Statistical significance as used in economics has weak theoretical justification. In particular it merges statistical and substantive significance. The 182 papers using regression analysis in the American Economic Review in the 1980s were tested against 19 criteria for the accepted use of statistical significance. Most, some three-quarters of the papers, did poorly. Likewise, textbooks in econometrics do not distinguish statistical and economic significance. Statistical significance should not be the focus of empirical economics.

Economists view talk as cheap and culture as insignificant. Yet humans are talking animals, talking in their markets. The talk probably matters: why else would the human animals bother doing it? The usual economic view of the talk is that it issues orders and conveys information. Workers at GM are ordered to report for work tomorrow; credit ratings are conveyed. Economic analysis takes these parts of the talk into account without fuss. Production theory can be viewed as the theory of one mind issuing orders. Much of game theory is concerned in one way or another with information (though game theory, as Joseph Farrell (1995) and others have found, requires more than bits of information). But issuing the orders and conveying the information does not account for all of the talk. The third part of the economic talk is persuasion.