The set of logical fallacies caused by meandering from one inference to another in the same essay

Below is an excerpt from Chapter 2 of my book, “Basic Statistics for Causal Inference: Visual and Historical Illustrations from the Social, Health, and Life Sciences,” which is currently a work in progress.

Chapter 2 makes the argument that a set of logical fallacies are caused by arguments that meander from one inference to another. This is because different kinds of inferences require different kinds of arguments. Chapter 2 provides an overview of the kinds of inferences we tend to make in higher education. The goal is to teach people to recognize and distinguish these arguments from one another to avoid falling prey to these logical fallacies. My argument is that these fallacies are extraordinarily common; rigorous inferences are rare. This results in most of what we read being bull$shit. To test this, I went to the first editorial I saw in the New York Times and I describe the logical fallacies that follow from this flawed meander.

Just to give an example, today, on March 16^th, 2026, I visited the New York Times Opinion page, knowing that the first Op-Ed (Opinion-Editorial) I would click on would be a perfect example of this type of twist-and-turn meander from one logical goal to another. The first political editorial was written by Matthew Yglesias. Yglesias is well-known for branding himself as a moderate. He tends to direct his energy to criticizing Democrats for taking stances that are unpopular among Republicans, such as defending trans youth, the rights of asylum seekers, and identity politics (i.e., Diversity, Equity, and Inclusion, “DEI”).[1]

In this essay,[2] he suggested that in some states, the Democratic Party has a “toxic” brand. Here, I won’t take issue with the label; instead, my argument is that Yglesias is not making a careful inference related to what constitutes the label “toxic.” A serious essay with this as its logical goal will define the meaning of the word toxic with great care. It would answer the question: how would we know that something is ‘toxic?”

Presumably, toxic has a definition and would include a way to distinguish something that is toxic from something less distasteful – or differently distasteful. See, for example, CJ de Silva’s explanation of the difference between a douche, an asshole, and a jerk, each of which constitutes attributes that are differently objectionable.[3] She defines each clearly and even uses a Venn diagram to define characteristics of people who are douches and jerks, but not assholes (and every other combination, including what it means to be all three). What makes this a good interpretive analysis is her care in identifying examples in each category so that a reader can agree or disagree with the relevant examples or labels.

A terrific example of this kind of work, often called “interpretivism,” is Isabel Wilkerson’s book Caste (2020), in which she argues that the problem in American politics is not racism; rather, the American political system is a “caste” system. She identifies fourteen pillars of caste and reveals how the U.S. racial caste system shares characteristics of the anti-Jewish caste system in Nazi Germany and the Indian caste system, neither of which has anything to do with race. Again, my praise is not based on whether I believe or agree with the argument; it is for the careful use of evidence in this book. The evidence she cites in this book could support causal or moral inference, but she stays focused on her interpretivist goals.

WebMD’s definition is: “A toxic person is someone who brings conflict and negativity to your life. They’re often controlling, manipulative, and even abusive.” Reading between the lines, it seems that Yglesias does not define toxic in this way. It seems that Yglesias defines “toxic” as “poisonous,” which is reasonable. Presumably, voters avoid toxic candidates, like, well, the plague. What this means is that Yglesias is warning Democratic primary voters and Democratic candidates to avoid taking issue positions that would make a candidate seem “toxic” to moderate voters, thereby helping the Republican win.

To understand which issues would cause voters to believe that a candidate is “toxic,” a researcher could undertake an opinion poll that, rather than asking them their view on a topic, could present respondents with a set of previously identified unpopular policy positions and ask whether they would refuse to vote for a candidate who supported that policy (but would otherwise be agreeable). This kind of study might count as reliable evidence for how we would know an issue makes a voter abandon a candidate as toxic.

Nonetheless, if you read this essay, what you learn is that the word “toxic” is defined by any policy position that (supposedly) caused a Democratic candidate to “lose their statewide election.”

Despite his stated goal of describing a toxic brand, he provides no evidence for what defines toxicity; instead, he directs his energies into making a causal inference. Explaining why Democrats lose statewide elections (like the Senate or the governor) is a general predictive causal theory of election outcomes. In other words, he meanders from an interpretivist essay focused on defining what constitutes toxicity to a causal argument. This is a problem because a causal argument uses different evidence than an interpretation of what issues are actually considered “toxic” in voters’ minds.

