| Abstract: | The paper identifies some basic shortcomings of contemporary accounting research. Aside from issues related to how researchers pick topics—they are generally too remote from pre‐existing realities—the main points concern the standard paradigm: explaining a dependent variable Y in terms of X, the primary variable of interest, while controlling for Z. The paper argues that just because the t‐statistic related to X is significant does not mean that X helps to explain Y. To address this issue requires a goodness‐of‐fit analysis that evaluates the incremental contribution of X. Such tests can show that X effectively acts as noise though X's t‐statistic is significant. Incremental goodness‐of‐fit analyses would potentially have dramatic consequences on research because rejection of the null would now take place much less often. The paper also considers problems associated with ordinary least squares, with specific emphasis on the scaling of dollar amount variables in linear cross‐sectional settings. |
