The big question with the unbalanced design is, is the apparent nonindependence between/among classification variables due to random loss of scores in some cells or due to those variables actually being related in the population of interest -- and, related to that, what to do about it -- the usual solution is just to ignore the variance that is confounded (although eliminating it from the error term).
Cheers, Karl L. Wuensch -----Original Message----- From: Stuart McKelvie [mailto:[email protected]] Sent: Thursday, December 15, 2011 6:46 AM To: Teaching in the Psychological Sciences (TIPS) Subject: RE: [tips] SPSS Question for Statistical Tipsters Dear Mike, Thanks for your helpful information. I also thank Jim Clark for looking at at printout. The distinction seems to hinge on whether weighting is taken into account. Sincerely, Stuart _____________________________________________________ Sent via Web Access "Floreat Labore" "Recti cultus pectora roborant" Stuart J. McKelvie, Ph.D., Phone: 819 822 9600 x 2402 Department of Psychology, Fax: 819 822 9661 Bishop's University, 2600 rue College, Sherbrooke, Québec J1M 1Z7, Canada. E-mail: [email protected] (or [email protected]) Bishop's University Psychology Department Web Page: http://www.ubishops.ca/ccc/div/soc/psy " Floreat Labore" _______________________________________________________ ________________________________________ From: Michael Palij [[email protected]] Sent: 14 December 2011 19:17 To: Teaching in the Psychological Sciences (TIPS) Cc: Michael Palij Subject: Re: [tips] SPSS Question for Statistical Tipsters I have not seen an SPSS source on this point but there are a few discussions about it by SAS users. One source is the following: http://support.sas.com/onlinedoc/913/getDoc/en/statug.hlp/glm_sect34.htm What SAS calls "LS means" or "Least Squares means" is equivalent to SPSS' estimated marginal means (emm). Ordinary means are referred to as "arithmetic means"; see the following for the distinction: http://onbiostatistics.blogspot.com/2009/04/least-squares-means-marginal-means-vs.html If you have a "balanced" design (e.g., a factorial design with constant N for each cell), then the LS/emm means will be the same as the arithmetic means. In an unbalanced design, the type of sums of square calculation you use can produce differences between the two (e.g., Type I SS or sequential SS vs.Type III SS or unweighted means). See the following for a worked example: http://www.public.iastate.edu/~dnett/S402/wlsmeans.pdf -Mike Palij New York University [email protected]<mailto:[email protected]> ------------------- Original Message ------------------------ On Wed, 14 Dec 2011 12:09:11 -0800, Jim Clark wrote: Hi I have not observed this problem and, as Stuart observed, generally see EMMs that differ from originals when covariates are involved. One thing to check might be whether it occurs with unequal ns per condition / cell? Could SPSS be estimating effects controlling for any confounding between factors? If you have an example of this issue, I would be happy to see it. Take care Jim >>> Stuart McKelvie <[email protected]<mailto:[email protected]>> >>> 14-Dec-11 12:38 PM >>> Dear Tipsters, I have puzzled over this question for a long time. When conducting an ANOPVA in SPSS, we have two options for obtaining means in each condition. 1. Click on descriptives. This gives means, standard deviations and sample sizes for all main effects and interactions. They appear at the beginning of the printout. 2. Under options, you can specify which means you want. This gives means and standard errors for the effects and interactions that you specify. They appear at the end of the printout. My question is: Why are these means sometimes different from the ones in the descriptives in 1? I think I have seen the term "estimated marginal means", with a reference to covariates, but if there are no covariates (it is ANOVA not ANCOVA), I do not understand this. I would have thought that a mean was a mean in ANOVA. --- You are currently subscribed to tips as: [email protected]<mailto:[email protected]>. To unsubscribe click here: http://fsulist.frostburg.edu/u?id=13510.2cc18398df2e6692fffc29a610cb72e3&n=T&l=tips&o=14893 (It may be necessary to cut and paste the above URL if the line is broken) or send a blank email to leave-14893-13510.2cc18398df2e6692fffc29a610cb7...@fsulist.frostburg.edu<mailto:leave-14893-13510.2cc18398df2e6692fffc29a610cb7...@fsulist.frostburg.edu> --- You are currently subscribed to tips as: [email protected]. To unsubscribe click here: http://fsulist.frostburg.edu/u?id=13060.c78b93d4d09ef6235e9d494b3534420e&n=T&l=tips&o=14902 or send a blank email to leave-14902-13060.c78b93d4d09ef6235e9d494b35344...@fsulist.frostburg.edu --- You are currently subscribed to tips as: [email protected]. To unsubscribe click here: http://fsulist.frostburg.edu/u?id=13090.68da6e6e5325aa33287ff385b70df5d5&n=T&l=tips&o=14905 or send a blank email to leave-14905-13090.68da6e6e5325aa33287ff385b70df...@fsulist.frostburg.edu
