ElMaestro
★★★

Denmark,
2020-08-14 22:54
(1322 d 02:05 ago)

Posting: # 21882
Views: 3,572
 

 EMA dataset I, can someone post results? [General Sta­tis­tics]

Hi all,

can someone post REML results for dataset I of EMA? For example a WNL output and/or SAS output of variance components and fixed effects.
I will be using it to verfiy my own scripts in R, or at least do some direct comparison. :-)
Thanks in advance if possible.

Pass or fail!
ElMaestro
PharmCat
★    

Russia,
2020-08-15 00:09
(1322 d 00:50 ago)

@ ElMaestro
Posting: # 21883
Views: 2,896
 

 EMA dataset I, can someone post results?

❝ Hi all,


Hi!

SPSS Code:
MIXED var1 BY sequence period formulation
  /CRITERIA=CIN(90) MXITER(8000) MXSTEP(200) SCORING(1) SINGULAR(0.0000000000001) HCONVERGE(0,
    ABSOLUTE) LCONVERGE(0, ABSOLUTE) PCONVERGE(0.000000001, ABSOLUTE)
  /FIXED=sequence period formulation  | SSTYPE(3)
  /METHOD=REML
  /PRINT=G 
  /RANDOM=formulation  | SUBJECT(subject) COVTYPE(CSH)
  /REPEATED=formulation  | SUBJECT(subject*period) COVTYPE(DIAG)
  /EMMEANS=TABLES(formulation) COMPARE ADJ(LSD).



Information Criteria
-2 Restricted Log Likelihood   530.145
Akaike's Information Criterion (AIC)   540.145
Hurvich and Tsai's Criterion (AICC)   540.354
Bozdogan's Criterion (CAIC)   563.528
Schwarz's Bayesian Criterion (BIC)   558.528
The information criteria are displayed in smaller-is-better form.   



Estimates of Covariance Parametersa
      
Parameter      Estimate   Std. Error
Repeated Measures   
   Var: [formulation=1]   0.117394208   0.019006700
   Var: [formulation=2]   0.202118093   0.029362559
formulation [subject = subject]   
   Var: [formulation=1]   0.686257768   0.122572131
   Var: [formulation=2]   0.727599985   0.136391910
   CSH rho   1.000000000b   0.000000000
      
b This covariance parameter is redundant.         


Random Effect Covariance Structure (G)
      [formulation=1] | subject   [formulation=2] | subject
[formulation=1] | subject   0.686257768   0.706626593
[formulation=2] | subject   0.706626593   0.727599985
Heterogeneous Compound Symmetry      
   


Pairwise Comparisons
               
Mean Difference Std. Error   df        Sig.c   
(I-J)                                                90% CI Lower Bound   90% CI Upper Bound
                  
 0.145*            0.047         209.440         0.002         0.069         0.222
-0.145*            0.047         209.440         0.002        -0.222          -0.069



Julia code:
using ReplicateBE
julia> ReplicateBE.rbe!(df4, dvar = :var, subject = :subject, formulation = :formulation, period = :period, sequence = :sequence, g_tol = 1e-10)


       
Bioequivalence Linear Mixed Effect Model (status:converged)
Hessian not positive defined!
Rho ~ 1.0!

-2REML: 530.145    REML: -265.072

Fixed effect:
──────────────────────────────────────────────────────────────────────────────────────────
Effect           Value        SE          F           DF        t          P|t|
──────────────────────────────────────────────────────────────────────────────────────────
(Intercept)      7.77467      0.141753    3008.17     82.0026   54.8467    2.11367E-66*   
sequence: 2      0.0204707    0.197223    0.0107733   74.735    0.103795   0.91761       
period: 2        0.0446377    0.0636371   0.492019    214.16    0.701441   0.483789
period: 3        0.00659021   0.0644219   0.0104648   189.589   0.102298   0.918629
period: 4        0.0727201    0.0639714   1.29222     214.259   1.13676    0.256909
formulation: 2   -0.145464    0.0465012   9.78552     208.081   -3.12818   0.00201093*
──────────────────────────────────────────────────────────────────────────────────────────
Intra-individual variance:
formulation: 1   0.117394   CVᵂ:   35.29   %   
formulation: 2   0.202118   CVᵂ:   47.33   %

Inter-individual variance:
formulation: 1   0.686258
formulation: 2   0.7276
ρ:               1.0        Cov: 0.706627

formulation: 1 / formulation: 2
Ratio: 115.66, CI: 107.1 - 124.89 (%)
formulation: 2 / formulation: 1
Ratio: 86.46, CI: 80.07 - 93.37 (%)
ElMaestro
★★★

Denmark,
2020-08-15 00:34
(1322 d 00:26 ago)

@ PharmCat
Posting: # 21884
Views: 2,882
 

 EMA dataset I, can someone post results?

