Different Box's M results in JASP and SPSS

MaddieP

Hi everyone,

I have been getting different results in JASP and SPSS for Box's M test.

For example:

In SPSS, X2 = 28.033, p = .390

• df1 = 21 AND df2 = 3670.442

In JASP, X2 = 22.339, p = .380

• df = 21

In this case, other MANOVA results were the same in both SPSS and JASP. However, I have also experienced times when the results of the f-tests were also different.

Has anyone else ever had these problems?

EJ

Hmm. Could you attach an example with concrete data? JASP uses R under the hood -- have you looked to see whether the common procedure in R gives a different result from SPSS?

EJ

JohnnyB

Hi @MaddieP,

Could you please provide a data set where you obtain different results? I just checked against the heplots package (https://search.r-project.org/CRAN/refmans/heplots/html/boxM.html), and get the same results in JASP.

Kind regards

Johnny

MaddieP

@JohnnyB @EJ

https://drive.google.com/file/d/1kIy5gEa5WOjmfvdQtI-I5Jo1cakckEmW/view?usp=share_link

JohnnyB

Thanks @MaddieP , what was the model you used? in terms of (in)dependent variables.

MaddieP

@JohnnyB We used location and year as independent variables and "affectstrength" and "familySES" were the dependents! Thanks so much.

We have also noticed differences in factor analysis results (for a different data set). Not sure if you have ever seen this either.

JohnnyB

Hi @MaddieP ,

Thanks! The JASP computation (https://github.com/jasp-stats/jaspAnova/blob/ae3f25859919c9a91a19832ae8f09065399f2c84/R/manova.R#L246) is based on the BioTools R package, but also produces the same results as the HePlots package:

> heplots::boxM(cbind(FamilySES, AffectStrength) ~ factor(Location)* factor(Year), data = dd)

   Box's M-test for Homogeneity of Covariance Matrices

data: Y

Chi-Sq (approx.) = 22.339, df = 21, p-value = 0.3802

I am not sure what code SPSS uses, so I cannot say why their result differs.

Kind regards

Johnny

JTPetter

Hello Maddie,

the X2 you reported for the SPSS output is not actually the X2 statistic, but the Box's M statistic. SPSS does not use a Chi-Square distribution but an F distribution to determine the significance of the Box's M statistic. So SPSS does not report an X2 statistic but an F statistic (= 1.056 for your dataset). For some background on how to go from M to either F or X2, see here: https://real-statistics.com/multivariate-statistics/boxs-test/boxs-test-basic-concepts/. I could not find a quick answer on which way is preferred, but it seems that using the Chi-Square Distribution for Box's M is more common.

Hope this helps!

Best,

Jonas