Summary: Fallacy theory is popular among skeptics, but it is in serious trouble. Every fallacy in the traditional taxonomy runs into a destructive dilemma which I call the Fallacy Fork: either it hardly ever occurs in real life, or it is not actually fallacious.
Why do people believe weird things? Why is there so much irrationality in the world? Here’s a standard answer from the sceptic’s playbook: fallacies. Fallacies are certain types of arguments that are common, attractive, persistent, and dead wrong. Because people keep committing fallacies, so the story goes, they end up believing all sorts of weird things.
In popular books about skepticism and in the pages of skeptical magazines such as this one, one commonly finds a concise treatment of the most common types of fallacies. The traditional classification is widely known, often by its Latin name: ad hominem, ad ignorantiam, ad populum, begging the question, post hoc ergo propter hoc. Some of them are more obscure, such as ignoratio elenchi, affirming the consequent, secundum quid, and ad verecundiam, better known as ‘argument from authority’. Most of them date back to the days of Aristotle, others are relative newbies, like the slippery slopefallacy, the genetic fallacy or – for obvious reasons – the reductio ad Hitlerum.
Such lists serve a pedagogical purpose. By learning the most common types of reasoning errors, you will avoid making them yourself, and become better at spotting them when others do. It’s a kind of inoculation against irrationality. If only people would learn the list of fallacies, the world would be a far more rational place!
Except this neat little story is wrong.
Skeptical about fallacies
I used to teach a course in critical thinking at Ghent University. As behooves a good skeptic, I first presented my students with the usual laundry list of fallacies, after which I invited them to put the theory into practice. Take a popular piece from the newspaper or watch a political debate, and try to spot the fallacies.
I no longer give that assignment. My students became paranoid! They began to see fallacies everywhere. Rather than dealing with the substance of an argument, they just carelessly threw around labels and cried “fallacy!” at every turn. But none of the alleged “fallacies” they spotted survived a close inspection.
Were my students to blame? I had to confess that, when I did the exercise myself, looking for clear-cut fallacies in real life, I came away mostly empty-handed. Perhaps because my students didn’t find any clear instances of fallacies, they started to make them up? So I turned to the classics. The Demon-Haunted World (1996) by Carl Sagan, perhaps the most celebrated work in the skeptical library, has a special section on reasoning fallacies, like many other books in the genre. But although Sagan duly lists all the usual suspects, he never puts them to work in the rest of the book. His treatment comes across as perfunctory, and he hardly gives any examples from real-life pseudoscience. Like many other skeptics, Sagan just invents some toy examples, which are easy to knock down but don’t actually correspond to real-life arguments. It seems that Sagan is paying lip service to fallacy theory, but has no use for it in his actual debunking work.
But if real life abounds with fallacies, why do skeptics like Sagan have to invent toy examples to make their point?
For a long time, philosophers and argumentation theorists have tried to define the different types of fallacies, mostly by using some (semi-)formal argumentation schemes. The attractive idea behind such an approach is that it would allow for swift and easy identification of common reasoning errors, in a wide range of contexts. But alas, the hopes of fallacy theorists have been frustrated. Definitions and schemas have become more complex and unwieldy over time, making them less fit for practical use. Yet, in spite of this, most authors continue to hold on to the notion that you can identify good and fallacious arguments based on their formal structure. The trick is just to find the right analysis.
I’ve now come to believe that this whole idea should be thrown overboard. Together with my colleagues Fabio Paglieri, an argumentation theorist, and Massimo Pigliucci, a dyed-in-the-wool skeptic, we recently published a paper in the journal Argumentation (2015) in which we explain what we think is wrong with fallacy theory. Not just with a particular definition of a given fallacy, but with all of them. Here’s the nub of the problem: arguments that are deemed ‘fallacious’ according to the standard approach are always closely related to arguments that, in many contexts, are perfectly reasonable. Formally, the good and bad ones are indistinguishable. No argumentation scheme can succeed in capturing the difference, separating the wheat from the chaff.
In our paper, we develop a destructive dilemma for fallacy theorists which we call the Fallacy Fork. In this dilemma, fallacy theorists are forced to choose between two options, neither of which is appealing. Take any fallacy from the list. Now we have two options:
(1) We characterize our fallacy by means of a deductive argumentation schema. For instance, in the case of post hoc ergo propter hoc, we use the following definition: “If B follows A, then A is the cause of B”. For a deductive argument to be valid, the conclusion needs to follow inexorably from the premises. In this case, that is clearly not true. By the standards of deductive logic, any argument instantiating that schema is fallacious. Now the good thing about this approach is that it has normative force. There’s no negotiating with deductive logic. The problem, however, is that we hardly ever find such clear-cut errors, presented in deductive form, in real life (see further). This is the first prong of the Fallacy Fork.