Thanks PharmCat - again,

your help here is very valuable to me.
:-)

Pass or fail!
ElMaestro
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2020-08-15 17:00
(1321 d 07:59 ago)

@ ElMaestro
Posting: # 21885
Views: 2,865
 

 EMA dataset I (PHX/WNL 8.1)

Hi ElMaestro,

second opinion:

Banded No-Diagonal factor Analytic (f=2)

FA0(2) 10 iterations
-2REML LL   504.28342
AIC         526.28342
BIC         566.72771
lambda(1,1)   0.85299473
lambda(1,2)   0.8284068
lambda(2,2)   8.3425233E-07
Var(wR)       0.20211806
Var(wT)       0.11739425

Hypothesis Num_DF  Denom_DF      F            p
int          1    74.740719  6165.0288       0
sequence     1    74.725365     0.010773289  0.91761054
treatment    1    207.73496     9.785515     0.0020112901
period       3    192.94942     0.54470318   0.65228368

Diff        Diff_SE       DF
0.14546428  0.046501237  207.73496
PE 1.1565764 (90% CI: 1.0710441, 1.2489393)


Banded No-Diagonal factor Analytic (f=1)

FA0(1) 6 iterations
-2REML LL   504.28342
AIC         524.28342
BIC         561.05095
lambda(1,1)   0.85299474
lambda(1,2)   0.82840679
Var(wR)       0.20211804
Var(wT)       0.11739424

Hypothesis Num_DF  Denom_DF      F            p
int          1    74.74072   6165.0288       0
sequence     1    74.725366      0.010773289 0.91761054
treatment    1   207.73498       9.7855157   0.0020112894
period       3   192.94944       0.54470322  0.65228366
Diff        Diff_SE       DF
0.14546428  0.046501236  207.73498
PE 1.1565764 (90% CI: 1.0710441, 1.2489393)


Heterogenous Compound Symmetry

CSH 8 iterations
-2REML LL   504.28342
AIC         526.28282
BIC         566.72712
cshSD(1)      0.85274447
cshSD(2)      0.82831743
cshCorr       1.0005616
Var(wR)       0.20252423
Var(wT)       0.11754154

Hypothesis Num_DF  Denom_DF      F            p
int          1    71.172734  6165.1799       0
sequence     1    71.158119     0.010757919  0.91768302
treatment    1    75.532826     9.8154921    0.0024629865
period       3   111.51635      0.54613599   0.65176269

Diff        Diff_SE       DF
0.14545681  0.046427791   75.532826
PE 1.1565678 (90% CI: 1.0705169, 1.2495357)


A correlation >1 is funny.

Compound Symmetry

CS 4 iterations
-2REML LL   504.51200
AIC         504.51200
BIC         561.28753
csDiag       -0.00133219
csBlock       0.70154912
Var(wR)       0.20618088
Var(wT)       0.11635790

Hypothesis Num_DF  Denom_DF      F            p
int          1    75.415756  6210.7664       0
sequence     1    75.400233     0.010613701  0.91821838
treatment    1    76.346313     9.8473171    0.0024171494
period       3   113.68812      0.54662827   0.65141337

Diff        Diff_SE       DF
0.14515515  0.046256578   76.346313
PE 1.1562189 (90% CI: 1.0705101, 1.2487890)


Unstructured

UN 4 iterations
-2REML LL   504.28282
AIC         526.28282
BIC         566.72712
un(1,1)       0.72717314
un(1,2)       0.70673981
un(2,2)       0.68610978
Var(wR)       0.20252423
Var(wT)       0.11754154

Hypothesis Num_DF  Denom_DF      F            p
int          1    74.737874  6165.1798       0
sequence     1    74.722533     0.010757919  0.91766913
treatment    1    75.567795     9.8154921    0.0024626145
period       3   111.56199      0.54613599   0.65176227

Diff        Diff_SE       DF
0.14545681  0.046427791   75.567795
PE 1.1565678 (90% CI: 1.0705174, 1.2495351)


