Since it is not a bug it /must/ be a feature! [Software]

posted by Helmut Homepage – Vienna, Austria, 2016-03-20 01:34 (3247 d 05:25 ago) – Posting: # 16122
Views: 4,518

Hi Mittyri,

❝ And I have a question about Studentized residuals (Outlier analysis for 4-Period study, EMA): why are we filtering the predicted values as a result of LME object 'LME Ref variability' only for periods 1 and 2 (2 sequences only for 2x2x4)?

Filter->(nseq=1 and period=2) or (nseq=2 and period=1) or (nseq=3 and period=1)


Don’t get confused by the red part: The template handles also the partial replicate where we have three sequences. If you have 4-period full replicate data, only the green part will be evaluated. Navigate in the Object Browser to

Analysis
└─ ABEL
   └─ Outlier Analysis
      ├─ Prep Steps
      │  ├─ Complete reference data_Result
      │  ├─ LME Ref variability_Residuals
      │  ├─ LME Ref variability_Final Variance Parameters
      │  ├─ Subjects by Sequence and Period
      │  ├─ Join _ Residuals and Nk
      │  └─ Join MSE_MSintra
      └─ Studentized Residuals → Step 1 Filter

In the Options below check ☑ Retain Intermediate Results and execute.
In Results:Intermediate Results:Step 1 Filter.Results Worksheet you see the results of all subjects who had two administrations of R (at this point incomplete data are already excluded) coming from either the first period (in sequence RTRT) or the second (in sequence TRTR). Works as designed. ;-)

❝ What about periods 3 and 4?


The mean of residuals within particular subjects should be zero. Therefore, if one is examined the other one is not informative since it is only of the opposite sign (|Res1| = |Res3| ∨ |Res2| = |Res4|).
Check above the object LME Ref variability_Residuals where you have both raw residuals (before studentization: Predicted–Observed), f.i. for the EMA’s “dataset I”:

Subject Period Sequence Observed Predicted  Residual
   1       1     BABA   7.734541  7.472175  0.262366
   1       3     BABA   7.204848  7.467214 –0.262366
   2       2     ABAB   7.859143  7.721813  0.137330
   2       4     ABAB   7.788792  7.926122 –0.137330

It was my decision to specify the filter this way (similar to Chapter 8 of Chow & Liu for the 2×2 crossover*). I could have filtered for the second occasions of R (coming from period 3 or 4) as well.
With my setup you should get two subjects tagged as potential outliers (#45 with –5.25 and #52 with +3.21). Modify the filter to (nseq=1 and period=4) or (nseq=2 and period=3) and see what happens: Same subjects tagged, same |studentized residuals| but with opposite signs compared to the original setup. ч.т.д.

BTW, expectations {μ, σ²} of studentized residuals are {0, 1}. On my machine (Win7 64bit, both with the 64 and 32bit versions of Phoenix/WinNonlin 6.4.0.768) I get 9.577·10–16 and 1.014 for this dataset. Not that bad.



Dif-tor heh smusma 🖖🏼 Довге життя Україна! [image]
Helmut Schütz
[image]

The quality of responses received is directly proportional to the quality of the question asked. 🚮
Science Quotes

Complete thread:

UA Flag
Activity
 Admin contact
23,376 posts in 4,912 threads, 1,667 registered users;
179 visitors (0 registered, 179 guests [including 12 identified bots]).
Forum time: 06:59 CET (Europe/Vienna)

The great tragedy of Science – the slaying
of a beautiful hypothesis by an ugly fact.    Thomas Henry Huxley

The Bioequivalence and Bioavailability Forum is hosted by
BEBAC Ing. Helmut Schütz
HTML5