Born Inquisitive
A blog of independent thinking and evidence-based inquiry

The Fallacy of Generic Thinking

March 4, 2022
5,531 words (~27 minutes)
Tags: fallacies statistics psychology philosophy third-person

Generic thinking is unquantified belief about large populations of individuals, a profound and terribly mistaken fallacy that has little use other than prejudice.

Table of Contents


What unites the many types of fallacies that this blog discusses is that they would appeal to a mind afflicted with what can be termed “generic thinking.” For instance, the article “How to Scrutinize a Statistic” discussed both the need to evaluate practical significance and how this is typically done by quantifying an effect size. The absence of summarizing an effect size was described as a fallacy of “generic comparisons.”

The phrase “generic comparison” is actually a neologism invented by this blog, but the word “generic” was used deliberately to allude to what is termed “generic generalization,” a phrase established in the academic literature.

Generic thinking comprises both generic generalizations and generic comparisons and encompasses all beliefs about an inherently quantitative subject, but that are held without any quantification. This often occurs when someone is thinking about large populations of individuals. For instance, social issues occur, by their very nature, in large populations of individual human beings. Thus, all unquantified social criticism is generic thinking.

Generic Generalization

In a generic generalization, a property is predicated of a subject that denotes a population of individuals, but no quantification is given.

“All swans are white,” “some swans are white,” “no swans are white,” “approximately 74.2% of swans are white” are all beliefs about a subject – denoted by “swans” – that includes a large number of individuals. All of these statements include some kind of quantification, namely “all,” “some,” “no,” or “approximately 74.2%.” However, the statement “swans are white” is a generic generalization because it is unquantified.

A body of scholarship pertaining to generic generalizations exists in both academic philosophy and academic psychology. Generic generalizations have been studied by philosophers because they can seem like a logic puzzle. For instance, statements like “dogs have four legs,” “ducks lay eggs,” “mosquitoes carry the West Nile virus” seem like they have the spirit of truth to them.1 However, they are all false for a portion of the population they discuss.

While it is true that most dogs have four legs, there are some dogs happily scampering around on three legs. Only female ducks who have reached sexual maturity even have the ability to lay eggs, so “ducks lay eggs” is only true for less than half the population it describes. Mercifully, a very small proportion of mosquitoes actually carry the West Nile virus.

A three-legged dog walks on a grass lawn on a sunny day.
“Three Legged Dog” by Sangudo is licensed under CC BY-NC-ND 2.0 and is not modified.

Resistance to Falsification

Much of the philosophizing around generic generalizations has been an attempt to find some way of interpreting generic generalizations that would allow the ones that seem plausible to be true and the ones that are patently false to be false. (Leslie and Lerner 2016) This highlights precisely the main issue with generic generalizations: they seem to defy truth conditions.

In particular, generic generalizations are resistant to falsification. Quantified beliefs have conditions under which the holder of the belief can learn that the belief is false, if it is indeed false.

If someone who asserts “all swans are white” encounters a single black swan, then the individual knows the belief is false. Similarly, presenting just one white swan to someone who asserts “all swans are black” demonstrates to this person that the belief is false.

To demonstrate that the assertion “some swans are white” is false without any uncertainty is more difficult, since it requires observations of all the swans in the world. However, this is just a practical matter. In principle, it is known how to demonstrate the belief is false, even if demonstrating it is false would require more work than most are willing to commit.

Statements such as “approximately 72.4% of swans are white” can be demonstrated false, given a specified hypothesis test level. This is slightly more complicated by the fact that for a continuous variable any such hypothesis can be arbitrarily rejected given a large enough sample size and enough statistical power, so exact statistical thresholds would have to be stipulated. Once these are specified, however, such hypotheses can be evaluated as to their truth or falsity according to standard methods.

