topic Re: interpret the PLS regression in Statistical Procedures

interpret the PLS regression

andy_wk — Tue, 05 Aug 2014 15:17:59 GMT

I'm giving you a bit of background before asking my question. I've done a univariate PLS regression where I came out with many models.

My boss asked to interpret the PLS regression for the best model.

As I've never done a PLS before, I don't know how to interpret that.

What does he mean by interpretation?

Thanks for any leads you will be able to give me.

Cheers.

Re: interpret the PLS regression

SteveDenham — Tue, 05 Aug 2014 16:41:53 GMT

Can you share code and output? Without those, interpretation becomes pretty difficult. I think by interpretation your boss is asking you--"So I see this equation here. What can you tell me about it?"

Steve Denham

Re: interpret the PLS regression

andy_wk — Wed, 06 Aug 2014 12:13:30 GMT

Hi Steve,

I will do that tonight. I'm working on something else that needs my full attention, or almost full

Cheers.

Re: interpret the PLS regression

PaigeMiller — Wed, 06 Aug 2014 14:31:19 GMT

I've done a univariate PLS regression where I came out with many models.

Normally, when you do a PLS, you get one model, so this needs explanation.

says

--"So I see this equation here. What can you tell me about it?"

Often, the interpretation of a PLS model refers to loadings and not the final model equation

Re: interpret the PLS regression

SteveDenham — Wed, 06 Aug 2014 14:56:48 GMT

Very true. But there is a model equation, and that has an interpretation--including the new independent variables that are constructed and the loadings of the measured variables onto the components (as the simplest PLS for univariate responses is regression on principal components).

I am guessing that the OP has done a lot regressing a single variable on a battery of observed variables, and thus came up with "many models". Kind of misses one of the main advantages in PLS, but I think that is what was done.

Steve Denham

Re: interpret the PLS regression

PaigeMiller — Wed, 06 Aug 2014 17:47:59 GMT

I have never seen an interpretation of the final PLS model equation; I have only seen interpretations of the loadings and the score plots. Maybe I'm reading the wrong journals.

As I pointed out, he needs to explain in more detail what he actually did (or wants to do). I was not guessing what you just said you were guessing.

Re: interpret the PLS regression

andy_wk — Thu, 07 Aug 2014 09:16:07 GMT

Hi Everyone,

@Steve : I hope it makes sense. Lots of Box-Jenkins in the program.

I'd be honest, I did not write it and I'm newcomer , feeling like a fraud sometimes , in the statistical domain.

http://pastebin.com/vKpwH6ZV

Re: interpret the PLS regression

SteveDenham — Thu, 07 Aug 2014 13:19:54 GMT

Could you take out the 'noprint' option in the PROC PLS calls, and then share _some_ of the output? Frankly, I am a lousy macro programmer, and I am not really sure what PLS is doing here. Some output (perhaps sanitized) will help in determining just what is being done and reported. I know it would make interpretation easier.

Steve Denham

Re: interpret the PLS regression

PaigeMiller — Thu, 07 Aug 2014 15:59:04 GMT

I agree with Steve.

We need you to explain this to us, what you are doing and what you are trying to do, in Plain old English. You know, give us a clear problem statement, and description of the approach you are taking, as if you were writing the abstract to a technical paper. We don't want code, we want understanding.

I do see you have two consecutive PROC PLS blocks, and the code in each is slightly different ... but I have no idea why two PROC PLS blocks with slightly different code are used in the analysis, and to tell you the truth, I doubt I would ever approach it that way (but maybe with more understanding, it might make sense). I also don't see any Box-Jenkins, which to my way of thinking, at this point in time, is probably a good thing.

Re: interpret the PLS regression

andy_wk — Mon, 11 Aug 2014 13:09:04 GMT

Hi Paige and Hi Steve,

I will do that but most likely this weekend as it will be assumption this Friday in Europe and I need to deal with another more pressing issue.

Best,

Andy