Bivariate and Multivariate - different results

[#1]

Dear all,

I have some troubles understanding the different results I get when performing a Bivariate or a Multivariate analysis on the same data.

1) Here I performed a Bivariate analysis between two variables. The result I get is correlation = 0. The three points stand on the same vertical line.

2) If I add one variable and perform a multivariate analysis, JMP automatically selects the REML method. This time, when checking the correlation between the same two variables as above (SOT and LasP), the result is equal to 0.9623!

 3) Finally, if I compute a multivariate analysis on several variables including those mentioned above, JMP automatically selects the Pairwise method. The correlation between LasP and SOT is equal to 0 (as it was in case 1).

I thought that the result of a couple of variables in the multivariate computation and the result of fit y to x on the same two variables should have been the same (maybe a small difference due to the different formula)… I am confused on how to interpret these results.

May anyone please provide some clearance, and explain me how to choose on what method to rely, or tell whether I am doing a mistake in one of the three calculations?  

Regards,

Sara

[#2]

There are different methods of estimating the correlation coefficient to suit different situations and conditions. Different methods can produce different results for the same data. The pairwise correlation the Pearson correlation coefficient. The other two methods are described here.