Thus, quantified beliefs that are false can be demonstrated as false. The same is not the case for generic generalizations. If someone asserts “dogs lay eggs,” how many viviparous dogs must be observed before this belief is demonstrated false? Similarly, how many two-legged ducks must be brought to someone who asserts “ducks have four legs” before the holder of the belief must relent? The stubborn can merely retort “not all dogs lay eggs” or “I didn’t mean all ducks have two legs.” Generic generalizations thus create a threshold that can always be conveniently moved to justify whatever arbitrary belief is desired.

Such resistance to falsification is a nontrivial matter because bigoted attitudes are apt to be construed in terms of generic generalizations. For instance, Leslie (2017) uses “Muslims are terrorists” and “Blacks are rapists” as examples of stereotypes formulated as generic generalizations. While these are negative stereotypes, positive stereotypes such as “men are good at math” or “women are empathic” are also often construed as generic generalizations. Thus, generic generalizations are tools for all sorts of category-based prejudice.

To those concerned about bigotry, this resistance to falsification is a troubling property of generic generalizations because it means that once a stereotype has been formulated as a generic generalization in someone’s mind, it is that much more difficult to convince the holder of the stereotype to change the belief.

If someone were to assert “most Muslims are terrorists” or “80% of Blacks are rapists,” these beliefs could be demonstrated false with the appropriate empirical observations and analyses. When someone asserts the generic generalizations “Muslims are terrorists” or “Blacks are rapists,” it does not matter how many Muslims are not terrorists or Blacks are not rapists, the generic generalization is resistant to falsification exactly because it is unquantified.


In addition to being resistant to falsification, generic generalizations are not informative because, given a large enough population, an arbitrary number of confirmatory examples can always be found.

For instance, generic generalizations such as “humans are terrorists” or “humans are rapists” seem odd. However, they have a similar amount of actual information as the more specific generic generalizations “Muslims are terrorists” and “Blacks are rapists.”

Such generic generalizations about “humans,” “Muslims,” or “Blacks” assert that, among some population of billions of individuals, there is some unspecified number of individuals who are terrorists or who are rapists. However, this is true of any large enough population of human beings.

The generic generalizations “humans are terrorists” or “humans are rapists” are conspicuous in just how vague and uninformative they are. However, this is precisely the same issue with all generic generalizations. The “humans” versions of these generic generalizations are merely more obvious in illustrating this defect.

Given a large population of humans, there will be among them some unspecified number of terrorists, rapists, charitable givers, volunteers, slalom skiers, stamp collectors, or anything else for that matter. This is a reasonable expectation a priori before any observations are made. Indeed, if a claim were made that a population of millions of human beings did not contain a single rapist, charitable giver, or stamp collector, such a claim would be treated incredulously. Generic generalizations are therefore simply not informative.

Empirical Study

The fallacy of generic generalizations is not just a philosophical concern, thanks in large part to the work of developmental psychologist Susan Gelman, who discovered that people become sensitive to generic generalizations at a young age. (Gelman, Star, and Flukes 2002)

Psychological research in this area has gone on to demonstrate the fallacy of generic generalizations empirically. Cimpian, Brandone, and Gelman (2010) conducted an experiment that found a discrepancy between the flexible truth conditions under which generic generalizations were accepted and the implied prevalence with which generic generalizations were interpreted.

A pool of 71 undergraduate student volunteers at two public universities were randomly assigned to two groups, one of which was given generic generalizations, and the other, a control group, was given generalizations that were quantified with the word “most.” Test subjects in groups were further randomly assigned into subgroups given either a truth conditions task or an implied prevalence task.

Each of the test subjects were given 30 generalizations supposedly about animals on a remote island. The animals were actually fictional. Previous research had used generalizations about real animals, which might have allowed test subjects’ prior knowledge to contaminate the results, hence Cimpian, Brandone, and Gelman used fictional animals.

Those in subgroups with the truth conditions task were given prevalence levels of either 10%, 30%, 50%, 70%, or 90% for the property described in the generalizations, then asked if the generalization was true or false. For instance, test subjects were given statements such as “30% of morseths have silver fur,” then asked if a generalization, such as “morseths have silver fur” or “most morseths have silver fur,” was true.

