I have 77 data from Extension worker and the issue for investigation is to identify the factors that influence the performance of extension worker. Dependent variables are binary in nature and there are 6 independent variables with both binary and other scales of measurement.
Moreover, in the case of the dependent variable, I have five different questions that I would like to run separately in this model.
Is it OK for running a binary logistic regression?
As a rule of thumb the minimum sample size in such cases is 50+8*p where p is the number of predictors (independent variables)
With p=6 you need a sample size of at least 98. Beware of multicollinearity.
In my article entitled "The relative standard error as an index for checking the reliability of regression coefficients" I have described complications in situations like yours. Please see examples 4 and 5. The article is on my Research Gate profile.
You say dependent variables that means you have more than one, do you have more than one? What are those 5 questions?
Yes, in dependent variable I have five different questions in different issues.
This is not exactly the right question. The minimum number is 2 or 3 per treatment. However, the outcome from such a small sample size is not much better than a random number generator. So the response to your question is that you need a sample size sufficient to have sufficient power to answer your question with the appropriate level of accuracy. Of course there are many studies that are done and published where sample size was dependent on budget and workload. A program like GPower can help you work through this http://gpower.hhu.de/. You might also consider reading this: http://journals.plos.org/plosmedicine/article?id=10.1371/journal.pmed.0020124 and the more recent articles that cite this work.
Do you have a reasonable prediction/confidence interval on the predicted probability for each question?
As a rule of thumb the minimum sample size in such cases is 50+8*p where p is the number of predictors (independent variables)
With p=6 you need a sample size of at least 98. Beware of multicollinearity.
In my article entitled "The relative standard error as an index for checking the reliability of regression coefficients" I have described complications in situations like yours. Please see examples 4 and 5. The article is on my Research Gate profile.
since your objective is to identify the variables that influence performance use the standardized regression coefficients (beta-weights) for interpretations.
There's not an unique approach for this question.
However, after conducting your analysis, you should check the fit of the model, if variables increasing/decreasing odd ratio of dependent binary fact/variable are theoretically accurate and, of course, the uverall success rate (0-100%) of your model.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size