Response to a Reader Question
In reply to my latest essay, “Collosal Failure: 100 Years of Viewpoint Diversity Activism,” a colleague from a different discipline offered a set of questions that I think is important, and illustrative of a line of thought that is very salient to discussions of the issues raised in my essay.
Below, I present their comment in its entirety, and then my response to their observations. I am sharing in part because the response ended up essentially comprising an essay in length, with a lot of important observations – and even a study design I would very much like to pursue (experimental collaborators, hit me up!). And also because they, themselves, subsequently expressed interest in finding a way to share their comment and my response for feedback and engagement with others in their discipline. So this represents an easy way for them to accomplish that goal, and for me to record these ideas, which are highly-relevant to the subject of the main essay, for posterity.
“This is another great installment in your series on how the two concepts of diversity should be embraced together. On a theoretical level, one reason I think focusing on views makes more sense (at least in the epistemic context of the academy) is that epistemic arguments for racial, sexual, background diversity etc. always seem to depend on claims of correlations between those variables and viewpoints. Sometimes those claims can hold up empirically, but sometimes less so. But in any case, the claim that ‘feature X is associated with certain views’ means that sex, race etc. are always at one remove from beliefs. Viewpoint diversity doesn’t have that problem, since it looks directly at viewpoints, which is what it seems to make sense to want to maximize in intellectual contexts. Empirically, there’ some support for the idea that maximizing viewpoint diversity rather than focusing on surface characteristics does have more of an effect on the creativity of teams (see Wang et al. 2019).”
Why to differences in identity commitments matter?
I think it may be helpful to begin by being clear about the core claim being made by the identity commitments camp. Their intent isn’t to use race, gender, etc. as a proxy for certain views – or to select for any particular view(s), using these demographic characteristics as a mechanism for determining who holds those views. If it were the case that race, gender, etc. are merely being used as proxies for particular viewpoints (as insinuated), then you are right — it would probably be more efficient and effective just select for those viewpoints directly.
But this is not actually the argument that the identity commitments folks are typically making (indeed, as noted in the essay, many from this camp are somewhat disinterested in ideological diversity. Consequently, they wouldn’t be trying to use identity commitments as a proxy for ideological diversity). Instead, what they are trying to select for are people who have a different set of life experiences, a different set of relationships, a different set of perceived interests, etc. relative to the majority and/or a different relationship to the social phenomenon under investigation. These can help fill blind spots and correct misperceptions.
We’ll start with an intermediary case between identity and ideological commitments: religion. There is a huge difference between studying and cataloging religious beliefs, for instance, as an atheist who thinks religion is all a bunch of bunk but can dutifully recite historical facts about various traditions, doctrinal positions where they exist, etc. — versus being a believer who has had religious experiences, is part of a religious community, etc. Their respective understandings of religion, its significance, how it operates in the world, the relationship between doctrine and practice, etc. will tend to be very different if they are a believer v. not… and indeed, perhaps also if they are part of one faith tradition versus another. But notice, the key difference here that I am drawing attention to is not the ideological component (i.e. their specific beliefs), but rather their direct experience of religion and the ways that shapes their understanding and perception of this social phenomenon, their relationship to said phenomenon, how they talk about it, etc.
A similar kind of thing holds for race, gender, sexuality, etc. And it makes a huge difference with respect to scholarship. Again, this has nothing to do with any member of a given racial, gender, sexual, socioeconomic group holding any particular (political, moral, metaphysical) view. You wouldn’t be selecting for any particular view in the case of diversifying along the lines of, say, gender, race or sexuality. You would merely be seeking to incorporate people whose priors, social networks, perceived interests, commitments, life experiences, etc. would tend to vary systematically from those of the majority.
Now, one should not ex ante presume on this account that the priors, commitments, perceived interests, social networks, life experiences, etc. of members of different racial, gender, sexual groups vary in any specific way from the majority (there is not something like a monolithic ‘black experience’ or ‘gay experience’ or ‘female experience’ – just as William James noted with respect to religion, there are a variety of experiences of being black or gay or female. And even common experiences often have differential effects on different members of a population as a result of various contextual factors. Moreover, people’s relationships to various identities evolve over time, as I’ve explored in greater detail elsewhere.
Indeed, if there were a static and monolithic ‘black experience,’ for instance, you actually wouldn’t need many black people taking part in social research. Any token example could serve as a stand-in for all. Higher levels of diversity matter in part because even within different groups, while there are shared challenges, experiences, etc., there is also a huge amount of variation within groups. Conversely, there are many points of commonality between members of different groups. We should want to capture as much of all this as possible.
