Hello, I have carried a binary logistic analysis. The independent variables were dichotomous and continuous variables. The dependent variable was binary of course. Now when I ran the analysis the results where as follows;significant pvalue , OR not equal to 1 and wide confidence interval. How can I interpret the results?
a) all factors with significant p value are significantlily associated with the outcome and the or and confidence interval were due to outliers or small sample size.
B) I cannot say the factors are significantly associated with outcome based on p value alone.
the other question is there is any way I could run the analysis differently?
thank you in advance
Neither a) nor b) are correct.
You can say that the inclusion of the predictors with very small p-values statistically significantly increased the likelihood of the data under the model in with all other predcitors are included.
A small comment on Jochen's answer : it depends on which kind of p-value you consider. If it correspond to likelihood ratios tests, Jochen's interpretation is correct. If it corresponds to Wald test of coefficient, then it means that this coefficient is significantly different from 0 in the model. It may be slightly different especially in case of logistic regression models…
Note that this is very subtle distinction, since ideally the two should be equivalent [at least for predictors associated with a single parameter], but relying on asymptotic distributions instead of (unknown) exact ones breaks down this equivalence (just it also breaks down the equivalence between the significance of the Wald test and the [quite often, profiled] confidence interval containing or not 0)…
- Badges
- Science method