There are some problems with the evidence Yglesias uses for his causal inference (which also apply to the New York Times Editorial Board Opinion he cites). First, he states a causal inference as certain, which is a logical fallacy in and of itself because a causal inference is virtually never a verifiable fact. He says: “We know that, in general, more moderate candidates tend to do better.” The problem is that, in a rigorous test of this hypothesis, Adam Bonica, Kasey Rhee, and Nicolas Studen find that “gains associated with ideological moderation are relatively modest” and are “secondary” to turnout effects.[4] Turnout effects may be amplified by progressive stances, which is a problem for Iglesias’s argument.

These editorials rarely engage with the entirety of the political science that looks at the effects of campaigns on election outcomes, which finds that very little of what is said during a campaign makes any difference whatsoever; even the scholars who make the case that specific campaign strategies change electoral outcomes argue that the effects of a campaign strategy may (sometimes) be “more than minimal” (Lewis-Beck and Tien 2018):303. Moreover, in our age of polarization, very few voters are crossover voters (Republicans who vote for Democrats, or vice versa). So, in other words, most of the evidence in political science scholarship undermines his causal reasoning. The problem is that if word gets out about this, it could undermine the job security of campaign strategists and pundits (like Yglesias).

The other gaping logical fallacy in this OpEd (and the essay Yglesias cites) is that they both provide a few anecdotes about what candidates who lost their elections said during their campaigns. Anecdotes virtually never serve as reasonable evidence for causation. Worse, these anecdotes of progressive statements are presented as THE definitive cause of their having lost the election. Causal certainty based on anecdotal evidence is an ugly combination.

Another fallacy that this kind of analysis can contain is called the “Texas sharpshooter fallacy.” When pundits wait until after an electoral loss and then dig up evidence to propose reasons the candidate lost, they are falling prey to this fallacy. Likewise, the sharpshooter shoots the arrow FIRST and THEN draws the target around his arrow, proclaiming victory, somewhat akin to the certainty in this essay.

After he casually claims victory vis-à-vis his causal conclusion, Yglesias’s essay meanders into a moral or normative critique of statements made by progressive Democrats. A moral critique of progressivism is a totally acceptable type of essay. A serious essay like that would begin with a first principle assumption explaining why a particular (not a laundry list) policy preference is morally or ethically a problem.

One important aspect of causal reasoning is that researchers are warned to leave their normative opinions at the door when making a causal inference, to avoid what we call “motivated reasoning.” Motivated reasoning is akin to the sharpshooter fallacy because it leads people to avoid considering counterarguments and counterevidence; it helps lull people into moral (and causal) certainty.

But since Yglesias’s audience is targeted at Democratic voters and candidates, he makes a reasonable assumption that everyone agrees that electing Democrats is a “good.” This makes his causal reasoning a little more logically connected to his normative critique of taking up progressive political stances. The logic is reasonable: if we want Democratic candidates to be elected, they should do things that are likely to help them succeed.

The problem is that things that cause good things can be nevertheless bad, and vice versa. This is part of why the casual meander from causal reasoning to normative inference is problematic. In this case, Yglesias complains that Democratic candidates pledge their support for the rights of transgender people, asylum seekers, and policies that help make people of color more represented in higher education and the workplace. Republican voters may not support these policies, but Democratic voters overwhelmingly do not find these issues “toxic,” as most Democratic voters are in favor of (most of) these policies.[5]

Thus, he cannot automatically assume that Democratic voters necessarily want their candidates to abandon these issues on the platform. This is where the casual meander from defining toxicity to causal inference to normative critique turns into an enormous gaping set of logical fallacies, rendering his essay into something that is entirely useless. It is not persuasive that these issues are toxic. It is not persuasive of the causal inference that moderation causes electoral outcomes. And it is not persuasive that abandoning vulnerable people in society is morally good.

