Hi all, I have been advised to learn and use the new procedure in SPSS, linear autoregressive mixed models. I have looked at what docs are available, and I'm having trouble understanding when I can use this analysis:
In this psychology experiment on exectuive processing, subjects say whether consecutive numbers presented on screen are odd/even (OE), or >5/<5 (HL). Sometimes the same task is performed on consecutive trials and sometimes subjects switch from one to the other. On trials following a switch, RT is slower, and this switch cost is thought to reflect executive processes redirecting attention between tasks. On some trials we have them perform the task on a number in memory rather than the one onscreen. So we have a repeated measures, 2x2x2 design. N = 20 in all cells. Order of trials randomized. Additionally, we record the trialnumber as a non-factor, for purposes of record-keeping. Data looks like: subject switch? task intern? trial# ReactionTime(DV) 1 N OE I 1 854 1 S OE E 2 456 1 S HL E 3 657 1 S OE I 4 767 1 N HL I 5 645 . . . . . . . . . . . . . . . . . . 1 S OE E 160 675 2 N OE I 1 576 2 S OE E 2 756 2 S HL E 3 654 2 S OE I 4 879 2 N HL I 5 459 . . . . . . . . . . . . . . . . . . 2 S OE E 160 765 and so on for each subject. We are interested in two approaches: i) what are main effects and interactions of switch, task, and intern? ii) it may be that performance degrades over time due to fatigue. The counterbalanced order should take care of this to an extent, but it would be nice to know how performance changed over time. More importantly it would be good to know if change over time was different in different cells. I suspect a mixed models autoregression approach using trial as an index might be called for in ii), but is it OK to use the same analysis for question i) which is not concerned with change over time? Is this type of analysis ONLY appropriate when the time covariate is involved? No doubt I have munged the terminology and confabulated terms from SPSS with actual statistical terms, my apology. One reason I would like to use the mixed models approach on i) if possible is not really statistical, it would save enormous time as (I think) it obviates the necessity of reoragnizing the entire data file into the structure required by SPSS for repeated measures analyses, and since I will be doing many such analyses, the savings would amount to days at least. Thanks for any pointers.... Jim Watch out for spam block . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
