On the Discomfort with the Accuracy Criterion: Reassessing Puddifoot’s Criticism of the Dual Factor View

1. Introduction

Focus on stereotypes and their negative effects has increased in the legal domain in recent years. Although the battle against the discriminatory effects of stereotypes has a long history in some legal systems such as Sweden and the United States—where the term “stereotype” has appeared in legal debates and decisions since the 1960s1 and the 1970s,2 respectively—in other legal systems the term has only emerged very recently.3 In the legal context stereotypes may produce their negative effects in several ways. First, they may hide behind norms that at a first sight might seem not only harmless but also beneficial for the stereotyped group. For instance, the norm that gives some financial aid to women who are the only caregiver of an elder member of her family. At first glance, the norm appears to introduce a benefit for women. However, it “hides the pervasive and mutually reinforcing stereotype that women are responsible for performing (unpaid) family care, and men are responsible for providing their families with financial support” (Franklin 2010, 139). Second, stereotypes may influence the assessment of evidence. For instance, they often play a significant role in how judges evaluate the credibility of victim testimony in cases of gender-based violence. Here I have in mind stereotypes such as “a victim of rape must try to resist and escape in all possible ways, but she is to be suspected of fabricating allegations of rape if she has the presence of mind to collect evidence after the assault” (CEDAW/C/82/D/148/2019) or “women accuse their partners of rape in order to obtain an economic advantage, such as keeping the family home” (Asensio et al. 2010, 87).

Given the pervasive negative effects of stereotypes, a widely shared view holds that almost all stereotypes are harmful and must be addressed using available legal instruments.4 However, some scholars argue that we should draw on empirical sciences and use statistical advancements to distinguish between stereotypes with and without statistical support.5 Although legal scholars seem to agree that the accuracy or statistical criterion improves our understanding of how stereotypes work, there is growing discomfort regarding some of its applications (Arena 2022; Ghidoni and Morondo Taramundi 2022). The discomfort stems mostly from the awareness that there are wrongs brought about by stereotypes that are not captured by that criterion, such as stereotype threat, disregarding individuality, and reinforcing discriminatory practices.

In her book, Katherine Puddifoot (2021) provides additional arguments supporting the discomfort with the accuracy criterion. What sets Puddifoot’s critique apart from previous analysis is her focus on the epistemic dimension of stereotypes and stereotyping (i.e., the application of a stereotype to an individual):6 “the various ways that stereotyping makes people susceptible to making errors in their perception and judgment” (2021, 5). Puddifoot is interested in a specific kind of judgements, to wit, those about the traits of an individual: for instance, whether an individual Afro-American is a drug dealer or whether a particular woman is an expert in STEM disciplines, etc.

Puddifoot identifies two previous approaches to the epistemology of stereotyping: the Single Factor view and the Dual Factor view. Both theories address “whether the application of the stereotype increases or decreases the chance of an accurate judgement being made” (2021, 31). According to the Single Factor view, stereotyping produces an epistemic benefit when the applied stereotype reflects some aspect of social reality. In contrast, the Dual Factor view holds that two conditions must be met for an epistemic benefit to arise: the stereotype must accurately reflect reality by aligning with precise statistical information, and the act of stereotyping must be sensitive to unambiguous individual information. Puddifoot claims that both the Single and the Dual factor views fall short of what we need in order to develop a full theory of the epistemology of stereotyping. I will focus here on her criticisms against the Dual factor view.

Puddifoot presents her objections and her perspective in several steps. She begins by offering a pragmatic argument in favour of what she terms a non-normative conception of stereotypes, which holds that our definition of a stereotype should not include the condition that stereotypes are inaccurate or distorting. She then advocates for a Multifactorial view regarding the criteria for the epistemic evaluation of stereotyping. Subsequently, she claims, on the one hand, that even stereotypes with statistical support can cause epistemic harm and, on the other hand, that having false stereotypes (that do not reflect some aspect of social reality) may also be epistemically beneficial—that is, they may increase the likelihood of making an accurate individual judgement (see also Saul’s discussion in this issue of EuJAP). While I share Puddifoot’s discomfort with the accuracy criterion, in the remainder of the paper, I will present some critical considerations regarding certain steps in her arguments.

