⚠️ Limitations & Caveats

Priors are educated guesses: The starting probabilities (“priors”) are based on general reasoning and public data, not precise statistics for this case. Try adjusting them to see how your assumptions affect the outcome!
Evidence weights are subjective: The likelihood sliders represent how strongly each piece of evidence supports or contradicts each hypothesis. There’s no universal scale—use the rubric above as a guide, and experiment with different values.
Assumes evidence pieces are independent: For simplicity, the model treats each piece of evidence as if it’s unrelated to the others. In reality, some evidence may be connected (e.g., timestamps and phone records). This can make the model overconfident—combine related items or adjust weights down if you suspect overlap.
Not all factors are quantified: Some important aspects (like emotional state or motive) are hard to express as numbers. This model focuses on the evidence that can be reasonably quantified.
No real-world validation: This tool is for learning and exploration. It hasn’t been tested against actual trial outcomes. Use it to understand the process, not to predict verdicts.
Sensitivity to assumptions: Small changes in priors or evidence weights can lead to big changes in the final probabilities. Try moving the sliders to see how robust (or fragile) the conclusions are.

This site is intended to help us explore how reasonable people, starting from different assumptions or weighing evidence differently, can reach very different conclusions about the same case. By making our reasoning explicit, we can better understand how and why viewpoints diverge—even when everyone is acting in good faith. This is an educational exploration of Bayesian reasoning, not a verdict predictor.

Bayesian Analysis of the Karen Read Case

Post-Verdict Analysis

Case Outcome: Not Guilty of Murder, Convicted of OUI

Verdict Date: June 18, 2025

Verdict:

Not Guilty: Second-degree murder, Manslaughter, Leaving the scene of an accident
Guilty: Operating Under the Influence (OUI)

Sentence: 1 year probation, 24D alcohol education program

Jury Deliberation: 5 days (21+ hours)

Key Factors in the Verdict: The jury determined there was reasonable doubt regarding the fatal incident, leading to acquittal on the more serious charges. The conviction for operating under the influence was based on evidence presented during the trial. The defense's arguments about investigative flaws and evidentiary issues were influential in the outcome.

This page explains the Karen Read case using Bayesian logic—a way of updating our beliefs as new evidence comes in. Think of it like detective thinking: start with what you believe, and change your mind a little or a lot depending on the clues.

🤔 What Are We Trying to Figure Out?

Who (or what) caused John O'Keefe's death?

We consider three main possibilities:

H1 (Prosecution Story): Karen Read hit John with her SUV and left him outside.
H2 (Defense Story): John was hurt or killed inside the house and Karen didn't do it.
H3 (Mixed Scenario): Maybe something in between happened—maybe Karen hit him accidentally, or someone else was involved and evidence was changed or hidden.

🎲 What Is Bayesian Thinking?

Imagine you're guessing who stole a cookie. You might first guess your little brother (because he does it often), but then you see your dad has chocolate on his shirt! That's a clue. So you update your guess.

Bayesian thinking works like this:

Start with what you think is likely (prior belief)
Get a new clue (evidence)
Update your belief based on how well each explanation fits the clue

1. Start with a Prior

We looked at how often people are hurt by their partners. That's not super common when there's no history of violence.

Chance Karen did it (H1)

10%

Prior probability

Chance someone else did it (H2)

60%

Prior probability

Chance of mixed scenario (H3)

30%

Prior probability

Adjust the sliders to set your priors:

H1: Karen Did It: 10% H2: Someone Else: 60% H3: Mixed: 30%

Total must equal 100%. If it doesn't, values will be scaled automatically.

