Content
Summary
Oliver's blood problem demonstrates that data consistent with a hypothesis does not necessarily provide evidence in its favor. By comparing the likelihood of blood traces under the hypothesis that Oliver is a contributor versus not, the analysis shows the data actually provide evidence against his involvement. This occurs because the unexplained rare blood type is more likely to be found in the general population when Oliver is excluded, creating a factor of two advantage for the alternative hypothesis.
Intuition
Intuitively, one might think that because Oliver has type 'O' blood which matches one of the crime scene traces, this supports the idea that he was there. However, Bayesian reasoning requires us to compare how likely the entire set of data is under two competing scenarios. If Oliver is the source of the 'O' blood, we are left with the burden of explaining the rare 'AB' blood solely by chance in the general population. If Oliver is not the source, we have two random people to explain both traces, and because there are two ways to assign the blood types to these two people, the probability of observing the data actually increases. Thus, the presence of Oliver's common blood type makes the rare blood type harder to explain away if he is the only suspect, whereas his absence allows for more probabilistic flexibility.
Key equations
Posterior odds equal prior odds multiplied by the likelihood ratio (Bayes factor).
Probability of finding one 'O' and one 'AB' blood type in a random pair, accounting for two orderings.
Derivation
- Define Hypothesis A: Oliver is a contributor. Define Hypothesis B: Oliver is not a contributor.
- Under Hypothesis A, Oliver accounts for the 'O' trace. The remaining data is the 'AB' trace. The probability of finding a random person with 'AB' blood is 1% (0.01). So, p(D|A) = 0.01.
- Under Hypothesis B, Oliver is not a contributor. The data consists of two random people with 'O' and 'AB' blood. There are two orderings: (O then AB) or (AB then O).
- Calculate p(D|B): The probability is 2 * P(O) * P(AB) = 2 * 0.6 * 0.01 = 1.2% (0.012).
- Calculate the Bayes factor (likelihood ratio): p(D|A) / p(D|B) = 0.01 / 0.012 ≈ 0.833.
- Since the Bayes factor is less than 1, the data favor Hypothesis B (Oliver is not a contributor) over Hypothesis A.
Worked example
Determine if the blood evidence supports the claim that Oliver contributed to the crime scene, given: Crime scene has one 'O' (60%) and one 'AB' (1%) trace. Oliver has 'O' blood.
- Identify the two hypotheses: A (Oliver is contributor) and B (Oliver is not contributor).
- Calculate likelihood under A: Oliver provides the 'O' blood. The 'AB' blood must come from a random person. Probability = 0.01.
- Calculate likelihood under B: Two random people provide the blood. They can be (O, AB) or (AB, O). Probability = 2 * 0.6 * 0.01 = 0.012.
- Compare likelihoods: 0.01 (A) < 0.012 (B).
- Conclusion: The data are more likely if Oliver is not a contributor.
Answer The evidence is against Oliver's involvement (Bayes factor < 1).
Common mistakes
- Assuming that because Oliver's blood type matches one of the traces, the data must be evidence in his favor.
- Failing to account for the two possible orderings (O then AB, or AB then O) when calculating the probability of the data if Oliver is not a contributor.