Firstly, I will introduce a methodological concern regarding the discussion of the normative versus non-normative conception of stereotypes. My main aim here is to clarify what appears to be a (fragment of a) conception of philosophical method underlying the way Puddifoot evaluates arguments put forward by other proponents of the non-normative view. Secondly, I will raise doubts about the extent to which the additional factors identified by Puddifoot's Multifactor view genuinely challenge the Dual Factor view. My central argument is that, while Puddifoot highlights important critical issues within the epistemology of stereotyping, the Dual Factor view provides the tools to address these concerns. Thirdly, I will critically examine the argument that not having stereotypes, or having egalitarian stereotypes that do not reflect some aspect of social reality, improves our chances of achieving certain epistemic goals regarding the perception of individual traits. I will argue that having egalitarian stereotypes would not eliminate the pitfalls of stereotyping; rather, what is needed is a form of epistemic sensitivity to new information. Finally, I will briefly introduce a hidden factor of deception: the normativity of stereotypes, i.e., the fact that some stereotypes are used to impose specific roles on members of a social group. This characteristic of stereotypes has been recognised by psychologists (e.g., APA 1991; Burgess & Borgida 1999) and political theorists (e.g., Appiah 2000). However, in my view, it has not received sufficient attention in the debate on the epistemology of stereotyping.

2. The philosophical status of the terminological move

There is a story often shared among Genoa University’s PhD candidates about two analytic philosophers who were very good friends: one Argentinian and the other Italian. Both shared a love for Italian wine and food but disagreed on one issue—Italian bread. According to the Argentinian, Italian bread was simply bad. One day, the Italian philosopher, in a mildly nationalist gesture, offered the Argentinian a piece of bread without revealing its origin. The Argentinian tasted it and, satisfied, declared: “It’s very good”. Triumphantly, the Italian exclaimed: “It’s Italian!” To which the Argentinian wittily replied: “Then it’s not bread”.

Beyond the question of whether the story may be considered humorous, it serves to raise an important point: does a terminological move qualify as a philosophical move? This question is particularly relevant in the context of Puddifoot’s approach to the debate between non-normative and normative conceptions of stereotypes. According to non-normative conceptions, stereotypes may be accurate, whereas normative conceptions hold that stereotypes are always inaccurate and distorting. While Puddifoot favours non-normative conceptions, she argues that the debate is ultimately a matter of framing or terminology:

From the way Puddifoot engages in the discussion between normative and non-normative conceptions of stereotypes, it appears that she treats the terminological move as a philosophical move. For example, she considers the strategy of pointing out all the accurate stereotypes to be a bad argument against the normative view:

But is this terminological move truly a philosophical move—that is, one that contributes to our understanding of a concept? The answer to this question is undoubtedly tied to one’s preferred approach to conceptual analysis. While this is not the place to fully address that issue, it seems to me that a terminological move does not constitute a method for advancing a conceptual proposal. The success of a conceptual proposal can be evaluated based on counterexamples, logical coherence, and ordinary language intuitions, among other criteria, but not on its ability to preclude the possibility of a terminological move.7

For example, a first type of argument might consist in pointing out intuitive examples of the concept excluded by the proposed analysis. Jussim et al. (2009), proponents of the non-normative view, have highlighted certain social attitudes found to be accurate—such as sex distributions in various occupations or beliefs about demographic differences between African Americans and other Americans. Even though these examples fall outside of the scope of a normative concept of stereotypes, they are intuitive examples of stereotypes that have been even identified as such by advocates of the normative view.8

A second type of argument might involve considerations about how well a conception aligns with ordinary language usage. For example, the fact that people often use the term “stereotype” as an accusation is frequently cited as an ordinary language-based argument in defence of the normative view (Puddifoot 2021, 21). However, the accusation implied by the term “stereotype” is not always, nor necessarily, based on the inaccuracy of the attitude. At times, the accusation may instead arise from other failures, such as disregarding individual information, being overly generalised, or violating the duty to treat others as individuals. For instance, the statement “Women do not ride buses” is accurate in nearly all cities in Argentina and widely recognised as such. Nevertheless, the attitude expressed by this statement would still be labelled a stereotype. Thus, adopting the non-normative view does not rule out the possibility of the term retaining its negative expressive content

