Type II sum of squares ANCOVA

[Joost van Ginkel]

Dear list,

 

I am currently trying to program (M)AN(C)OVA for a project that I am working on. I am experiencing some problems programming the type-II sum of squares. For simple cases (e.g., two-way designs, no covariates) the literature that I found explains pretty well how they are computed, but for higher-order designs with covariates and possible interactions between the covariates and the factors (or among covariates), I found no literature. When trying to generalize the mechanism that the literature gives for simple designs I have trouble reproducing the results that SPSS provides. With trial and error I managed to reproduce the type-II sum of squares of some of the effects, but I can’t seem to find any general mechanism that could underlie the computations. I have looked at the algorithms in SPSS as well (including the references that were used) but that didn’t make things much clearer, either. Is there anyone who can tell me (preferably at a conceptual level) how SPSS calculates the type-II sum of squares in such MAN(C)OVAs with covariates and interactions of covariates (including the sum of squares of the intercept since I can find no literature on that either). Or can anyone recommend any literature on this topic that would give me the answer? Thank you very much in advance.

 

Best regards,

 

Joost van Ginkel

 

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD

[Kirill Orlov]

Did you consult with the SPSS Algorithms document? It is found in Help  menu and, as pdf file, somewhere in the intallation folder.


===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD

[Joost van Ginkel]

Yes I did, but it wasn’t really insightful. Both the algorithms document in SPSS and the corresponding references assume a lot of pre-knowledge from the part of the reader (or at least more pre-knowledge than I have). Besides, I think it would be more helpful if someone explained it to me at a conceptual level, rather than at the technical level, like the algorithms document does. In other words: per effect, what other effects (covariates, factors, interactions among covariates, interactions among factors, interactions between covariates and factors) does the type-II sum of squares correct for?

 

From: Kirill Orlov [mailto:[hidden email]]
Sent: woensdag 18 november 2015 15:06
To: Ginkel, Joost van; [hidden email]
Subject: Re: Type II sum of squares ANCOVA

 

Did you consult with the SPSS Algorithms document? It is found in Help  menu and, as pdf file, somewhere in the intallation folder.

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD