Whether one accepts if human reasoning is
logical or not depends greatly on what one accepts
logic to be. Early experiments on human logical problem solving (e.g., Wason,
1966) showed that people were very much poorer than might be expected at
logical reasoning. However, later research has shown that the context in which such problems are phrased can greatly affect performance. Phrasing the problem thematically produces better performance than if it is phrased abstractly (Wason and Shapiro,
1971). It seems now that the apparent poor human reasoning, is just a product of
evolution passing on processing biases that aided our survival. From this angle, human reasoning can be seen as predictable, as it seems to be a direct product of evolution, but it does not necessarily follow "logic". In this essay I have taken "reasoning" to mean " logical reasoning", rather than "statistical reasoning".
It is not unreasonable to expect that humans are able to think "logically" (by "logic" I mean "propositional logic"), after all, one might wonder how mankind developed logic in the first place if there were not some logic aspect of the human brain. As stated by Pinker (1998), there is something "peculiarly compelling, even irresistible, about P, P implies Q, therefore Q". In discussing the seeming intrinsicness of logic, Pinker provides supporting evidence from language. All languages have logical terms such as "not", "and", "some", "equivalent", and "opposite", and in English and the several other languages studied, children can use "and", "not", "or", and "if" appropriately before the age of three.
However, we know that people do not always use logical reasoning correctly. Pinker, continuing his discussion from a linguistic point of view, highlights the point that logical words in everyday life can be ambiguous. For instance, "or" could be taken to be inclusive or exclusive - i.e., "A, or B, or A and B", or "A, or B, but not A and B". another example would be that in formal logic "some A are B" means "at least some", or "some and perhaps all". However, in ordinary speech, it is usually taken to mean "some but not all" (Woodworth and Selles, 1938). Woodworth and Sells (1938) also posited the idea of "atmosphere effects", where premises that include "all" (premises are universal), people tend to accept a more universal conclusion - also including "all" - even if this does not necessarily follow.
Woodward and Selles concluded that people are not logical. But is this fair comment? Chapman and Chapman (1959) disagree, claiming that it is not that people are illogical, rather they misinterpret premises. This interpretation does not conflict with Wooward and Selles' findings, and it is a point of view similar to that subsequently adopted by Henle (1962), who said that people reason in accordance with standard logic, except they either misunderstand the task or fail to accept the premises. An example of the second of these points would be that many people have a problem with the argument; "The moon is green. If the moon is green, then Mars is striped. Therefore, Mars is striped". This argument is valid, but people may not be able to see this due to the fact that the Moon is not green, nor is Mars striped, and nor would there be any reason to believe that one might infer the other.
Most of the research on people's ability to use logical reasoning has focussed around Wason's (1966) Selection Task. In the original version of the task, people are presented with four cards with a letter on one side and a number on the other and the rule "if a card has an A on one side, then there is a 4 on the other side". The cards are presented so that the sides that can be seen have A, D, 4, and 7 on them. Participants must then decide which of the cards must necessarily be turned over in order to find out if the rule is true. Broken down into logical terminology, the rule is "if P, then Q", and the A card is "P", the D card is "¬P", 4 is "Q", and 7 is "¬Q". Logic dictates that a statement can only be verified if no counter-example can be found. So, the correct solution to the problem is picking the A card and the 7 card - i.e., picking the P and ¬Q cards. If the A card has any number on the back apart from 4, the rule is false, and if the 7 card has an A on the back, the rule would also be false. No matter what is on the reverse of the D or 4 cards, it tells us nothing about whether the rule is true or false. However, 40% of people choose the P card (correctly) and also the Q card (incorrectly). A further 30% make the same choice, but also choose the ¬Q card. Only 4% choose the correct combination of P and ¬Q.
Cheng et al. (1986) found that even after 40 hours of lectures on logic, performance on this task was no different than it was before. Evidence like this might suggest that people are not capable of logical reasoning. But, Wason did find that once participants were allowed to turn over the cards freely, they could see what the correct solution to the problem should be. So, people do have some insight into the logic of the problem. Furthermore, Wason and Shaprio (1971) found that if the same logical problem was presented using thematic materials, performance improved. This time, the rule was "every time I go to Manchester, I travel by train" and the cards said "Manchester" (P), "Leeds" (¬P), "Train" (Q), and "Car" (¬Q). 60% of people now gave the correct solution. This particular study has not replicated well, but others using thematic versions of the original Wason problem have - for instance, Johnson-Laird et al.'s (1972) Postal Rule Problem ("if the letter is sealed, then it has a 50 lire (or 5d) stamp on it"), and Griggs and Cox (1982) using the Drinking Age Rule Problem ("if a person is drinking a beer, then that person must be over 19 years of age").