Even though I find these arguments convincing, the point here is that it is this type of argument, rather than a terminological move, that determines the outcome of a conceptual debate. Otherwise, if a terminological move were to count as a philosophical move, then pointing out the existence of stereotypes such as “African Americans are musical” or “Women are caregivers” would have to be regarded as a bad argument against the claim that only attitudes associating negative traits qualify as stereotypes. As the supposedly humorous story about Italian bread illustrates, in the context of conceptual analysis, a terminological move represents philosophical surrender. While it is always an available option, it amounts to raising the white flag.

3. The scope of the Dual Factor View

One key point in Puddifoot’s book is the criticism of two principled approaches to stereotyping that she labels the Single and the Dual factor views. Principled approaches purport to identify in which cases stereotyping may have epistemic quality, in the sense of increasing our chance of making an accurate judgement about the traits of an individual. In that context, Puddifoot distinguishes between a stereotype that reflects some aspect of social reality and a stereotype that fits accurate statistical information. An example of the distinction may be found in the difference between the stereotype “women are not good at math” and the stereotype “Chilean young women get lower results than Chilean young men in math exams for University admission” (see del Río et al., 2016). Given the accuracy of the second stereotype, the first one reflects some aspect of social reality even if itself is not accurate. On this basis she distinguishes between the Single and the Dual factor view. The Single factor view claims that the only factor that determines the epistemic quality of an act of stereotyping “is the extent to which the stereotype reflects social reality” (32). The Dual factor view claims that the epistemic quality of an act of stereotyping is determined by two factors: “the accuracy of the stereotype and the nature of the available evidence about the individual (whether or not it is high quality and unambiguous)” (32). Puddifoot does not purport to deny the importance of the factors identified by these two approaches. On the contrary, she challenges both approaches by showing that there are further factors that determine the epistemic quality of an act of stereotyping. These factors are: (i) Stereotype Accuracy; (ii) Stereotype Relevance; and (iii) Response to Case-Specific Information.

The first factor, as Puddifoot notes, provides a challenge only to the Single factor view. There are stereotypes that, even if they reflect some aspect of social reality, given that they do not fit statistical information, may lead a person to form a distorted judgement of an individual. The stereotype about the relationship between women and math in Chile is a good example of that risk. I agree with this point; my concern here will be with the two remaining factors as they purport to represent a challenge to the Dual factor view. The point is not to deny the importance of these further factors, but to comment on their value as a criticism against the Dual Factor view.

The second key factor when assessing the epistemic quality of stereotyping is relevance: “individuals are not appropriately sensitive to contexts in which any statistical information that might be encoded in a stereotype is relevant to the judgement” (46). Puddifoot introduces a psychological explanation of this lack of sensitivity linked to the way in which stereotyping is triggered: wounded ego and the desire to justify the current social system. Given these psychological causes of stereotyping, it is often the case that bearers of stereotypes would apply them even when irrelevant. And, when the stereotype is irrelevant, its application does not increase the chance of an accurate judgement being made.

The first difficulty in assessing this challenge to the Dual Factor view is that some examples (the black man who committed a minor traffic violation and a black person treated as threatening (46-47)) are based on inaccurate stereotypes. So, those cases show indeed that relevance is important, but do not challenge the Dual Factor view. The second, and key, difficulty, is that there is only a generic definition of “irrelevance” as “influenc[ing] judgements about other aspects of social reality” but from the examples there seem to be several possible interpretations of the meaning of relevance.