(2) We characterize our fallacy in a way that captures real-life arguments. In order to do so, we need to abandon our strict deductive approach. We need to relax our definitions and add some qualifiers and nuances. For post hoc ergo propter hoc, this might go as follows: “If B follows shortly after A, and we can think of a plausible causal mechanism linking A and B, then A is probably the cause of B.” This definition is a bit more cumbersome, but it is much closer to the kind of arguments people make in real life. By casting our net wider, we catch many more fish. But now we have another problem on our hands: is our argument still fallacious? In other words, is every instantiation of the scheme wrong?
Let’s see how the most famous fallacies fare when confronted with the Fallacy Fork.
Post hoc ergo propter hoc
Every skeptic is familiar with the saying: correlation does not imply causation. To think otherwise is to commit the post hoc ergo propter hoc (or cum hoc) fallacy. The website Spurious Correlations has collected some outrageous examples, with fancy graphs: there is a clear correlation between margarine consumption and divorce rates, and between the number of people who drowned by falling in a pool and the number of films featuring Nicholas Cage (per year). Is there a mysterious causal relationship between these events? If I was ill yesterday and feel better today, to which of the myriad possible earlier events should I attribute my improved condition? That I had cornflakes for breakfast? That I watched a movie with Nicholas Cage? That I was wearing my blue socks? That my next-door neighbor was wearing blue socks?
Not even the most superstitious person believes that correlation automatically implies causation, or that any succession of two events A and B implies that A caused B. There are just too many things going on in the world, and not enough causal connections to account for them. In its clear-cut deductive guise, the post hoc ergo propter hoc inference is a fallacy, to be sure, but hardly anyone makes it in real life. This is the first prong of the Fallacy Fork. So what about the kind of post hoc arguments that people do use in real life? (Pinto 1995) As it turns out, many of those are not obviously mistaken. It all depends on the context.
Imagine you eat some mushrooms you picked in the forest. Half an hour later you feel nauseated, so you put two and two together: “Ugh. That must have been the mushrooms”. Are you committing a fallacy? Not as long as your inference is merely inductive and probabilistic. Intuitively, your inference depends on the following reasonable assumptions: 1) some mushrooms are toxic 2) it’s easy for a lay person like you to mistake a poisonous mushroom for a healthy one 3) nausea is a typical symptom of food intoxication 4) you don’t usually feel nauseated. If you want, you can show the probabilistic relevance of all these premises. Take the last one, which is known as the base rate or prior probability. if I am a healthy person and don’t usually suffer from nausea, the mushroom is most probably the culprit. If, on the other hand, I suffer from a gastro-intestinal condition and I often have bouts of nausea, my post hoc inference will be less strong.
Indeed, almost all of our everyday causal knowledge is derived from such intuitive post hoc reasoning. For instance, my laptop is behaving strangely after I accidentally dropped it on the floor; some acquaintances un-friended me after I posted that offensive cartoon on Facebook; the fire alarm goes off after I light a cigar. As Randall Munroe (the creator of the web comic xkcd) once put it: “Correlation doesn’t imply causation, but it does waggle its eyebrows suggestively and gesture furtively while mouthing ‘look over there’.” Most of the time these premises remain unspoken, but that cannot be a problem per se. Practically every form of reasoning in everyday life, and even in science, contains plenty of hidden premises and skipped steps.
So how about the post hoc arguments that we hear from quack therapists and other pseudoscientists? Someone takes a dose of oscillococcinum (a homeopathic remedy) for his flu, and he feels better the next day. If he attributes this to the pill, is he committing a fallacy? Not obviously, or at least not on formal grounds. It all depends on the plausibility of a causal link, the availability of alternative explanations, the prior probability of the effect, etc. Dismissing any such inferences as post hoc ergo propter hoc fallacies is just a knee-jerk reaction. The real problem with homeopathy is that there is no possible physical mechanism, because of the extreme nature of the dilutions, and because randomized clinical trials have never demonstrated any effect whatsoever. But appealing to post hoc reasoning by itself is not fallacious. We do it all the time when we’re taking real medicine and conclude that it “works for us”.
