When we have a large sample size but it caused multicollinearity, can I use PLS to solve this problem ?
Hope someone can give me a solution
Firstly I have performed multiple linear regression but it caused multicollinearity ( VIF value was very high ).
When we have a large sample size but it caused multicollinearity, can I use PLS to solve this problem ?
Multicollinearity can be one of the problems for PLS. Firstly you can try to understand the root cause of the multicollinearity issue & address it e.g. identifying the duplicate independent variables / indicators & dropping them, combining the similar independent variables / indicators if make sense, recheck literature review whether the instrument for various variables / constructs are adopted / adapted correctly etc.
On the sample size for PLS, you can refer to the following book:
Hair, J. F., Hult, G. T. M., Ringle, C. & Sarstedt, M. (2013). A Primer on Partial Least Squares Structural Equation Modeling (PLS-SE). Sage Publications, Inc.
Minimal PLS sample size can be determined by:
Sample size does not cause multicollinearity. Possibly what you mean is that you finally got a sample size large enough that you could detect the multicollinearity that has always been present in your data.
http://blog.minitab.com/blog/understanding-statistics/handling-multicollinearity-in-regression-analysis
Essentially the choice is to leave some factors out, or try to derive a new set of variables that are uncorrelated. If the problem is not too bad, then you can also consider ignoring the problem. Clearly defining "not too bad" is not easy.
I agree with Timothy. I do not understand fully your question. What means exactly for you "large data size"? Why did you computed just PLS?
Anna
Dear Khamphone,
I guess we are talking about Partial Least Squares regression and what you mean is that the data matrix you are trying to analyze has a higher number of variables than samples. In this case, you can definitely use PLS. PLS calculates, starting from your original descriptors, a set of new latent variables, ORTHOGONAL (I.E. INDEPENDENT), maximising the covariance between your X (predictor matrix) and your Y (response matrix). Multicollinearity is then overcome. Mind that such latent variables constitute both a good summary of X and valuable predictors of Y.
Hope this helps.
Best regards,
Raffaele
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size