5

I get the same p-value for a one-tailed and two-tailed exact binomial test for some combinations of x, n, and p. Does anyone know why this happens? What formula does R use to calculate the p-value for two-tailed binomial tests?

>binom.test(x=2,n=9,p=.1, alternative="greater")

Exact binomial test

data:  2 and 9
number of successes = 2, number of trials = 9, p-value = 0.2252
alternative hypothesis: true probability of success is greater than 0.1
95 percent confidence interval:
0.04102317 1.00000000
sample estimates:
probability of success 
             0.2222222 

> binom.test(x=2,n=9,p=.1, alternative="two.sided")

    Exact binomial test

data:  2 and 9
number of successes = 2, number of trials = 9, p-value = 0.2252
alternative hypothesis: true probability of success is not equal to 0.1
95 percent confidence interval:
0.02814497 0.60009357
sample estimates:
probability of success 
             0.2222222
Cliff AB
  • 20,980
user109857
  • 53

1 Answers1

4

This hypothesis test is based on the so-called Clopper-Pearson interval. The one sided-test ''greater'' computes the probability to have a value at least 2 under a binomial density:

size<-9
x<-2
p<-0.1
# compute probability that outcome is >= 2, i.e. x:size
p.value<-sum(dbinom(x:size, prob=p, size=size))

p.value

If you do a two-sided test, you look at the probability that the outcome is equal to 2 (dbinom(x=2,prob=p,size=size)) and you compute the probability of all outcomes with a probability lower than or equal to that:

all.probs<-dbinom(0:size, prob=p, size=size)
# probability that the outcome is x=2
prob.x<-dbinom(x=x, prob=p, size=size)

# probability of all outcomes that have a probability lower than or equal to the probability of the outcome 2 (prob.x) 
p.value<-sum(all.probs[all.probs <= prob.x])
p.value

If you draw the plot (plot(0:size, all.probs) ) you will see that there are no outcomes with a probability smaller than or equal to that for the outcome 2 in the left queue.

If you do the same with size<-30 you will see that there are such outcomes in the left queue (and that for n=30 test.binom will give different results for one- and two-sided tests).