5 replies to this topic

[vaishnavi ganti]

Hello All:

How would one represent the DV vs PRED plot for log additive error model results. 

Since the results are log transformed .

or am i missing something?

Thank you in advance 

Edited by vaishnavi ganti, 26 October 2016 - 09:51 PM.

[mittyright]

Dear Vaishnavi,

 

Sorry, maybe I'm late and you've found a solution. I think you're asking how to present the plots in back-transformed style, right?

One can see the note when log-additive residual error is selected in Built-In mode:

 

This model log-transforms observations and predictions only if there is just one observe statement in the model <...> The residuals worksheet reflects this.

 

It could be inconvenient, especially for VPC purposes. So you need to add another observation to the model to get not-transformed observations. Or to exp the values in the residual sheet to transform from log-domain. 

 

BR,

Mittyright

[vaishnavi ganti]

Dear Vaishnavi,

 

Sorry, maybe I'm late and you've found a solution. I think you're asking how to present the plots in back-transformed style, right?

One can see the note when log-additive residual error is selected in Built-In mode:

 

This model log-transforms observations and predictions only if there is just one observe statement in the model <...> The residuals worksheet reflects this.

 

It could be inconvenient, especially for VPC purposes. So you need to add another observation to the model to get not-transformed observations. Or to exp the values in the residual sheet to transform from log-domain. 

 

BR,

Mittyright

Yes I did figure it out . Thank you for your reponse

[Ana Henry]

For those interested, there was an update in the newest version of Phoenix,  Phoenix 7.0, with respect to this, now

 

Predicted observations from simulating a log-additive model are reported in non-log-space, allowing for comparisons with predictions (QC 15385).

[mittyright]

Dear Ana,
 

Predicted observations from simulating a log-additive model are reported in non-log-space, allowing for comparisons with predictions (QC 15385).

 

You are right about simulated data, PHX7 produces it correctly with log-additive error.

But if we are talking about residuals and VPC, the output is still the same (in log scale). That's why a note is given (see above)

Please also see the project attached. I was looking for a workaround to get the VPC in back-transformed scale. 

 

I suspect there are some reasons to present values in Residuals sheet in log domain.

 

I think this sentence in the User's guide is incorrect: "The residuals output and plot output are transformed back so they are not in log-space."  Not always

 

Hope it helps,

Mittyright

Attached Files

•  Smith et al 2000_LogWorkaround.phxproj   707.6KB   452 downloads

[MojDa01]

Hi there!

 

I am working on a PK model with a log-additive error model, too. Is there a way to back-transform the conc. values for plotting purposes (without adding forther observe() statements)?

Above a suggestion was to exp the residual sheet to transform from log-domain... where do I find this sheet? Is this the "Residuals" sheet in the "output data" of the results??

 

Thanks in advance!

Daniel

 

P.S.: I am using Phoenix 8.0 (64 bit)

Edited by MojDa01, 14 May 2018 - 10:31 AM.