Rebirth of Reason


Qualitative Predictions
by Joseph Rowlands

All too often people think about predictions as being quantitative.  A physics problem might ask you to predict how far a bullet will fly given an initial velocity and position.  Economic forecasters will try to predict the change in inflation, or unemployment, or the trade surplus.  Engineers will make predictions about how much force is a building can take before it collapses.  In each case, the prediction includes a measurable result.

A different kind of prediction is the qualitative prediction.  These kinds of predicts don't have the same kind of precision that quantitative predictions do.  In fact, precision is a term that doesn't really apply to qualitative predictions.  They don't include information about degrees.  They deal with primarily in alternatives, but not the degrees of the alternatives.  A qualitative prediction about a bullet being shot might predict the direction, or the shape of the bullet trajectory, or that it will move faster than some other moving object, or that it won't hit a target.  Each of these omits degrees.

Even when qualitative predictions are recognized, there is a bias against them as being unscientific.  A quantitative prediction has significantly more detail.  That shows the understanding of the behavior is very good, it makes testing the scientific principles more refined, and the more detailed predictions can be better utilized since more information is available. 

Qualitative predictions, in contrast, have significantly more variance.  It could indicate that the understanding of the causal relationship isn't as clear.  Any tests to validate it are not as meaningful because so many different outcomes would count as a success.  And finally, the qualitative prediction is lacking so many details that it can't be utilized well.

Take an engineering prediction about how a new feature in a computer will affect performance.  A prediction that says only that performance will go up or go down is not as useful as a prediction that provides the exact amount performance will go up or down.  The feature may have other costs or benefits, and the engineer may desire to weigh the total costs and benefits to see if a tradeoff is worthwhile.  Quantitative predicts, if accurate, are better than qualitative predictions.

The important qualifier, though, is the word 'accurate'.  There are areas where quantitative predictions are not really possible.  A qualitative prediction may work fine, but the quantities involved are either unknown or unknowable.  Under these circumstances, the quantitative prediction is not superior.  It is vastly inferior.  In a misguided attempt to provide measurements or degrees, the accuracy of the prediction is discarded.

Economics is one such example.  Qualitative predictions are possible.  If the price of a good increases, the quantity demanded will decrease or stay the same.  That's because when a product increases in price, its net value is reduced and some customers will find that other products are now superior.  They would divert their spending, reducing the quantity demanded for this product.  Of course, it is possible that if the price increase was small enough, or the product was so much better than the next most valuable product based on every customer's preferences, they may not change their spending habit after all.

Instead of predicting a potential decrease in quantity demanded, assuming all else stays the same, why not predict the exact amount the quantity will decrease?  Wouldn't that make the prediction better?

The problem is that the information needs to make such a prediction is unknown or unknowable.  How will a change in price affect quantity demanded?  That depends on the personal evaluations of each of the market participants.  This is not information that's available to the predictor.  The information is unknown.  A demand curve is a conceptual tool, not a real graph that producers can use.  At any point in time, the quantity demanded for a product at the current price can be known, but there is no information about any other price point.  The information isn't available.

Even worse, the information may be unknowable in principle.  Asking people how they would change their behavior is not always a good reflection of how or if they will.  To get information at different price points, the only sure way to know is to change the price and see what happens to the quantity demanded.  Even this is not information that can be assumed to apply in the future.  Demand may change for many other reasons.  New customers may act in a different way, health reasons may make some people change their purchases, fashions or trends could occur, or they might just get upset that the store keeps changing its prices.

To make quantitative predictions where they are not appropriate, people are forced to make assumptions that they know are not accurate, but that allows for quantitative predictions.  Economists may assume that people's preferences won't change over time, or that different groups of people will have the same preferences, or that changes measured in other markets can be averaged and that will provide a good enough estimate in this market. 

There are countless assumptions that can be made that remove the unwanted changes or effects in an economic situation.  But each assumption is created to ignore a real factor.  The process of making the prediction quantifiable is the process that reduces the accuracy of the prediction.  It's a tradeoff.  You can make very accurate qualitative predictions, or you make your predictions less and less reliable in order to gain more and more detail.

There are many areas where qualitative predictions are superior to quantitative predictions.  Economics is just one example.  Moral predictions is another.

Moralities that are concerned with consequences are interested in making predictions.  A moral principle can be used to guide your action by indicating the likely consequences.  If you are dishonest, you can expect several possible consequences.  First, if the person you lied to finds out, they will be upset and have less trust for you in the future.  Since that is an undesirable option, it is likely that you will try to preserve a lie so that the truth isn't discovered.  Since a lie conflicts with reality, there are many possible indications that your statement was a lie.  An observant person, or someone who stumbles upon the right information, is likely to discover your lie.  To prevent that, you'll have to try to distort new information to reflect the lie.  And you'll need to spend mental effort remembering the lie, who has been fooled by the lie, and who has information that could deflate the lie.  And finally, you have to not be fooled by the lie yourself, always remembering the truth and your misrepresentation of it.

These are qualitative predictions.  Some of them may not even happen.  The lie may be quickly forgotten and won't trouble you.  The probability that you'll need to deal with these is impossible to predict as you can't possibly know every factor involved.  But you do know that these are consequences that you wouldn't have to deal with otherwise.

Some level of quantitative prediction might be possible here.  Certain lies, like a very big lie, may be much more likely to have these effects.  If you are on a date with someone and tell them you are a doctor when you definitely aren't, it's very likely to come up again if you continue to see this person.  This kind of prediction still does not assign numbers or exact measurements to the predictions, but it does provide some information about the degree.  It is more likely to be a problem then a little lie.  Even this can be viewed as a qualitative prediction, though.  Saying the you predict the odds of an event being higher than the odds of another event is a qualitative prediction again.

This isn't to say that a quantitative prediction wouldn't be more useful.  If you could accurately predict the results of your choices, you would be able to choose more wisely.  The superiority of the qualitative prediction is in its accuracy, not in its degree of detail.

Again, one could try to make assumptions that allow a quantitative prediction.  If many possible scenarios exist, you could choose to ignore many of them.  That might make it possible to assign some numbers to the prediction.  But again, this is at the expense of accuracy. 

By making false assumptions in order to enable a quantitative prediction, you are choosing an illusion over reality.  The illusion is that you can make rigorous, detailed predictions, but it comes at the cost of genuine accuracy.
Sanctions: 36Sanctions: 36Sanctions: 36Sanctions: 36 Sanction this ArticleEditMark as your favorite article

Discuss this Article (7 messages)