Dear friends,
I am looking for a procedure to calculate sample sizes/evaluate the power in logistic regression models with 3+ predictors which are partly binary partly continuous variables. Can someone help me with references/software?
Thanks in advance!
Guilherme
Dear Edward,
I do not think GPower can help me with logistic regression. All the tests I see listed there refer to linear regression. Some guys such as Vittinghoff & McCulloch say it may be fine to have EPV = 10 or even less, but in their simulations the smallerst sample sizes were 128. In my case I have only n = 62 and the proportion of hits in the DV = 50%.
Hello Guilherme. One of the posters in the Stack Exchange thread below did find something about logistic regression in the G*Power documentation:
- https://stats.stackexchange.com/questions/162294/multiple-logistic-regression-power-analysis
Bear in mind too that rules of thumb about events-per-variable (EPV) are about avoiding over-fitting of the model, not about ensuring adequate power to detect some minimally important effect. Re EPV rules of thumb, I like Mike Babyak's discussion in this article:
- https://people.duke.edu/~mababyak/papers/babyakregression.pdf
HTH.
Hi All,
In G*Power go to z tests and you will find Logistic regression as an option. However, I’m not certain it is setup to help with your particular problem, but worth a look.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size