2. Evaluate the Evidence

Here's how each piece of evidence fits with each story:

Clue	Fits H1 (Karen)?	Fits H2 (Others)?	Fits H3 (Mixed)?	Comments
Body found on lawn	Maybe	Likely	Likely	Could have been moved
No blood in her car	Unlikely	Very likely	Likely	No crash signs inside car
Injuries don't match car hit	Unlikely	Likely	Possible	Injuries match falling or being hit, not run over
Tail light pieces near body	Medium	Low	Medium	Some think they were planted
Phone tracked inside house	Low	High	Medium	His phone was still in the house after Karen left
Police video gaps	Low	Medium	High	Raises suspicion
Police story changes	Low	High	High	Defense says there's a cover-up
Karen was emotional	Medium	Medium	Medium	Could mean guilt or confusion
New timeline in 2nd trial	Low	High	High	Supports idea that he was hurt inside, not outside

3. What If It's Partially True? (H3)

Real life isn't always clear-cut. What if:

Karen bumped him by accident but wasn't fully responsible?
Or someone else hurt him, and people covered it up later?

These ideas blend parts of both H1 and H2. That's H3, the "middle-ground" theory.

4. What Happens When We Add It All Up?

Adjust how strongly the evidence supports each scenario. The sliders let you weigh different pieces of evidence:

How to interpret the weights: ⓘ

0–1×: Weak evidence
5–7×: Strong evidence
10×: Very strong/conclusive
Values in between represent varying degrees of support

💡 Tip: Play with the sliders and priors to see how your beliefs change! Bayesian reasoning is about making your assumptions explicit and seeing how conclusions shift.

How Bayesian Updating Works:

Start with a prior belief → update with evidence → get a new belief (posterior)

Show an Example: Numeric Walkthrough

Example: Suppose your prior for H1 is 20%, and you rate a piece of evidence as 80% likely if H1 is true, and 40% likely if H2 is true.

Prior odds for H1:H2 = 20:80 = 0.25
Likelihood ratio = 80/40 = 2
Posterior odds = 0.25 × 2 = 0.5
Posterior probability for H1 = 0.5 / (0.5 + 1) = 33%

This is how a single piece of evidence updates your belief!

As you adjust the sliders, the chart below will update to show how different weightings of the evidence affect the probabilities of each scenario, given your initial belief set above (your priors).

H1: Karen did it

10%

likelyhood given evidence

H2: Someone else did it

60%

likelyhood given evidence

H3: Mixed scenario

30%

likelyhood given evidence

🎯 What Should the Jury Think?

The jury isn't being asked, "Did Karen do it?" They're being asked: "Are you sure—beyond a reasonable doubt—that she did?"

Bayesian answer: No. Even if H1 is possible, it's the least supported explanation. H2 and H3 together account for 90% of the likely explanations.

✅ Final Thoughts

Bayesian logic helps us:

Think like detectives
Update our beliefs with new facts
Avoid jumping to conclusions

In this case:

H1 (Karen did it): least supported
H2 (someone else did): strongest support
H3 (something in between): possible but less likely than H2

Unless much stronger proof appears, Bayesian thinking says: reasonable doubt remains.

🎛️ Try It Yourself: Adjust the Priors

Use the sliders above to set your own starting assumptions. The chart below shows how the outcome changes when the evidence is applied (held constant).

The Right Side of Wrong

Is Karen Read Guilty of John O'Keefe's Death?

⚠️ Limitations & Caveats

Bayesian Analysis of the Karen Read Case

Post-Verdict Analysis

Case Outcome: Not Guilty of Murder, Convicted of OUI

🤔 What Are We Trying to Figure Out?

🎲 What Is Bayesian Thinking?

1. Start with a Prior

Chance Karen did it (H1)

Chance someone else did it (H2)

Chance of mixed scenario (H3)

2. Evaluate the Evidence

3. What If It's Partially True? (H3)

4. What Happens When We Add It All Up?

H1: Karen did it

H2: Someone else did it

H3: Mixed scenario

🎯 What Should the Jury Think?

✅ Final Thoughts

🎛️ Try It Yourself: Adjust the Priors

More Links

Official & News Evidence Galleries

Key Retrial Developments (2025)