly enjoy the boost in their reputation. In contrast, individuals with intermediary ability suffer from a change in the thresholds
$ {E}_D $
as they stand to lose the most in terms of chances of joining the elite. When
$ \left.\frac{\partial {W}_D(\underset{\_}{\theta },\Delta )}{\partial \Delta}\right|{}_{\Delta =0}>0, $
$ \left.\frac{\partial {W}_D(\overline{\theta},\Delta )}{\partial \Delta}\right|{}_{\Delta =0}>0, $
and
, a case consistent with the evidence in Besley et al. (Reference Besley, Folke, Persson and Rickne2017), average group-D individuals oppose changing the thresholds, whereas individuals at the top and bottom of the ability distribution support such policies.
For the disadvantaged group, in contrast, the direct effect of the policy is to increase the odds of joining the elite, whereas the indirect effect is a reduction in social reputation. Individuals in the middle of the ability distribution see the greatest gain because their chances of social success increase most; individuals with very low and very high abilities reject the reform due to their loss in reputation.
Proposition 2 relies on the size of the elite being fixed, like for quotas or affirmative action, otherwise group-D members would not care about the reform. It also depends on individuals having some information about their ability (see Remark B.1 in Supplementary Appendix B.2). Yet individuals do not need to perfectly know their ability as I assume here. When the random variables are normally distributed, I document the same within group splits as described in Proposition 2 as long as individuals receive a sufficiently precise signal about their ability (see Proposition B.3 in Supplementary Appendix B.2). Overall, the analysis in this subsection reveals that reducing inequalities in access to the elite can easily generate split within identity groups.
CONCLUSION
This article suggests two possible rationales for the rising anger among white men: information provision about systemic biases in their favour and policies helping disadvantaged group members. Both, I show, can cause a backlash from white men. This backlash does not arise because white men are fundamentally racist or sexist. Rather, it comes from the loss white men experience when they care about their social status and their social reputation.
The two rationales of white men’s anger that this article describes are not undistinguishable. Information provision can hurt all white men regardless of their social status. Policies that decrease the threshold for disadvantaged individuals can split white men between those with high and low ability benefiting from those reforms and those in the middle losing from them. This offers an opportunity to differentiate between these two causes. I do so in Supplementary Appendix D with data from the British Election Study in the United Kingdom as well as the GSS and the Cooperative Congressional Election Survey in the United States.
I use survey items that measure opposition to social changes (whether too much is being done for minorities or women, whether whites/men are discriminated against or not advantaged, whether racial problems are rare, whether Blacks are responsible for their own advancement, see the dataverse for more details; Wolton Reference Wolton2024). I compare attitudes across different educational levels (no high school, high school and some university, Bachelor degree and above) and across different age groups (under 25, 26–64, over 65). If information is the main cause of white men’s anger, then the difference between white men and other respondents should remain almost constant across all groups as per Proposition 1. If changes in policies matter more, differences in attitudes between white men and women and minority respondents should be significantly lower for individuals with no high school (a proxy for low ability) and with university degree (a proxy for high ability) than for individuals with a high school degree (a proxy for intermediate ability) as per Proposition 2. We should also expect over-65 white men to have similar attitudes as women and minorities in the same age range since older white men are less likely to suffer from the direct effect of policies favoring disadvantaged groups.
A few findings emerge from the analysis. Across all educational levels and all age groups, white men are more opposed to social changes than other survey respondents. Besides a few exceptions highlighted in Supplementary Appendix D, the difference in attitudes between white men and women or minorities is largely unaffected by education or age. In particular, university education reduces the likelihood that a respondent expresses unease with social changes, but the gap between white men and other survey respondents generally is the same as for respondents with a high school diploma. While only a first step, the empirical results so far indicate that the impact of information is at least as large as the effect of policy changes.
The setup presented here can be extended in multiple directions. The information individuals receive could be strategically communicated, rather than exogenous. In Supplementary Appendix C.1, building on Alonso and Padró I. Miquel (Reference Alonso and Padró I. Miquel2023), I study a model with strategic senders from the same group as the receivers or from the opposite group. There, individuals react much more negatively to bad news communicated by in-group senders, which may explain why white men sometimes feel betrayed by their peers. The information individuals receive could also be more complex to interpret. In Supplementary Appendix C.2, I study a simplified model in which both the threshold values and distribution of ability in group-D are unknown. Some signals increase the reputation of individuals from both groups as they indicate a high bar for disadvantaged group members and a high deservedness of dominant group members. Other news polarize the two groups just like in Proposition 1. I only consider the (rational) causes of white men’s anger and do not study the (possibly irrational) consequences of this sentiment for domestic electoral politics (e.g., combining the present approach with Schnakenberg and Wayne’s Reference Schnakenberg and Wayne2024 analysis of anger and conflict dynamics). Finally, I distinguish between white men from other individuals. Doing so, I have merged gender and race identities. Decomposing white men’s anger across these identity lines is a promising avenue for future research.
SUPPLEMENTARY MATERIAL
To view supplementary material for this article, please visit https://doi.org/10.1017/S0003055424000959.
DATA AVAILABILITY STATEMENT
Research documentation and/or data that support the findings of this study are openly available at the American Political Science Review: https://doi.org/10.7910/DVN/B3P41O.
Acknowledgements
The author thanks Dominic Rohner, Roland Hodler, the editors, and three anonymous reviewers for very helpful comments. All remaining errors are the author’s responsibility.
CONFLICT OF INTEREST
The author declares no ethical issues or conflicts of interest in this research.
ETHICAL STANDARDS
The author affirms this research did not involve human participants.
Open access
Comments
No Comments have been published for this article.