Skip to content

The Puzzling Persistence of Gender Discrimination in Insurance


Greta Krippner is Associate Professor of Sociology at the University of Michigan.

This post is part of a symposium on the law and political economy of insurance. Read the rest of the posts here.

** ** **

Across most lines of insurance, men and women in the United States can expect to pay different prices for access to coverage or receive substantially different benefits from their policies. Insurers’ reliance on gender to price insurance presents an anomaly in the contemporary era: gender classifications have been progressively eliminated from most other market domains, including employment, housing, and credit. In fact, one is hard pressed to find another major economic institution in which overt gender discrimination is legally sanctioned, making insurance an orphan in the civil rights revolution that has transformed how men and women transact in the marketplace over the last half century. This anomaly is even more striking when we observe that these practices have been vigorously contested in the period since the 1980s, with organizations such as the National Organization for Women (NOW) lobbying for federal legislation that would ban gender discrimination in insurance markets. When such efforts failed, these organizations brought gender discrimination lawsuits against major insurers such as Metropolitan Life, Mutual of Omaha, State Farm, and the Equitable Life Assurance Society. 

What explains the stubborn persistence of gender inequities in insurance markets, even in the face of significant resistance? In a forthcoming article to be published in the American Journal of Sociology, I argue that insurers’ persistent gender discrimination can be understood through the ways in which gender comes to be embedded in the tools used to price risk. Building on early work by feminist theorists that treated organizational routines inside firms as configured by gender difference, as well as on more recent scholarship in economic sociology that attends to the “devices” that enable transactions in the market, I suggest that insurers’ technologies of risk classification can usefully be considered as constituting a gendered market device. By a gendered market device, I refer both to the concrete tools that organize exchange and the cultural meanings that give shape to material practices in markets. Accordingly, delineating a gendered market device requires elaborating the set of practical techniques that “select” gender as a pricing variable and identifying the cultural logic that “assembles” these techniques into a loosely coherent system. 

The broader cultural logic that binds gender to insurers’ pricing tools, I argue, is supplied by legacies of mutualism that have long shaped the evolution of insurance as a social institution. Mutualism refers to the practice of embedding risk in solidaristic groups to provide social protection to group members who experience misfortune. While collectively provisioning against risk has always been part of human societies, it has involved widely varying social practices: from ancient burial societies among the Greeks and Romans, to the custom in medieval guilds of placing a box in the workshop to collect contributions for workers who became ill or disabled, to more familiar forms of risk sharing associated with trade unions and the institutions of the modern welfare state. Techniques of risk sharing as practiced by insurers are derived from these mutual traditions, and it is these aspects of insurance pricing technologies that I suggest lead insurers to privilege gender as a pricing variable. That discriminatory practices in insurance markets are rooted in mutual traditions may come as a surprise, but it should not: through the ages, the extension of social protection has often involved forms of hierarchy and exclusion, an issue I have explored in other writings.

The first, and most foundational, of these insurance pricing technologies is what insurers refer to as “class-based pricing,” meaning that insurers attach risks to groups, not individuals. This is at one and the same time an expression of the mutual idea and a mathematical requirement of predicting unknown future contingencies. To assign a price to risk, insurers first create classes composed of individuals who hold certain risk-relevant characteristics in common. Individuals who belong to a given risk class are believed to have the same exposure to accident, illness, or death. Risks are socialized within these groups: insurers examine the prior loss experience of each such group to estimate likely future losses, sharing their cost equally among group members who pay premiums into the insurers’ reserve fund. Because risks are calculated on the basis of group experience, insurers’ method is probabilistic. That is, insurers cannot predict which particular individual will have an accident, become ill, or die. But given a large enough group, insurers can predict with reasonable accuracy that some number of individuals will have an accident, become ill, or die over some period of time. Notably, because risks can only be calculated on the basis of groups, classification is unavoidable, as some criteria must be identified to sort individuals into classes. While the predictors of risk are potentially limitless, insurers traditionally rely on a finite number of demographic variables to define risk classes, including gender.

That classification forms the basis of insurance pricing immediately sets up a conflict with the anti-classification principle embedded in civil rights law. Critically, feminists who contested insurers’ discriminatory pricing in the 1980s and 1990s took issue not with the practice of classification per se, but only with insurers’ reliance on classifications prohibited by civil rights law, such as gender. In fact, this distinction was essential to NOW’s litigation, as feminists did not intend to extend their opposition to the use of gender as a pricing variable to a broader rejection of the (class-based) logic of insurance. Insurers had difficulty grasping such nuances, however, and tended to conflate feminists’ objections to gender classifications with a refusal of the practice of classification in general. This conflation reflected the fact that, for insurers, gender classifications were not easily dislodged from the apparatus used to price risk. 