Those assigned to subgroups with the implied prevalence task were presented with a generalization, then asked to estimate the prevalence level of the property described in the generalization. For instance, test subjects were given a generalization, such as “morseths have silver fur” or “most morseths have silver fur,” then asked “what percentage of morseths do you think have silver fur?”

Among the generic generalizations group, the average prevalence level at which a generalization among the truth conditions task subgroup was accepted was 69.1%, but the average prevalence at which a generalization was interpreted by the implied prevalence task subgroup was 95.8%.

In contrast, the control group given “most” generalizations did not have such a discrepancy. The average prevalence at which the truth conditions task subgroup accepted a “most” generalization was 76.9%, and the average prevalence at which the implied prevalence task subgroup interpreted a “most” generalization was 78.0%.

Cimpian, Brandone, and Gelman also found that a property being dangerous (e.g., “this fur sheds particles that get lodged in your lungs and make it impossible to breath”) or distinctive (e.g., “no other animals on this island have this kind of fur”) exacerbated this effect, especially at lower levels of prevalence.

The study went on to conduct additional experiments to investigate another control group given “some” generalizations, whether the effect would be seen with obviously temporary conditions (e.g., muddy feathers or broken legs), and whether dangerous or distinctive properties would exacerbate the effect independently of each other.

This was a randomized, controlled experiment using an opportunity sample, implying a scope of inference described in the lower right cell of Table 1 in the article “How to Scrutinize a Statistic.” Thus, it can be concluded that the discovered fallacy (i.e., the discrepancy between truth conditions and implied prevalence) is caused by the generalizations being generic. However, these results cannot be inferred as estimating quantities in any larger population. Thus, this experiment conclusively shows that it is indeed the genericness that causes the fallacy, but it does not shed any light into how common this fallacy is in society at large.

Cimpian, Brandone, and Gelman discuss the implications of their findings.

The discrepancy between generics’ truth conditions and implied prevalence is not just an arcane bit of experimental data. . . . [G]enerics are a powerful means of manipulating public opinion. Since these generalizations are legitimized even by scant evidence, their truth is rarely questioned. Yet, after they become part of accepted discourse, they take on a life of their own, turning what may have originally been a nuanced, contextualized fact into a definitive pronouncement: A few cases of successful school voucher programs morph into “School vouchers work”; a few salient incidents at nuclear power plants become “Nuclear power plants are dangerous”; and so on.

This asymmetry has immediate relevance to stereotyping as well. . . . Consider, for example, a sentence such as “Boys are good at math.” Due to generics’ flexible truth conditions, this statement may be accepted based on very little evidence . . . . Once believed to be true, though, this statement may imply that being good at math is a normative, near-universal fact about boys, which may in turn have a powerful effect on one’s perceptions and behavior. (Cimpian, Brandone, and Gelman 2010, 6.1.2)

Thus, generic generalizations provide a false bridge across which faulty induction can travel. Indeed, the authors’ examples highlight that this faulty induction provides a mechanism for even anecdotes to be falsely generalized.

Bare Plurals and Elision

Because generic generalizations are resistant to falsification, uninformative, and have been found empirically to lead to flawed thinking, this article interprets generic generalizations as mere fallacies with no merit. Therefore, this article recommends that everyone strive to remove generic generalizations from their thinking.

This is a value judgment that implies that the philosophical program of attempting to salvage generic generalizations is not worthwhile. One might protest, what about the generic generalizations that feel like there is a ring of truth to them, such as “dogs have four legs,” “ducks lay eggs,” and “mosquitoes carry the West Nile virus?”

Such examples are called “bare plurals,” in which the subject of a sentence is a plural noun that is not proceeded by any modifier word. However, not all uses of bare plurals are truly generic generalizations.

