Surprise Disconnection: When the Link You Expect Isn't There

Inspired by: recurring Humboldt’s Home theme of unexpected breaks in systems

Introduction: When the Link You Expect Isn’t There

We live in an age of networks, models, and correlations. Our instinct is to see patterns everywhere. But just as often, reality confounds us: connections we assume to be strong turn out to be weak, or nonexistent. This is what I call a surprise disconnection—when expectation and evidence part ways. In science, policy, medicine, and daily life, these moments remind us that systems are rarely as smooth or linear as they appear.

Correlation and Its Discontents

The classic caution in science is simple: correlation does not equal causation. Yet humans are wired to conflate the two. If two events happen together, we lean toward thinking one caused the other. This cognitive shortcut once helped our ancestors survive—if eating a certain berry was followed by illness, best to assume the berry was poisonous. But in a world of complex, layered systems, the instinct can lead us astray.

Case Study: Tylenol and Autism

A prominent and controversial example is the ongoing debate about acetaminophen (Tylenol) and autism. Several observational studies have reported associations between prenatal or early childhood exposure to Tylenol and later autism diagnoses. But association is not proof. Confounding factors—such as the reason for taking the medication in the first place, genetic susceptibility, or environmental exposures—make the question complex. To date, researchers have not established a clear causal mechanism. Still, the correlation was enough to spark public concern and lawsuits. Here lies the disconnection: people assume the link is solid, while scientists remind us the evidence is incomplete. This gap between public perception and scientific rigor erodes trust. The Tylenol-autism case illustrates both sides of the coin—our need to investigate potential harms, and our equal need to avoid prematurely elevating correlation to causation.

Famous Misfires: When Correlations Fooled Us

History is full of examples where correlations looked persuasive but later crumbled under scrutiny: - Coffee and Heart Disease: Early studies suggested coffee drinkers had higher rates of heart disease. Later research showed the apparent risk was explained by smoking, which was more common among coffee drinkers at the time. - Vaccines and Autism: A now-retracted study falsely linked the MMR vaccine with autism. The claim relied on anecdotal correlations, not robust data. - Hormone Replacement Therapy: Observational studies suggested hormone therapy in postmenopausal women reduced heart disease risk. Randomized controlled trials later revealed the opposite—an increase in cardiovascular risk. Each case shows the same lesson: correlation can be an important clue, but not the final word.

Everyday “Spurious” Correlations

Some correlations are so obviously coincidental that they make us laugh—but they’re powerful teaching tools: - Ice cream sales and drowning deaths both rise in summer. The confounder? Hot weather. - Divorce rates in Maine once closely tracked with margarine consumption. Neither causes the other. - The number of films starring Nicolas Cage once correlated with swimming pool drownings. Pure chance. - Cheese consumption per capita has matched rates of death by bedsheet entanglement. Nonsense, but statistically neat. - Shoe size and reading ability correlate in children—not because big feet make better readers, but because older children both read better and have bigger feet. These examples stick in the mind because they reveal how seductive numbers can be. A chart with matching lines can feel more convincing than a paragraph of logic. The lesson is that plausibility and mechanism matter as much as numerical correlation.

Hidden Connections: When We Underconnect

The opposite problem also matters: missing a real but subtle link. For decades, the tobacco industry insisted smoking and lung cancer were merely correlated. They funded research to highlight any alternative explanation. It took overwhelming epidemiological and mechanistic evidence to prove causation. Another example: leaded gasoline. Crime rates in the U.S. fell dramatically in the 1990s, and researchers later showed a strong correlation with the phase-out of leaded gasoline two decades earlier. Lead exposure impaired brain development and impulse control in children, fueling crime rates years later. Here, the correlation pointed to a causal mechanism only uncovered after careful study. Climate change offers another case. Weather feels personal and local—was today hotter or colder than last year? Climate, however, is a statistical pattern of decades. Missing this connection fuels the common refrain: “If it’s snowing, how can the planet be warming?” Surprise disconnection arises when people fail to see how patterns at scale differ from personal experience.

When Corridors Fail, When Metrics Mislead

Surprise disconnections occur not only in data but in design. In ecology, wildlife corridors are meant to reconnect fragmented habitats. But if they are plotted by human maps rather than animal behavior, they may look like connections on paper while functioning as dead ends in practice. In hospitals, infection-prevention campaigns sometimes boost compliance with hand sanitizer use—at least as recorded on clipboards—but infection rates stay flat. The connection between metric and outcome was weaker than assumed. The system “looked connected” but wasn’t.

The Double Risk: Overconnect or Underconnect

Surprise disconnections take two main forms: - Overconnection: Believing two things are causally linked when they aren’t (as in the Tylenol-autism debate, or in conspiracy theories that feed on pattern-seeking). - Underconnection: Missing a genuine but non-obvious linkage, such as how climate change intensifies wildfire risk, or how poverty contributes to chronic disease through stress pathways. Both errors distort our decisions. One leads to wasted effort or misplaced fear; the other blinds us to real vulnerabilities.

Systems Thinking as Antidote

How do we guard against surprise disconnection? The answer lies in systems thinking. It means stepping back, looking for multiple layers of evidence, and resisting the seduction of single-variable stories. Scientists do this by designing experiments, testing mechanisms, and looking for reproducibility. Citizens can do it by holding both skepticism and openness at once: asking not just “Is there a link?” but also “What kind of link, under what conditions, and with what evidence?”

Sidebar: Quick Guide to Correlation vs. Causation

- Correlation: Two variables change together (e.g., ice cream sales and drowning deaths both rise in summer). - Causation: One variable directly influences the other (e.g., smoking causes lung cancer). - Confounder: A hidden third variable drives both (e.g., hot weather drives both ice cream sales and swimming, explaining the drowning correlation). - Lesson: Look for mechanism, replication, and statistical controls before inferring cause.

Classroom Prompts

1. Why do humans so often mistake correlation for causation? Share examples from your own life. 2. How would you design a study to test whether acetaminophen truly affects neurodevelopment? 3. What’s the funniest spurious correlation you can find or imagine? 4. Can you think of a time when a missing connection in a system caused a breakdown (e.g., supply chains, school planning, technology)? 5. What are the dangers of overconnection versus underconnection in public policy? 6. How might systems thinking help us spot both false bridges and hidden links more effectively?

Sources

Avella, J., et al. (2021). Paracetamol use during pregnancy and child neurodevelopment: A systematic review. European Journal of Pediatrics.

Bauer, A. Z., et al. (2021). Paracetamol use during pregnancy — a call for precautionary action. Nature Reviews Endocrinology.

Hernán, M. A., & Robins, J. M. (2020). Causal Inference: What If. Chapman & Hall/CRC.

Ioannidis, J. P. A. (2005). Why most published research findings are false. PLoS Medicine.

Keim, B. (2012). Lead exposure linked to crime rates. Wired Science.

Lenz, W. (1962). Thalidomide and congenital abnormalities. Lancet.

Rosenhan, D. (1973). On being sane in insane places. Science.

Rosenfeld, S. (2004). Correlation Is Not Causation. Science News.

Spurious Correlations project, Tyler Vigen: https://www.tylervigen.com/spurious-correlations

Surgeon General’s Advisory Committee (1964). Smoking and Health: Report of the Advisory Committee of the Surgeon General of the Public Health Service.

U.S. Environmental Protection Agency. (2016). Benefits of the phaseout of leaded gasoline on crime and health.