DASHBOARD LEARN MENU

Learn VEE Mathematical Stats 4 4.1 4.1.2 One-Tailed Tests

One-Tailed Tests

As mentioned, a two-tailed test differs from a one-tailed test by the number of tails included to form the
rejection region. This is driven by a researcher's specific interest. In the case of a fair coin, both extremes of
too few or too many heads would question whether the coin is fair. Thus, a two-tailed test is appropriate for
that scenario. However, there are situations where only one extreme is of concern: too extreme to the left, or
too extreme to the right, but not both.

For example, consider a factory claiming that the mean number of defective manufactured products per day is
4. The factory would typically be concerned if the mean is in fact much higher than 4, but not concerned if it is
lower than 4. A one-tailed test is more suited in this scenario.

Before proceeding further, note that numerous different combinations of the null and alternative hypotheses fall
under a one-tailed test. Despite these nuances, the core flow of a one-tailed test is the same. In general, to
determine which type of test applies, we compare the two hypotheses to identify the appropriate rejection
region.

Right-Tailed Tests
Let's continue using our trivial setup. A possible set of hypotheses involving parameter μ and constant d is

H0 : μ ≤ d H1 : μ > d

An extreme x in the right tail of the normal distribution would support H1 more than H0 , but an extreme x in
the left tail would not. This means the rejection region is in the right tail, given by

x≥b ⇒ z≥c

As expected, the critical value is determined by the significance level, where

Pr(Z ≥ c ∣ H0  is true) = α

As seen in the graph below, α is not split into two regions. Therefore, for the same α , the critical value here
will be different from the critical value in the two-tailed case.
Although H0 does not specify an exact value for μ , use the exact point that separates the two hypotheses,
i.e. use μ = d .

In summary, here are the changes for this case:

• The rejection region only involves the right tail of the sampling distribution.

• The significance level is Pr(Z ≥ c ∣ H0  is true) .

• The decision logic is:
◦ If z ≥ c for μ = d , then z ≥ c will hold for all other possible μ values in H0 . Therefore, reject
H0 .
◦ If z < c for μ = d , then the test statistic is not extreme enough for some subset of possible μ
values in H0 ; there's not enough evidence to deem H0 unreasonable. Therefore, do not reject
H0 .
• The p -value is Pr(Z ≥ z ∣ H0  is true) .

Let's try a few examples. The first is a variation on Example 4.1.1.

EXAMPLE 4.1.2

It is believed that the mean age of licensed drivers in 2017 is 43.7. A researcher thinks that the mean age is
actually higher. One licensed driver's age is observed to be 60 in 2017. These ages are normally distributed
with variance 80.
Test whether the researcher is justified at the 5% significance level.

SOLUTION

Letting μ represent the mean age of licensed drivers in 2017, the null and alternative hypotheses are

H0 : μ = 43.7

H1 : μ > 43.7

Although H0 differs from the form we introduced, this is still a right-tailed test. This is because H1 would be
supported by observing an extreme older age, thus justifying the researcher's claim. On the other hand,
observing an extreme younger age would not justify the researcher's claim.

Next, determine the critical value. Since α = 0.05 ,

Pr(Z ≥ c) = 0.05

The graph indicates that c is the 95th percentile of Z . From the Z -table, c = 1.645 .

Next, calculate the test statistic.

x−μ 60 − 43.7
z= = −− = 1.82
σ √80
Since 1.82
> 1.645 , the test statistic falls in the rejection region, and thus we reject the null hypothesis of
μ = 43.7 in favor of μ > 43.7 at the 5% significance level.

COACH'S REMARKS

Recall that for Example 4.1.1, H0 was not rejected at the same 5% significance level. This is because when
including both tails, the driver's age of 60 is not among the "top 5% of extreme values".

However, when including only the right tail, 60 is among the "top 5% of extreme values", thus H0 is rejected
here.

