Deviance residuals - graph

[Rose Matthews]

Hello there

I don’t know if this is the right platform but thought i’d try

Research question bx: we have allied health professionals (AHP) who do specific operations inserting a specific implant into a hip (dynamic hip screw for trauma)
Question: Do allied health professionals have more experience in terms of less failures of the implant over time.

Stats used:
I used a cox regression model to determine if the more experience a AHP has the less likely they are to get a failure of the implant

I plotted a cox regression , adjusted for predictors using backward selection.

Happy with my model.

I then wanted to see if

1. Model fits PH
2. Model has a good fit in terms of cox-Snells Residuals
3. Check the linearity of continuous variables with Martingale residuals
4. Check the model of fit again with Deviance residuals <---my main question

Final model:

Stcox experience age gender anaestethictype
gen insample2= e(sample) 
drop if insample2 = 0  //using those observations which are only to be used by final model 

//No 1: PH 
predict sch*, scaledsch
scatter sch6 _t || lfit sch6 _t   //schoefield residuals for gender 
count if sch >0.1 //8508
count if sch6 <0.1 //987

With regards to 1

• All fit PH hazards except gender which is odd as in the univariate analysis it looked PH ok.
• Tbh I thought of still including as the schoefield residuals graph looks fairly horizontal - I’m not sure if i’m committing a foul here! (see above)

With regards to 2, I’ve got a nice graph (good news) - cox - snell

With 4, fitting deviance residuals

I don’t understand why I have some centred around 0 and some centred above ? Range 1-4

• Our local statistician is also unsure

Just fyi,

There are no missing deviance residuals
Here’s my code

Stcox experience age gender anaestethictype
predict mg, mgale
predict xb, cb
scatter mg, xb
predict dev, deviance
scatter dev xb

Any other sources of enlightenment?