Now when presented with the fact that there are a variety of black experiences, some erroneously infer that if we can’t identify a monolithic impact of ‘blackness’ on knowledge production, then it must actually be meaningless whether or not someone is ‘black.’ Hence, we actually don’t have to worry about racial diversity in scholarship: if we can’t directly infer that people would hold particular views, experiences, etc. in virtue of their race, then ‘race’ must be something of an empty signifier (ditto with other identity commitments, mutatis mutandis). A couple problems with this line of thought:
First, such critics would be obliged then to say that ‘religion’ is also meaningless (again, as William James noted, experience of religion is similarly varied). Indeed, a comparable argument could be made against ‘conservatism’ or other constructs of this nature: one would be hard pressed to find universal (necessary and sufficient) conditions or impacts for almost any of these kinds of social phenomena. Consequently, on this account, most dimensions of ideological diversity would be rendered inconsequential alongside most dimensions of identity diversity. There would be no point to organizations like Heterodox Academy, because the idea of referring to incorporating ‘conservative perspectives’ would be just as meaningless as ‘black perspectives’ – and for the same reasons.
But actually, this line of thought is wrong in both cases. Wittgenstein more-or-less put this thinking to bed some time ago: just because something can’t be exactly specified doesn’t mean it’s empty or meaningless. That just isn’t the way language works, and it certainly isn’t the way the world works (i.e. it is not the case that if we can’t fit some phenomenon into neat little analytical boxes, it actually ceases to exist or matter ‘in the world’).
We can talk meaningfully of race, gender, religion, conservatism, etc. and their impacts while recognizing that people’s experiences of all of these things will vary and the impact of these commitments on knowledge production will also vary between individuals and across contexts. This variance does not reduce the significance of diversity, but rather, makes it all the more important to incorporate a range of black perspectives, Christian perspectives, conservative perspectives, female perspectives, etc. Focusing on patterns of distribution, we can also speak meaningfully about (and empirically demonstrate) the ways life experiences, social networks, perceived interests, etc. for one population tend to vary systematically from those of another population. These characterizations can quite reliable/ explanatory/ predictive in practice, even if they do not well-describe literally every single person within some group.
Perspectives v. beliefs
Diversity of perspectives is not the same thing as diversity of beliefs. For instance, you can also have diversity of perspectives without diversity of beliefs: people taking different paths to arrive at the same conclusions. This is probably the fantasy of many people advocating for demographic diversity, or attending to identity commitments, while ignoring ideological diversity and ideological commitments.
My challenge to them is that, actually, if you increase diversity you tend to get both. You often do get diversity of perspectives and you also generally increase diversity of beliefs. Because, for various reasons, as is widely captured in polling, etc., there are significant differences in beliefs about various issues that trend along group lines. For instance, blacks and Hispanics tend to be significantly more socially conservative and religious than whites. So if demographic diversity advocates really want to be inclusive and empowering with respect to what minorities think and feel, they will have to account for that fact: by increasing demographic diversity, they aren’t just bringing different perspectives, but also typically different beliefs. This has not been sufficiently accounted for by those who think demographic diversity is important but are not concerned with ideological diversity. This was my main argument of an earlier essay (here), with respect to the demographic diversity crowd.
This essay, by contrast, essentially argues to the identity commitments crowd that diverse perspectives are not enough — with respect to knowledge production, you also need to have people who actually hold different beliefs (not just people who arrived at the same beliefs by different means, or express the same beliefs in different ways, or emphasize different facets of the same general beliefs, or who explore different dimensions of something in different ways, but working from the same basic ideological framework). To the ideological diversity camp, in turn, I argue that diversity of beliefs are not enough — you also need diversity of perspectives (for the reasons described above and throughout).
On ‘skin deep’ differences
Your comment seems to be defining identity commitments as ‘surface characteristics’ while ideological commitments are implied to be something different. Two issues:
First there is the implication, again, that race, gender, etc. have no significant impact. They are just meaningless differences in appearance – superficial attributes. This is incorrect. Take sex and gender. There are important hormonal differences between men and women that influence cognition, emotion, reaction, etc. in HUGE ways. There are also important differences in social expectations, relationships (i.e. how people treat you, who you form relationships with, what types of relationships you form with them) and experiences between men and women that contribute to significant differences in how interests and capacities are valued, utilized and developed. The culmination of these and other differences result in men and women tending to have different ‘cognitive styles’ and differences in preference tendencies, as I explored here. These different experiences, relationships, expectations, etc. can lead, and often have lead, women to approach issues with a different set of perspectives as compared to men – often contributing to different research priorities, asking different questions, asking them in a different way, interpreting evidence differently, as compared to men — explored a bit more here.
