MANOVA with a varying covariate on three levels

[Peter Watson-7]

 

Dear all,

 

I wonder if someone can illuminate me on using MANOVA?

 

I have been using the MANOVA command (which lives on in syntax) for using a varying covariate (varying within subject) having three levels on a repeated measures factor with three levels (with no between subjects group factor). (I know I could use a mixed procedure but am exploring whether one could use simple regressions to obtain the same result as MANOVA for those who are more comfortable for example using EXCEL to do their analyses and might not want to use a more overtly stats package).

 

The orthogonal pairs of linear combinations printed by MANOVA are 0.707*y1 - 0.707*y3 (cy1), 0.408*y1 -0.816*y2 + 0.408*y3 (cy2) and 0.707*x1-0.707*x3 (cx1) and 0.408*x1 -0.816*x2 + 0.408*x3 (cx2). Now my understanding here was that one can obtain the same results  (sums of squares on MANOVA) by performing two regressions the first with cy1 as outcome, cx1 as predictor and the second with cy2 as outcome and cx2 as predictor and then adding the two sums of squares together (since they are orthogonal) to get the SS outputted by MANOVA for the regression (SS for the covariate with the intercept present), factor (SS for the intercept with the covariate present) and error (Residual SS with both intercept and covariate in model). This works for a repeated measures factor with three levels without a covariate and for a repeated measures factor with two levels with a varying covariate but I cannot get the SS to agree for when the repeated measures factor has three levels with a varying covariate. Perhaps it’s not possible to combine simple regressions to get the MANOVA SS for three level + RM with varying covariates?

 

I am obviously missing something but can't think what.

 

I can't find anything on this on the internet. I did find a reply from David Nichols in 1997 to the SPSSX-L list but this related only to the two level case.

 

So… any thoughts gratefully received,

 

With many thanks

 

Peter

 

---------------------------------------------------------------------

 

Peter Watson

MRC Cognition and Brain Sciences Unit

15 Chaucer Road

Cambridge

CB2 7EF

 

Direct line:  [phone]

Line (via reception): [phone] x801

Fax:  [phone]

 

Email secured by Check Point

[Art Kendall]

This is one of my soapbox posts, apologies to those of you who are tired of my making this point.

Using spreadsheets for stat can be like using a hammer to drive a screw.

IF I RECALL CORRECTLY any accounting system that uses computer spreadsheets cannot pass ISO certification. This is largely due to the lack of an audit trail.

I do use Excel Pro or tables in WordPerfect several times a week.
Sometimes it is practical to enter data via a spreadsheet and then read it into a stat package e.g., SPSS, for comparing double entry and other quality checks. [This is because spread sheets are usually available, but it not cost effective to provide  stat package to every person who will do data entry.]

Although these critiques are about EXCEL the same considerations can apply to other spread sheets.
For an overview of the issues
A video can be found at

http://www.spss.com/events/event.cfm?E_ID=2921&Country=US

A pdf can be found by clicking
<The Risks of Using Spreadsheets in Data Analysis>


For more technical review see.

http://www.pages.drexel.edu/~bdm25/excel-intro.pdf

http://www.pages.drexel.edu/~bdm25/excel2007.pdf

http://www.pages.drexel.edu/~bdm25/excel-rng.pdf

an instructional link is
http://www.umass.edu/statdata/software/handouts/excel.html.
On 2/28/2014 8:24 AM, Peter Watson-7 [via SPSSX Discussion] wrote:

 

Dear all,

 

I wonder if someone can illuminate me on using MANOVA?

 

I have been using the MANOVA command (which lives on in syntax) for using a varying covariate (varying within subject) having three levels on a repeated measures factor with three levels (with no between subjects group factor). (I know I could use a mixed procedure but am exploring whether one could use simple regressions to obtain the same result as MANOVA for those who are more comfortable for example using EXCEL to do their analyses and might not want to use a more overtly stats package).

 

The orthogonal pairs of linear combinations printed by MANOVA are 0.707*y1 - 0.707*y3 (cy1), 0.408*y1 -0.816*y2 + 0.408*y3 (cy2) and 0.707*x1-0.707*x3 (cx1) and 0.408*x1 -0.816*x2 + 0.408*x3 (cx2). Now my understanding here was that one can obtain the same results  (sums of squares on MANOVA) by performing two regressions the first with cy1 as outcome, cx1 as predictor and the second with cy2 as outcome and cx2 as predictor and then adding the two sums of squares together (since they are orthogonal) to get the SS outputted by MANOVA for the regression (SS for the covariate with the intercept present), factor (SS for the intercept with the covariate present) and error (Residual SS with both intercept and covariate in model). This works for a repeated measures factor with three levels without a covariate and for a repeated measures factor with two levels with a varying covariate but I cannot get the SS to agree for when the repeated measures factor has three levels with a varying covariate. Perhaps it’s not possible to combine simple regressions to get the MANOVA SS for three level + RM with varying covariates?

 

I am obviously missing something but can't think what.

 

I can't find anything on this on the internet. I did find a reply from David Nichols in 1997 to the SPSSX-L list but this related only to the two level case.

 

So… any thoughts gratefully received,

 

With many thanks

 

Peter

 

---------------------------------------------------------------------

 

Peter Watson

MRC Cognition and Brain Sciences Unit

15 Chaucer Road

Cambridge

CB2 7EF

 

Direct line:  [phone]

Line (via reception): [phone] x801

Fax:  [phone]

 

Email secured by Check Point

---

If you reply to this email, your message will be added to the discussion below:

http://spssx-discussion.1045642.n5.nabble.com/MANOVA-with-a-varying-covariate-on-three-levels-tp5724668.html

To start a new topic under SPSSX Discussion, email [hidden email]
To unsubscribe from SPSSX Discussion, click here.
NAML