Getting 0% OOD Optimality for my design script

[neeraj85]

Thanks again.

I just need a confirmation on one thing..

You mentioned in the previous mails that "increasing the design also means that you will ....... efficiency per choice task will actually go down" & "that the D-error hasn't got much meaning".

I executed the most recent script shared by you for 2 scenarios and following were the results:

1. Rows=30; block=2; (which comprise of less efficient choice tasks) and the simulation stabilized at a D-error ~ 0.66

2. Rows=15 (which comprise of more efficient choice tasks than the former) and the simulation stabilized at a D-error ~ 1.33

So, should I still consider the second design to be more efficient even if the D-error is worse off than the first case? Please confirm.


Also, which metric helps us in deciding that the second design is more efficient?


Thanks
Neeraj

[neeraj85]

Thanks again.

I just need a confirmation on one thing..

You mentioned in the previous mails that "increasing the design also means that you will ....... efficiency per choice task will actually go down" & that the D-error hasn't got much meaning.

I executed the most recent script shared by you for 2 scenarios and following were the results:

1. Rows=30; block=2; (which comprise of less efficient choice tasks) and the simulation stabilized at a D-error ~ 0.66

2. Rows=15 (which comprise of more efficient choice tasks than the former) and the simulation stabilized at a D-error ~ 1.33

So, should I still consider the second design to be more efficient even if the D-error is worse off than the first case? Please confirm.


Also, which metric helps us in deciding that the second design is more efficient?


Thanks
Neeraj

[neeraj85]

Hi,

A gentle reminder regarding the query that I posted the last week.


Thanks
Neeraj

[Michiel Bliemer]

design 1 with 30 rows has D-error = 0.66, which means that the determinant of the AVC matrix is 0.66^4 = 0.189.
design 2 with 15 rows has D-error = 1.33, which means that the determinant of the AVC matrix is 1.33^4 = 3.129.

Scaling design 2 to 30 rows will yield an AVC matrix in which all elements are divided by 2. This means that the determinant of the AVC matrix with 30 rows will be (1/2)^4 * 1.33 = 0.195.
Then the D-error is 0.195^(1/4) = 0.665.
(if I did all the math correctly).

So in your case the efficiency of the choice tasks in the design with 15 rows is the same as the design with 30 rows.

[Michiel Bliemer]

Or to make the computation more simple, you can divide the D-error by half, 1.33 / 2 = 0.665.

[neeraj85]

Thanks a lot for explaining the concept..

I didn't come across the same in the Ngene manual.
Just a small correction to the mathematical calculation.

AVC matrix with 30 rows will be (1/2)^4 * 1.33 = 0.195. --> This should be AVC matrix with 30 rows will be (1/2)^4 * 3.129 = 0.195 (3.129 is the determinant of the AVC matrix for 15 rows)


Thanks
Neeraj

[Michiel Bliemer]

That is right, Thanks for the correction.