Why is it the case that when there is multicollinearity, the regression coefficients become indeterminate and their standard errors become infinite?
Asked
Active
Viewed 32 times
0
-
2I vote to reopen this. The OP says coefficients are indeterminate meaning the design matrix is not full rank. That's different than just "highly correlated" variables as per the flagged duplicate answer. There are a few different numerical approaches and ways of doing inference. – AdamO Nov 21 '18 at 15:49
-
@AdamO The question doesn't ask about doing inference. – whuber Nov 21 '18 at 17:29
-
@whuber beg pardon. I should have said "...calculating estimates". – AdamO Nov 21 '18 at 17:30
-
@AdamO Fair enough. But the question still asks about a specific phenomenon that I believe is addressed (at several different levels) in the duplicates. It does not ask about calculating anything. If it were changed to such a form, it's likely our many threads on model building or some of them on identifiability in regression models would cover it, so I am reluctant to guess how the question might be modified or even whether it should be. – whuber Nov 21 '18 at 17:34