This is due to the messy nature of natural languages. In particular, human beings are prone to make their communication more concise by elision. Sometimes just a few sounds are left out as in the case of a contraction, but whole words, phrases, and clauses can be elided to be inferred from context.

Commonly, a bare plural is used when “all” quantification is implied. For instance, in this very article, it is written that “generic generalizations are resistant to falsification.” This is not, however, a generic generalization about generic generalizations. From context, it should be gleaned that “all” quantification is implied, as in “all generic generalizations are resistant to falsification.”

If one went about one’s day putting “all” in front of each and every universally quantified statement, it would sound strange to native English speakers. In many contexts such as this one, using the word “all” might be used for emphasis, but it is not necessary for the English language to function.

Once one accepts that natural languages function with such elision, then whether or not any given bare plural is a generic generalization becomes a question, not an automatic assumption.

When someone says “ducks lay eggs,” there is typically social context surrounding this statement from which a more complicated thought can be inferred, such as “the system of reproduction used by ducks is oviparous.”

For instance, a child might ask if a duck found in a pond will give birth to little chicks that can swim about, to which an adult might reply, “no, ducks lay eggs.” Given context, the longer statement about reproductive systems is implied, but rarely do English speakers talk in this way, and instead this is shortened to “ducks lay eggs.”

Furthermore, one could assert “a dog without genetic aneuploidy and absent any physical trauma, allowed to develop in a salubrious environment including nutritional needs, grows into an organism with four legs.” That is a lot of words, which can be elided to the simple statement “dogs have four legs” given social context.

Similarly, those who claim to have gotten West Nile virus from an ant bite might be curtly told, “no, mosquitoes carry the West Nile virus.” It is likely in this case the proposition “mosquitoes are the only kind of organism that function as a vector for West Nile virus transmission in humans” has simply been shortened to “mosquitoes carry the West Nile virus.”

How to Avoid the Fallacy

One might protest that perhaps all generic generalizations could be explained as instances of this kind of elision. Such a protest is consistent with this article’s recommendation for how to avoid the fallacy of generic generalizations. The remedy is straightforward: whenever a generic generalization is noticed, it should be replaced by a quantified generalization.

If there is indeed elided quantification lurking beneath a supposed generic generalization, then making the quantification explicit demonstrates that the belief is not truly a generic generalization. If no quantification can be produced, then the belief is indeed a generic generalization, and the holder of the belief should reflect upon the flaws in this way of thinking.

Generic Comparison

In a generic comparison, it is asserted that one group has “more” or “less” of some attribute than another group, but no further quantification is specified.

As discussed previously, the phrase “generic comparison” is not as established as the phrase “generic generalization.” Indeed, while generic generalizations are studied by philosophers and psychologists, the phrase “generic comparisons” encompasses criticisms that are sometimes discussed by empirical researchers and applied statisticians, but are not as widely discussed outside of this niche. However, these criticisms are just as relevant, if not more so, whenever thinking about issues involving large populations of individuals, such as social issues.

Unlike a generic generalization, which is generic in precisely one way, there are two ways in which a generic comparison can be generic. A generic generalization fails to describe the proportion of the population for which its predicate is true. When “swans are white” is asserted, it is not clear if this means that 2% or 50% or 99% of swans are white. Similarly, if it is settled that, by any reasonable measure, one group has “more” of something than the other, a generic comparison leaves unspecified how much “more” one group has than the other.

Additionally, there are many ways to summarize how a variable is distributed in a population, and different ways might lead to different directions of “more” and “less.” Thus, generic comparisons can sometimes be ambiguous about which direction the “more” should go.

This section first addresses the issue of ambiguous direction, which is peculiar to generic comparisons, before discussing the issues stemming from unspecified magnitude, which are similar to the issues with generic generalizations.

