negative ETA shrinkage

On Sun, Sep 5, 2010 at 8:19 AM, David Ternant <[email protected]> wrote: > Dear Maurice, > > Negative shrinkage means that your SD is bigger than your estimated omega. > And your negative shrinkage is near 0 (<-1%). Therefore, I would take it for > a null shrinkage. > > Best Regards, > > David > > Chenguang Wang a écrit : > > Dear all, >> I have found a negtive ETA shrinkage result in my model output. I did a >> simple simulation with 1-compartement model via iv bolus administration, and >> then refitted the simulated data with the same model and initial values. In >> the NONMEM output (of NONMEM 7), I got ETAshrink(%): -2.0123E-01 >> -2.3271E-01. From the definition of ETA shrinkage: 1-SD(eta*)/omega, eta* is >> the EBE estimated eta, there is no constraint preventing ETA shrinkage going >> below zero. Could somebody give me an interpretation of negtive ETA >> shrinkage? >> Thanks in advance! >> Maurice >> > > > > -- Navin Goyal

________________________________ From: [email protected] [mailto:[email protected]] On Behalf Of Navin Goyal Sent: Sunday, September 05, 2010 11:52 AM To: Chenguang Wang; [email protected] Subject: Re: [NMusers] negative ETA shrinkage Hi Maurice, While using NM7, it also depends on what estimation method was used ... for eg. the SAEM method sometime outputs incorrect shrinkage values. (this is also mentioned in the intro supplement to NM7). Did you try to manually estimate shrinkage using the individual omega values ? Does that still result in negative estimates ? I have also noted negative shrinkage values while using other methods in NM7..... but if I estimate them manually they are different from what is reported in NM7 output and positive. PsN estimates the shrinkage using the individual Omega values and reports it (example in NM6) but with NM7, PsN reports shrinkage values calculated by NM7. Hope this is of some help. Regards Navin On Sun, Sep 5, 2010 at 8:19 AM, David Ternant <[email protected]> wrote: Dear Maurice, Negative shrinkage means that your SD is bigger than your estimated omega. And your negative shrinkage is near 0 (<-1%). Therefore, I would take it for a null shrinkage. Best Regards, David Chenguang Wang a écrit : Dear all, I have found a negtive ETA shrinkage result in my model output. I did a simple simulation with 1-compartement model via iv bolus administration, and then refitted the simulated data with the same model and initial values. In the NONMEM output (of NONMEM 7), I got ETAshrink(%): -2.0123E-01 -2.3271E-01. From the definition of ETA shrinkage: 1-SD(eta*)/omega, eta* is the EBE estimated eta, there is no constraint preventing ETA shrinkage going below zero. Could somebody give me an interpretation of negtive ETA shrinkage? Thanks in advance! Maurice -- Navin Goyal