Reverse regression

From Machinelearning
Jump to: navigation, search

Reverse regression refers to a problem in statistical inference where a conclusion is reversed depending on the variable being adjusted.

The example given by Pearl is that of salary discrimination: comparing the salaries of equally qualified men and women, we see that men earned a higher salary, but comparing the qualifications of equally paid men and women, we see that men are more qualified (p. 426).[1]

There is some discussion of this on LessWrong.[2]

Relation to Simpson's paradox is discussed in http://ect.bell-labs.com/who/tkh/publications/papers/reg.paradox.TAS.pdf

Also see https://concretenonsense.wordpress.com/2013/02/16/defense-against-the-dark-arts-the-reverse-regression-effect/

History

Judea Pearl says the "controversy" "occupied the social science literature in the 1970s".[1]

References

  1. 1.0 1.1 Judea Pearl. "Epilogue: The Art and Science of Cause and Effect" in Causality: Models, Reasoning, and Inference. 2nd ed. Cambridge University Press. 2009.
  2. https://www.greaterwrong.com/posts/FGsoajyYJRczieqh5/understanding-simpson-s-paradox/comment/JyrGhmYj8xSHgPqst