Ambiguous Direction

What makes comparisons between groups more complicated than comparing individuals2 is that comparison between groups is a comparison of distributions, rather than point values. For instance, when comparing the height of two people, who can be labeled “A” and “B,” there will be a single point value for the height of person A, e.g., 68 inches, and a single point value for the height of person B, e.g., 75 inches. In this case, it is true that person A is shorter than person B, and false that person A is taller than person B.

On the other hand, if “A” and “B” represent groups of thousands of people, there is not a single point value that captures all the variation in the thousands of heights of people in group A or a single point value that can represent the thousands of heights of people in group B. In this case, the two distributions must be compared.

Unless the two distributions are completely disjoint, some people in group A will be taller than some people in group B, and some people in group B will be taller than some people in group A. It is not simply true or false that group A “is taller” than group B or group A “is shorter” than group B.

One could calculate and compare the arithmetic means of the heights in the two groups, but arithmetic mean is just one measure of the central tendency of a distribution. There are other measures of central tendency, such as the median, which also have their merits. More importantly, there are other properties of a distribution that are just as relevant, such as the dispersion or spread, the skewness, and to a lesser extent, the kurtosis.

Even with just the two basic statistics of mean and median, it is possible for the mean of group A to be greater than the mean of group B, but the median of group A to be less than the median of group B. Simplifying the heights example to involve only 30 people instead of thousands of people in each group leads to Table 1.

Table 1: The randomly generated heights of two groups of 30 people each.

Person Group Height (in)
1 A 51.0
2 A 51.8
3 A 53.3
4 A 53.8
5 A 54.4
6 A 54.4
7 A 55.1
8 A 55.5
9 A 55.7
10 A 55.7
11 A 55.8
12 A 56.0
13 A 56.1
14 A 56.2
15 A 56.5
Person Group Height (in)
16 A 56.5
17 A 56.6
18 A 57.2
19 A 57.3
20 A 58.1
21 A 65.2
22 A 66.1
23 A 66.3
24 A 66.8
25 A 67.5
26 A 67.6
27 A 67.7
28 A 67.9
29 A 68.3
30 A 69.3
Person Group Height (in)
31 B 53.7
32 B 55.2
33 B 55.7
34 B 55.8
35 B 56.1
36 B 56.5
37 B 56.6
38 B 56.7
39 B 56.7
40 B 56.8
41 B 56.9
42 B 57.3
43 B 57.4
44 B 57.4
45 B 57.9
Person Group Height (in)
46 B 58.4
47 B 58.6
48 B 58.8
49 B 58.8
50 B 59.4
51 B 59.4
52 B 59.6
53 B 59.8
54 B 60.4
55 B 60.4
56 B 60.4
57 B 60.7
58 B 61.2
59 B 61.9
60 B 65.3

A quick calculation shows that the mean height of people in group A is 59.3 inches, which is greater than the mean height of people in group B of 58.3 inches. However, the median height of people in group A is 56.5 inches, which is actually less than the median height of people in group B of 58.1 inches.3

This ambiguity can occur even between two of the most basic statistics used to describe a distribution: mean and median. Since there are many more statistics that can be calculated, there are many more opportunities for such ambiguities to occur. A similar phenomenon, but with means and totals, was described in this blog’s discussion of missing denominators.

The important takeaway is that it always flawed to think about a group having “more” or “less” of something than another group. Assertions such as these are outside the realm of truth or falsity. Rather, it is correct to think about whether a specific statistic – such as a mean, median, total, proportion, rate, ratio, etc. – for one population is greater or lesser than that statistic for another population. Assertions about differences in statistics between two groups are true or false.

Unquantified Magnitude

For any given variable, it is extremely unlikely that it would have the exact same distribution in two different populations. Thus, when comparing two groups, any variable is going to have a greater mean (or other statistic) for one group than for the other.

Because nearly every variable has at least a trivially different distribution between two groups, simply asserting that a mean (or other statistic) is greater for one group than the other is largely uninformative. Given enough precision, no two means will be equal.

