Background to my interest in teaching children with autism

But, you see, that can create problems too. You might think that the more information you have on a person, the more information you have. Actually, that’s not true. When you add together bits of information to draw a composite conclusion, a funny thing happens. The more information you try to include, beyond a rather limited amount, the poorer your ability to predict and the poorer your ability to create therapeutic effects. Now, everybody knows that what has just been said is both flat wrong and utterly mad. However, strange to say, a good principle to consider in most things you do is: ‘less is more’. Felicity kept forgetting this principle.
The principle involved starts with the observation that each bit of information concerning anything contains both information and error in it – for example, in making a prediction. The information part is narrowly concerned with the issue being predicted. The error part may be related to all sorts of other issues it might predict – so that different bits will contain different ‘kinds’ of errors. The first bit of relevant data or information used contributes all of its information and all of its error – so that, by itself, it predicts imperfectly, but as well as it predicts. The second bit of relevant data, also containing information and error, contributes that part of its information which is not yet accounted for by the first bit of data (i.e., less than all its information), and that bit of its error not yet included – which may be a quite different ‘kind’ of error from that in the first variable, and so may add all of its error. By the time we add in the third, fourth and fifth bits of relevant data, more and more of the informational elements have already been included. But, since the error parts can relate to all sorts of other predictions, the contribution of error from each additional variable continues to increase the amount of prediction error. The point of diminishing returns from adding information is reached very quickly in dealing with ‘uncorrelated’ variables – that is where the variables may predict what you’re after, but also predict all sorts of other different things. With ‘uncorrelated’ variables, the optimum level of prediction is usually achieved using no more than 5 to 7 separate variables. The point of diminishing returns is reached a little later when the variables being used are ‘correlated’ – that is, all commonly predict the same kind of thing. With ‘correlated’ variables, the optimum level of prediction is usually achieved using about 16 (12 to 20) variables. Above or below these numbers of variables any predictive efficiency tends to decline. Now that sounds like numerological nonsense to anybody with any brains. But it isn’t.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s