 |
 |
Dear Elton, At 12:23 22/01/99 -0800, Elton Ngan wrote: | We have a two group single scan design with age as a covariate of no | interest. | | What is the appropriate design for this analysis? | | The groups single scan per subject design does not allow for the | covariate to be enter. | | The Multistudy covariate design does not allow for entry of any | contrast between the studies. | | The multistudy conditions and covariate design ( 1 condition per study) | results in the suitable design matrix parameters " 2 conditions +0 | covariate + 1 Block + 1 confound". with the two condition being Study1 | scan 1 and study 2 scan1. | But when I enter the contrast as -1 1 and 1 ,-1 | I get F statistics map but the contrast are not enter | and we get the error message | "Divide by Zero... | ... error in .....spm_spm_ui.m | on line 192 if bMCond, nCond = spm_input ([sStudP,'# of conditions? | ??? Error evaluating call back string" At 15:55 22/01/99 -0800, Elton Ngan wrote: | I probably should have added that we have 18 subjects per group | (therefore the divide by zero error is not the result of insufficient # | of scans - 3 parameters, having 2 degrees of freedom, giving 34 residual | df (36 scans)) Indeed the "Compare groups: 1 scan per subject" design in SPM96 doesn't offer the option of including a covariate. Attempting to use the "MultiStudy" designs gets you into problems, as you found. In your case the contrasts you entered are illegal for a normal multi-study design with repeated measurements, and SPM discards them, leaving you with no contrasts! You can get round this in two ways: You can edit spm_spm_ui.m to ask you for covariates, by changing last line of the "DesDefaults" matrix on line 103 (spm_spm_ui.m v1.15.1.2) so that the sixth element (bAskCov) is 1 rather than 0, so the line becomes: 1, 1, 0, 0, 6, 1, 0, 123 ]; %-4 Alternatively (& easier), use the "Single-subject: conditions and covariates" design. Choose two conditions, corresponding to the two groups. Read group for "condition" when prompted to enter scans. (Note that the Compare groups: 1 scan per subject" design is almost the same as using the "Single-subject: conditions only design, except that the opton of grand mean scaling by group is offered in the former.) Note that if there is a marked difference in the average age between the two groups, i.e. if the covariate is correlated with group membership, then the group and covariate effects are confounded. In this case you must interpret the effect of either with caution, preferably including plots. Hope this helps, -andrew +- Dr Andrew Holmes ------------------ mailto:[log in to unmask] -+ | ___ __ ___ Robertson Centre for Biostatistics | | ( ,) / _)( ,) Boyd Orr Building, | | ) \( (_ ) ,\ University Avenue, | | (_)\_)\__)(___/ Glasgow. G12 8QQ Scotland, UK. | +----------------------------------------- http://www.rcb.gla.ac.uk/ -+ %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%