Here, I am not arguing that Yglesias is wrong. It may be that moderation helps Democratic candidates win elections, but this does not mean that moderation is morally good. It may be that moderate policies are good, and if they are, it does not matter whether they cause election outcomes. It may be that Republican voters find progressive stances so morally objectionable that they consider them “toxic,” but this does not make them actually “toxic” in an intersubjective analysis of the meaning of the word toxic. My argument is that he is not using evidence well. Because he is not focused on a particular inference, he is not careful about any of them.

What a careful causal analysis should consider

Causal inference requires careful thinking about counterfactuals. It requires paying attention to the bulk of the evidence, not just a single study (and especially not a basket of anecdotes).

It requires careful “intersubjective” measures, meaning that an analyst should avoid confusing their own definition of toxicity with an objective one. I suspect Yglesias is not paying attention to this, since the so-called toxic issues are the pet issues he constantly complains about. He seems to lack curiosity about measurement entirely.

A causal inference requires a consideration of reverse causality, confounding (i.e., in this case, a careful examination of any possible common causes of an ideological stance and election outcomes), and careful attention to the underlying mechanisms (whether measurable or not). A good causal inference requires investigation into the conditions (e.g., temporal or spatial attributes associated with a particular election) that may influence whether factors are more or less likely to cause outcomes.

One positive aspect of his essay is that Yglesias cites a study by reputable scholars, David Broockman and Joshua Kalla, that does something like this. I won’t provide an entire overview of the study, but in their abstract, they state their conclusion that [c]andidates moving towards the other party win some voters but lose others who preferred their party’s  position—producing small aggregate effects.” Yglesias summarizes their findings and argues that progressive Democrats support unpopular policies, which will certainly lead them to lose elections, which poorly represents the study’s emphasis on small effects.

Yglesias’ response may be that he is only talking about unpopular policy positions and, on aggregate, the effects might therefore be larger. The problem is that he takes this experimental evidence as overwhelmingly persuasive, despite the fact that it may lack what researchers call “external validity,” meaning it is not entirely conclusive about how causality actually works in the real world, outside an experimental setting. This is why valid inferences must be attuned to the entirety of the evidence; relying on a single study with a single methodological approach, particularly as though the evidence is certain, is problematic.

This essay does not mention the possibility that running for office in a state with about the same number of Democrats as Republicans (considered “battleground” states) makes candidates more moderate, so, in essence, Yglesias might be falling into a reverse-causality trap. He should consider the possibility that something might be causing the candidates to take the stances they take AND to win (or lose) elections. A careful causal inference requires examining all the causes of voter attitudes and choices to identify the underlying mechanism that leads a voter to change their vote, a topic that has not garnered strong consensus in the political science literature.

Candidate moderation may affect funding for these candidates (which may have affected outcomes during the primaries and contributed to whether a state has a more moderate versus progressive candidate), a mechanism distinct from voters finding their positions “toxic.” There are also temporal effects, including but not limited to economic pressures, which are known to affect electoral outcomes.

From an effect size perspective, what tends to drive public opinion and therefore electoral outcomes is a causal model I talk about in Chapter 4, called “The Public as Thermostat,” (Wlezien 1995), in which the public becomes more conservative when policy is perceived as liberal and vice versa. Indeed, the Public as Thermostat theory has been corroborated dozens of times, yet most consultants and candidates were educated before the first publication, so they do not use this model to predict when being more or less progressive would help their candidacy. This means that when the president is a Democrat, progressives may be more likely to be rejected by the public than in years when a Republican is in office. Since Yglesias is promoting moderation while Trump is president, this public-as-thermostat effect predicts that moderation will be less successful in 2026 than in 2024.

In general, what causes election outcomes in some years may not work in others. There are time-related factors that can alter causal structures during particular election years, making campaign statements important in shaping outcomes. And of course, candidates also have opponents who are also doing and saying things, which are also contributing to their electoral success (though the effects of either are at most likely to be barely “more than minimal.”) Indeed, the last time a Democrat won a statewide election in Texas, the Republican candidate said something stupid about rape being like bad weather and suggested that victims should just sit back and enjoy it, which may have affected that close race (Tolleson-Rinehart and Stanley 1994): 69. Many argue that this pro-rape stance was so toxic that it caused some Republicans to vote for the Democratic candidate for governor. Is this true? Maybe. It is very difficult to nail down causality in a particular instance.

