An animated guide to the differences between Bayesian Research and Frequentist Research
— and why it matters (a lot) for real decisions.
A busy intersection with no traffic light has averaged 100 accidents per year for a long time, with natural variation ranging between 90 and 110 accidents.
A traffic light is installed. The following year, accidents fall to only 50 accidents — a dramatic reduction of 50 incidents.
Two statisticians will look at the very same data
but they will ask very different questions.
Same starting data. Same outcome (50 fewer accidents).
Live simulation — cars entering intersection each year
The frequentist view — only outcomes visible, causes hidden
The Frequentist asks a binary question:
"Is this reduction (50 fewer accidents) too large to be explained by random variation (chance) alone?"
It works by establishing a null hypothesis — if the light had no effect, the reduction would be due to chance alone.
There would be between 90-110 accidents per year AFTER the light was installed.
The reduction of 50 accidents is so far outside that range, it cannot be due to chance alone.
THE STATISTICALLY SIGNIFICANT VERDICT: 95% confident the reduction was not due to chance.
But here's the catch: the frequentist approach can only see that accidents decreased — not why.
It's like counting crashes without determining the specific cause each time (or at all).
The traffic light wasn't installed in a vacuum. It wasn't the only change or the only thing affecting traffic at any given moment. Many things change at an intersection over time.
Each of those things can have an affect on what happens at that intersection. These other factors — called confounders — may have ALSO contributed to fewer accidents.
Without looking at each of/all of those things, the frequentist cannot know how much of an affect each one had. The frequentist does not know specifically what caused how much of the change.
The frequentist lumps all of these together. It says: something caused fewer accidents, and it probably wasn't random.
The Bayesian approach goes further by modeling each factor's contribution. Accounting for the confounders all along the way will reveal the direct correlation — isolating exactly how much of the change was caused by the traffic light itself.
Bayesian view — each incident's cause is labeled and visible
The Bayesian asks: "Given everything we know about this intersection, what is the probability the traffic light specifically caused the reduction — and by how much?"
It integrates prior knowledge — everything we have already learned about historical data on weather, traffic patterns, road conditions, and more — into a statistical model that weighs each factor at every moment throughout the study.
The model is also updated/informed by everything we learn during the study, adjusting to integrate all factors impacting the outcomes along the way.
The result isn't a single number. It's a probability distribution: a range of how many accidents the light likely prevented, at different confidence levels. The levels of confidence are not arbitrary - they are defined by the impact of each particular factor at any time.
90% probability that the traffic light directly prevented 40 of the 50 fewer accidents.
This exact Bayesian methodology was applied in a 10-year study (2014–2023) published in the American Journal of Infection Control (AJIC) in January 2026.
Researchers replaced standard hospital surfaces with preventive biocidal surfaces (EOScu) — patient overbed tables, bed rails, and horizontal surfaces — with a goal of reducing healthcare-associated infections (HAIs).
Previous frequentist studies showed 65–85% fewer infections — but couldn't specifically isolate how much of the 65% to 85% was directly from the copper impregnated surfaces versus other hospital improvements happening simultaneously.
The Bayesian model showed EOScu was directly responsible for cutting infection rates in half.
Not just correlated. Caused.
"Ten years of longitudinal data, combined with advanced statistical modeling, provides a level of evidence we rarely see in infection prevention research." — Dr. Chetan Jinadatha, lead investigator
95% probability that EOScu directly caused a 49% reduction in ALL HAIs over the ten years studied.