News and Views

Nature 424, 1004-1005 (28 August 2003) | doi:10.1038/4241004a

Neuroscience: Of mice and mentality

Steve Blinkhorn


Evidence of a general learning ability in mice — that there is a good correlation between an individual's performance in tasks that make different processing demands — suggests a parallel with humans.

Why isn't everyone a genius? Argue it whichever way you like, people manifestly differ in the ease with which they learn, and, whatever the reason, we have to provide for these differences in education, training and employment. But for the study of the elementary processes of learning, psychologists have traditionally looked to laboratory animals: rats, mice and pigeons. The practice of using inbred strains eliminates as far as is possible the influence of individual differences in the subject pool, but has led to the neglect of the study of these differences for their own sake.

Now Louis D. Matzel and collaborators, in what in the future may be seen as a seminal study, have reported the tentative identification of a general learning ability, varying from one individual to another, in mice (J. Neurosci. 23, 6423–6433; 2003). Their work invites independent replication, not least for its 'crossover' character: it remains a rarity for research to combine the skills of the laboratory experimental psychologist with the correlational methods that dominate the study of individual differences.

There are two major elements in the design of the study that should be noted. First of all, Matzel et al. used animals from an outbred strain of laboratory mice, on the basis that they show greater behavioural variability — the very opposite of the usual requirement for minimum inter-individual variation (although we have no indication of how laboratory strains might differ from wild populations). Second, five tasks were chosen that differ greatly in their sensory, motor, motivational and information-processing demands, and in which the possibility of transfer of learning from one task to another is minimal.

For instance, the passive avoidance of sensory overload and the active running of a Lashley maze to gain a food reward require quite different information processing and involve different control modalities (aversive as compared with appetitive), different motor activity, and so forth. The other learning tasks were odour discrimination, fear conditioning and a water maze. The animals showed marked variation from individual to individual in all the tasks. The authors also collected some additional measures, including three aspects of free-field behaviour (such as speed of running), the number of faecal pellets deposited during free-field activity, and body weight.

Of the five tasks, the highest correlation found, 0.47, was between performance in the Lashley maze and on the passive avoidance task. To a physical scientist used to seeing near-exact correspondence, a correlation of 0.47 is a paltry thing. But to a psychologist looking at two such different kinds of task, it is the kind of result that raises the hairs on the back of your neck, and sets the pulse, if not racing, then up to the canter.

In fact, the pattern of correlations for individual mice between all five learning tasks is entirely positive, and although not all of the positive correlations are high, they are sufficient to define a single factor in a factor analysis, which loads the five learning tasks plus the proportion of time spent in the more open areas of the free field. Body weight and number of faecal pellets have no systematic relationships to the other variables, and overall free-field activity and speed of running, although highly correlated with each other, have unsystematic associations with the other variables.

So, have we barked our collective scientific shins against a rodent version of IQ? And do we have here an insight into the nature and origins of human intelligence? The 'ability to learn' is a relatively uncontroversial member of the long list of proposed definitions for intelligence, and Matzel et al. have provided some plausible evidence — from impressively controlled experiments — of an influence on learning that cannot be explained by common elements in the tasks. When it comes to the size of the factor, this study yields estimates in the same broad range as has been suggested for the influence of g, the general-intelligence factor in humans.

But the sample size, at 56, is a shade uncomfortably low for a factor-analysis study of ten variables: the standard errors of the correlations are of the order of 0.14, so some care is needed in weighing the existence of pattern among the correlations, and in tolerating the fact that some of them are not significantly different from zero, in a conventional bivariate significance test. (The authors kindly made their raw data available to me for independent checks, to ensure that their results are not the outcome of particular choices of factor-analysis methods.)

This is a dilemma for any researcher stepping over the venerable boundary between experimental and correlational approaches to psychology, and Matzel and his collaborators are to be congratulated on striking the right note of caution in reporting their findings. Spearman's announcement of his discovery of the human general-intelligence factor in 1904 was far less cautious in tone, although from a modern perspective his sampling was less controlled, his sample sizes were of the same order of magnitude, and his experimental controls were, by any standards, rather poor.

A promising pilot study, then? Rather more than that. First of all, we now have five carefully described tasks that can serve to define a reference paradigm, and an example of care in experimental design that should set a benchmark for further work. Second — and this is not immediately obvious from the published paper — there are suggestive patterns buried in the correlation matrix that could be rather productive in determining the next steps, in particular with regard to the relationship between learning and emotionality. And finally, the researchers have performed a signal service in demonstrating how the distinctive skills of what Lee Cronbach once called "the two disciplines of scientific psychology" can combine to open lines of enquiry, and to shed new and refreshing light on the underpinnings of differences in behaviour.