R help - May 2011 - predict 'expected' with eha package

I am unsure what is being returned, and what is supposed to be
returned, when using 'predict' with "type='expected'"
for an aftreg
survival model.  The code below first generates a weibull  model, then
uses predict to create a vector of the linear predictors, then
attempts to create the 'expected' vector, which is empty.  The final
two steps in the code generate a lognormal model with the same data,
and the same empty 'expected' vector.

My expectation had been that 'expected' would generate the same
transformed dependent variable output as predict with a survreg model
using type='response'.  Since my 'real' data is left-truncated
and
right-censored I cannot use survreg, and I wanted to investigate the
output from eha.

Thanks in advance!

Mike
> data(mort)
> aftreg(Surv(enter, exit, event) ~ ses, data = mort)
Call:
aftreg(formula = Surv(enter, exit, event) ~ ses, data = mort)

Covariate          W.mean      Coef Exp(Coef)  se(Coef)    Wald p
ses
           lower    0.416     0         1           (reference)
           upper    0.584    -0.348     0.706     0.089     0.000

log(scale)                    3.603    36.698     0.065     0.000
log(shape)                    0.331     1.392     0.058     0.000

Events                    276
Total time at risk         17038
Max. log. likelihood      -1391.3
LR test statistic         16.1
Degrees of freedom        1
Overall p-value           5.91578e-05
> m1 <- aftreg(Surv(enter, exit, event) ~ ses, data = mort)
> head(predict(m1, type='lp')) ## produces output
        1         2         3         4         5         6
-0.347853  0.000000 -0.347853  0.000000  0.000000 
0.000000
> head(predict(m1, type='expected')) ## is this correct?
numeric(0)
> m2 <- aftreg(Surv(enter, exit, event) ~ ses, dist='lognormal',
data = mort)
> head(predict(m2, type='expected')) ## is this correct?
numeric(0)


from eha (the survival and rms packages are not an option for my
'real' question, since I have left-truncated right-censored data

Mike,

On Sat, May 21, 2011 at 10:15 PM, Mike Harwood <harwood262 at gmail.com>
wrote:
> I am unsure what is being returned, and what is supposed to be
> returned, when using 'predict' with
"type='expected'" for an aftreg
> survival model.
I'm sorry, but there is no 'predict.aftreg'. The generic
'predict'
"works" because aftreg objects have a secondary class 'coxph'.
This is
probably a mistake (or two) that should be corrected.

Thanks,

G?ran
> The code below first generates a weibull ?model, then
> uses predict to create a vector of the linear predictors, then
> attempts to create the 'expected' vector, which is empty. ?The
final
> two steps in the code generate a lognormal model with the same data,
> and the same empty 'expected' vector.
>
> My expectation had been that 'expected' would generate the same
> transformed dependent variable output as predict with a survreg model
> using type='response'. ?Since my 'real' data is
left-truncated and
> right-censored I cannot use survreg, and I wanted to investigate the
> output from eha.
>
> Thanks in advance!
>
> Mike
>
>> data(mort)
>> aftreg(Surv(enter, exit, event) ~ ses, data = mort)
> Call:
> aftreg(formula = Surv(enter, exit, event) ~ ses, data = mort)
>
> Covariate ? ? ? ? ?W.mean ? ? ?Coef Exp(Coef) ?se(Coef) ? ?Wald p
> ses
> ? ? ? ? ? lower ? ?0.416 ? ? 0 ? ? ? ? 1 ? ? ? ? ? (reference)
> ? ? ? ? ? upper ? ?0.584 ? ?-0.348 ? ? 0.706 ? ? 0.089 ? ? 0.000
>
> log(scale) ? ? ? ? ? ? ? ? ? ?3.603 ? ?36.698 ? ? 0.065 ? ? 0.000
> log(shape) ? ? ? ? ? ? ? ? ? ?0.331 ? ? 1.392 ? ? 0.058 ? ? 0.000
>
> Events ? ? ? ? ? ? ? ? ? ?276
> Total time at risk ? ? ? ? 17038
> Max. log. likelihood ? ? ?-1391.3
> LR test statistic ? ? ? ? 16.1
> Degrees of freedom ? ? ? ?1
> Overall p-value ? ? ? ? ? 5.91578e-05
>> m1 <- aftreg(Surv(enter, exit, event) ~ ses, data = mort)
>> head(predict(m1, type='lp')) ## produces output
> ? ? ? ?1 ? ? ? ? 2 ? ? ? ? 3 ? ? ? ? 4 ? ? ? ? 5 ? ? ? ? 6
> -0.347853 ?0.000000 -0.347853 ?0.000000 ?0.000000 ?0.000000
>> head(predict(m1, type='expected')) ## is this correct?
> numeric(0)
>> m2 <- aftreg(Surv(enter, exit, event) ~ ses,
dist='lognormal', data = mort)
>> head(predict(m2, type='expected')) ## is this correct?
> numeric(0)
>
>
> from eha (the survival and rms packages are not an option for my
> 'real' question, since I have left-truncated right-censored data
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>


-- 
G?ran Brostr?m