GOF error message

[steven]

I am doing a bootstrap GOF for a multi-strata model and having an interesting (frustrating time). During each iteration, an error message appears as follows:

Error Code: 34 [DATA WIDTH ERROR]
Subsystem: DBFCDX
Error Subcode: 1121 (Datawidth error)
Function: FIELDPUT
Requesed type: NUMERIC
CallStack:
SIMWINDOW: SIMULATIONS (Line: 87)
SIMULATIONTHREADNT (Line: 2)
MYTOPAPPWIN:DOSIMULATION LINE: 58)
MYTOPAPPWIN:BOOSTRAPGOG (LINE: 83)
APP: START LINE: 39)

When I click the ignore button, the program proceeds (goes on to next iteration) and the results seem to look OK. Of course, I have to ask it to ignore the error for each iteration.....so I have to babysit the program.

Suggestions about how to deal with this are most welcome.

S.

[cooch]

I am doing a bootstrap GOF for a multi-strata model and having an interesting (frustrating time). During each iteration, an error message appears as follows:

I'm guessing you haven't read Chapter 5 in it's entirety. If you have, you'd have come across the sections which do not recommend the bootsrap GOF - it has been supplanted by either (i) U-CARE (for fully-time dependent MS models), or (ii) the median c-hat approach in MARK.

Ditch the bootstrap and try either of those...

Also, for MS models, be aware of the potential for local minima. Unlike simple multinomials with a single likelihood ridge, MS models can periodicall (often?) have local minima issues - again, discussed in some detail in Chapter 9.

[steven]

Thanks for the reply. Of course, I have read chapter 5, although perhaps not in a way that maximized my understanding! I do note that you wrote (page 5 - 29):

However, be advised that no GOF test is perfect - the median GOF approach is a "work in progress", and its performance compared to other GOF approaches has not been fully evaluated in all situations (e.g., multistate models.

Perhaps this has been revised. Either way, I'd still like to understand why I am getting this error message. Suggestions are welcome.

[cooch]

Thanks for the reply. Of course, I have read chapter 5, although perhaps not in a way that maximized my understanding! I do note that you wrote (page 5 - 29):

However, be advised that no GOF test is perfect - the median GOF approach is a "work in progress", and its performance compared to other GOF approaches has not been fully evaluated in all situations (e.g., multistate models.

Perhaps this has been revised. Either way, I'd still like to understand why I am getting this error message. Suggestions are welcome.

Translation - lots of work done, but not published - yet. Comparisons between U-CARE and media c-hat for fully time-dependent models show they are comparable, so using median c-hat for general models which are not fully time-dependent is *likely* to be pretty robust.

So, I'd suggest (i) using median c-hat, and then (ii) following the suggestions on p. 35 of chapter 5.