 |
 |
>Date: Fri, 16 Mar 2001 17:39:08 -0500 >To: Steve Berman <[log in to unmask]> >From: Badreddine Bencherif <[log in to unmask]> >Subject: Re: Thanks for your response >Cc: >Bcc: >X-Attachments: > >Dear Steve: >When you do the analysis a mat files is created (SPMcfg.mat) >do the following at the Matlab prompt > >1)Load SPMcfg.mat >then >2) xGX,rg >this gives you raw values >type >3)xGX.gSF >this gives you the scaling factor (as you see they are all DIFFERENT). >Use an excel spreadsheet and multiply each element of xGX,rg by >xGX.gSF ( #1 by #1 , 2 by 2, .... etc) >then If you have choosen the 50 default for your analysis, you >should get 50 at each data point. >I suspect this is the normal behavior you should expect although I >do not have much experience with proportional scaling. > >Hope I answered your question >Didine > >ps: When using AnCova, and some scaling let say mean scaling to 1, >the scaling factor is the same for all values. > > > >>At 02:43 PM 3/16/01 -0500, you wrote: >>>>Date: Fri, 16 Mar 2001 14:42:15 -0500 >>>>To: "SUYENOBU,BRANDALL" <[log in to unmask]> >>>>From: Badreddine Bencherif <[log in to unmask]> >>>>Subject: Re: more bad global scaling factors >>>>Dear Brandall: >>>>suppose a volume data can be shown as 8 bits (or min 0 max 255). >>>>and for this volume data has the real max (computed by Matlab or >>>>CTRL ALT C with Analyze) is for example 4.3554 and min 0 >>>>The scaling factor for the set of data 0.01708 since 255*0.01708= 4.3554. >>>> >>>>Now in the statistical analysis for scaling of global mean to 50 >>>>(CBF) or other physiologically relevant values. How it is done >>>>lets global means of 5 subjects be 3, 3.7, 4.2, 5, 5.1 (you can use >>>>your values since I do not have global means for CBF I just choose >>>>these values randomly) >>>>first compute average of 3, 3.7, 4.2, 5, 5.1 = 4.26 then divide each >>>>value by 4.26 >>>>3/4.26= 0.77, 3.7/4.26=0.86, 4.2/4.26=0.98, 5/4.26=1.17 and >>>>5.1/4.26=1.19 (the average of these new values then is 1) >>>>then multiply each of these values by 50 you obtain 38,73, 43.42, >>>>49.29, 58.68 and 59.85. >>>>their average is now 50. They are scaled so their new mean is 50. It >>>>does not change your data at all. >>>>In this case the scaling factor is 11.73 since 3 *11.73=38.73, >>>>3.7*11.73=43.42, 4.2*11.73=49.29, 5* 11.73= 58.68 and 5.1*11.73=59.85. >>>>Hopefully I answered your question although someone may be able to >>>>make it shorter. >>>>Didine >> >>Dear Didine, >>Thanks for your response to our question. While your explanation >>is clear, shouldn't the global column of the SPM "Files and >>Covariates" page list the final *50 values (38.73, 43.42, 49.29, >>58.68 and 59.85 in your example)? It always seemed to do this in >>SPM 96, so we were surprised when all the global values listed were >>nowhere near averaging 50 when we used proportional scaling to 50 >>(default) in SPM99. Could it have anything to do with using a >>subject-by-condition interaction design? I didn't use this model >>in SPM96. >>Thanks again, -- Badreddine Bencherif, MD Department of Radiology Division of Nuclear Medicine Johns Hopkins University School of Medicine 601 N. Caroline St. / JHOC 4230 Baltimore, MD 21287-0855 Phone : (410) 614-2787 Pager : (410) 283-2050 Fax : (410) 614-1977 email : [log in to unmask]