Omitted variable vs Multicollinearity

Multicollinearity: correlation among independent variable -> correct by dropping an independent variable

Omitted correlated variable (so no multicollinearity) -> coefficient biased and inconsistent.

So I deduce that according CFA having multicollinearity is bad and omitting a correlated independent variable (which is requested to “remove” Multicollinearity) is bad as well.


No only when an important variable is ommitted, which would have been essential to your model, that’s a misspecification.

You cure multicollinearity by ommitting a variable, not the most important one, but one which is useless because it has so large correlation with another.

I have not seen (or not remember) good questions on this issue though, so the above is my deduction.

Thanks Moosey,

yes but do deduct that you “cure multicollinearity by omitting not the most important one…” is not really helpful do understand what’s really going on…

I mean, I just saw a question where was stated that omitting a correlated variable is misspecification and so coefficients are biased and inconsistent. Then, on the same problem, the indep. variables where correlated and so multicollinearity was present --> one variable has to be omitted. Which seems to be contraddictory… and they said nothing about a more important variable than others…was so in general.

I hate how che CFA curriculum is explained and written. Too general and not clear.

Yes in several points it is. Honestly I don’t know in real life how we exclude variables in order to have a better model, I’ve always worked with models which were ready and I never had to question their multicollinearity or whatever. I think this is not the scope of Level II.

in real life, yes you’re going to have make subjective choices of whether you omit an ind variable to reduce multicollinearity or to keep them to improve R2/adjusted R2.

however, cfai never leaves anything for chance. you’ll not be asked to choose between getting rid or keeping an ind variable.

yes but, they are saying something which in the end isn’t really true!

I think CFAI is not being thorough in the subject, just that. Econometrics and modeling variables is not as simple as the 200 or 300 pages we read in L2.

PhD people in quantitative economics or statistics are well deep in the theory and practice of this issues and procedures. At the end, it is the choice of the researcher whether doing A or B (or even Z) to a model. We don’t need to get to this level at the moment (for the exam). You can further improve your knowledge on those subjects out of the CFA Program, but prepare to spend a lot of time and effort.

I am just saying that CFA is not clear enough and so they should probably avoid to make some statements like: “Omitting a correlated variable leads to biased coefficients”.

To make a statement like this they shoudl be more clear (for example saying that omitting the MOST important variable) or not saying it at all.