Empirical Foundations for Economic Analysis

The social scientist, unfortunately, seldom can run controled experiments. We can’t get everything else in the economy to stand still while we lower income taxes 10 per cent to see just what would happen. Nor can we get at the results by putting a few people off in a closed room and lowering their income tax 10 per cent. So the social scientist is never as sure and precise as the physical and natural scientist can be.

But it’s easy to overstate the differences, too. Economics is increasingly an empirical science, which is building up a vast body of quantitative knowledge about the behavior of economic units and the interactions among them.

How, for example, can we be sure about what proportion of their new disposable income consumers will spend in the event of a tax cut? The answer is, there is no way to be sure. But through intensive empirical analysis of consumer behavior in the past, we can greatly increase our confidence in the prediction we may make. And modern statistical techniques make it increasingly possible for us to use past information as a basis for predicting the future.

Suppose we speculate (“hypothesize”) that the amount people spend on goods and services in any year will depend on the disposable income they have in that year. We hypothesize that con- sumer spending is a “function of” disposable in- come. That is, we hypothesize that consumer spending depends on, or is predictably related to, the disposable income they receive. Economists might write the functional relationship:

C = f(DI)

where C stands for consumption, f for “a func- tion of” (perhaps with a value of .95 in this case), and DI for disposable income.

Suppose now that we get records showing the disposable incomes and consumption spending of a large number of families over the past ten years. Looking at these records, we find that many families have Went about .95 of their dis- posable incomes on consumption in most years, but there are lots of exceptions. For example, young families seem consistently to spend more than .95, families in their SO’s spend less, and re- tired families spend more. (These differences make sense if you think about it; young families are just starting up, buying new durable goods, raising babies, and the like, so they are able to save little; older families, once their homes are established and the children raised, find it easier to save out of their incomes; while retired families have reduced incomes and spend down past sav- ings.) We also find that in years when incomes have risen rapidly, the percentage spent on con- sumption falls below .95. (Again, this seems rea- sonable, because it takes time for people to adjust their spending to new higher incomes.) And so we might examine many other special forces at work. But over all, for the average of all families in periods of reasonably stable, prosperous times, con- sumption hovers around .94 to .96 of disposable income.

In addition, we might observe a sample of families this year to see how they behave. Looking at them one at a time, we find wide diversity. But again, on the average, they seem to come out at about the .95 level for this year.

What could we infer from this statistical analysis? We should have to be careful, because we have seen exceptions to our general presump- tion. Certainly we could not safely predict the behavior of any particular family without know- ing a lot about that family. But on the imaginary evidence cited just above, we would be increas- ingly comfortable in saying that, other things equal, in reasonably stable, prosperous periods consumers as a group will spend about .95 of any new disposable income on goods and services, given a reasonable amount of time to adjust to the new income.

This is an oversimplified example, but it sug- gests the way in which we must go about building up reliable quantitative information on economic behavior. Actually, as we shall see in Part Two, con- sumer behavior is a good deal more complex than this, and we need a more elaborate theory to ex- plain and predict it satisfactorily. So it is with most other parts of economic life. Economics, like any other empirical science, must continually de- velop new theories, test them out against the  world, and reformulate them in the light of em- pirical evidence.

Prediction with Economic Theories

Thus far we have been talking mainly about a simple economic relationship—how much fam- ilies will spend on consumption out of an addi- tional dollar of disposable income. If we return now to the more complex issue of the total eco- nomic results of a tax cut like the one in 1964, the problem is more complex. Here we need a theory with several subparts and a complex set of inter- actions among them. The more convinced we are about the reliability of our understanding of the subparts of the system, the stronger our faith can be in our predictions about the consequences of the tax cut on employment and income.

How can we test how good a total theory is? There are two ways. One is to examine the assumptions on which the theory rests and the internal logic that it builds on those assumptions. The other is to make a pragmatic test of how well it actually predicts in the real world the variables in which we are interested (here employment and income). Both are valuable.

In the first approach, we ask two questions: (1) Do the assumptions of the theory correspond to the reality to which it is being applied? (2)Is the internal reasoning of the theory logically cor- rect? If both these conditions are met, the theory should be useful for explanation and prediction. But there are problems. Most important, the world is so complex that it’s hard to be sure that our assumptions are correct and that they are the only relevant ones for our problem. Remember that in the tax-cut example above, some econo- mists suggested that leaving the money stock out of the analytical model left out the really impor- tant causal factor.

As a practical matter, therefore, many econo- mists now rely more on the second test—how well does the theory actually predict? Economists thus try a model put to see what it predicts, and then check the prediction against the real world they are trying to explain. If the prediction is poor, they distrust the model, no matter how beautiful its logic. If a model based on dubious assumptions gives good predictions, they are inclined to use it, but only tentatively. For then they are uncer- tain about whether it predicts well because it is sound, or merely by chance. Alas, we can never be absolutely sure a model is “correct,” even if it has predicted correctly in a number of past situa- tions. It is always possible that the result may be the consequence of some other set of factors that we haven’t thought about and don’t have in our model. But the more frequently a model predicts correctly, the more confident of it we have a right to be.

Lastly, we can never be sure of our predic- tions for another reason—we can never know that the future will be like the past. Even though con- sumer spending of disposable income has been extremely consistent in the past, it may change tomorrow. It may change because consumers sim- ply decide to behave differently. Or, more likely, it may change because of a change in other cir- cumstances which affect consumer behavior. For example, international tensions may grow, and in such an uncertain world consumers might decide to save more than in the past.

All these warnings are important. There are many uncertainities about our understanding of the economic world. But much of economic theory has met both the tests discussed above. And there is no uncertainty about the path to better under- standing. It is the path of economics as an em- pirical science, which involves an interacting pat- tern of basic empirical research, the development of better analytical models, and continuing checks on the predictive power of the new models. Hap- pily, modern economics is making rapid progress as an empirical science, and much of this progress underlies the chapters to come.

One last introductory comment concerning theory and practice. You often hear someone say, “That’s good theory, but it doesn’t work in prac- tice.” ‘What the speaker must mean, if he is talk- ing sense, is either that it is not good theory be- cause it doesn’t help explain the real world, or that it is good theory but it is being inexpertly used. A theory (the use of an abstract model) is good only when it does work in practice—that is, when it is useful in helping to understand the problem to which it is being applied.