Helmert Contrasts

MattPeterson

First off, I'd like to say "thank-you"! JASP is as easy to use as SuperANOVA , and SPSS is still as dreadful to use as when I first started using it on a VAX system 25 years ago

My big question has to do with Helmert contrasts. The results I'm getting in JASP are not matching the results I'm getting in SPSS.

JASP:

Comparison         Estimate    SE       t      p
L1 - L2,  L3, L4     0.488    0.148   3.300   0.002
L2 - L3, L4          0.101    0.139   0.725   0.473
L3 - L4              0.008    0.121   0.064   0.950

SPSS:

          Contrast               Type3 SS   df   MeanSquare   F      Sig.
Source    Level 1 vs. Later        5.497     1     5.497     5.833   .032
          Level 2 vs. Later        0.299     1     0.299     1.759   .209
          Level 3 vs. Level 4      0.003     1     0.003     0.005   .946
Error     Level 1 vs. Later       11.213    12     0.934   
          Level 2 vs. Later        2.036    12     0.170 
          Level 3 vs. Level 4      7.772    12     0.643

This is a repeated-measures anova with 1 IV and 4 levels. I’ve double checked both the descriptives and the omnibus ANOVA, and both JASP and SPSS give the same results, with the only difference being the Helmert contrasts (hand-coded Helmert contrasts give the same result as the built-in contrast).

MattPeterson

I went and did the contrasts by hand, and came up with the same answers as JASP. Out of curiosity, I tried doing the contacts using a series of students-t tests in Excel, and came up with the same answer as SPSS.

I think I had a similar issue back in the days of SuperAnova: SuperAnova calculated contrasts correctly, but SPSS did not.

EJ

Hi Matt,

JASP borrows most of its functionality from R packages. I have not yet gotten a response from the person who implemented this in JASP, but I'm certain this comes from an R package. And it's nice to hear that it corresponds to what you did by hand.

Cheers,
E.J.