MANIPULAR VOCÊ É FÁCIL!

Summary

The video argues that statistics can mislead without anyone falsifying the numbers. By selecting, describing, or repeating true data in particular ways, a communicator can guide people toward conclusions the data do not actually establish. The narrator calls this social engineering through numbers: the vulnerability lies not necessarily in the data, but in predictable biases in human thinking.

Why people are susceptible

• The brain often uses quick mental shortcuts when making decisions under uncertainty. These shortcuts are efficient, but can produce systematic errors called biases.
• Anchoring: An initial number can influence later judgments, even when it is unrelated to the question. The video describes an experiment in which a randomly selected roulette number shifted people’s estimates of how many African countries belonged to the UN.
• Confirmation bias: People give more weight to evidence that supports what they already believe and discount evidence that challenges it. In a study about the death penalty, participants on opposing sides read the same studies, but each group found the evidence supporting its own position more convincing—and became more entrenched.
• Seeing evidence does not necessarily bring people with opposing beliefs closer together. It can instead reinforce their existing views.

Techniques for misleading with real data

1. Cutting or selectively showing information

A result can look very different depending on the dates or starting point chosen. An investment might be described as having risen sharply from a low point while having lost value over a longer period. Showing only successful examples also hides people who tried the same thing and failed.

Survivorship bias occurs when analysis focuses only on those who made it through a process. The video uses Abraham Wald’s wartime aircraft analysis as an example. Bullet holes on returning planes marked places where a plane could be hit and still survive. The lack of holes in the engines suggested that planes hit there often did not return. The narrator says this meant armor should protect areas with few holes on the surviving planes.

A chart’s scale can also distort a comparison. Bars representing 35 and 39 appear similar if the axis starts at zero, but dramatically different if it starts at 34.

2. Changing what a statistic is said to measure

A number measuring one thing can be described as if it measured something else—for example, changing “arrests” into “crimes committed.”

The narrator’s example that “black people are 13% of the US population and 51% of homicide arrests” illustrates how a figure about a specific crime and arrests can be generalized into a claim about all crime or all individuals. The video notes that arrests are not the same as offenses committed, and that policing and arrest patterns can differ across groups. It also presents victimization survey figures and argues that differences between groups exist while the justice system may widen them.

A hidden variable can explain an apparent relationship. The video cites a study of young people in Chicago in which differences in violence narrowed when researchers compared people living in similar circumstances. Neighborhood, family structure, and other social factors were offered as explanations.

The narrator cautions against using group statistics to assign guilt to individuals. A comparison with homicide suspects by sex is used to show how the same statistical framing can feel different when applied to a familiar group.

3. Leaving out the denominator or base rate

Relative increases can sound alarming without the underlying absolute risk. In the video’s pill example, the risk of thrombosis rose from roughly 1 in 7,000 users of older pills to 2 in 7,000 users of newer pills. That is a doubling, but the absolute risk remains small.

The narrator says reporting the relative increase without the absolute figures contributed to many women stopping the pill, followed by nearly 13,000 additional abortions in England and Wales the next year. The warning concerned a real risk, but the way the number was presented could distort how people understood it.

Large counts also need context: a number such as 1,000 means something different depending on the population it comes from. Averages can conceal the distribution, too. In the video’s example, one very wealthy person raises the average income of a room, even though the average describes none of the people in it.

4. Aggregating groups can reverse the apparent pattern

The video describes the 1973 UC Berkeley admissions figures. A higher overall acceptance rate for men than women initially seemed to show discrimination. When Peter Bickel examined admissions by department, women had similar or sometimes higher acceptance rates within departments. The narrator attributes the overall difference to application patterns: women applied more often to highly competitive departments, while men applied more often to departments with more available places.

This is called Simpson’s paradox: a pattern in combined data can disappear or reverse when the data is separated into relevant subgroups. The narrator stresses that this analysis does not prove discrimination never occurred; it shows that the overall figures alone did not establish the apparent conclusion.

A similar issue can arise when a third factor influences both a treatment and an outcome. The video’s hormone-replacement example says observational studies found fewer heart problems among women taking hormones, but a randomized study found higher risk in the hormone group. The narrator suggests that the women who chose hormone therapy may already have had healthier habits.

