Confessional evidence has a long, controversial and notorious history. Past cultures have regarded it as the ultimate form of evidence, requiring no corroboration. It has also been frequently associated with grievous abuse of power. In modern times, the extraction of confessions under duress has been viewed with considerable scepticism, and is now inadmissible in English law. Yet while many defence lawyers are sceptical of the evidential value of confessions, the police and judiciary generally remain convinced of their importance. Confessions play a significant role in an estimate 20 per cent of all cases. In this paper, I show that a Bayesian analysis of confessional evidence supports those who adopt the sceptical view. A potentially dangerous counter-intuitive situation can arise unless confessional evidence is assessed by a jury in the appropriate way. Specifically, for a confession to contribute to the evidence of guilt, the prosecution must prove that the probability that an innocent person would confess to the crime is undoubtedly less than the probability that a guilty person would confess under the same conditions. That is,
Prob(Confession given guilt) > Prob(Confession given innocence)
If this is not shown to be the case, then the confession is at best useless, and at worst may actually provide evidence of innocence of the accused. The error of assuming that confessional evidence can never reduce the probability of guilt is what we may call the Interrogator's Fallacy.
How can such a fallacy arise ? Recent psychological research by Gudjonsson suggests that for certain types of cases, the above inequality may indeed be reversed. He has shown that the psychological traits of interrogative suggestibility and compliance are important in determining how individuals cope with interrogation. Essentially, those with high scores for these two traits are less able to resist interrogation. Interrogative suggestibility has, moreover, been found to correlate with a number of cognitive and personality metrics, including reduced IQ and lack of assertiveness, while compliance appears to be linked to avoidance of conflict and obedience to authority. Given such correlates, the perpetrators of anti-establishment crimes, most notably terrorism, are clearly unlikely to score highly for suggestibility and compliance. As such, terrorism cases appear to run a considerable risk of a reversal of the inequality given above, and thus to the counter-intuitive situation of a confession of guilt actually contributing to evidence of innocence. It is interesting to note that many of the most notorious miscarriages of justice do indeed centre on terrorist cases in which confessional evidence played a key role.
Confessions with corroborative evidence
Given the reluctance of English courts, at least, to accept unsupported confessional evidence, the paper goes on to consider the case of confessions backed by corroborative evidence. Again, the situation is not as straightforward as it may at first appear.
Suppose initial evidence of guilt, A, is followed by further evidence of guilt, B. For the combination of evidence to increase the probability of guilt, we require
Prob(G | A & B) > Prob(G | A)
Now, it is easy to fall into the trap of believing that the mere emergence of the new evidence of guilt, B, must imply a greater probability of guilt, because only with the guilty is both A and B both likely to point to guilt. Symbolically, this argument is equivalent to
Prob(G | A & B) >> Prob(G | A) as Prob(A & B | G) >> Prob(A & B | ~G)
However, corroborative evidence does not simply "tot up" in this manner, and the above reasoning is undermined by a subtle case of transposition of conditioning. Specifically, Bayes's Theorem shows that the new evidence B only constitutes further evidence of guilt if
Prob(B | A & G) > Prob(B | A & ~G)
In other words, the mere existence of B is not sufficient to provide further evidence of guilt. Rather, B's relationship to both the original evidence A and guilt G must first be assessed.
To take a concrete example, A could be a confession of murder, and B the discovery of a body resulting from the confession; in this case the key inequality is clearly likely to be met, and the probability of guilt has clearly been boosted by the new evidence. However, if A is evidence that could be planted on a suspect by police, and B is the resulting confession, then the inequality does not necessarily hold: an innocent person can be persuaded to confess if they become sufficiently confused or brow-beaten by the police.
In conclusion, confessional evidence remains fraught with danger, even when supported by corroborative evidence.