I saw a meme once of a stern-looking cop leaning into a car window. The caption: “Looks like you’ve had a bit too much to think.”
In Part 1 we covered fallacies that dodge the point. In Part 2 we covered fallacies that rig the question before you can answer it. Bad Evidence fallacies are sneakier than both, because they actually look like they’re doing the work. They come with numbers. Studies. Historical precedents. Personal testimony. They feel like proof. They just aren’t.
These are arguments that mishandle or fabricate evidence. They don’t dodge the question or rig the frame. They answer the question with data that doesn’t actually support the conclusion. And they’re harder to catch, because you have to check the evidence instead of just the structure.
Cherry Picking — selecting only the evidence that supports your conclusion while ignoring the rest.
Your brother-in-law swears his all-bacon diet is healthy because he lost ten pounds. He doesn’t mention his cholesterol is through the roof, his doctor told him to stop, and the ten pounds came back in three weeks.
The Center for American Progress has been building a case around the “3.5% rule,” a concept from Harvard professor Erica Chenoweth’s research. The idea: when 3.5% of a country’s population mobilizes peacefully, the government falls. CAP keeps pointing to the No Kings protests (8 million people would be about 2.4% of the U.S. population) and treating the gap as a countdown to political transformation. What they leave out: Chenoweth’s research studied campaigns to overthrow authoritarian regimes (the Philippines in 1986, Serbia in 2000, Madagascar in 2002). Political scientist Kyle Matthews specifically noted the rule “does not apply to change in liberal democratic states.” Chenoweth herself calls it a “rule of thumb,” not a law, and has acknowledged that resistance effectiveness has been declining worldwide since 2010. The 3.5% number is real. The conclusion CAP draws from it is not supported by the research they’re citing. They just don’t show you the parts that say so.
Conservatives cherry-pick too. A Fox News investigation into No Kings funding found that some organizing groups had received money from foundations connected to George Soros. They ran with it. Snopes later found no evidence of direct financial connections between Soros and the protests themselves, only that Open Society Foundations had given grants to Indivisible over the years. Picking the one financial thread that sounds sinister while ignoring the documented picture of a broad coalition isn’t investigative journalism. It’s curation.
Is it Cherry Picking? You cite five studies and ignore three that contradict you. Yes. You cite the most relevant study and it happens to support you. No — the test is whether you looked for contradicting evidence and hid it.
Hasty Generalization — drawing broad conclusions from too little data or unreliable data.
Your friend eats at a restaurant once, gets food poisoning, and tells everyone the place is a health hazard. One visit. One bad night. Maybe that cook got fired the next morning.
No Kings organizers claimed 8 million people attended the March 28 protests. Most news outlets reported this number. Buried in the coverage was a single sentence: “Independent verification of the figures was not immediately available.” At the St. Paul flagship event, the state patrol initially estimated 50,000. The Department of Public Safety later revised it to 100,000. Organizers claimed 200,000. In Rhode Island, police estimated 20,000; organizers had hoped for 50,000. In Missoula, Montana Free Press estimated 3,500; one Facebook commenter claimed 15,000; others looked at the aerial footage and said it wasn’t even 1,000. For the October 2025 round, organizers estimated 7 million, while an independent data journalism effort put it at 5 to 6.5 million. From numbers like these (where organizer estimates routinely double or triple independent counts) comes the sweeping headline: “America has spoken.”
Trump does the same thing. His “Million MAGA March” in 2020 drew somewhere between 11,000 and 15,000 people. He called it a million. His inauguration crowd claims were debunked on camera. Both sides inflate because crowd size feels like legitimacy. It isn’t. It’s just crowd size, assuming anybody counted honestly. Which apparently nobody does.
Is it a Hasty Generalization? “I talked to twelve people at the rally and they all agreed, so the country agrees.” Yes. “A poll of 1,500 randomly selected adults shows 58% disapproval.” No — that’s a properly constructed sample.
False Cause — assuming that because B followed A, A caused B. (The Latin name is post hoc ergo propter hoc, but you don’t need Latin to spot it.)
Every time you wash your car, it rains. You have not discovered weather control. You’ve discovered that you only remember the times it rained.
