Log in
RegisterEMA crippled method - R peculiarities [RSABE / ABEL]
Hi ElMaestro,
Exactly. This is what the EMA wants us to do. At least in the context of calculation of the intra-subject variability for the reference.
But also in the evaluation of 3-treatment-3-period or 4-treatment-4-period they request in the same spirit " ... In studies with more than two treatment arms (e.g. a three period study including two references, one from EU and another from USA, or a four period study including test and reference in fed and fasted states), the analysis for each comparison should be conducted excluding the data from the treatments that are not relevant for the comparison in question."
Simple Answer: Someone (my boss, my sponsor, my spouse, a NLYW ...) has placed an order to me to do it. And I have to respond under threat of punishment
.
This may be true. But my SAS code tells me further that the difference R1-R2 (and the 90% CI) is not estimable!
And it seems not only a type III issue and not a SAS phenomenon:
(PKdata contains the data for Reference only, factors created prior to lm() call.)
Note the varying df's for period!
The same happens in models with intercept. Weird I think. The same strong dependence on the order of the effects I had noted already in this post.
Ways out? Abandon EMA's crippled method?
❝ Did you just mean to prune away everything related to T1 and T2 and then run an analysis*?
Exactly. This is what the EMA wants us to do. At least in the context of calculation of the intra-subject variability for the reference.
But also in the evaluation of 3-treatment-3-period or 4-treatment-4-period they request in the same spirit " ... In studies with more than two treatment arms (e.g. a three period study including two references, one from EU and another from USA, or a four period study including test and reference in fed and fasted states), the analysis for each comparison should be conducted excluding the data from the treatments that are not relevant for the comparison in question."
❝ It becomes more and more unclear to me what you actually wish to accomplish. I still haven't got the faintest clue why you wish to evaluate Ref against itself.
Simple Answer: Someone (my boss, my sponsor, my spouse, a NLYW ...) has placed an order to me to do it. And I have to respond under threat of punishment
.❝ If you use drop1(C, test="F") to get type III SS then you get basically the anova you pasted above. But we are only interested in the effect estimates not the source of variation, and the residual stays the same. So forget the type III SS approach - it is not necessary.
This may be true. But my SAS code tells me further that the difference R1-R2 (and the 90% CI) is not estimable!
And it seems not only a type III issue and not a SAS phenomenon:
(PKdata contains the data for Reference only, factors created prior to lm() call.)
# Anders order of effects
> model1 <- lm(log(Cmax) ~ 0 + tmt2 + subject + period + sequence, PKdata)
> anova(model1) # type I sum of squares
Analysis of Variance Table
Response: log(Cmax)
Df Sum Sq Mean Sq F value Pr(>F)
tmt2 2 8839.6 4419.8 22175.1817 < 2.2e-16 ***
subject 76 119.3 1.6 7.8755 2.798e-16 ***
period 1 0.4 0.4 2.0046 0.1612
Residuals 71 14.2 0.2
Sequence gone!
> model2 <- lm(log(Cmax) ~ 0 + subject + period + sequence + tmt2, PKdata)
> anova(model2)
Analysis of Variance Table
Response: log(Cmax)
Df Sum Sq Mean Sq F value Pr(>F)
subject 77 8958.6 116.345 583.7277 <2e-16 ***
period 2 0.8 0.386 1.9383 0.1515
Residuals 71 14.2 0.199
Sequence gone! tmt2 vanished (coef() for tmt2R2=NA)!Note the varying df's for period!
The same happens in models with intercept. Weird I think. The same strong dependence on the order of the effects I had noted already in this post.
Ways out? Abandon EMA's crippled method?—
Regards,
Detlew
Regards,
Detlew
Complete thread:
- Ref. vs. Ref. in partial replicate design d_labes 2012-04-25 15:52 [RSABE / ABEL]
![[Mix]](https://static.bebac.at/img/mix.png "open in Mix view")
- Ref. vs. Ref. in partial replicate design again d_labes 2012-10-10 10:56
- Ref. vs. Ref. in partial replicate design ElMaestro 2012-10-10 12:39
- Ref. vs. Ref. in partial replicate design d_labes 2012-10-10 16:26
- Ref. vs. Ref. in partial replicate design ElMaestro 2012-10-10 17:16
- R vs. R in partial replicate design d_labes 2012-10-11 09:43
- R1 vs R2 ElMaestro 2012-10-11 10:26
- R1 vs R2 d_labes 2012-10-11 12:01
- R1 vs R2 ElMaestro 2012-10-11 12:09
- R1 vs R2 d_labes 2012-10-11 12:46
- treatment df ElMaestro 2012-10-11 13:38
- Missunderstanding d_labes 2012-10-11 16:34
- Missunderstanding, crippled ElMaestro 2012-10-11 18:12
- EMA crippled method - R peculiaritiesd_labes 2012-10-12 14:54
- All is good, just forget ANOVA - you don't need it! ElMaestro 2012-10-12 16:25
- Just forget ANOVA. What next? d_labes 2012-10-15 08:35
- Just forget ANOVA. What next? ElMaestro 2012-10-15 10:04
- Just forget ANOVA. What next? d_labes 2012-10-15 08:35
- All is good, just forget ANOVA - you don't need it! ElMaestro 2012-10-12 16:25
- EMA crippled method - R peculiaritiesd_labes 2012-10-12 14:54
- Missunderstanding, crippled ElMaestro 2012-10-11 18:12
- Missunderstanding d_labes 2012-10-11 16:34
- treatment df ElMaestro 2012-10-11 13:38
- R1 vs R2 d_labes 2012-10-11 12:46
- R1 vs R2 ElMaestro 2012-10-11 12:09
- R1 vs R2 d_labes 2012-10-11 12:01
- R1 vs R2 ElMaestro 2012-10-11 10:26
- R vs. R in partial replicate design d_labes 2012-10-11 09:43
- Ref. vs. Ref. in partial replicate design ElMaestro 2012-10-10 17:16
- Ref. vs. Ref. in partial replicate design d_labes 2012-10-10 16:26
Ing. Helmut Schütz