If, on a charge of murder, an accused claims to have been mistaken as to whether he was under attack – that he thought the person he shot and killed (who it transpires was his girlfriend) was imminently about to attack him, the veracity of this claim would seem to be undermined by evidence that the girlfriend screamed loudly before she was shot. The question arises whether this fact, that the girlfriend screamed, must be proved beyond a reasonable doubt? More importantly, must each individual piece of evidence of screaming be proved beyond a reasonable doubt? Is this not what an accused is entitled to in a criminal trial?
Virtually anyone who is old enough to watch TV will tell you that the standard of proof required of a prosecution, in a criminal trial, is proof beyond a reasonable doubt.
Lawyers will explain that this means that every requirement of the criminal offence in question must be proved beyond a reasonable doubt. They will also explain that the case against an accused and all the evidence against the accused must be considered in its totality. That is, that courts are not permitted to take a piecemeal approach to the evaluation of evidence. Many will say that all evidence against an accused person must be proved beyond a reasonable doubt. This, they will say, all follows.
What is often not recognised is that it also does not follow from the ultimate onus of proof (that the prosecution must prove its case beyond a reasonable doubt) that all evidence must be proved beyond a reasonable doubt. One possible source of this error may lie in the adoption in our law of “the cardinal rules of logic” pronounced by our former Appellate Division (now the Supreme Court of Appeal) in the case of S v Blom (1939 AD 188). These cardinal rules of logic were set out as a guide to how courts ought to draw inferences in the context of circumstantial evidence.
Conventionally, evidence is thought of as falling into two discrete and mutually exclusive categories: direct evidence and circumstantial evidence. On one hand, direct evidence, such as the testimony of an eyewitness, requires that no inferences are supposedly necessary from this sort of evidence. Direct evidence is conventionally thought of as evidence which, if true, is direct evidence of the fact in issue. Circumstantial evidence on the other hand is considered to be evidence which, if true, is not direct evidence of the fact in issue. Instead, it is evidence of a fact from which an inference must be drawn to the fact in issue. Thus where a question of fact before the court is whether the accused shot and thereby killed the victim, direct evidence may take the form of a witness testifying to the fact that s/he saw the accused shoot the victim. In the alternative, circumstantial evidence would be evidence such as that of a witness who testifies that s/he saw the accused who was armed with a gun go into the room where the victim was shot dead, that s/he heard a bang, and saw the accused exit the room shortly thereafter. This sort of evidence is circumstantial in that it requires that inferences must be drawn from the evidence (of what the witness saw) to the fact in question – did the accused shoot and kill the victim?
A first point worth observing is that circumstantial evidence is often considered to be weaker evidence and a so-called circumstantial case is one which is considered to be more easily refuted – at least relative to the case in which the prosecution will produce direct evidence. However, it is not the case that circumstantial evidence is necessarily weaker than direct evidence. The classic example is that of the persuasive value of fingerprint evidence (which is circumstantial evidence) for the purpose of identification, compared with that of an eyewitness.
Beyond that, as is the case in respect of many attempted distinctions in law, the distinction between circumstantial and direct evidence is not sacrosanct and breaks down on analysis. It breaks down at least in the sense that it is incorrect to imagine that direct evidence or indeed any evidence is free from requiring the court to draw inferences (Zeffertt and Paizes, The South African Law of Evidence, p 99). This is best illustrated by the following example (drawn from Wigmore on Evidence): The accused (X) is charged with the murder of the victim (Y) by shooting. X denies that s/he shot Y. The prosecution calls a witness, who testifies that s/he saw X arguing with Y, that s/he saw X produce a firearm, and point it at Y, heard a loud bang, and saw Y fall to the ground. The defence explains that it has only one question for the witness: whether the witness saw the bullet strike Y.