First, relevance may be interpreted as “the individual is within the scope of the stereotype”. This seems to be a possible interpretation of relevance in the case of the black man who committed a minor traffic violation and received the application of the stereotype associating Black people with crime and the case of the woman who entered a university to work in STEM disciplines and received the application of the stereotype associating men more strongly than women with scientific expertise. In both cases, it cannot be claimed that the stereotypes are irrelevant in the sense of influencing judgements about other aspects of social reality because in both cases the judgements are about the aspect to which the stereotype refers: crime and scientific expertise. The irrelevance there seems to stem from the fact that both individuals are not within the scope of the more accurate formulation of the stereotype—e.g., “Black people under certain contextual conditions are prone to certain kinds of crimes” or “Women with specific characteristics have less scientific expertise than men”.

Second, relevance may also be interpreted as “non-spurious statistical correlation”. It seems to be a good interpretation of relevance in the case of stereotypes regarding Black people or career women. In the first case, stereotypes of Black people while slavery was legal in the United States, such as that they are happy, childlike, and affectionate, are irrelevant in order to associate one of those traits with an individual Black person because they express a spurious correlation. A similar interpretation of relevance may follow from the examples of stereotypes about career women as lacking creativity given their monolithic lifestyle.

Finally, relevance may even be interpreted as “argumentative force”. It seems to be a good interpretation of irrelevance in the case of the debate on the changes to the US criminal justice system regarding the punishment attached to the possession of crack and powder cocaine. Within this debate, the stereotype strongly associating Black people with crime is irrelevant because it has no argumentative force regarding the question “whether specific sentences are appropriate for particular crimes” (48).9

The important point here is that the Dual factor view would agree with the claim that an act of stereotyping has epistemic quality (produces epistemic benefits) when it is applied to an individual that belongs to the group associated by the attitude with certain trait and when that association expresses a non-spurious relationship, i.e., when it has statistical basis. The reason for that agreement is that both requirements follow from the condition of accuracy. The stereotype according to which Chilean women got lower results than men in math exams for the admission to universities has statistical basis and will produce epistemic benefits only if it is applied to a woman that fits the description: a Chilean woman that has taken the math exam.

The last factor is response to case-specific information. The point here is “whether or not the application of the stereotype leads diagnostic case-specific information to become inaccessible or distorted” (48). Puddifoot introduces several findings in psychology showing that stereotypes may distort case-specific information in different ways that leads to misperception of the individual traits:

I agree with Puddifoot to the extent that these factors illustrate the complexity of the task of gathering individual information and assessing the traits of an individual. However, given the emphasis that the Dual factor view puts on case-specific information, it seems to me that many of the cases of stereotyping pointed out by the examples would be considered by the Dual factor view as lacking epistemic quality, given that the ambiguity of individual information was not adequately ruled out. Take for instance the example of the Black female barrister who suffers the application of the stereotype associating barristers with being male and White. This stereotype, while being accurate (according to the example it fits statistical information), may produce epistemic costs given that people will fail to notice that a Black female may not share the characteristic attributed by the stereotype to the group and therefore will fail to notice that she is a barrister.

It may also produce epistemic costs given that people will tend to take as non-ambiguous (“she is a defendant”) information (“a Black woman entering the court building or the courtroom”) that is actually ambiguous (“she is either a defendant or a barrister”). However, from my point of view, this seems to be a case in which individual information was not taken into consideration in the right way. Remember that one of the conditions that, according to the Dual factor view, must be satisfied for the act of stereotyping to bring epistemic benefits is to take into consideration non-ambiguous individual information. This means that individual information must be carefully considered to rule out ambiguity before applying the stereotype. In this example, the problem was that the ambiguity of the information (“a Black woman entering the court building or the courtroom”) was not ruled out before applying the stereotype. If this is correct, then the Dual Factor View provides the resources to determine that, in this case, the act of stereotyping did not bring about epistemic benefits.

