6 replies to this topic

[Shu-Pei Wu]

Hi Pharsight modellers,

 

I just wonder why we get correlation matrix only for the WNL5 Classic modeling mode but not the new Phoenix model when using naive pooled method?

Is that because the fundamental algorithm difference?

 

Thanks,

Best,

 

Shu-Pei

[Shu-Pei Wu]

I found the covariance matrix in one of the output when doing naive pooled, and I guess we can convert to correlation matrix manually.

Just wonder is there a reason that the correlation matrix is not a "default" output.

 

Thanks,

Best,

 

Shu-Pei

[serge guzy]

Dear colleague
In the new upcoming version, you will get thetacorrelation.
for the meantime you can calculate it from thetacovariance which has the error matrix in terms of variances and covariances.

The correlation terms (diagonal elements) are computed as follws:

corri,j=covari,j/(sdi*sdj)

correlation = covariance divided by the product of the standard deviations.
The standard deviations are the square root of the variances, variances being the diagonal elements of thetacovariance.

best
Serge

Edit from SDavis:
Below is a Screenshot from the 1.4 Release that will be publicly available within the next couple of weeks

Attached Thumbnails

• 

[Shu-Pei Wu]

Thanks for the updating info, Serge,

 

I also have other question regarding to the weighting feature.

 

I couldn't find the feature when using PML model using naive pooled like classic model.

such as weighting as 1/Y, 1/Y*Y etc..

 

Tons of thanks for your response.

Best,

 

Shu-Pei

[Simon Davis]

Shu-Pei,

The analogous function is residual error in the Phoenix model engine

here is a table to compare them with the WNL classic weighting functions.

Simon

Attached Thumbnails

• 
• 

[Shu-Pei Wu]

Simon, Thank you so much for the table.

It is very helpful!!

 

Best,

Shu-Pei

[Robert Stratford]

Dear Simon,

 

I do not have the additive + multiplicative residual error model option in NLME 1.2, yet the screen shot shown in your answer has this option. Is this screen shot showing NLME 1.3, which is not yet available?

 

Kind regards,

 

Robert