Two-tailed tests
In a nutshell
The procedure of conducting two-tailed hypothesis tests is very similar to that of one-tailed tests. The main difference is that the relevant probability is compared to half of the significance level.
How to conduct a two-tailed test
To conduct a two-tailed test, follow this procedure.
procedure
1. | Define the test statistic and model it with the appropriate binomial distribution. |
2. | Write down the null and alternative hypotheses. |
3. | Assume the null hypothesis to be true. |
4. | Calculate the relevant cumulative probability and compare it with half of the significance level. - If the probability is less than half of the significance level, reject the null hypothesis.
- If the probability is greater than half of the significance level, do not reject the null hypothesis.
|
5. | Write a conclusion based on the context of the question. |
Deciding the relevant probability
Because the test is two-tailed, the alternative hypothesis does not indicate the inequality for the relevant probability. Instead, use the mean:
The mean of a binomial distribution with parameters n and p is np.
If the observed value is less than the mean, then the relevant probability will have a ≤ sign.
Example 1
A casino is suspected of using weighted dice. A particular die is tested for whether or not it is biased towards or against landing on 1. The die is rolled 30 times and it lands on 1 only once. Test, at the 10% significance level, whether or not the die is biased.
Define the test statistic and write down the hypotheses:
Let X be the number of times the die lands on 1.
X∼B(30,p)
H0:p=61
H1:p=61
Assume the null hypothesis to be true:
p=61⇒X∼B(30,61)
Calculate the relevant probability:
The mean of this binomial distribution is np=30×61=5. Because 1<5, the relevant probability is P(X≤1).
P(X≤1)=0.0295
Compare this with half of the significance level:
0.0295<0.1÷2=0.05, so reject the null hypothesis.
Conclude based on the context of the question:
There is sufficient evidence to suggest that the die is biased at the 10% level of significance.
Note: It is also possible to find the critical values of the test and see whether or not the observed value lies in the critical region, but the procedure above is quicker.