Perhaps the most infamous among the fallacies is the argumentum ad hominem. The principle is quite simple. If you are assessing the merits of someone’s argument, you should not attack his or her personal background or motives. If you play the man instead of the ball, you are guilty of ad hominem reasoning. But are things so simple?
Let’s trot out the Fallacy Fork again. If your ad hominem argument take a deductive form, then of course it is invalid. Even a broken clock is right twice a day. Take this argument: “Researcher A is in the pocket of the pharmaceutical industry, therefore it follows that A’s study is flawed”. If the “therefore” is intended to be deductive, then clearly the argument is invalid. But how often do you encounter ad hominemarguments in this strong form?
So we move to the second prong of the Fallacy Fork. Take the following, weaker version of the same ad hominem argument: “Researcher A published a study on the efficacy of a certain antidepressant, but he’s in the pocket of the pharmaceutical company that manufactures the drug. Therefore, we should take his results with a large grain of salt. Better to have an independent team replicate the study.” Now this sounds a lot more reasonable. The second argument is non-deductive and ‘defeasible’, which means that it is inconclusive and up for revision. Almost all arguments in real life are like that. The fact remains that the argument has an ad hominem structure. But should we really dismiss it on those grounds?
In reality, we cannot do without ad hominem reasoning. This is because the fabric of human knowledge is deeply social. Virtually everything we know derives from what other people have told us. Only a fraction of the knowledge we possess is supported by the evidence of our own senses. The rest is, literally, hearsay. Life is too short to investigate everything by yourself. No wonder we are very sensitive to the reputation and trustworthiness of our sources (Sperber et al. 2010).
Many ad hominem arguments, in most contexts, are therefore perfectly reasonable. Much depends on factors that cannot be captured in the formal argumentation scheme: psychological assumptions about prejudice and bias, the past track record of our sources, the relevance of personal background for the issue at hand, background knowledge about hidden agendas. In the courts, ad hominem arguments are standard fare. Experts and witnesses can be discredited and censured because of a hidden agenda, bias, or conflict of interest. Naturally, it is logically possible that even a biased witness might be offering an honest testimony. But courts are not schools of logic.
In The Demon-Haunted World, Carl Sagan unwittingly illustrates the problem with fallacy theory. In his section about ad hominem reasoning, Sagan writes – perhaps bending over backwards to show his good will – that even skeptics are sometimes guilty of ad hominem reasoning, as in the following example: “The Reverend Dr. Smith is a known Biblical fundamentalist, so her objections to evolution need not be taken seriously” (Sagan 1996, 212). The little vignette – as usual in these discussions – was dreamt up by Sagan himself. It’s just a pedagogical straw man, easy to knock down.
But actually, unless Sagan’s argument is meant to be deductive (first prong), it is not fallacious at all (second prong). If we know that the good Reverend is an evangelical Christian, who dogmatically clings to a literal reading of Scripture, of course this will color our judgment about her arguments against evolutionary theory. I’d go even further: pragmatically speaking, this fact alone is reason enough to dismiss her arguments, and not to waste any further time on it. It’s simply naïve to think we have an obligation to scrutinize the arguments of every single crank. In an ideal world perhaps, with unlimited time on your hands, but not in this one. So ad hominem arguments are indispensable for navigating our way through a social world.
None of this is to deny that, logically speaking, even a die-hard creationist could conceivably level a good argument against evolutionary theory. If you think that the Reverend’s argument must be wrong, given her evangelical faith, you are making an error of deductive logic. But let’s be honest: if some Jehovah’s Witnesses hand you a pamphlet with “scientific” arguments against Darwin, are you going to give them your full attention, lest you succumb to ad hominem reasoning?
If we adopt deduction as the norm of rationality, the whole of science goes out of the window. Science is based on trust and reputation, because empirical evidence is deeply testimonial. Researchers have to report their affiliation, funding sources and possible conflict of interests, and fraudsters are harshly punished. We want to know who they are, and we want them to know that their reputation is at stake. Imagine if Science and Nature were to publish anonymous papers about certain revolutionary discoveries made in unnamed labs. Would anyone be inclined to take them seriously?
To be sure, some ad hominem arguments are uncalled for and distract from the issue at hand. But where to draw the line? Again, this depends on the specific context, which cannot be captured in an argumentation scheme. A possible rule of thumb is this one: “If it’s possible to play the ball, don’t play the man. If not, play the man.” But even this pragmatic rule will only get you so far. There’s no neat formula for distinguishing good ad hominem arguments from bad ones.
