Binary logistic regression analysis - bootstrapping and reporting of the results.

#1
Dear all,
I need our advice regarding bootstrapping of binary logistic regression.
I have following questions:
- the sample size is 305 people. How many random samples are most appropriate? Some say 1000 and some 2000, but what is thelogic behind choosing the sample size?
- How can i get/is it possible to get bootsrapped OR, Hosmer and lemeshow's p value, Nagelkerke's R square and predictive probabilities for building the ROCs?
- What values shoud i repport in the paper?

I run analyses in SPSS.

Thank you for the answer.

Best regards.
 

hlsmith

Less is more. Stay pure. Stay poor.
#2
You just need to have enough samples for the estimates to converge. So if you run 1000 and 10000 and get the same thing you are there. I usually go say 10000 so it is easy to get the 2.5 and 97.5 values for estimate intervals.

Of note, you typically use your original estimates and the BS for the precision estimates. So you can get the BS values for anything in the logistic output!
 
#3
You just need to have enough samples for the estimates to converge. So if you run 1000 and 10000 and get the same thing you are there. I usually go say 10000 so it is easy to get the 2.5 and 97.5 values for estimate intervals.

Of note, you typically use your original estimates and the BS for the precision estimates. So you can get the BS values for anything in the logistic output!

Thank you for the answer.
So if i understand it now well:
- B coefficietnts and OR has to be reported from the original Log. regression model.
- S.E., P values and 95 % CI should be reported according to the BS models.

Is it correct?

I will rerun the analyses with 10 000 samples and compare the results with the models i have.
 

hlsmith

Less is more. Stay pure. Stay poor.
#4
That is it as I know it! Given modern computers they should be able to handle 10k samples no problem.