"Donald Burrill" <[EMAIL PROTECTED]> schreef in bericht news:[EMAIL PROTECTED] > On 22 Aug 2003, Pingu wrote in part:
<snip> > In terms of these category labels (1) - (4), your two factors are > dialling type (VD or TTS) and background noise level (noisy or quiet), > which I'll abbreviate as DT and BNL: > > 1) Voice dialling (VD) + quiet DT = VD BNL = quiet > 2) VD + noise DT = VD BNL = noisy > 3) TTS dialling + quiet DT = TTS BNL = quiet > 4) TTS dialling + noise DT = TTS BNL = noisy > > (although I presume you've coded those category values as (1,2) or > (0,1): for example, DT = 1 if VD, DT = 2 if TTS, and so on.) > > One thing you ought to do, fairly early on, is to find the means and > variances (or std. deviations) for each of your four cells [(1) to (4) > above], for each dependent variable you want to analyze. Of course, the > means are what you're mostly interested in, but you need to know if the > variances are reasonably similar in all four cells. As Jos remarked, > this may not be true in your case. If it be reasonable to assume > approximately equal variances, tests based on the ANOVA within-cell > variance will be more stable (and possibly more sensitve) than your > t-tests are, owing to the increased number of degrees of freedom > associated with the error variance -- and your tutors are correct. Not the within-cell-between-subjects variances, but the within-subjects variances are relevant for the comparison of the treatments. These can be estimated by first calculating differences (more generally contrasts) between treatments within subjects, and subsequently calculating variances of these differences (or contrasts) over subjects. Equivalently, by calculating analyses of variance, including subjects as a blocking factor in the model. > If the variances are distinctly different -- suppose, e.g., that > variances in the two noisy conditions are similar and much larger than > variances in the two quiet conditions, which themselves are similar -- > then you may need to take that fact into account; possibly by using the > t-test based on different variances for comparisons between "noisy" and > "quiet", Calculation of variances, based on differences (contrasts), accounts for differences in variance between treatments. > and using the t-test based on equal variances for comparisons > between "TST" and "VD". At this point you probably actively need the > advice of your tutors, or at least of someone in the same room with you > and your SPSS protocol and output. > > > On the data i collected i did the following (all t tests): > > And did you adjust the significance level of the t-tests for the > multiple tests (e.g., via the Bonferroni correction)? There is no need to do so. All sensible tests of differences may be considered to be planned beforehand. > > 1) speed of successful dials, VD vs TTS, with quiet > > 2) speed of succesful dials, VD vs TTS, with noise > > > > etc... for aspects A, B, C listed above. > > > > For D (effect of background noise) i did: > > > > 1) Speed of succesful dials, TTS in quiet vs TTS in noise > > 2) Speed of succesful dials, VD in quiet vs VD in noise > > etc... for error rates and speed of dialling including errors > > < snip, the rest > . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
