Monday, 8 May 2017

Preferences over statistical and economic significance

It's been a couple of years since I read Ziliak and McCloskey's "The Cult of Statistical Significance", but I must have put aside this paper by Erik Thorbecke (Cornell) at the time to read later (I don't see an ungated version, but it appears it might be open access), and I just ran across it again last week. In the paper, Thorbecke looks at Ziliak and McCloskey's argument that economics researchers should focus more on economic significance (what Z&M term "policy oomph"), and less on statistical significance (I reviewed the Ziliak and McCloskey book here, a couple of years ago).

What struck me about Thorbecke's paper though was that he demonstrated the difference in preferences between economists with a greater preference for economic significance, and economists with a greater preference for statistical significance, using indifference curves. Which is timely, given that we only recently covered the consumer choice model in ECON100. Here's the key figure from his paper:


The two 'goods' over which economists preferences are defined in this model are economic significance (on the x-axis) and statistical significance (on the y-axis) (for more on the distinction, see my earlier post on the Ziliak and McCloskey book). The upper panel (A) shows economists with a greater preference for statistical significance. Notice that the indifference curves are relatively flat. We know from ECON100 that the slope of the indifference curve is equal to the ratio of marginal utilities (-MUx / MUy). In this case of economists with a greater preference for statistical significance, the marginal utility of x (economic significance) is relatively low and the marginal utility of y (statistical significance) is relatively high (since these economists would prefer more statistical significance, rather than economic significance), so -MUx / MUy is a small number (i.e. a flat curve).

The lower panel (B) shows economists with a greater preference for economic significance. Notice that the indifference curves are relatively steep. In this case of economists with a greater preference for economic significance, the marginal utility of x (economic significance) is relatively high and the marginal utility of y (statistical significance) is relatively low (since these economists would prefer more economic significance, rather than statistical significance), so -MUx / MUy is a large number (i.e. a steep curve).

Notice that, despite the difference shapes of the indifference curves, both groups of economists would prefer more economic significance and more statistical significance. That is, both groups prefer indifference curves further up and to the right. They just differ in terms of which of those two 'goods' is more important.

Where's the budget constraint? I don't think there is one, at least not in the sense of a continuous line like we see in the basic consumer choice model from ECON100. However, there may still be a trade-off between economic significance and statistical significance in the 'preferred' model that we report in any given research paper. And the economists with the preferences in Panel (A) would be more likely to prefer the model with more statistical significance, while the economists with the preferences in Panel (B) would be more likely to prefer the model with more economic significance.

Who is right, of the two groups of economists? They both are. In the model, the two groups of economists make decisions based on their own preferences, maximising their utility (by attaining the highest possible indifference curve - the highest level of utility). In his paper, Thorbecke concludes:
Ultimately, the determination of economic importance is an issue that can only be approached within a specific context and should be left to the (subjective) judgments of individual researchers.
Which would be based on the shape of their individual indifference curves.

No comments:

Post a comment