A serious essay would consider these effects, but Yglesias is not interested, really, in causality; he is just trying to convince Democratic candidates to moderate, likely because it is his policy preference. Instead, this essay’s true purpose is to list the stances of candidates as “things I disagree with.” The argument is: You know why Democrats lose? Because they take stances I don’t like. “Hey, Democratic candidates! Do and say the things that I like, and this will get you elected!” I don’t think this is necessarily why Iglesias (and the NYT editorial board’s essay that Yglesias cites) ignores these considerations and meander from one kind of inference to another. My entire point here is that most editorials like this use evidence poorly because their stated inferential purpose is (perhaps) a sham.

What makes this essay more of a sham is that Yglesias knows that Republican voters (and Independents) are also reading his essay. Spending a ton of energy undermining Democrats’ moderate brand by complaining about Democrats who take up progressive policy positions is likely counterproductive because this essay is foisting this “toxic” brand on the party label. This renders this essay no more than a basket of anecdotes that represents a tactic that psychologists call “FUD,” a manipulation technique that inspires fear, uncertainty, and doubt, which also tends to undermine the normative goal he is seeking, which is to apply a moderate brand to Democrats and get moderate Democrats elected.

So, toxic is in the headline because of its likely effectiveness as a fear tactic: “Hey voters! Don’t support progressive candidates in the primary, or you will lose to the Republicans! Hey candidates! Beware of being associated with supporting progressive (i.e., “toxic”) issues, or you will lose your election!” In this way, Yglesias is being controlling and manipulative, creating conflict among Democrats, which is, according to WebMD, ironically, “toxic” behavior. As always, he sidesteps the question entirely of whether moderation is actually morally defensible. Its only good is to cause election outcomes, and if this causality is brought into question, then moderation has no moral value without a moral defense.

I am picking on Yglesias, which is a little unfair, because my critique applies to virtually every political journalist, regardless of ideological predisposition. Very few of them make principled moral arguments. Very few of them take causality seriously. Most meander from blame to causality to meaning-making (e.g., defining words like “toxicity”), with no sensitivity to the fact that, logically, these different kinds of arguments require different kinds of evidence. The more important point I am making is that I could seek out the New York Times’ Opinion page today (which happens to be March 16^th, 2026), and the first opinion piece I click on does exactly what this book teaches you not to do.

How the bulk of unserious, careless nonsense affects relying on AI

Importantly, this is why no one should use AI to help them learn to write. First, AI is slop. It uses lofty words like “complex interplay” that communicate nothing of value. But they also train on essays like these editorials, which use evidence poorly. If students (and those in their later professions) want to be recognized for excellence, they must learn to be creative, say something that has never been said, and then use evidence well, in a way that argues for a particular inference that readers may find surprising yet useful. Arguments are not facts; they must be capable of being wrong, a concept called “falsifiability.”

If you want to say something new, something nonobvious, relying on AI models that have trained on everything that has already been said is not a good strategy. Arguments should synthesize persuasive evidence in accordance with predetermined logical rules; each kind of argument uses its own rules, which is why they must be kept separate. AI trains on essays that often meander from one logical inference to another, rendering these essays useless slop.

[1] https://www.slowboring.com/p/the-two-kinds-of-progressives

[2] “The Democratic Brand Is Toxic in Too Many States” https://www.nytimes.com/2026/03/16/opinion/democrats-senate-moderate.html

[3] https://medium.com/@cjdesilva/3-kinds-of-guy-youll-meet-in-your-life-b8537c29a22a

[4] https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5172049

[5] https://www.pewresearch.org/short-reads/2025/02/26/americans-have-grown-more-supportive-of-restrictions-for-trans-people-in-recent-years. Democrats and Independents are also overwhelmingly supportive of the rights of asylum seekers: see https://www.refugeesinternational.org/statements-and-news/majority-of-u-s-likely-voters-support-access-to-asylum-at-the-u-s-southern-border/