EMA Method A

-2REML LL   218.14281
AIC         382.14281
BIC         659.29439
Var(w)       0.15999518

Hypothesis       Num_DF  Denom_DF        F           p
int                 1       217    109374.16        0
sequence            1       217         0.24365135  0.62208048
treatment           1       217         9.78364171  0.0020020741
period              3       217         0.78064221  0.50589993
subject(sequence)  75       217        17.844668    0

Diff        Diff_SE       DF
0.14547367  0.046508691  217
PE 1.1565873 (90% CI: 1.0710567, 1.2489481)


EMA Method B (residual df like SAS DDFM=CONTAIN)

VC 3 iterations
-2REML LL   510.34005
AIC         526.34005
BIC         555.75408
Var(w)       0.16010032
Var(b)       0.70693799

Hypothesis       Num_DF  Denom_DF        F           p
int                 1       217      6161.8922      0
sequence            1       217         0.01197525  0.91296138
treatment           1       217         9.8646416   0.0019195216
period              3       217         0.82881025  0.47928487

Diff        Diff_SE       DF
0.14608818  0.046513007  217
PE 1.1572982 (90% CI: 1.0717074, 1.2497247)


EMA Method B (Satterthwaite’s df)

VC 3 iterations
-2REML LL   510.34005
AIC         526.34005
BIC         555.75408
Var(w)       0.16010032
Var(b)       0.70693799

Hypothesis       Num_DF  Denom_DF        F           p
int                 1     74.737599  6161.8922      0
sequence            1     74.720846     0.01197525  0.91315361
treatment           1    216.93862      9.8646416   0.0019195905
period              3    217.11838      0.82881025  0.47928403

Diff        Diff_SE       DF
0.14608818  0.046513007  216.93862
PE 1.1572982 (90% CI: 1.0717073, 1.2497248)


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
PharmCat
★    

Russia,
2020-08-16 02:57
(1320 d 22:03 ago)

@ Helmut
Posting: # 21886
Views: 2,856
 

 EMA dataset I (PHX/WNL 8.1)

Hi Helmut!

❝ -2REML LL 504.28342


I suppose Phoenix REML includes/exludes some constant.

❝ A correlation >1 is funny.


I think CSH is most fair structure. But if otimize rho without limitaion or link function result cab be very srange :rotfl:
ElMaestro
★★★

Denmark,
2020-08-20 00:50
(1317 d 00:10 ago)

@ Helmut
Posting: # 21891
Views: 2,696
 

 EMA dataset I (PHX/WNL 8.1)

Hi Hötzi,

thanks for WNL outputs with various covariance structures.

❝ A correlation >1 is funny.


In a way you can compare it with the CoA of a reference standard which is labeled with purity = 101.2 %. It does not contain 101.2% - I guarantee it :-D - but that may stil be the most viable estimate.
It is all about how we estimate / measure it. We are not guaranteed to get the same model maximum likelihood if you keep the rho controlled between (exactly) -1 to (exactly) 1 with any iterative method regardless of your tolerance setting on the x- or y-scale.

Pass or fail!
ElMaestro
Helmut
★★★
avatar
Homepage
Vienna, Austria,
2020-08-20 03:05
(1316 d 21:54 ago)

@ ElMaestro
Posting: # 21892
Views: 2,702
 

 EMA dataset I (PHX/WNL 8.1)

Hi ElMaestro,

❝ ❝ A correlation >1 is funny.


❝ In a way you can compare it with the CoA of a reference standard which is labeled with purity = 101.2 %. It does not contain 101.2% - I guarantee it :-D - but that may stil be the most viable estimate.


That’s another cup of tea. I guarantee that it can contain exactly 101.2% indeed – regardless what the CoA says. ;-)

❝ It is all about how we estimate / measure it. We are not guaranteed to get the same model maximum likelihood if you keep the rho controlled between (exactly) -1 to (exactly) 1 with any iterative method regardless of your tolerance setting on the x- or y-scale.


Here you are right despite some weirdos laments.* The largest correlation in my 500 simulated data sets was for #168 with 7.285 (‼)…


  • Schütz H. An extremely strange observation? [letter]. Europ J Drug Metabol Pharmacokin. 2004;29(1):69–71. doi:10.1007/BF03190576.

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
UA Flag
Activity
 Admin contact
22,957 posts in 4,819 threads, 1,636 registered users;
74 visitors (0 registered, 74 guests [including 9 identified bots]).
Forum time: 00:00 CET (Europe/Vienna)

Nothing shows a lack of mathematical education more
than an overly precise calculation.    Carl Friedrich Gauß

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