Thus, generic comparisons are resistant to falsification. If someone asserts that group A makes more money on average than group B, a study which finds that, to the nearest hundred dollars, members of group A have a mean annual income of $46,300 and members of group B have a mean annual income of $46,300 does not conclusively demonstrate that this assertion is false. The person making the assertion could simply demand more precision with the expectation that the means, calculated to the dollar, would be different, such as with mean incomes of $46,382 and $46,311.

Even if the mean annual incomes of the two groups were found to be $46,382 and $46,382, there might be a difference if precision were increased to the cent. Indeed, since a mean is a ratio between a sum and a count, this could continue ad nauseum to an arbitrary level of precision.4

While generic comparisons are in this way resistant to falsification, they are not quite as resistant to falsification as generic generalizations. With a generic generalization, as long as at its predicate is true of least one member of its subject population, it resists falsification. With a generic comparison, assertions have a direction, i.e., the “more” or “less” component. If this were asserted to be in the wrong direction, then the generic comparison could be readily falsified.5

As long as the direction of the generic comparison is correct, however, the generic comparison is resistant to falsification. Like generic generalizations, generic comparisons that are resistant to falsification are also highly uninformative. For instance, asserting that group A makes more money on average than group B is uninformative. It could be that the mean annual income in group A is $32,950 versus a mean income in group B of $28,552. It could, however, be that the mean annual income of group A is $157,800 versus a mean income in group B of $28,552.

Because these kinds of generic comparisons are uninformative, they can be used to reach specious conclusions in flawed analysis, a practice all too commonly seen in social or behavioral research. The general scheme is to claim that the mean (or other statistic) of variable X is greater for group A than for group B because the mean (or other statistic) of variable Y is greater for group A than for group B. For instance, someone might claim that group A has a higher mean income than group B because the mean hours worked per week is higher for group A than for group B.

It might very well be the case that the differences in the number of hours worked per week between the two groups explains a substantial portion of the difference in average income between the two groups, but the same could be said for any other variable because, for any variable, it is extremely unlikely that the mean of the variable would be exactly equal between the two groups. Thus, by the exact same logic, anything can explain the difference in mean income between group A or group B.

Quantification of the difference in mean hours worked per week between the two groups and of the difference in mean annual income between the two groups might cast doubt on such an explanation. For instance, if it is discovered that the mean hours worked per week is 42.7 for group A and 41.3 for group B, but the mean annual income is $74,920 for group A and $39,500 for group B, such a small difference in mean hours worked per week seems unlikely to explain such a large difference in mean income.

Therefore, quantification might be withheld by deceptive analyses intended to make an explanation seem more plausible than it actually is. Alternatively, it might be withheld by flawed, lazy analyses that do not know any better or do not invest enough time or resources in the investigation.

Even with quantification, simply comparing differences in mean (or other statistic) between groups is itself insufficient for judging how much of the differences in one variable is explained by differences in another variable.6 However, quantification of means (or other statistics) is at least a step in the right direction and away from generic thinking.

How to Avoid the Fallacy

The answer to the fallacy of generic comparisons is similar to the answer to the fallacy of generic generalizations: all that is needed is quantification. However, quantifying a generic comparison is much more nuanced than quantifying a generic generalization.

When quantifying a generic generalization, typically all that is needed is a proportion, in particular, the proportion of the subject population for which the predicate of the generalization is true. For instance, specifying “72.4%” in “approximately 72.4% of swans are white” is relatively straightforward.

Various ways to provide summary differences between groups were given in the discussion of effect size. While an effect size is a helpful initial summary of a difference between groups, it is just that, an initial summary and not a complete description of the difference. How to quantify differences between groups is a deep enough topic to merit its own treatment, but is beyond the scope of this article.

Further Investigation

Generic thinking occupies a central position in this blog’s discussion of flaws in reasoning about evidence, and there are many opportunities for further investigation. Many of the fallacies discussed in their own articles either are specific kinds of generic thinking, or are supported by or derived from generic thinking. In order to present an alternative to generic comparisons, there is an opportunity for further investigation into the subtleties of comparing differences between groups.

