Are regression models useful?

Freedman, David, "Statistical Models and Shoe Leather," Sociological Methodology, 21, 1991, 291-313.

Freedman posed this and other interesting questions regarding the frequently used regression technique in 1991, although regression model has been extensively used since Yule (1899). Yet the weakness of regression model being used to make causal inference is just pointed out and Freedman raise the concern to evaluate the results of various applications.

Freedman doubt the capability of regression model in making causal inference and even its role in controlling confounding variables. There has been illustrations with the use of regression models, with the development of two stage least square, latent variable approach, specification testing and so on, yet Freedman question their validity of assumption of the models since there is effort carried towards this direction.

Some examples of abuse of regression techneque have been given in the article and then he directed the reader to some appropriate application of the regression models, including one of his work in 1987.

In the end, Freedman argued that it is time to ask questions like, are models usful, is it possible to differentiate sucessful from unsucessul use of models, and how to test and evaluate the models?

This paper caught my eyes when I was surfing Professor Ken Chay's (at Berkerley) website for his applied econometric course.