Hello Everybody! I feel very good to see the communication here regarding use of appropriate sample size for using binary logistic regression. My sample size is 390 cases for dependent variable (150 in "YES" category & 240 in "NO" category ) My Independent variables are 10. So, by applying the formula N= 10*k/ p, there is one confusion, I am calculating the "p"
Question: 1 Would somebody help me with in how to calculate "p"=????????
Question: 2 if i focus on 20 to 1 ratio of valid cases to Independent variables,then in my case; i have 390 cases and 10 independent variables, SO, 390/10= 39:1 this is more than 20 to 1. Is it right??????????
you ask *why are you estimating sample size if you already have a sample?* My understanding is that Syed wants to estimate the power of his study considering the actual N (or the projected sample size). If the power is not acceptable at the alpha level chosen, he will increase N by adding more subjects or continue measurements.
salve!
Dear Syed,
There is a formula for sample size for binary logistic regression in http://www.statstutor.ac.uk/resources/uploaded/sample-size.pdf. It is based on the relative probabilities of the two choices.
p is that of the level of significance chosen in your study, usually 0.01 or 0.001 depending of the domain.
you ask *why are you estimating sample size if you already have a sample?* My understanding is that Syed wants to estimate the power of his study considering the actual N (or the projected sample size). If the power is not acceptable at the alpha level chosen, he will increase N by adding more subjects or continue measurements.
salve!
In using multivariate analysis , Kerlinger and Pedhazur recommend using 30 observations per independent variable for the stability of Beta Weights. Since you have 390 observations and 10 independent variables then, I would say that you did really fulfill the requirements of having an appropriate sample size and you better worry in your analysis about violating the assumptions underlying the error term
I think he wants to know if his sample size is big enough for the size of his model, 10 predictors. I think Kerlinger overdoes it. However if you use 30 x 10 predictors is 300 < 390 so I agree with Abdulrazzak. I would have said 10 but in any case you're fine. His comment about the assumptions is very important.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size