Lessons learned? [Regulatives / Guidelines]
Hi all,
quick note:
There is a sequence effect to be fit, no doubt about that.
The fact that we boil TRTR and RTRT down to RR is not the direct culprit here as treatment is not considered. But as was said by Helmut there is an interaction or nesting of period in sequence. So just as is the case with subejcts, whenever period is in the Muddle, any type III anova goes bonkers.
Compare the two sequential ones:
Remember earlier discussions about three-treatment designs where EMA occasionally does not like that the residual is polluted by an irrelevant treatment A when comparison of treatment B and C is done. In those cases the dataset is pruned down to a "pseudo" two-treatment, two-sequence, two-period dataset and analysed the usual way. We could do the same to periods here and get rid of the ugly 'issue', I think. I don't think this will change the residual SS (which is what is important), however, but I haven't done it or thought it fully through.
quick note:
There is a sequence effect to be fit, no doubt about that.
The fact that we boil TRTR and RTRT down to RR is not the direct culprit here as treatment is not considered. But as was said by Helmut there is an interaction or nesting of period in sequence. So just as is the case with subejcts, whenever period is in the Muddle, any type III anova goes bonkers.
Compare the two sequential ones:
Muddle=lm(log(b$Data)~factor(b$Period) + factor(b$Sequence))
anova(Muddle)
Analysis of Variance Table
Response: log(b$Data)
Df Sum Sq Mean Sq F value Pr(>F)
factor(b$Period) 3 1.57 0.52326 0.5752 0.6322
Residuals 146 132.81 0.90968
Muddle=lm(log(b$Data)~factor(b$Sequence) + factor(b$Period))
anova(Muddle)
Analysis of Variance Table
Response: log(b$Data)
Df Sum Sq Mean Sq F value Pr(>F)
factor(b$Sequence) 1 0.088 0.08801 0.0968 0.7562
factor(b$Period) 2 1.482 0.74088 0.8144 0.4449
Residuals 146 132.813 0.90968
Remember earlier discussions about three-treatment designs where EMA occasionally does not like that the residual is polluted by an irrelevant treatment A when comparison of treatment B and C is done. In those cases the dataset is pruned down to a "pseudo" two-treatment, two-sequence, two-period dataset and analysed the usual way. We could do the same to periods here and get rid of the ugly 'issue', I think. I don't think this will change the residual SS (which is what is important), however, but I haven't done it or thought it fully through.
—
Pass or fail!
ElMaestro
Pass or fail!
ElMaestro
Complete thread:
- EMA replicate surprise d_labes 2011-08-23 13:01 [Regulatives / Guidelines]
- EMA replicate surprise ElMaestro 2011-08-23 13:15
- Big bäng CV d_labes 2011-08-23 13:58
- Big bäng CV ElMaestro 2011-08-23 15:02
- Period DFs d_labes 2011-08-23 15:28
- Period DFs Helmut 2011-08-23 15:36
- Period DFs ElMaestro 2011-08-23 16:26
- Period DFs ElMaestro 2011-08-23 16:58
- Period DFs Helmut 2011-08-24 00:44
- Lessons learned? d_labes 2011-08-24 11:24
- Lessons learned? Helmut 2011-08-24 13:51
- Lessons learned?ElMaestro 2011-08-28 21:34
- Lessons learned? d_labes 2011-08-24 11:24
- Period DFs - EMA code literally in R d_labes 2011-08-24 10:47
- Period DFs - EMA code literally in R ElMaestro 2011-08-24 13:04
- call for expert’s opinions Helmut 2011-08-24 14:01
- call for expert’s opinions jdetlor 2011-11-01 03:05
- The finance sector ElMaestro 2011-11-02 09:48
- call for expert’s opinions jdetlor 2011-11-01 03:05
- call for expert’s opinions Helmut 2011-08-24 14:01
- Period DFs - EMA code literally in R ElMaestro 2011-08-24 13:04
- Period DFs Helmut 2011-08-24 00:44
- Period DFs d_labes 2011-08-23 15:28
- Big bäng CV ElMaestro 2011-08-23 15:02
- Big bäng CV d_labes 2011-08-23 13:58
- EMA replicate surprise ElMaestro 2011-08-23 13:15