4. On the epistemic quality of an egalitarian attitude

According to Puddifoot, stereotypes that reflect aspects of social reality may nevertheless produce epistemic costs due to the different ways, listed in the previous section, in which stereotypes may distort case-specific information. Given these effects, Puddifoot claims that “there can be epistemic benefits associated with lacking these beliefs or other social attitudes, and even from having false social beliefs or other social attitudes” (73). One kind of false social beliefs that may bring epistemic benefits is “egalitarian social attitudes”, that is, social attitudes that represent the world as being egalitarian. To illustrate this point, Puddifoot introduces the example of Tim, who grew up in a society where men and women were equally well represented in science but now lives in one where they are not. Given the conditions in which he was raised, Tim holds an egalitarian attitude, believing that men and women are equally represented in science. While this belief was accurate in his home country, it does not reflect the reality of the society he currently lives in (in this case, the UK). According to Puddifoot, this attitude, which falsely represents the world as more egalitarian than it really is, may yield epistemic benefits—most notably, the benefit of “avoiding the distortion of case-specific information” (76). For instance,

That is why “the possession of [egalitarian] attitudes can protect against various epistemic errors, e.g., misinterpreting ambiguous evidence, having distorted memories, and so on” (2021, 75).

Let’s consider a further example to evaluate the argument. As mentioned above, in Chile the generalization that women get lower results than men in the math exams for admission to universities is an accurate generalization, i.e., it fits statistical information. Therefore, according to the Dual Factor View, applying this generalization to a specific female Chilean student who recently took the exam—associating her with the trait of having obtained a lower score than men—should increase our chances of making an accurate judgment about her individual traits. However, Puddifoot argues that this is not always the case, as even accurate stereotypes can produce effects that reduce our chances of making accurate judgments. For instance, suppose that after lunch with colleagues, the stereotyped student correctly splits the bill among all participants but then miscalculates the tip for the waiter. Due to the stereotype, we are more likely to remember the second event and not the first, leading to inaccurate judgments about her mathematical abilities. Now imagine a math professor who grew up in Argentina, where university admission does not require mandatory math exams. He holds egalitarian views about the distribution of math ability between men and women and later moves to Chile. According to Puddifoot, his egalitarian outlook would improve his chances of remembering all relevant information and forming accurate judgments about individuals, as he would not be influenced by the Chilean inegalitarian stereotype. And this would hold true even if, as in the example, the stereotype is statistically accurate.

However, is it necessarily the case that egalitarian attitudes bring epistemic benefits? Selectiveness is an inherent consequence of viewing circumstances through the lens of a generalization encoded in a stereotype. The use of a generalization creates a blind spot—namely, the facts excluded by the generalization. As Frederick Schauer puts it: “In focusing on a limited number of properties, a generalization simultaneously suppresses others, including those marking real differences among the particulars treated as similar by the selected properties” (Schauer 1991, 21–22). If selectiveness results from applying a generalization, then adopting a different or opposing generalization merely introduces another blind spot. Holding an inaccurate egalitarian attitude about the distribution of mathematical abilities may lead the math professor to overlook information that aligns with the accurate inegalitarian stereotype. For instance, he may be condescending toward a student with weak math skills or, as Puddifoot suggests, may forget the tip miscalculation incident. The point here is that the distortion of case-specific information is a consequence of basing the assessment of individual traits on a generalization, regardless of whether it has egalitarian or inegalitarian content.

Moreover, an individual with an inegalitarian attitude—that reflects the reality of the society where she was raised but inaccurate in her current society—may actually avoid distortions caused by an accurate egalitarian attitude that misrepresents case-specific inegalitarian information. Consider a Chilean-born and raised math professor who moves to Argentina. Her inegalitarian attitude accurately reflects her upbringing in Chile but is inaccurate in Argentina. However, if she encounters a female student with weak math abilities, an egalitarian attitude might distort her perception of the student, decreasing the likelihood of making a correct judgment. A similar issue arises, following Puddifoot’s example, with a UK-born physicist who travels to Tim’s home country and meets a female colleague with poor expertise.

It is certainly true that an egalitarian attitude can lead to better interpersonal interactions, mitigating the negative effects of discrimination. However, this does not necessarily mean that having egalitarian attitudes improves our chances of making accurate judgments about an individual’s traits. A person with an egalitarian bias may make the same errors as someone whose attitude reflects social reality, even if that reality is inegalitarian. The risks of ignoring relevant inegalitarian information, misinterpreting inegalitarian facts as unambiguously egalitarian, or forming distorted egalitarian memories of inegalitarian events remain present.

