Dear researchers, I'm currently working with robust regression variable selection in the presence of multicollinearity problem. Actually, I want to test my proposed method with different degrees of multicollinearity. Say the correlation of some covariates equal to 0.99, 0.97, 0.95 . Unfortunately, I couldn't find real data having this problem. I will be highly appreciated if you help me to find such data.
Hassan -
The US Energy Information Administration (EIA, http://www.eia.gov, [email protected]) might have data you can use.
When I worked there, I lead development of estimation for missing data for many cases, using regression 'prediction,' where usually only one (very good) regressor was needed, but there were some cases of multiple regression, so it occurs to me that one possibility for you might be as follows, where you might find some collinearity cases which might possibly be useful for you: Say for hydroelectric generation, for a given State (a larger geographic region would be better, but they likely would not know what to give you if you don't just ask by State, though I can give you an attachment below that might help), you could ask for monthly sample data of generation by plant (for y), and past corresponding annual census generation data by plant (x1), and nameplate capacity by corresponding plant (x2). There may be some months and States where x1 and x2 may be highly correlated.
One example for combining State data may be found by looking at examples in the following (Oregon, Washington, and Idaho, or California and Nevada):
https://www.researchgate.net/publication/261586154_Using_Prediction-Oriented_Software_for_Survey_Estimation
Anyway, you might want to see what data the EIA will show to you. There may be many other possibilities, but the one above occurred to me as at least an example.
However, I do not know that you will commonly find the high correlations in these data that you seek. Maybe you won't find any. So depending upon what data they may give you, it may or may not be worth your efforts to try to use it, unless you are not able to find friendlier data somewhere else.
If you would post the findings you might eventually have, I'd like to know.
Cheers - Jim
Article Using Prediction-Oriented Software for Survey Estimation
- Badges
- Science topic
- Similar topics
- Instrumentation Engineering
- Instruments