The point of the illustration is only to draw one’s attention to the fact that even given what appears to be a straightforward case of direct evidence, one must nevertheless draw inferences. The point ultimately is that all evidence requires a court in considering its verdict to draw inferences from the evidence. Zeffertt and Paizes explain that: “All evidence requires the trier of fact to engage in inferential reasoning.” (The South African Law of Evidence, p 99). Some evidence requires fewer inferences, this would be traditionally so-called direct evidence whereas other evidence, traditionally circumstantial evidence, will require more inferences. Nevertheless the point must be observed that the court is never free of drawing inferences and therefore the rules that govern the drawing of inferences govern the court in its ultimate evaluation of all evidence.
Thus we are left with the ultimate problem: how is a court to evaluate the evidence? We must recall that the standard of proof required is beyond a reasonable doubt. So does this require that every piece of evidence that the court considers significant, be tested against this standard (of a reasonable doubt) and only if it passes can the court rely on it, whereas if it fails, the court must discard it? For many lawyers this is obviously what a court must do. It follows from the onus of proof, and something more – it follows from the so-called “cardinal rules of logic” from the case of S v Blom (above).
The “cardinal rules of logic” in Blom represent the law on the drawing of inferences in criminal trials. They state that (to paraphrase):
1. The inference sought to be drawn must be consistent with the proved facts.
2. The proved facts must exclude all other inferences except the one sought to be drawn.
The second rule seems to be little more than a restatement of the standard of proof in a criminal trial. The inference must be the only reasonable inference – presumably, otherwise it may not be drawn. The problem in criminal cases, is that in its ultimate assessment of guilt, all evidence relies upon inferences, and so all evidence requires the application of the rules of Blom. Each inference must be the only reasonable inference – right?
This would seem to fit well with a case in which say, the accused (X) is charged with murder of a victim (Y) if the only piece of evidence is eyewitness testimony in which the witness testifies that s/he saw X shoot Y, and that, upon immediate inspection, the witness had found Y to be dead. Let us assume (again) that X denies that s/he shot and killed Y. X claims further to have never met the victim and to have been nowhere near the scene of the crime at the relevant time.
Since the evidence of the witness is the only evidence against the accused, the prosecution’s case depends on it. If this intermediate fact represents “an indispensable [link] in a chain of reasoning toward an inference of guilt” (from the Australian case of Sheperd v The Queen (1990) 170 CLR 573 at 579) the strength of the prosecution’s case is, of course, only as strong as the link in the chain. If this piece of evidence falls below the standard of proof beyond a reasonable doubt, a reasonable doubt must exist in respect of the accused’s guilt, and s/he is entitled to an acquittal. Just as a chain is only as strong as its weakest link, so must all the links (evidence) be proved beyond a reasonable doubt. Thus, in the example given, the only piece of evidence, the testimony of the witness, should be subjected to the ultimate standard of proof: is it true beyond a reasonable doubt? But what if the evidence is less straightforward. If for instance there was only evidence that:
1. Someone resembling X was seen in the vicinity of the crime scene. The eyewitness says he is 90% sure it was X;
2. The ballistic expert is able to say, because of damage to the bullet, that the likelihood that the bullet was fired from X’s gun is 80%;
3. A partial fingerprint is discovered at the scene which yields only enough points of similarity with the accused’s right thumb, so that the forensic expert can only be 70% sure this print is that of X.
Blom requires that, at least on the second “cardinal” rule of logic, the inference sought to be drawn must be the only reasonable inference. But what if none of the possible inferences are reasonably possible and what if there are other reasonable inferences? If we assume that we adopt, nominally, a 99% standard of proof as proof beyond a reasonable doubt, what if none of these pieces of evidence exclude all other reasonable possible inferences:
1. There is a 10% (reasonable) possibility that the eyewitness identified the wrong person;
2. A 20% (reasonable) possibility that gun that fired the fatal shot is not that of the accused; and
3. A 30% (reasonable) possibility the fingerprint is not that of the accused.
None of these items of evidence stand up to the standard of a reasonable doubt, and, on the 2nd rule in Blom, no inferences can be drawn at all. Blom seems to demand ‘chain-reasoning’. Bellengère and Palmer (et al), state:
“Even where the court may draw a reasonable inference to the exclusion of all other reasonable inferences, it must nevertheless assess that inference to be persuasive beyond a reasonable doubt before it can rely on the inference to sustain a criminal conviction” (Oxford, The Law of Evidence In South Africa, p 202).