EXAMPLE 4.1.3

X is normally distributed with mean μ and variance 22. The following set of hypotheses are considered:

• H0 : μ = 5
• H1 : μ = 10

With a single observation and at the 5% significance level, calculate the power of this test.

SOLUTION

This time, both H0 and H1 differ from the form we introduced. Nevertheless, this is still a right-tailed test. This
is because the μ hypothesized by H1 is larger than the one by H0 .

First, determine the critical value in the unit of X . For α = 0.05 , we previously saw that c = 1.645 .
Therefore,

b−5
−− = 1.645 ⇒ b = 12.716
√22
Hence, the rejection region is x ≥ 12.716 .

It is implied that the power is calculated based on H1 . This produces the answer of

12.716 − 10
Pr(X ≥ 12.716) = Pr (Z ≥ −− )
√22
= Pr(Z ≥ 0.58)
= 1 − Pr(Z < 0.58)
= 1 − 0.719
= 0.281

Left-Tailed Tests
Now consider the hypotheses

H0 : μ ≥ d H1 : μ < d

An extreme x in the left tail of the normal distribution would support H1 more than H0 , but an extreme x in
the right tail would not. This means the rejection region is in the left tail, given by

x≤a ⇒ z ≤ −c

where the significance level is

Pr(Z ≤ −c ∣ H0  is true) = α

Similar to the previous case, use μ = d in the calculations when assuming H0 as true.

In summary, here are the changes for this case:

• The rejection region only involves the left tail of the sampling distribution.

• The significance level is Pr(Z ≤ −c ∣ H0  is true) .

• The decision logic is:
◦ If z ≤ −c for μ = d , then z ≤ −c will hold for all other possible μ values in H0 . Therefore,
reject H0 .

◦ If z > −c for μ = d , then the test statistic is not extreme enough for some subset of possible μ
values in H0 ; there's not enough evidence to deem H0 unreasonable. Therefore, do not reject
H0 .
• The p -value is Pr(Z ≤ z ∣ H0  is true) .

EXAMPLE 4.1.4

The mean total cholesterol for overweight individuals is said to be at least 250mg/dL. A drug chemist thinks the
mean total cholesterol is actually lower. An overweight individual is observed to have a total cholesterol of
208mg/dL. Assume total cholesterol for these individuals is distributed normally with variance v .

At the 3% significance level, determine the largest value of v that would result in supporting the chemist.
SOLUTION

Letting μ represent the mean total cholesterol of overweight individuals, the null and alternative hypotheses
are

H0 : μ ≥ 250

H1 : μ < 250

Since α = 0.03 ,

Pr(Z ≤ −c) = 0.03

The graph indicates that c (as opposed to −c ) would be the 97th percentile of Z . From the Z -table,
c = 1.88 . In turn, the critical value −c = −1.88 .

To support the chemist (i.e. reject H0 ), the test statistic must be less than or equal to the critical value.
Remembering that μ = 250 for the test statistic,
 208 − 250
≤ −1.88
√v
42
⇒− ≤ −1.88
√v
 42 

⇒ ≥ 1.88
√v
The largest value of v that would result in rejecting H0 is approximately 499 .
42
⇒ √v ≤
1.88 ■
⇒ v ≤ (22.3404)2

The following table summarizes the three variants of a hypothesis test, including the critical value formulas in
terms of zp , the 100p th percentile of Z . Remember that z without a subscript is the test statistic, and α is
the significance level.

Left-Tailed Two-Tailed Right-Tailed

Reject H0 if z ≤ −c |z| ≥ c z≥c

Critical Value −z1−α z1− α
2
z1−α

MORE INFORMATION

For left-tailed tests, we denote the critical value as −c , not c . Other resources may use different notation.
For example, they may use only one symbol to represent the critical value, regardless of which variant is being
discussed.

Ultimately, it is simply a different way to express the same thing. If you do consult other resources, pay close
attention to how things are defined.
Discussions
Previous Lesson Next Lesson
Watch 4.1.1 Type I and Type II Errors, Power of a Test Watch 4.1.2 One-Tailed Tests