More powerful tests of a theory of choice of accounting methods and the effect of changes in these choices on equity values are provided. The power increase comes from efficiently aggregating results across studies. One conclusion is that at least six variables common to more than one study have explanatory power. These variables are managerial compensation, leverage, size, risk, and constraints on interest coverage and dividends. Another conclusion is that the posterior probability that the theory taken as a whole has explanatory power is close to one. This conclusion includes the effect of variables that only appear in one study.

This paper examines cross-sectional analysis procedures common to many market-based accounting research papers. Both the economic and econometric properties of ‘levels’ and ‘returns’ studies are discussed. Topics covered include the relations between the accounting studies and cash flow valuation models, the role of expectations of accounting variables, deflators, spurious inference, risk adjustment and its relation to growth, size and leverage, residual dependence, dependence among explanatory variables, and the effect of scale differences across firms. Major conclusions are that market value is the correct deflator in returns studies, and that levels and returns studies are economically but not econometrically equivalent.

Value maximization requires either that knowledge is transferred to those with the right to make decisions, or that decision rights are transferred to those who have the knowledge. A tradeoff of knowledge transfer costs and control costs is required. Characteristics of firms' investment opportunity sets (IOSs) that affect knowledge transfer costs and control costs are identified. Testable predictions about the relations between these characteristics and firms' decentralization decisions are developed and tested. The evidence presented is consistent with our predictions and is robust to different ways of measuring variables.

A number of recent research papers use two-stage procedures in lieu of a single multiple regression, in some cases purportedly as a solution to colinearity among independent variables. We demonstrate that, since collinearity is inherently a data problem rather than a statistical problem, no partitions of dependent or independent variables, orthogonal or otherwise, can provide insights into the relative influence of collinear variables. For the class of linear unbiased estimators this follows directly from the Gauss-Markov Theorem, but we demonstrate some of the results in detail as an aid to interpreting particular papers.