It was suggested that familiar materials could act to remind people of similar problems they have experienced or seen solved in the past - the "memory cueing hypothesis". However, this has been falsified by d'Andrade's Sears' Problem; "if any purchase exceeds $30, the receipt must have to signature of the department manager on the back". People mostly find the correct solution even though most have never experienced this rule before. It is neither something intrinsic about using realistic material either, as Evans (1979) discovered by using the rule, "every time I eat haddock, I drink gin", that this elicited poor performance.
Cheng and Holyoak's (1985) "pragmatic reasoning schemas" began to get nearer to current thought on the issue of human reasoning. They proposed that we use three sets of generalised rules in reasoning - permission schema, causal schema, and evidence schema. Cheng and Holyoak only really developed their permission schema, as they believed it related to the selection task. The schema was something along the lines of; "if one is to take action 'A', then one must first satisfy precondition 'P'". Cosmides (1989) provided an alternative view - "social contract theory". The rationale behind this idea was that the best way for the species to survive was that if an individual offered help, then help should be expected in return in the future. Altruism is not a good evolutionary trait, as in a competitive environment, an altruistic animal will be unlikely to survive, whereas a selfish animal will quickly become more successful. However, there is seemingly selfless behaviour, and, according to Cosmides, and her colleague Tooby, this can only be explained by there being reciprocation of helping - or "social contracts". So how does this serve to explain the results of the selection task? The problem with social contracts is that they are vulnerable to exploitation by cheaters, who may gladly accept help but offer nothing in return. So, for social contracts to have evolved, humans must have some kind of cognitive ability that allows them to remember who has taken and whether they gave in return. Such a "cheat-detector algorithm" has previously been proposed by Trivers.
Social contract theory has not been unopposed though. Many of the facilitatory contexts in which performance of the selection task improves do not involve socially exchanged costs and benefits. Grigg's drinking law problem is an example. Being 19 is not a cost that is paid for being able to drink beer. Another example is that of Manktelow and Over (1990). In their rule - "if you clean spilt blood, you must wear rubber gloves" - cleaning up spilt blood is not a cost of wearing rubber gloves. But, on the other hand, there is evidence that humans have so-called "cheat detectors". Gigerenzer and Hug (1992) found that given the rule; "if an employee gets a pension, he/she has worked for 10 years", people's choice as to whether they picked P and ¬Q or ¬P and Q depended on whether they imagined the problem from the perspective of the employer or the employee. People tended to choose cards that allowed them to detect if the person who's perspective they were taking had been cheated.
In the new tradition of such thinking, the more recent view of why people seem to reason in the way they do is due to what we as humans have needed to do, not what logic might dictate we do. The classic example of this is Hempel's (1965) Raven Paradox, which states that if "one wants to test the hypothesis that 'if it is a raven, then it is black,' one is much better off inspecting ravens and checking whether or not they are black, than inspecting non-black things and checking whether or not they are ravens". This is, in essence, a kind of selection task. Propositional logic suggests that we should check all "non-black things", which is impractical in a real environment. If one were to go about this problem in real life, it would be much easier to check all ravens. Therefore, people may well pick cards in order to maximally reduce their uncertainty over the rule, rather than picking to find a counter-example. This is known as "confirmation bias" and is widely seen in human reasoning experiments. Oaksford and Chater (1994) related it strikingly to the selection task by pointing out that if a person were to choose cards with the goal of reducing uncertainty, then they would pick in the order - P, Q, ¬Q, and ¬P. In the 30 studies they collated the results of, they found that 89% choose P, 62% choose Q, 25% choose ¬Q, and 16% choose ¬P, thus following the order of selection they proposed.
Is this logical? Perhaps not according to propositional logic, but it is rational if not logically deductive. Indeed, Anderson (1990) proposed that behaviour is rational if it is optimally adapted to its environment. So, from this perspective, human reasoning is both logical and predictable. It is logical because it follows another kind of rationality than is followed by propositional logic - a more "evolutionary" rationality. It is also predictable in that the way we reason could be predicted by survival pressures on the species and evolution.
References:
Pinker, S (1998), How The Mind Works
My lectures!