5. Framing identical facts differently

The same information can prompt different reactions depending on whether it is expressed positively or negatively. In a medical decision study described in the video, doctors were more likely to choose surgery when told that 90% of patients survived the first month than when told that 10% died during that period—even though the figures describe the same outcome.

This is the framing effect: presentation influences how information feels and how people decide.

6. Repeating a claim

The illusory truth effect is the tendency for a statement to seem more credible after repeated exposure, partly because it becomes familiar. Repetition can make a claim feel true even when it contradicts what someone previously knew. Social media can amplify this effect when the same claim circulates across many accounts or groups.

Further examples of statistical reasoning errors

• The Sally Clark case: Pediatrician Roy Meadow testified that the odds of two natural infant deaths in one family were about 1 in 73 million, multiplying the estimated probability of one death by itself. The video says this wrongly treated the deaths as independent, despite the children sharing family and environmental factors.
• The narrator also identifies a prosecutor’s fallacy: treating the probability of evidence under one explanation as though it directly answers which explanation is more likely. The relevant comparison should have been between competing explanations, including natural deaths and murder.
• The Royal Statistical Society later said the figure lacked a statistical basis, and Clark’s conviction was overturned in 2003. The narrator says she spent more than three years in prison and died in 2007.

Why knowing about bias is not enough

People are generally better at spotting bias in others than in themselves. This is called the bias blind spot.

Intelligence or familiarity with bias does not automatically remove it. Likewise, knowing the correct answer may not protect people from the effects of repeated claims.

The narrator acknowledges that the video itself uses framing and selective examples, including a provocative title and cases where errors were eventually uncovered. No one can list the unknown cases that were never discovered.

Practical method for evaluating a statistic

When encountering a claim, the video recommends asking:

• What exactly does this number measure? For example, arrests, convictions, reported offenses, or offenses committed?
• What information may have been omitted? Check the time period, comparison group, unsuccessful cases, and relevant subgroups.
• What is the denominator? Look for the absolute risk or rate, not only a percentage increase or a raw count.
• Is the total hiding a different pattern within subgroups? Consider whether a factor such as department, location, or social conditions changes the comparison.
• Could a third variable explain the relationship? Correlation alone does not establish causation.
• How is the same information framed? Try restating it in an equivalent positive or negative form.
• Have I heard this claim repeatedly? Familiarity is not proof.
• Do I want the claim to be true? If so, deliberately look for evidence that could prove it wrong.
• Can I inspect the original sources? The narrator says citations are provided so viewers can check the evidence themselves rather than relying on trust in the presenter.

Speakers and sources featured

Speaker

• Copas — the video’s narrator and presenter; no guest speakers appear in the supplied subtitles.

Named researchers, experts, and cases discussed

• Amos Tversky and Daniel Kahneman — heuristics, biases, and anchoring.
• Lord, Ross, and Lepper — the study of attitudes toward the death penalty.
• Abraham Wald — analysis of damage to returning aircraft and survivorship bias.
• Peter Bickel — analysis of the UC Berkeley admissions data.
• Emily Pronin and Stanford researchers — the bias blind spot.
• Roy Meadow and Sally Clark — the expert testimony and wrongful-conviction case discussed in the video.

Studies, institutions, and data sources cited or described

• Stanford researchers’ 1979 death-penalty study.
• Columbia University mathematicians working for the US military, including Wald.
• University of California, Berkeley admissions data from 1973.
• Studies of hormone replacement therapy, including an observational study and a randomized study; individual researchers are not named in the subtitles.
• A 1982 study presenting doctors with differently framed descriptions of lung-cancer treatment; researchers are not named in the subtitles.
• Research on the illusory truth effect, including a study on whether knowing a fact protects against repetition; researchers are not named in the subtitles.
• Harvard research following young people in Chicago; individual researchers are not named in the subtitles.
• FBI arrest statistics, US population data, US homicide figures, marijuana-arrest comparisons, and victimization surveys discussed in the video.
• The 2022 Brazilian census and Brazilian prison statistics.
• The Royal Statistical Society of the United Kingdom.