Moreover, it is not just things like race and gender that are ‘surface characteristics’ (i.e. things that can be quickly and reliably inferred by looking at someone quickly or through brief interaction). The same actually holds for religion, political ideology, socioeconomic status, sexual orientation – you name it.
People often give indicators, both subtle and obvious, intentional and unintentional, for all of these things — and they can be consistently (if imperfectly) inferred from people rather quickly. We can pretty quickly and reliably infer many people’s religion, political ideology, socioeconomic background, sexuality from the ways they present themselves and very brief interactions. We’re good at doing this, when we choose to turn our attention to making these distinctions. And there are obviously lots of evolutionary reasons it makes sense that we would be good at doing this.
And not just people! A recent study showed that AI could predict people’s political orientation with extremely high accuracy simply by analyzing pictures of their faces. In short, all of these much-discussed identity and ideological commitments are in a deep sense, surface characteristics.
What’s more, these impressions are not merely superficial. Instead, they tend to correlate reliably (if imperfectly) with people’s personality characteristics, behaviors, etc. As Lee Jussim argues extensively, stereotype accuracy is perhaps one of the most well-studied, widely-replicated and statistically significant effects in all of social psychology (accessible summary of this point here). This is not to ignore or trivialize the many instances where stereotypes are inaccurate, and the harm that can sometimes come from negative stereotypes. The point is merely that we are actually pretty good at inferring a lot about people from their ‘surface characteristics’ – including with respect to their ideological commitments.
So I would reject the definition that some of these (race and gender are what you seemed to have in mind) are ‘surface characteristics’ while ideological commitments are not. I would also reject any implication that those attributes described as ‘surface characteristics’ are just meaningless traits that people happen to have or not, but are more-or-less irrelevant with respect to how people approach various problems, or how they act and interact in the world, how they are perceived and treated by others, etc.
An Experimental Design
As it relates to whether or not ideologically diverse teams perform better at completing tasks than ones which include greater gender or racial diversity — there are actually a few big problems with imputing what you seem to want to impute about knowledge production, and about the relative significance of ideology v. identity commitments as it relates thereto.
One important factor is that people have higher levels of discomfort, and change their behaviors and expression, when put into groups with people of different races in a way that is different than when they happen to be in the presence of people from their same racial/ ethnic group who happen to subscribe to different ideologies, especially if they don’t know that they are in the presence of people with different ideologies. See, for instance, the phenomenon of competence downshifting among white liberals when they are addressing people of color, etc.
Hence, many of these performance differences probably relate to greater group cohesion in groups with less demographic diversity, which permits more trust, less fear of conflict, a freer flow of information and ideas, etc. That is, the apparent difference is a function of the discomfort and self-censorship many engage in when they are in a room of people who are obviously of a different group than them, rather than having anything to do with the relative epistemological virtues of identity v. ideological diversity per se (the different kinds of knowledge / critique each brings to the table).
I would be willing to wager actual money that if these exercises began by having people, say, announce to the group who they voted for in the last election – which may make clear that many in the group were Trump supporters for instance – then researchers tested how effectively they performed as a group, you would almost certainly get much different results.
For a true apples-to-apples comparison, participants should perhaps be instructed to wear a t-shirt and cap for the candidate they voted for in the last election — so that literally every time group members saw their colleague, their political leanings were thrust in their face in an unavoidable way. This is how it is for race and gender, after all – it is difficult to see someone without noting their apparent race or gender.
Now, if someone carried out such an experiment (having people perform group tasks while making it unavoidably obvious who they voted for in the last election) and found that there was a substantial difference in group performance in ideologically diverse teams when these differences were made unavoidably perceptible in the same way racial or gender differences often are, versus when they are not made unavoidably perceptible – this would suggest that a good deal of the apparent performance difference between ‘ideologically diverse’ groups v. ‘demographically diverse groups’ may be due to factors unrelated to the informational value of identity v. ideological differences per se – but rather, they are an artifact of the extent to which various forms of difference are unavoidably perceptible or not, obvious or subtle – and the effects of this on information sharing, cooperation, comfort, etc.
If there was no difference between the ideologically explicit condition and the control group (where ideological differences were not made to be pronounced), and especially if both conditions outperformed demographically diverse teams, then one would be better able to make the kind of inference you want to make with less threat of an obvious confound. But I suspect we would find that the major difference is actually the extent to which group differences are made salient, rather than the nature of the differences themselves.
Here it is worth noting that scholars conducting these kinds of comparisons tend not only to be aware of this huge possible confound, but they regularly acknowledge it in their work (including in the piece you cited. More on that soon): they aren’t apples to apples comparisons of race or gender to ideology unless participant ideology is presented to participants in the same way as participant race or gender.
