Visual inference for model checking

Visual inference for model checking

We strive to specify models that resemble data collected in studies or observed from processes. One way to check whether the model is a reasonable abstraction of reality is to display the data in the model space, such as residual plots for linear models. While these plots are well-behaved for simple models, such as linear regression with uncorrelated errors, this is not the case for more-complex models. For example, residual plots for multilevel models often show patterns that are artifacts of the model-fitting process, and are not indicative of a model deficiency. This talk will outline how visual inference can be utilized during model validation for multilevel models, and how this approach can be generalized to other models. I will also discuss how these techniques have informed how I teach model validation to undergraduate students.


adam loy

July 31, 2019