To understand why this would be the case, a quick lesson in the mathematics of mutuality is necessary: because individuals together “occupy” the class defined by the intersection of characteristics selected as predictors of risk, there must be a sufficient number of appropriately-classed individuals to accurately predict losses for each group. For example, if gender and smoking are two factors considered most important in determining longevity, then the insurer requires observations on smoking mennonsmoking mensmoking women, and nonsmoking women to test statistical differences between these groups. As a result, the task of statistically validating differences between groups becomes exponentially more difficult as predictors are added to insurers’ models. Accordingly, class-based pricing necessitates relatively simple classification schemes, relying on the use of a few, broad classifiers that can generate adequate sample sizes to produce reliable statistical estimates. Here it would be difficult to improve on gender as a classifier that neatly divides the population into two approximately equal groups, efficiently populating the cells of the actuarial table. It is telling in this regard that racial classifications were unilaterally eliminated by insurers when subject to a similar civil rights challenge in the 1950s. Racial classifications do not evenly sort into broad groupings, as racial groups do not comprise equal segments of the population, and hence they offered insurers a much less useful predictor of risk.

A second feature of insurance pricing also derived from mutual traditions that operates to privilege gender over other available classifiers concerns insurers’ fixation on what they refer to as the “problem of subsidy.” The basic idea here is that – much like traditional mutual aid societies that restricted membership to those who shared the same occupation, religious affiliation, place of residence, and so on – risk classes are constructed as internally homogeneous groups containing individuals presumed to be roughly equal in life chances. Critically, the assumption that members of a class represent the “same” risk – defined as the average loss experience of the class – enables equivalent contributions to the reserve fund used to compensate individuals who experience a loss. By contrast, if individuals representing different risk exposures are combined in the same class, lower-risk individuals would necessarily “subsidize” higher-risk individuals. From the perspective of insurers, such subsidies violate basic norms of fairness and threaten to erode the social bond forged by the insurance principle. 

While insurers tend to discuss subsidy as though it is a purely technical consideration, in reality the problem of subsidy is an inherently sociological one. That is, the only thing that separates “risk sharing” – the basic purpose of insurance – from the pernicious danger of “subsidy” is the presence of recognizable groups. Any given risk class will contain heterogeneity between individuals who represent higher and lower risks than the average for the class as a whole – this is implicit in what it means to calculate an average, which necessarily involves dispersion around a mean value. Critically, this heterogeneity only appears as problematic when it coincides with a legible social difference. 

The importance of legibility in this sense was clearly demonstrated when NOW introduced litigation in the 1980s that would have required auto insurers to adopt classifications based on mileage driven rather than on gender. Feminists argued that risk classes organized around gender indiscriminately mixed lower- and higher-risk drivers, as the true underlying risk of accident was associated with mileage driven, for which gender represented an imperfect proxy. The courts ultimately rejected NOW’s lawsuit, refusing to recognize “low-“ and “high-mileage” drivers as legitimate social groups, such that one could meaningfully “subsidize” the other. Rather than gender serving as an imperfect proxy for mileage, regulators argued, NOW was using mileage as a “proxy” for gender. In other words, the only reason that the alleged “subsidy” between low- and high-mileage drivers mattered – to courts, regulators, or the public – was because the former tended to be women and the latter men. In this manner, the logic of subsidy operated to anchor insurers’ risk classification schemes in categories – such as gender – that already organized and stratified social experience.

Taken together, I suggest that these aspects of insurance pricing help explain why gender classifications appear to be so entrenched in the apparatus used by insurers to sort and stratify risks. Gender is quite literally built into the infrastructure used to price risk – an infrastructure formed in part by a culture of mutualism that embeds risk in solidaristic social groups. While both the concrete material organization of insurance pricing and its larger cultural integument suggest the durability of gender discrimination, the account here should not be read as signaling that these practices are either inevitable or immutable. Indeed, the advantage of gender as a classifier to insurers – the ease with which gender sorts the population into broad groupings, and the way it also claims a certain ontological privilege over classifiers that may, in fact, be more causally proximate to risk – may now be eroding as gender is destabilized as a social category. It is noteworthy in this regard that California recently prohibited the use of gender classifiers in auto insurance in order to make insurance practices consistent with a recent law that allows state residents to indicate their gender as “nonbinary” for all official purposes. Accordingly, if the concept of gendered market devices allows insights into how market mechanisms construct hierarchies organized around forms of social difference, it may also allow insights into how these hierarchies can be deconstructed, as well.