calculating eta-squared for MANOVA with one IV

[Rayna Berlin]

I am trying to calculate eta-squared for a MANOVA with one factor/IV and multiple DVs, using the syntax below. I'm using my ouput to manually calculate eta-squared--or trying to, at least! I'm trying to calculate eta-squared by

 SSb/SSt (the sum of SSb plus SS error), which is the same formula for calculating partial eta-squared, and indeed my manual calculations match the output for partial eta-squared. However, I don't want partial eta-squared. I think I read somewhere that with one IV these will be the same, but the total is greater than one, which doesn't make sense for eta-squared.

 

Any ideas on what I'm doing wrong? I don't have a MANOVA example for calculating eta-squared, and clearly whatever I'm doing isn't right. I need to provide eta-squared, not partial eta-squared. With only one IV/Factor, the only values I have for SSt are SSb and SSerror, whereas my ANOVA example has SS values for two IVs and their interaction, and the SS for error is included as well.

 

GLM V1 V2 V3 V4 V5 V6 V7 BY GROUP

/METHOD=SSTYPE(3)

/INTERCEPT=INCLUDE

/PRINT=ETASQ OPOWER

/CRITERIA=ALPHA(.05)

/DESIGN= REG_DIS_O.

[Swank, Paul R]

Multivariate R squared is 1 – Wilks’ lambda.

 

Dr. Paul R. Swank,

Professor and Director of Research

Children's Learning Institute

University of Texas Health Science Center-Houston

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Rayna Berlin
Sent: Saturday, February 20, 2010 4:39 PM
To: [hidden email]
Subject: calculating eta-squared for MANOVA with one IV

 

I am trying to calculate eta-squared for a MANOVA with one factor/IV and multiple DVs, using the syntax below. I'm using my ouput to manually calculate eta-squared--or trying to, at least! I'm trying to calculate eta-squared by

 SSb/SSt (the sum of SSb plus SS error), which is the same formula for calculating partial eta-squared, and indeed my manual calculations match the output for partial eta-squared. However, I don't want partial eta-squared. I think I read somewhere that with one IV these will be the same, but the total is greater than one, which doesn't make sense for eta-squared.

 

Any ideas on what I'm doing wrong? I don't have a MANOVA example for calculating eta-squared, and clearly whatever I'm doing isn't right. I need to provide eta-squared, not partial eta-squared. With only one IV/Factor, the only values I have for SSt are SSb and SSerror, whereas my ANOVA example has SS values for two IVs and their interaction, and the SS for error is included as well.

 

GLM V1 V2 V3 V4 V5 V6 V7 BY GROUP

/METHOD=SSTYPE(3)

/INTERCEPT=INCLUDE

/PRINT=ETASQ OPOWER

/CRITERIA=ALPHA(.05)

/DESIGN= REG_DIS_O.

[MaaikeSmits]

I want to rapport the percentage of explained variance from the output of a MANOVA (SPSS 21.0) that I ran on a clustersolution with two clusters (clustermembership a independent VAR) and 10 dependent variables. In your earlier reply you stated that
1) multivariate r-squared is 1-wilk's lambda.

I first used this calculation, but now I read the following:

2) Multivariate eta squared = 1- wilks' lambda ^ (1/s). In which s is the number of levels of the factor minus 1 or the number of dependent VAR, whichever is smaller.

When I calculate the latter, then the result compares with the PARTIAL eta squared column that is shown in the Multivariate test SPSS table . I am wondering now which calculation I should use to inform me on the percentage variance in the dependent Variabels that is explained by the clustermembership.

Can I state:
* The first calculation provides the multivariate variant of R-squared (which can in the case be used interchangeable with eta squared?), meaning the proportion of the total variance (on all 10 dependent VAR) that is explained by the variabele (in my case clustermembership).
* The second calculation provides PARTIAL eta squared which differs from R-squared /eta-squared in the sense that this is the proportion of the total variance that is explained by the independent VAR (clustermembership) AND that is not explained by other variables in the analysis?


I would very much appreciate you help.