2
$\begingroup$

I am trying to compare the proportions before and after TWO interventions in a small sample.

Basically, 10 people had intervention A, and I want to compare proportions of their success rate before and after to the 10 people who had intervention B.

My null hypothesis is that intervention A is equivalent to intervention B with respect to how they change the success rate.

Should I just be doing a Chi-Squared to just compare the Delta Success rates for the two interventions?

Also, Is McNemar something I should be considering?

Thanks in advance!

$\endgroup$
  • $\begingroup$ Did I answer your question? $\endgroup$ – Antoni Parellada May 4 '17 at 11:54
2
$\begingroup$

The McNemar test is for pair data, and I presume that the subjects in intervention A are different from the ones in intervention B.

Although you could use a chi square test of homogeneity to test if the proportion of responders is the same under both interventions, the small counts may make the Fisher exact test more suitable.

Here is a simulation in R:

> Interv <- matrix(c(8, 10 - 8, 4, 10 - 4), nrow = 2)    
> dimnames(Interv) = list(Response = c("Success", "Failure"),
+                         Intervention = c("A", "B"))
> addmargins(Interv)
         Intervention
Response   A  B Sum
  Success  8  4  12
  Failure  2  6   8
  Sum     10 10  20
> 
> chisq.test(Interv)

    Pearson's Chi-squared test with Yates' continuity correction

data:  Interv
X-squared = 1.875, df = 1, p-value = 0.1709

Warning message:
In chisq.test(Interv) : Chi-squared approximation may be incorrect

You can see that the expected frequencies in two cells are below $5$:

> (addmargins(expect <- chisq.test(Interv)$expected))
         Intervention
Response   A  B Sum
  Success  6  6  12
  Failure  4  4   8
  Sum     10 10  20
Warning message:
In chisq.test(Interv) : Chi-squared approximation may be incorrect

Here is the Fisher test:

> fisher.test(Interv)

    Fisher's Exact Test for Count Data

data:  Interv
p-value = 0.1698
alternative hypothesis: true odds ratio is not equal to 1
95 percent confidence interval:
  0.6026805 79.8309210
sample estimates:
odds ratio 
  5.430473 
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.