Dear colleagues
A) I have a sample (n=132) and I performed a multiple logistic regression. I have 1 dependent variable (dichotomous variable: yes/no) and 6 predictors (2 continuous variables and 4 categorical variables) in the final model.
B) I have another sample (n=75) and I performed a multiple logistic regression. I have the same dependent variable and 6 predictors.
The two samples are totally independent from each other.
I have two questions:
1) How can I see that which model are significantly better than the other one? Higher Naegelkerke R2? Or how?
2) I have to plan a study with a sample size calculation? I would like to prove that the B method is better. G*Power is not optimal and cannot calculate the necessary sample size in this specific case.
Thank you in advance.
Zoltan
Please refer to:
1. Landau, S. and Everitt, B. S.(2004). A Handbook of Statistical analyses using SPSS.
2.Howitt, D. and Cramer, D.(2008). Introduction to SPSS.
Regards,
Zuhair
Thanks for answer Zuhair, but my question is referring mostly the sample size calculation. In other words: I have to pilot studies with binary logistic regression models and the dependent variable is the same in both. In both pilot studies I have the same independent variables except one. Based on the results I have to plan a new study in order to demonstrate that one method is better in predicting the dependent variable than the other. SPSS is not calculating any sample size on it. So, in other words, I have two logistic regression models (same dependent variables) and same independent variables except one. The final models in the two dataset are different, I mean the predictors are different. How can I calculate the sample size for the next study?
There is a branch of statistics that can be described as computer intensive methods. This includes methods like jackknife, bootstrap, randomization tests, permutation tests, and simulation. It also includes methods where one (or more) observations are left out of the analysis, the model is recalculated and is then used to predict placement of the observations that were left out.
1) Model accuracy questions are ideal for the leave one (or many) out approach.
2) Sample size is answered by simulation approaches. Given that the pilot study is correct, set up a random number generator that will give a population with the population values from your pilot study. How many samples do you need to achieve any desired level of accuracy. Of course the tricky part is that it is unlikely that the pilot study has the correct population values. So you might want to run several simulations with values that reflect your uncertainty about the accuracy in the pilot study.
This task is fairly easy if you know R. R is very good at this sort of thing. SAS will be a bit harder because this is not the sort of thing that you can just look at the Iris data example and type in. However, if you are good at SAS it should not be a problem. If your statistical package of choice does not have a programming feature, then you may need to talk to people in the statistics or computer science departments.
Interesting discussion here, is there a way I could calculation power of study for sample size of 90 in a cross sectional study?
Dear Olusegun, I have two dataset. One is comprising 75 patients (and it is involving the new method) and another database on 180 patients (and it is involving a new method). Let's say there are two logistic regression models predicting the Birthweight.
The question is that what should be the necessary sample size if I have both method in a mixed way in the newly/lately collected data. I can explain it in skype if you want.
Zoltan
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size