I’ll add, again, we usually can impute political differences reliably through surface characteristics and brief interactions if we turn our mind to figuring out say, what someone’s religion is or how they voted. But often we don’t turn our minds to these tasks – especially not in settings where we are trying to accomplish a task as a group. Which makes sense from the standpoint of social cohesion. The difference is that for race and gender, the phenotypical differences seem much more pronounced, such that we can’t help but noticing them. Whereas for many these other differences, we can choose to notice them or not (obvious exceptions being, for instance, an orthodox Jewish man in black garb, a nun, a covered Muslim woman, a Sikh wearing a turban, someone who is wearing a MAGA hat, etc. That is, even noticing ideological characteristics and differences can sometimes become unavoidable).
Finally, I’ll add that the enterprise of academic knowledge production (what my essay is about) is very different *kind of thing* than completing tasks in the corporate world, etc. (which most of those studies look at). But even just running with the premise of looking at group performance in a corporate setting, we can see that far from being trivial differences, actually these ‘surface level characteristics’ matter a lot. In many contexts they matter more than ideological differences (which are less immediately apparent), which may lead to the observed differences in group performance.
On the cited Wang et al. metanalysis
As it relates to the paper you shared, an important note in that very article, related to the arguments I made above about the confound, is available on p. 695:
“In multicultural teams, surface-level cultural attributes provide clear signals about cultural identity, thereby resulting in identity threat and fragmentation (Leung & Wang 2015), as well as intergroup conflict and withdrawal behavior (Stehl et al. 2010).”
That is, differential performance may be the result of the fact that differences are surface-level, i.e. readily apparent. Rather than it being the case that these differences are irrelevant, it could be the case that people taking part in these exercises pretty much always perceive them to be salient — and cannot not notice them.
In the discussion and conclusion sections, the authors go on to say that they have nothing concrete to offer empirically about the mechanisms behind the differences they observe, nothing about potential moderators, etc. That is, their study cannot rule out the possibility that the entire observed difference is due to the phenomena they described at the top about how surface-level attributes undermine social coherence, resulting in differential performance for groups with higher levels of ‘surface differences’ as compared to groups with less. All their paper finds, in practice, is that there seems to be a difference in performance. What that means (practically speaking) and why the difference exists are separate questions that their study cannot meaningfully address.
And it must be emphasized: I often see this paper rhetorically deployed to talk about the supposed benefits of ideological diversity – by which people seem to mean political or religious diversity – as compared to racial or gender diversity. Yet literally none of the studies they draw their metanalysis from seem to actually be about the effects of religious or political diversity in teams (and as a side note, more than a fifth of the papers they look at for the meta-analysis [9/44] were never even published in an academic-press book or peer-reviewed journal).
Indeed, their definition of ‘deep differences’ is… strange. Many of the studies they cited — highlighting the effect of ‘deep differences’ — focused particularly on ‘cultural diversity.’ By ‘cultural diversity’ many of the papers include things ethno-linguistic groups (such as being white, but of Hispanic origin), ethno-religious groups (being Jewish, for instance) as examples of ‘cultural’ differences rather than ‘racial’ or ‘ethnic’ differences (i.e. being non-white). This is then put in apparent juxtaposition with the effects of race and ethnicity per se.
Yet many Hispanics and Latinos (who typically are legally classified as white, and would end up checking ‘white’ on a survey, etc.) – and indeed many non Hispanics or Latinos — view being Hispanic or Latino as being part of a minority group, something akin to being black, for instance. So this is a case where researchers are trying to draw a line that many do not acknowledge in the real world. And given that their analysis is about how these differences function in the real world, that disparity in classification between the researchers and the participants actually matters a good deal.
Moreover, many of the cited studies use ‘nationality’ as a proxy for ‘superficial’ differences in race and ethnicity — despite the fact that nations often contain multiple races and ethnicities, as the authors themselves note. On the other end of the spectrum, there is a significant potential confound in many cases between, say, ethno-linguistic identification and nationality: different ethno-religious groups (which cited studies often define as a ‘deep difference’) overtly tie their language etc. to their national identity (which cited studies often code as a ‘superficial’ difference). So that’s all kind of a mess.
In any case, for those people who want to use this study to make the claims that political diversity is more practically substantial than, say, racial diversity — the analysis doesn’t speak to that at all. And indeed, as the authors themselves note, actually one of the reasons why we may observe the effects we do is actually because ‘surface level’ differences (such as race or gender) are quite socially substantial. That is, they are not trivial with respect to team performance. Quite the reverse: they significantly affect how people interact with one another, share information, etc. in group settings – apparently much more directly and concretely than other forms of diversity (be it for better or for worse).