Multivariate Analysis

[Statsstudent88]

Hello Stats pros,

I've a complete novice with stats, so sorry if this sounds muddled.

I've carried out two regression models for y (Oxygen change) and x1 (grain diameter), y and x2 (% organic matter). All the data is continuous.

I'm trying to find out the proportion of y variability each x value can explain.

So far I've carried out a Primary Component Analysis. The PCA seem to find the relationship between all the data (i.e the relationship between x1 and x2), and I'm not sure if this affects the output.

Am I doing the right thing, or is there another analysis I can use.

Thanks

Rich

[Dason]

http://en.wikipedia.org/wiki/Coefficient_of_determination

[mads_st]

First - there is no such thing as continous data.
Second - you have mistunderstood PCA (principal component analysis). PCA takes combinations of your data vectors so as to maximize the variance of this combination (basically): http://en.wikipedia.org/wiki/Principal_component_analysis.
PCA is not really novice statistics so I would advice against using stuff like that.
What is it that you want an answer to?

[Statsstudent88]

Hi,

Sorry I confused continuous data with ratio scale. The data are numerical and don't have arbitrary zero points.

I've got the R squared value (Coefficient of determination) for y~x2 and y~x1, and I understand that this is the amount of variability in my data that is accounted by the linear model I ran on the values.

I'm basically trying to find a way to combind the two models, to find if, for example, the variability accounted for by the y~x1 model, can explain any of the variablity not accounted for by y~x2. I hope to try and get a better understanding of the non accounted variablity.

I hope I've explained myself more clearly this time!

Thanks

Rich

[mads_st]

Well you can start with the model:

y = x1 + x2

I would also add a constant.

You can then compare this model to the restricted model

y = x1 (plus a constant)

[Statsstudent88]

Thanks alot mads_st, I didn't think it would as simple as that!

Cheers,

Rich