Hi:)
Trying to find a way to perform non inferiority analysis in one of the above softwares.
Couldn't find except in PASS which I don't have.
Any suggestions?
Thx!
Suggest you google non inferiority analysis. There are some helpful references on methods and software available for these methods which are somewhat known in statistics. I believe you can find programs that may be helpful to you as well. Some are freeware. Best wishes, David Booth
Hello Sagi,
Non-inferiority simply means that a 95% (typically) CI has a lower bound for the observed effect that is not smaller than the negative value of some chosen margin, delta. This is generally straightforward to determine, once you have the computed effect.
Consider a randomized study in which the target sample treatment is compared to a standard treatment, with n1 = n2 = 25 cases per group. Let the effect size be quantified as Cohen's d: (M1 - M2)/Sp (Sp is the pooled within-groups SD), and the margin chosen for the non-inferiority is +/- 0.25. If the observed ES is 0.35, then the 95% CI for the effect size is (-0.22, 0.91). Thus, the lower bound does not fall below -delta, and one concludes that the target treatment is not inferior to the standard treatment.
The CI for the ES, d, requires use of the non-central t distribution (unless d = 0, in which the ordinary t distribution will work for the CI). So for me, it's easiest to use a predefined function, d.ci, within the R library, psych.
R text and sample output:
# load the psych library
library(psych)
# specify observed ES, and the two group sizes
d.ci(0.35, n1=25, n2=25)
#
### output for this example:
# lower effect upper # [1,] -0.2165608 0.35 0.9095212
###
Good luck with your work.
David Morse thank you very much for the excellent answer.
95% CI is wonderful, but I need the p-value as well. Does the function you metioned enalbes to recieve it?
In addition, what if my outcome isn't parametric, meaning it is ordinary or binary? Is it possible to analyse by the d.ci package? I dougt because standard deviation isn't appropriate.
Thanks!!
A two sided confidence interval of 95 per cent has a p-value of .025. a 1 sided has p=.05
Best wishes, David Booth
BTW this assumes a sampling distribution symmetric about it's mean.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS