The authors describe a method of assessing the potential effect of errors in variables in the logit regression model with continuous-valued independent variables. They then apply the method to a model of lending discrimination in which measurement error would bias estimates of discrimination. Although the method cannot detect whether or not there are errors in variables, the authors show how to assess whether parameter estimates are sensitive to alternative assumptions about error variance.
Amanda L. Kramer
To receive a notification by email when Economic Quarterly is posted online or to order single copies of past issues, click on the links below (published online only since 2012).