BookmarkSubscribeRSS Feed
Ruhi
Obsidian | Level 7

Hi

I have a dataset for 16 examiners who took a test on correctly diagnosing the side of asymmetry in 6 landmarks on human body in anterior and posterior direction. I am trying to see which landmark ,asymmetry and direction combinations are most accurately answered by examiners.So there are 6 landmarks, two directions (ant and posterior) and different levels of asymmetry. Not all landmarks are tested on same level of asymmetry, which means if say landmark 1 has asymmetries 4,5,6, landmark 2 might have 2, 3, 4. and similarly other landmarks. So my level of asymmetries are 3,4,5,6,7,8 used randomly for testing students at different landmarks.

Now when I tried some techniques of making the model converge from the suggestions on other posts on similar convergence issues but still have issues in convergence. Please suggest some solution.

I reconverted my examiner id variable to make it numeric and then removed from the class statement after sorting by id. I also tried removing the direction from the model statement , but still model does not converge.

 

Thanks a ton!!!

 

proc glimmix data=dat1 noitprint;
  class absasym  direction landmark  ;
  model Autoscore(desc) = landmark*absasym*direction / dist=binary link=logit ddfm=kr;
  random intercept / subject=ID gcorr solution;
  nloptions technique=congra maxiter=1000 gconv=1e-4;
run;

 I want to run some lsmeans statements but seeing issues with convergence, I could not do so yet.

I used proc freq to see how is the missingness in different categories for landmark and absasym combinations. The table from that.

 

Table of Landmark by absasym
Landmark absasym
3 4 5 6 7 8 Total
ANT IC
0
0.00
0
0.00
48
6.25
16
2.08
16
2.08
16
2.08
96
12.50
ASIS
0
0.00
64
8.33
64
8.33
64
8.33
0
0.00
0
0.00
192
25.00
ITB
0
0.00
32
4.17
32
4.17
32
4.17
0
0.00
0
0.00
96
12.50
POST IC
0
0.00
0
0.00
32
4.17
16
2.08
48
6.25
0
0.00
96
12.50
PSIS
16
2.08
48
6.25
32
4.17
96
12.50
0
0.00
0
0.00
192
25.00
PT
64
8.33
32
4.17
0
0.00
0
0.00
0
0.00
0
0.00
96
12.50
Total
80
10.42
176
22.92
208
27.08
224
29.17
64
8.33
16
2.08
768
100.00
1 REPLY 1
SteveDenham
Jade | Level 19
Hi Ruhi,
I think you have a confounding problem, based on your description.  It appears from the contingency table that the variable "direction" is completely confounded with landmark (i.e., three of the landmark descriptions look like anterior and the other three are posterior).  However, since you are already fitting a means model, "direction" can be extracted later from the lsmestimate statement--once we get convergence.  So I would consider:

 

proc glimmix data=dat1 noitprint method=laplace;
  class absasym  landmark  ;
  model Autoscore(desc) = landmark*absasym / dist=binary link=logit;
  random intercept / subject=ID gcorr solution;
  nloptions technique=congra maxiter=1000 gconv=1e-4;
run;

 

See if this will converge.

 

Steve Denham

sas-innovate-2024.png

Join us for SAS Innovate April 16-19 at the Aria in Las Vegas. Bring the team and save big with our group pricing for a limited time only.

Pre-conference courses and tutorials are filling up fast and are always a sellout. Register today to reserve your seat.

 

Register now!

What is Bayesian Analysis?

Learn the difference between classical and Bayesian statistical approaches and see a few PROC examples to perform Bayesian analysis in this video.

Find more tutorials on the SAS Users YouTube channel.

Click image to register for webinarClick image to register for webinar

Classroom Training Available!

Select SAS Training centers are offering in-person courses. View upcoming courses for:

View all other training opportunities.

Discussion stats
  • 1 reply
  • 669 views
  • 1 like
  • 2 in conversation