2

Are there any general results on bounds for linear regression model coefficients if we change the data set? For example, if I learn a model with a few data points less, what will happen to the model coefficients?

I have tried to look at the algebra, but the design matrix inversion involved in finding the model coefficients seems to "block" finding useful bounds.

Note: this is not a "sensitivity analysis" question. The changes in the data could be "big".

Frank
  • 1,305
  • 1
  • 12
  • 17
  • 2
    The method of sequential updating I describe at http://stats.stackexchange.com/questions/6920/efficient-online-linear-regression/6923#6923 also works in reverse. From that you can deduce bounds for the changes in regression estimates as a function of the data that are added or deleted. – whuber Jan 18 '17 at 20:20
  • 1
    Influence functions tell you the exact change in a parameter estimate when a particular observation is left out. – Michael R. Chernick Jan 18 '17 at 22:25

0 Answers0