5. A hidden factor: Stereotype normativity

In the case of Price Waterhouse v. Hopkins, the American Psychological Association’s amicus curiae noted the importance of distinguishing between descriptive and normative stereotypes about women. The authors of the amicus curiae claimed that: “descriptive stereotypes characterize women in a way that undermines their competences and effectiveness; normative stereotypes label women whose behaviour is inappropriately masculine as deviant” (APA 1991, 16). After a decade of further research, this distinction has been generalized in order to distinguish between a descriptive and a prescriptive component of stereotypes:

When used descriptively, a stereotype aims to provide information about the world—it seeks to describe a state of affairs, specifically the traits of a group. Its direction of fit is stereotype-to-social-group. Therefore, descriptive stereotypes can be evaluated based on their accuracy—whether they accurately reflect aspects of social reality.

On the contrary, when used normatively, stereotypes seek to impose certain roles on members of a social group. In this case, the stereotype follows a social-group-to-stereotype direction of fit. This means that, for those who uphold it, a mismatch between the world and the stereotype is a reason to change the world rather than the stereotype itself. Since normative stereotypes do not claim to describe reality, they are not subject to empirical evaluation. Thus, in principle, it does not make sense to ask whether normative stereotypes are accurate or reflect social reality. This implies that an approach to stereotypes based solely on accuracy will overlook their normative dimension.

However, normative stereotypes can also have epistemic consequences, as norms about the behavior of certain social groups may distort perceptions of individual group members. The most dangerous effect is the inversion of the purpose of the epistemic endeavour, in the sense that the bearer of a normative stereotype may end up abandoning epistemic goals, limiting himself to disapproving of the behaviour of those who do not conform to the normative stereotype and constructing the facts in such a way as to make it possible to inflict some type of punishment. For example, someone who upholds the decent woman stereotype may, when faced with a victim who does not fit this ideal—i.e., a woman who does not behave according to its expectations—deny her the status of victim and dismiss the possibility of a crime, not based on evidence but as an expression of moral disapproval. In such cases, by transforming an epistemic inquiry into a normative judgment, the stereotype reduces the likelihood of an accurate assessment. A full theory of the epistemology of stereotyping should then take into consideration this further factor of distortion.

6. Summing up

In my opinion Puddifoot’s book is an essential reading for any legal scholar seeking a serious approach to stereotypes. These brief comments have focused only on the first chapters, but in the subsequent sections, Puddifoot develops further insights and arguments on the epistemology of stereotyping that also merit detailed discussion. The structure of my arguments here reflects my agreement with most of Puddifoot’s conclusions. I share her non-normative conception of stereotypes, even though I do not find her objections to alternative arguments for that same conception convincing. Like Puddifoot, I consider statistical accuracy a necessary—though not sufficient—criterion for assessing both the epistemic import of stereotypes and the epistemic quality of stereotyping.

Reading Puddifoot’s book, I have learned a great deal about how stereotypes can hinder epistemic inquiry due to various psychological factors that may lead their bearers to make incorrect judgments. My only disagreement concerns whether the existence of these factors constitutes an objection to the Dual Factor View. Relevance and the quality of individual or case-specific information are necessary criteria for assessing the epistemic import of stereotypes, but they are merely implications of the conditions set forth by the Dual Factor View.

The only fundamental disagreement concerns the benefits of egalitarian attitudes. As I have argued, some pitfalls of inegalitarian stereotypes stem from the fact that stereotypes function as generalizations. Since egalitarian attitudes are also generalizations, they may incur the same epistemic costs. Avoiding these costs requires a form of epistemic sensitivity—being open-minded and willing to revise one’s beliefs when reality proves them inaccurate.

Finally, I have introduced what appears to be an additional factor affecting stereotyping and contributing to epistemic costs: the normativity of stereotypes. This presents a challenge to the Dual Factor View, at least insofar as the accuracy criterion becomes ineffective in assessing the epistemic risks posed by stereotypes that do not aim to describe reality but rather to regulate the behavior of the target group.

Acknowledgments

Work on this paper has been supported by grant CEX2021-001169-M, funded by MICIU/AEI/10.13039/501100011033.