Dear colleagues,
I am struggling to calculate a minimum detectable odds ratio for a multivariate logistic regression model, with a binary categorical variable as the outcome (exposure to intimate partner violence or IPV). I have around 10 predicting variables (independent variables) in the model.
I go to STATA, go to sample power, precision, and sample-size analysis, then select odds ratio. There are only two options- and I need to select 2x2 tables because it is not a case-control study, is a cross-sectional study.
Then I get the screen below. I put "target odds ratio" with the actual sample size of 307, power of. 80% (0.8) and standard error of 0.05.
My question is what should I put for "stratum-specific group ratios" and "effect size"?
and another question is because this is not a binary logistic regression model, but multivariate logistic regression model, is this the correct way to calculate the minimum odds ratio? This means, in my model, I do not have the main predicting factor of the interest, but all the 10+ factors were equally important to detect any statistically significant associations.
finally, If I can get OR with p < .005, is it enough, without calculating the minimum detectable OR, or would I still need to get a minimum detectable OR for the predicting factors for this model, so that I can say that "even though the adjusted OR for exposure of X for IPV victimization was not statistically significant, it exceeded the minimum detectable OR." ?
FYI, I am doing this, because our initial sample size was 900, but I only have 570 samples, because of the COVID pandemic which shut down schools and our data collection had to be stopped.
My supervisor told me to justify RETROSPECTIVELY, that my sample size was enough to answer to my research questions.
Thanks for any tips you may provide.
Rinko
I am struggling to calculate a minimum detectable odds ratio for a multivariate logistic regression model, with a binary categorical variable as the outcome (exposure to intimate partner violence or IPV). I have around 10 predicting variables (independent variables) in the model.
I go to STATA, go to sample power, precision, and sample-size analysis, then select odds ratio. There are only two options- and I need to select 2x2 tables because it is not a case-control study, is a cross-sectional study.
Then I get the screen below. I put "target odds ratio" with the actual sample size of 307, power of. 80% (0.8) and standard error of 0.05.
My question is what should I put for "stratum-specific group ratios" and "effect size"?
and another question is because this is not a binary logistic regression model, but multivariate logistic regression model, is this the correct way to calculate the minimum odds ratio? This means, in my model, I do not have the main predicting factor of the interest, but all the 10+ factors were equally important to detect any statistically significant associations.
finally, If I can get OR with p < .005, is it enough, without calculating the minimum detectable OR, or would I still need to get a minimum detectable OR for the predicting factors for this model, so that I can say that "even though the adjusted OR for exposure of X for IPV victimization was not statistically significant, it exceeded the minimum detectable OR." ?
FYI, I am doing this, because our initial sample size was 900, but I only have 570 samples, because of the COVID pandemic which shut down schools and our data collection had to be stopped.
My supervisor told me to justify RETROSPECTIVELY, that my sample size was enough to answer to my research questions.
Thanks for any tips you may provide.
Rinko
Comment