 # Questions on output

4 replies to this topic

### #1 sjpan

sjpan

Newbie

• • Members
• • 5 posts

Posted 08 June 2022 - 02:27 AM

Dear all,

I have a few questions:

1. How is the stdev0 calculated in the Theta worksheet in the modeling output? How does one interpret that value?

2. What are the blue line and the red line in the IWRES vs IPRED plot? how are they generated (i.e., via calculation or?)

I cannot find the answers to these questions in the User's Guide. Does anyone know where I can find such information.

Thank you very much.

SP

### #2 Simon Davis

Simon Davis

• • • 1,230 posts

Posted 08 June 2022 - 06:23 AM

This should be in the online help but I agree it is hard to find so will check with our Documentation lead where this is;
<the blue line is a LOESS regression of all values, whereas the top redline is a LOESS of all residuals after being made positive, and the lower red line its' mirror.>
e.g.
https://onlinehelp.c...832_Plot_output

IWRES vs IVAR (Individual; Simple): Plot of individual weighted residuals (IWRES) versus the inde­pendent variable (IVAR, e.g., time). Ideally, the blue line should be at 0 and the red line (with its nega­tive reflection) should not show any fanning. Fanning indicates room for improving the distribution of residuals.

The stdev0 is the residual error of the observations i.e. CEps. so depends on your residual error model type, ie. additive you can think of this as an absolute value relative to the observations as you are assuming a uniform error;

error(CEps = 0.20532149859721)
observe(CObs = C + CEps)

whereas multiplicative assumes an error proportional to conc so CEps of 0.1 is suggesting 10%

error(CEps = 0.1)
observe(CObs = C * (1 + CEps))

Simon

PS The Phoenix Loess function uses the Cleveland-Grosse Bell labs fortran 77 implementation in Netlib. See http://www.netlib.org/a/loess for a description. The smoothing parameter, alpha, is determined automatically and depends on the number of points: if(n.gt.10) alpha=1 if(n.gt.10) alpha=.9 if(n.gt.15) alpha=.8 if(n.gt.20) alpha=.7 if(n.gt.50) alpha=.6 if(n.ge.100) alpha=.5 if(n.ge.200) alpha=.5 if(n.ge.500) alpha=.5 if(n.ge.1000) alpha=0.5 In most of the case, the default smoothing value is good enough. The span parameter in R gives the proportion of points in the plot which influence the smooth at each value. A scientific consultant got the same results for the plot using the R default span parameter = 2/3: As you can see from the plot, the data points are not distributed evenly, especially those sparse data at the beginning that the default Loess parameter value is not able to capture. Changing the smooth span parameter in R to = 0.08 gives a better result: There is no option to change the alpha parameter in Phoenix ? this is determined automatically based on the number of points. One suggestion would be to instead use linear regression to replot IPRED vs DV from the Residuals output worksheet.

Edited by Simon Davis, 08 June 2022 - 08:43 AM.

### #3 sjpan

sjpan

Newbie

• • Members
• • 5 posts

Posted 08 June 2022 - 07:26 AM

Hi Simon,

Thank you for the prompt response.

Can you please explain how CEps was calculated? Is it from individual residual values (IRES)?

Thank you very much.

SP

### #4 smouksassi1

smouksassi1

• • Members
•   • 200 posts
• LocationMontreal

Posted 08 June 2022 - 08:39 AM

Which method are you using the default FOCE-ELS ?

The epsilon is simultaneously optimized with the ELS:

in Phoeinx you get it as the stdev of epsilon it is an estimated parameter not a derived one

### #5 sjpan

sjpan

Newbie

• • Members
• • 5 posts

Posted 08 June 2022 - 08:59 AM

Thank you for the explanation.

We ran under the "individual modeling run options" and used the Max ODE.

Which method are you using the default FOCE-ELS ?

The epsilon is simultaneously optimized with the ELS:

in Phoeinx you get it as the stdev of epsilon it is an estimated parameter not a derived one

#### 0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users 