Dear all,
Michael and I updated the Template.
New:
- Works in Phoenix 6.4 and 7.0.
- Added 4-period 4-sequence designs:
- TRTR | RTRT | TRRT | RTTR
- TRTR | RTRT | TTRR | RRTT
Both are not recommended (confounded effects) and included only for completeness.
- Added the 3-period 2-sequence (partial) replicate design:
- TRR | RTR (aka “extra reference design”)
Not recommended since lacking period effects have to be assumed. Since T is not administered in the third period, the treatment comparison will be biased in the presence of true period effects. This design should be avoided and included only for completeness.
- TRR | RTR (aka “extra reference design”)
- Plots of standardized residual vs. subject and model predictions.
- Comparison of within-subject variabilities of T and R (full replicate designs only).
- Instructions included in the documents folder.
Changed:
- The name of both formulations can be specified (previously only the one of the reference).
The run-time is slightly better than in previous versions but needs patience. Since Phoenix swaps intermediate results1 to the system drive, it will take a while to complete. On my machine (with a fast solid-state drive) execution of the EMA’s reference data set I takes 2:15 minutes.2 If you have a hard drive, it may take substantially longer.
The template was developed in PHX 6.4.0.768 and tested in PHX 7.0.0.2535. The template was cross-validated with 26 data sets (all designs, up to 360 subjects) against SAS 9.4 (Methods A, B, C), R 3.4.1 (A, B), STATISTICA (A), and partly (Q&A data sets I and II; Methods A, B) against STaTa 14, SPSS 22, and JMP 10. As usual, use at your own risk.
When you execute the workflow Final Conclusions, the outlier analysis (boxplot, etc.) is not automatically performed. Navigate to Analysis > ABEL > Outlier Analysis and execute.
Reports of defects and experiences with Windows installed on a hard drive are welcome.
1 After execution 4.7 MB (!) in 300 (‼) files. 6 copies each of linmix.exe (292.0 kB) and Dforrt.dll (405.5 kB).
2 In R for comparison: 40 milliseconds…
PS: Balaam’s design (TR|RT|TT|RR) works but was not seriously tested. Should be avoided due to its terrible power characteristics!