A major conceptual error in the standard BLS and Lebergott unemployment estimates for 1930-43 is reported. Emergency workers (employees of government contracyclical programs such as WPA) were counted as unemployed on a normal-jobs-to-be-created instead of job-seekers unemployment definition. For 1933-41, the corrected unemployment levels are reduced by 2-3.5 million people and the rates by 4-7 percentage points. The corrected data show strong movement toward the natural unemployment rate after 1933 and are very well explained by an anticipations-search model using annual full-time earnings.

T his paper analyzes optimal formation of rational expectations where information is costly to obtain to utilize. Under these conditions, expectations normally formed by an estimator will be intermediate between those based on all available information and those which are optimal extrapolative predictors. 1 Since the costs of information are likely to vary less than the value of information, this model can explain systematic differences in expectations in markets, such as, labor, for which the value of accuracy of expectations differs between or among buyers and sellers. Similarly, expectations about the same variable may differ between markets where arbitrage is also costly. Section I extends a model due to Theil toanalyzetheeffectsofcostlyinformation. Illustrative applications are discussed in Section II .

This paper analyzes optimal formation of rational expectations where information is costly to obtain to utilize. Under these conditions, expectations normally formed by an estimator will be intermediate between those based on all available information and those which are optimal extrapolative predictors.1 Since the costs of information are likely to vary less than the value of information, this model can explain systematic differences in expectations in markets, such as, labor, for which the value of accuracy of expectations differs between or among buyers and sellers. Similarly, expectations about the same variable may differ between markets where arbitrage is also costly. Section I extends a model due to Theil toanalyzetheeffectsofcostlyinformation. Illustrative applications are discussed in Section II.

A major conceptual error in the standard BLS and Lebergott unemployment estimates for 1930-43 is reported. Emergency workers (employees of government contracyclical programs such as WPA) were counted as unemployed on a normal-jobs-to-be-created instead of job-seekers unemployment definition. For 1933-41, the corrected unemployment levels are reduced by 2-3.5 million people and the rates by 4-7 percentage points. The corrected data show strong movement toward the natural unemployment rate after 1933 and are very well explained by an anticipations-search model using annual full-time earnings.