Log in
RegisterOops. Oops. [Regulatives / Guidelines]
Dear Helmut!
Quite a time ago I had promised to come out with a simulation including subject variability and intra-subject correlation.
Due to spare time I could not until now consider this.
Before posting my results I would like to compare it to your code.
Inspecting it I have some doubts:
Must the EMA guidance interpreted in this way, i.e. analysing the dataset with 'pseudo' periods and sequences? Or should it interpreted the way which is employed in the Q&A for the replicate design (intra-subject variability of the reference). That means only massacre out the data concerning R2 but retaining the original periods and sequences.
At least the code for FMMSE has stolen my sleep
. According to the help page of
Check this out via
Or did I miss sumfink here?
Quite a time ago I had promised to come out with a simulation including subject variability and intra-subject correlation.
Due to spare time I could not until now consider this.
Before posting my results I would like to compare it to your code.
Inspecting it I have some doubts:
❝ # massacrating out everything related to R2
❝ # means that when T was before R1 we now have a seq 1 otherwise a seq 2
❝ seq2 <- as.factor(rep(c(1,1, 2,2, 1,1, 1,1, 2,2, 2,2), size/6))
❝ sub2 <- as.factor(rep(1:size, each = 2))
❝ per2 <- as.factor(rep(1:2, size))
❝ trt2 <- as.factor(rep(c(1,2, 2,1, 1,2, 1,2, 2,1, 2,1), size/6))
Must the EMA guidance interpreted in this way, i.e. analysing the dataset with 'pseudo' periods and sequences? Or should it interpreted the way which is employed in the Q&A for the replicate design (intra-subject variability of the reference). That means only massacre out the data concerning R2 but retaining the original periods and sequences.
❝ FullModel <- lm(log(y1) ~ 0 + seq1 + sub1 %in% seq1 + per1 + trt1)
❝ FMDelta <- mean(log(y1)[trt1==1])-mean(log(y1)[trt1==2])
❝ FMdf <- aov(FullModel)$df.residual
❝ FMMSE <- summary(FullModel)$sigma^2/FMdf
❝ FMlo <- 100*exp(FMDelta - qt(1-0.05,FMdf)*sqrt(2*FMMSE/size))
❝ FMhi <- 100*exp(FMDelta + qt(1-0.05,FMdf)*sqrt(2*FMMSE/size))
❝ FMCI <- FMhi - FMlo
❝ FMCVintra <- c(FMCVintra, sqrt(exp(FMMSE)-1))
At least the code for FMMSE has stolen my sleep
. According to the help page of lm.summary() "... sigma - the square root of the estimated variance of the random error" and IMHO then the MSE is sigma^2 without the division by the degrees of freedom!Check this out via
anova(FullModel)["Residuals","Mean Sq"]==summary(FullModel)$sigma^2 .Or did I miss sumfink here?
—
Regards,
Detlew
Regards,
Detlew
Complete thread:
- Anticonservativism?! ElMaestro 2010-02-06 17:10 [Regulatives / Guidelines]
![[Mix]](https://static.bebac.at/img/mix.png "open in Mix view")
- Conservativism? Helmut 2011-11-03 21:12
- In hindsight... ElMaestro 2011-11-03 21:43
- In hindsight... Helmut 2011-11-04 01:56
- In hindsight... ElMaestro 2011-11-04 08:41
- rlnorm simulates what? d_labes 2011-11-04 09:40
- Oops. Helmut 2011-11-04 16:05
- Oops. ElMaestro 2011-11-04 16:22
- Simulation of intra-subject variability d_labes 2011-11-07 11:16
- Simulation of intra-subject variability ElMaestro 2011-11-08 11:02
- Simulation of intra-subject variability d_labes 2011-11-07 11:16
- Oops. Oops.d_labes 2011-11-25 13:54
- Another Oops. Helmut 2011-11-25 14:23
- Oops. ElMaestro 2011-11-04 16:22
- Oops. Helmut 2011-11-04 16:05
- In hindsight... Helmut 2011-11-04 01:56
- In hindsight... ElMaestro 2011-11-03 21:43
- Simul Ants questions d_labes 2011-11-04 15:46
- Simul Ants questions ElMaestro 2011-11-04 16:06
- Simul Ants questions Helmut 2011-11-04 16:09
- Liberal Conservatives d_labes 2011-11-08 11:31
- Liberal Conservatives martin 2011-11-08 20:50
- Liberal Conservatives Helmut 2011-11-08 22:56
- intra-subject correlation martin 2011-11-09 09:01
- intra-subject correlation Helmut 2011-11-25 17:16
- intra-subject correlation martin 2011-11-09 09:01
- Liberal Conservatives Helmut 2011-11-08 22:56
- Liberal Conservatives martin 2011-11-08 20:50
- Conservativism? Helmut 2011-11-03 21:12
Ing. Helmut Schütz