The Center for American Progress claimed in its March 2026 analysis that “people-powered mobilizations in Minnesota successfully forced the Trump administration to reduce the number of ICE agents in the state.” Protests happened. ICE drew down. Therefore protests caused the drawdown. But ICE operations shift for dozens of reasons: legal challenges from the state attorney general, resource reallocation to other states, court injunctions, operational priorities, budget constraints. Minnesota’s AG filed lawsuits. Other states were demanding agents. Federal money was being redistributed. Attributing the drawdown to one cause (the one that makes your movement look effective) is exactly how false cause reasoning works. The protests might have contributed. Calling it a clean cause-and-effect victory is laundering correlation into proof.
Is it a False Cause? “Crime dropped after we installed cameras, so the cameras reduced crime.” Probably — you haven’t controlled for anything else that changed. “A controlled trial showed patients on the drug recovered 40% faster than the placebo group.” No — that’s controlled evidence with a baseline comparison.
Anecdotal Evidence — substituting personal stories for systematic data.
Your uncle smoked two packs a day and lived to 94, so cigarettes can’t be that bad. His experience is real. It’s also one person.
Media coverage of the No Kings protests runs almost entirely on anecdotes. NPR’s reporter at St. Paul noted that “anecdotally” people seemed less afraid to give their full names this time than during the January ICE surge. An American Federation of Teachers official said the protest gave him “a renewed sense of faith in the country.” A 21-year-old in San Diego said he’d felt “discouraged and depressed” but attending a march changed everything. These are real experiences. They’re also not evidence that the movement is achieving policy change, that democracy is under genuine authoritarian threat, or that eight million people share the same motivations. They’re evidence that the people telling the stories feel a certain way about what they experienced. Moving and misleading are not mutually exclusive.
Conservatives do this in the other direction. Interviewing one confused protester who can’t explain why she’s there and treating that clip as representative of eight million people is the same fallacy with different politics.
Is it Anecdotal Evidence? “My neighbor got fired after speaking out, so dissent is being crushed.” Yes — one case isn’t a pattern. “Department of Labor data shows a 15% increase in retaliatory termination complaints this quarter.” No — that’s systematic evidence.
Appeal to Ignorance — claiming something is true because it hasn’t been proven false, or false because it hasn’t been proven true.
You can’t prove there isn’t a teapot orbiting Jupiter. That doesn’t mean there is one.
“Independent verification of the figures was not immediately available.” That sentence appeared in coverage from multiple outlets after the March 28 protests. The 8 million claim was reported. The caveat was noted. Then everyone proceeded as if the number were established fact. The absence of a disproof became, functionally, proof. A week later, independent verification still hadn’t appeared. The number was still being cited.
The broader No Kings argument has a version of this: “You can’t prove Trump isn’t becoming a dictator.” He won an election, serves a constitutionally limited term, has had executive orders struck down by courts he then obeyed, and eight million people protested him on national television without a single arrest in New York City. That’s a lot of evidence pointing one direction. The case for “dictator” rests on the fact that you can’t conclusively prove he won’t become one someday. That’s not an argument. That’s a horoscope.
From the right: “You can’t prove the protests aren’t being run by foreign operatives.” The documented picture shows a coalition of domestic progressive organizations with traceable funding. The absence of a comprehensive audit doesn’t prove a conspiracy any more than the absence of a disproof proves a dictatorship. Not knowing everything is not the same as knowing something.
Is it an Appeal to Ignorance? “No one has proven this supplement doesn’t work, so it must work.” Yes. “This drug completed three clinical trials and showed no measurable benefit.” No — that’s evidence of absence, based on actual testing.
Every fallacy in this article does the same thing: it takes something that isn’t evidence and seats it in evidence’s chair. Cherry-picked data, inflated numbers, post-hoc causation, feel-good anecdotes, the absence of disproof. They all look legitimate until you check.
And here’s how this connects to the rest of the series. Bad evidence is the raw material. You can’t rig a question (Part 2) without cooked numbers to justify the framing. You can’t change the subject (Part 1) if the original evidence is solid enough to hold the room. Cook the books first, and everything downstream looks real.
The cop from the meme doesn’t need you to stop thinking entirely. He just needs you to stop checking.