However, if one regards the evidence as the “strands in a cable” (following Sheperd v The Queen (1990) 170 CLR 573 at 579), quite a different result follows. The strength of a cable or rope is, of course, made up of the combined strength of its strands – and each strand may be, by itself, quite weak. To pursue this analogy, pieces of evidence may be considered together, even if each, by itself, does not satisfy the standard of beyond a reasonable doubt, to support an inference. At once it will be obvious that this is contrary to the specific instructions in Blom. Yet, our courts are clear that in the drawing of the inferences they must take account of the totality of the evidence, and must not consider evidence on a piecemeal basis. (S v De Villiers 1944 AD 493; S v Reddy 1996 (2) SACR 1 (A); R v Mtembu 1950 (1) SA 670 (A)).
Is it that Blom is wrong, and if so, why has it not been ejected from our law? The problem is that Blom seems right, given that it seems to fit neatly with the criminal onus, and it is sometimes right – depending on the work that the inference is doing. Blom appears to demand “chain-reasoning” through-out, whereas, on occasion, “rope-reasoning” is more appropriate. If the inference sought to be drawn will carry an entire case, Blom would be an appropriate standard to employ. But if one is engaged with numerous intermediate facts, none of which are crucial, or necessary in and of themselves, then the “cardinal rules of logic” in Blom, can lead one into a logical fallacy. When one should adopt “chain-reasoning” (as directed by Blom), or “rope-reasoning” is not straightforward – it all depends on the work that the evidence has to do in a given case. Perhaps one method would be to always adopt “rope reasoning” on the analogy that if a single piece of evidence or fact is crucial, it equates with a single, and only strand of a rope – of a rope of one strand.
So, if for instance, on a charge of murder, the accused claims to have been mistaken as to whether he was under attack – that he thought the person he shot and killed (who it transpires was his girlfriend) was imminently about to attack him, and the veracity of this claim would seem to be undermined by evidence that the girlfriend screamed loudly before she was shot. For our purposes let us assume that, given the context, it would entirely undermine the accused’s claim of mistake. The question of fact as to whether she screamed becomes so crucial as to represent an ultimate issue which would seem to properly attract the requirement that it must be proved beyond a reasonable doubt. However, evidence (of intermediate facts) from which it may be inferred that she screamed need not pass the ultimate standard of proof beyond a reasonable doubt. If five witnesses claim to have heard a woman scream at about the relevant time on the night of the incident, each need not be true beyond a reasonable doubt before the court, considering this testimony together, and together with ballistic and pathology reports to the effect that she incurred a severe wound which would probably have caused her to scream before she was incapacitated, concludes that a woman screamed that night, before she was shot.
There is a substantial difference between raising some doubts, even reasonable doubts in respect of some of the evidence against an accused, and raising reasonable doubt in respect of the case against the accused.
 Assume that, for our purposes, proof of these (intermediate) facts beyond a reasonable doubt (99% on our assumption), the accused would be guilty. For those who may wonder what may be going on at the level of maths (although it is controversial whether using maths is appropriate at all), instead of needing all the factors to be true at the same time (in the case of ultimate issues/facts), when we are dealing with an intermediate fact though, where any number of pieces of evidence could prove the fact, the formula swings around – we now ask what are the chances of all of the pieces of evidence offered to prove that fact all being wrong (false) at the same time. Thus, items of evidence at 90, 80, and 70% produce a probability of 99.4% (0.1 x 0.2 x 0.3 = 0.006)
 Chain reasoning must also be carefully chosen and applied in case the “links” are not independent. (Tillers revision of Wigmore on Evidence IA (1983) at 1113, reported in Zeffertt and Paizes p 117.)
*I wish to thank Professor Andrew Paizes (co-author of Zeffertt and Paizes, The South African Law of Evidence) for his helpful comments on an earlier draft. I have amended this post in line with his suggestions. Nevertheless, any mistakes remain entirely mine.