The main thesis of our paper is that each and every fallacy in the traditional list runs afoul of the Fallacy Fork. Either you construe the fallacy in a clear-cut and deductive fashion, which means that your definition has normative bite, but also that you hardly find any instances in real life; or you relax your formal definition, making it defeasible and adding contextual qualifications, but then your definition loses its teeth. Your “fallacy” is no longer a fallacy.
Let’s briefly consider some other examples. Arguments from ignorance (argumentum ad ignorantiam), according to the standard view, are fallacious because of the following well-known bit of wisdom: “absence of evidence does not constitute evidence of absence”. But as a matter of fact, it often does, and people turn out to be attuned to this (Hahn and Oaksford 2007). Here is a perfectly decent argument from ignorance, which is even used by skeptics: “Recovered memories about satanic cults sacrificing babies are probably the product of confabulation and suggestion, because we have never found any material traces of these atrocities.” This argument is acceptable, as we argue in the paper, because the hidden premises are probabilistically justified (in particular, concerning the ‘likelihood’ of finding such evidence on the assumption that these cults exist).
The so-called genetic fallacy is a relative latecomer to the party, having being coined only about a century ago. It draws (negative) conclusions about X by pointing to the origins of X. It’s a close neighbor to ad hominem reasoning. In reality, once again, many such arguments are not fallacious at all. For instance, if you can explain how out-of-body experiences originate neurologically, or even induce them in the lab, you make supernatural explanations less likely. Deductively, such an argument is fallacious, but probabilistically, it has strong probative weight. Someone who dismisses this as a “genetic fallacy”, as spiritualists and parapsychologists often do, is just not getting the point. In Joseph Heller’s Catch-22, the protagonist Yossarian at some point presents the following argument: “Just because you’re paranoid doesn’t mean they aren’t after you”. Deductively speaking, Yossarian has a point. But the comment is funny precisely because, pace deductive logic, it is absurd. If a psychiatrist tells you that your friend suffers from paranoid psychosis, you will not take seriously your friend’s claim that that he is being persecuted by the CIA (even though it is logically possible!).
The fallacy of Affirming the Consequent, according to the standard story, goes as follows: “If A then B. / B. / Hence: A” Once again, the schema is deductively invalid, but many arguments instantiating it have strong probative value, depending on the circumstances. Such arguments are simply what is known as inferences to the best explanation. For instance: “my car starts if someone turns the ignition key. / I hear my car starting. / Hence: someone must have turned the ignition key”. Logically invalid, but pragmatically justified, given some probabilistic assumptions.
One final example: an argumentum ad verecundiam states that proposition P is true because some authority X has claimed that X is true. As we saw earlier, much of our knowledge is testimonial, and therefore based on authority. There is nothing wrong with arguments from authority, provided that you bear in mind certain questions: does the claim fall within X’s domain of expertise? Are there other experts available, and if so, do they agree? What is the track record of X when it comes to similar claims? Ironically, the problem with pseudoscientists, and conspiracy theorists in particular, is not that they rely too heavily on arguments from authority, but too little. They are excessively suspicious of authority. Anything coming from established academic institutions, or from the “Mainstream Media”, is immediately discredited. Indeed, the diagnostic label “argument from authority” is a convenient excuse for them to reject any form of respected authority.
Pure fallacies are rare specimens. They are found in logic textbooks, but rarely in real life. If you think about it, this makes sense. The function of reasoning, according to the argumentative theory of reasoning of Hugo Mercier and Dan Sperber (2011), is to convince other people, and to be persuaded in return. But arguments that are too blatantly fallacious cannot perform this function. Deductive versions of ad hominem or post hoc ergo propter hoc are just too easy to debunk. Your audience sees right through them. Why would you bother making them?
The odds of finding such paper tigers in real life are low. If you believe you did spot one, chances are that you missed something, or that the case is less than clear-cut. Perhaps you’ve exaggerated the intended strength of the argument, by portraying an inductive and defeasible argument as an example of deduction, by glossing over some tacit probabilistic assumptions, or by stripping the argument to its bare form and then knocking it down. In other words, perhaps you’ve built a straw man (itself a supposed “fallacy” for which no formal definition can be given). For instance, if you’re being uncharitable, you could flag my mushroom argument as a fallacy, by misconstruing it as a piece of deduction. But I’m not saying that the mushrooms must be the culprits, by dint of deductive logic. I’m just saying that they are the most likely cause of my nausea, given the circumstances.
