In 2016, Paul Smaldino and Richard McElreath published a striking and uncomfortable paper: “The Natural Selection of Bad Science.” It argues that science is not just failing in isolated pockets — it is evolving in a direction that systematically favors poor practices.
This is not because scientists are bad people. It’s because scientists are people inside an environment that rewards speed, flashiness, and positive results, regardless of whether those results are true.
This first post in our deep-dive series introduces the idea that bad science evolves, just like biological traits do — not through malice, but through selection pressures.
1. What Exactly Is “Bad Science”?
“Bad science” doesn’t necessarily mean fraudulent science or outright misconduct. It means science that:
-
uses underpowered studies,
-
relies on weak statistical methods,
-
employs p-hacking,
-
selectively reports only significant results, and
-
rarely replicates findings.
Such science can be performed by well-meaning researchers simply trying to survive in an academic ecosystem designed around publish-or-perish.
The replication crisis
The 2015 Open Science Collaboration attempted to replicate 100 psychology findings; only about 36% replicated. Similar failures have been seen in cancer biology and economics.
When replicability fails, it’s a sign that science is producing too many false positives, and doing so systematically.
2. Why Are False Positives So Common?
False positives arise naturally from noise, but the modern scientific ecosystem amplifies them.
The incentives look like this:
| Behavior | Reward |
|---|---|
| Publish flashy results quickly | Grants, tenure, fame |
| Take years to do a careful, high-power study | Very few rewards |
| Publish a null result | Often impossible |
| Do a replication | Actively discouraged |
This creates a pressure cooker in which the quickest way to generate publishable results is simply to lower methodological standards.
As Ioannidis famously argued in 2005, “Most published research findings are false” — not because scientists are bad, but because the system selects for false-positive-generating behavior.
3. Historical Anecdotes: When Incentives Tilt, Science Skews
The ESP Debacle
In 2011, psychologist Daryl Bem published a paper suggesting students could predict future events — ESP.
The methods were weak and statistically tortured, but the findings were novel and surprising. So they were published in a prestigious journal.
Why? Because novelty sells, even if the methods are flimsy.
The Brian Wansink “P-hacking Factory”
Wansink, a Cornell researcher, ran a social-nutrition lab famous for headline-grabbing results (“People eat more soup from self-refilling bowls!”).
His emails later revealed systematic data dredging — not fraud, but a culture where “find something publishable” trumped rigor.
These stories illustrate the paper’s thesis: labs that produce lots of positive results prosper, even if the results are fragile.
4. Smaldino & McElreath’s Insight: Science Evolves Like a Darwinian System
Here’s the key insight of the paper:
Research methods are transmitted culturally through labs, and labs that publish more quickly produce more “descendant” labs.
Just as biological traits that increase reproductive fitness spread, research behaviors that increase publication output spread — regardless of whether they uncover truth.
This is the heart of the argument.
Labs = organisms
With traits such as:
-
sample size norms
-
statistical approaches
-
replication habits
-
degree of rigor
Students & postdocs = progeny
They carry the lab’s practices to new institutions.
Publication success = reproductive fitness
Thus, science becomes an evolutionary system — and not a benign one.
If quick-and-dirty methods generate more papers per year, they become dominant in the population of labs. Over decades, methodological deterioration becomes inevitable.
5. The Model: Why Low-Power Science Wins
Smaldino & McElreath built computational models to test this idea. The models show that:
-
Labs that use low sample sizes can run more studies.
-
More studies = more chances for false positives (“significant results”).
-
Significant results = publications.
-
Publications = hiring, tenure, grants.
-
Successful labs produce more trainees → spreading their methods.
In evolutionary terms:
Low-rigor labs have higher fitness.
This is an uncomfortable conclusion.
Is this really how academia works?
Yes — and you can see it empirically.
-
Neuroscience has a median statistical power around 20–30%.
-
Ecology has chronically tiny sample sizes.
-
Biomedical research repeatedly fails in pharmaceutical replication checks (Amgen, Bayer).
These are not failings of individuals — they are signs of evolutionary pressure.
6. Why Replication Fails as Quality Control
Replication is supposed to act like the immune system of science. But it almost never does.
Why?
-
Replications are expensive.
-
Replications are discouraged.
-
Journals often reject replication papers.
-
Senior scientists retaliate against negative replications.
Thus, poor methods do not get “punished.”
Instead, they persist and propagate.
The model shows that unless replication is made incredibly common and highly rewarded, it cannot counteract the evolutionary drift toward bad science.
7. What This Means for the Future
If left unchanged, the system will continue to evolve toward:
-
lower power
-
higher false positive rates
-
more irreproducible results
-
faster publication cycles
-
increased pressure on young scientists
-
widening gap between published claims and reality
The paper is a warning: We are selecting for the worst kinds of science.
Unless incentives change, good methodology will go extinct in many fields.
Conclusion: A Crisis of Evolution, Not of Ethics
Smaldino & McElreath force us to confront a difficult truth:
The decline in scientific rigor is not caused by bad people, but by a bad system.
Science is evolving — and not toward greater reliability.
But evolution is not destiny.
In later posts, we’ll explore how to redesign incentives so that good science becomes the winning strategy again.
No comments:
Post a Comment