There is much in the field of psychology pertaining to generic thinking. A review of the psychological literature around the study of generic generalizations is an opportunity for further investigation. Generic thinking is believed to be related to the psychology of perceptions of social groups as entities themselves, so a review of the psychology of entitativity is another opportunity. This, in turn, may be related to the phenomenon of psychological essentialism, which presents yet another opportunity.


Generic thinking is an extremely flawed, but all too common, way of viewing the world. Purging generic thinking from one’s mind may seem like an aspirational goal. It might be unreasonable to expect anyone to be entirely free of generic thinking, but it is also unreasonable to expect any society to be entirely devoid of house fires, and this does not negate the fact that the desired number of house fires is none.

In particular, care should be taken when generic thinking rears its ugly head in social criticism.

First, this is because social criticism is one of the sorts of thinking when people are – supposedly – at their most serious, and while generic thinking in daily banter may be of little consequence, social criticism deserves better for its seriousness.

Second, as touched on earlier, social criticism is inherently quantitative; what separates social criticism from human interest stories is that while human interest might discuss the affairs of one or a handful of people, social criticism directly discusses the state of affairs among millions if not billions of individual people.

Third, generic thinking in social criticism serves little purpose but prejudice, and guarding against generic thinking is thus a way to guard against bigotry.

Quantification is proposed here as a remedy both to generic generalizations and to generic comparisons. While quantification does away with literal generic thinking, it is only a mechanism that presents itself whenever generic thinking is challenged.

Underlying generic thinking is a blunt, inflexible worldview that puts vast numbers of individuals into categories and thinks it useful to hurl around broad, vague ideas about these categories and all the many individuals in them.

Opposed to this are worldviews that realize any population of vast numbers of individuals exhibits variation, and that to understand large populations is to understand this variation. Quantification is merely the tool that such worldviews use to describe variation. Thus, while quantification is the answer to the letter of generic thinking, to counter the spirit of generic thinking is to expect, seek, understand, and describe variation in the world.


Cimpian, Andrei, Amanda C. Brandone, and Susan A. Gelman. 2010. “Generic Statements Require Little Evidence for Acceptance but Have Powerful Implications.” Cognitive Science 34 (8): 1452–82.
Gelman, Susan A., Jon R. Star, and Jonathan Flukes. 2002. “Children’s Use of Generics in Inductive Inferences.” Journal of Cognition and Development 3 (2): 179–99.
Leslie, Sarah-Jane. 2017. “The Original Sin of Cognition: Fear, Prejudice, and Generalization.” The Journal of Philosophy, October.
Leslie, Sarah-Jane, and Adam Lerner. 2016. “Generic Generalizations.” In The Stanford Encyclopedia of Philosophy, edited by Edward N. Zalta, Winter 2016. Metaphysics Research Lab, Stanford University.


  1. “Dogs have four legs” and “ducks lay eggs” are examples used in “Generic Generalizations” in the Stanford Encyclopedia of Philosophy (Leslie and Lerner 2016) and the West Nile virus example is used in “The Original Sin of Cognition” (Leslie 2017).↩︎

  2. In statistical jargon, here “group” is taken to mean a domain and “individual” is taken to mean an observational unit.↩︎

  3. This is simply due to the fact that medians are more resistant to outliers than means, and 30% of group A consists of exceptionally tall people.↩︎

  4. This is a simplification for illustrative purposes. This is not how statistical analysis works in practice. In reality, such an encounter would turn on how much power a study has and whether a difference is statistically significant.↩︎

  5. Real-world falsification is, of course, still contingent on a study having enough statistical power to detect the difference, but in principle, the falsification can be done.↩︎

  6. To determine how much of the variation in one variable is explained by another variable, both variables should be measured in the same observational units, and techniques such as linear models or analysis of variance (ANOVA) should be used.↩︎