None of this is to suggest that people don’t use bad arguments. But lazy and sloppy arguments are much more common than cut-and-dried fallacies. People often take pot shots when they are arguing, offering the barest outline of a weak or inconclusive argument, without spelling out their crucial premises and without making clear the structure and intended force of their argument (Mercier & Sperber 2017). In a cooperative context, the audience needs to reconstruct the speaker’s argument and fill in the blank spaces, preferably in a charitable way. It is true that this principle of charity can also be abused, for instance in the phenomenon of innuendo and plausible deniability, where the speaker hints at a dubious argument but then hides behind a cloak of ambiguity. For instance, I may not explicitly set up an argument to discredit someone’s reputation, but I can insinuate it and hope that the audience gets my drift (“We all know that X is being paid by the pharmaceutical industry”). A fortiori, once you have to take into account these ambiguities and subtle means of deception, the formal definitions of fallacies become completely unhelpful.
Language can bewitch us. If a word exists, we tend to assume that there must be something in reality to which it refers. Labels are meant to be slapped onto things, right? The traditional taxonomy of fallacies, with its portentous Latin phrases, creates the impression that here we are dealing with established and indubitable theoretical results. Does an argument exhibit structure X, following steps Y and Z? Then we blow our whistle: “fallacy”! Do we hear someone drawing causal inferences from successions of events? Post hoc ergo propter hoc! Is someone relying on authority to make a point? Ad verecundiam! Playing the man rather than the ball? Ad hominem!
Arthur Schopenhauer, in his sarcastic little book The Art of Being Right (1896), expressed the wet dream of all argumentation theorists: “It would be a very good thing if every trick could receive some short and obviously appropriate name, so that when a man used this or that particular trick, he could be at once reproached for it.” That would be a splendid thing indeed! Alas, the real world is a bit more complicated. The diagnostic labels of fallacy theory are much less useful for combating irrationality than is often assumed. The standard list, handed down through the ages since the days of Aristotle, is a blunt instrument in real-life discussions. Virtually every definition of a fallacy runs into the Fallacy Fork: either it singles out invalid arguments that rarely occur in real life, or it does apply to real life but turns out to be toothless.
By carelessly throwing around labels and crying foul at every turn, defenders of science and reason may actually harm their own cause. People may start to harbor sympathy for the targets of such unfair dismissal – in other words, for woo. And indeed, the complacency of skeptics is understandable. Precisely because doctrines like homeopathy and astrology have been debunked so many times in so many ways, and because it is so unlikely that its advocates will ever rehabilitate their cause, it is tempting for skeptics to become lazy and smug. If you have the truth on your side anyway, why not?
But that would be a pity indeed. Even bad ideas don’t deserve bad criticism. It’s time for skeptics and other fallacy buffs to get rid of fallacies.
(To be published in Skeptical Inquirer. The PDF of the original academic paper in Argumentation can be found here. Thanks to Nick Brown for proof-reading!)
Boudry, Maarten, Fabio Paglieri, and Massimo Pigliucci. 2015. “The Fake, the Flimsy, and the Fallacious: Demarcating Arguments in Real Life.” Argumentation29 (4):431–456. doi: 10.1007/s10503-015-9359-1.
Hahn, Ulrike, and Mike Oaksford. 2007. “The rationality of informal argumentation: a Bayesian approach to reasoning fallacies.” Psychological review 114 (3):704-732.
Mercier, Hugo, and Dan Sperber. 2011. “Why do humans reason? Arguments for an argumentative theory.” Behavioral and Brain Sciences 34 (2):57-74.
———-. 2017. The Enigma of Reason. Harvard University Press.
———-. 2017. The Enigma of Reason. Harvard University Press.
Pinto, Robert C. 1995. “Post Hoc Ergo Propter Hoc.” In Fallacies: Classical and contemporary readings, edited by Hans V. Hansen and Robert C. Pinto, 302-311. Pennsylvania: The Pennsylvania State University Press.
Sagan, C. 1996. The demon-haunted world: Science as a candle in the dark. New York: Random House.
Schopenhauer, A. 1896. “The Art of Being Right (Die Kunst, Recht zu behalten).” In: Wikisource. https://en.wikisource.org/wiki/The_Art_of_Being_Right.
Sperber, Dan, F. Clément, C. Heintz, O. Mascaro, H. Mercier, G. Origgi, and D. Wilson. 2010. “Epistemic vigilance.” Mind & Language 25 (4):359–393. doi: 10.1111/j.1468-0017.2010.01394.x.