Beruflich Dokumente
Kultur Dokumente
A t-test is a hypothesis test of the mean of one or two normally distributed populations. Several
types of t-tests exist for different situations, but they all use a test statistic that follows a tdistribution under the null hypothesis:
Test
Purpose
Example
1 sample t-test
college students?
2 sample t-test
a target value
paired t-test
target value
t-test in
regression output
of college GPA?
An important property of the t-test is its robustness against assumptions of population normality. In
other words, t-tests are often valid even when the assumption of normality is violated, but only if
the distribution is not highly skewed. This property makes them one of the most useful procedures
for making inferences about population means.
However, with nonnormal and highly skewed distributions, it might be more appropriate to use
nonparametric tests.
Use a 1-sample t-test to estimate the mean of a population and compare it to a target or reference
value when you do not know the standard deviation of the population. Using this test, you can do
the following:
Determine whether the population mean differs from the hypothesized mean that you
specify.
For example, the manager of a pizza business collects a random sample of pizza delivery times. The
manager uses the 1-sample t-test to determine whether the mean delivery time is significantly lower
than a competitor's advertised delivery time of 30 minutes.
The test calculates the difference between your sample mean and the hypothesized mean relative to
the variability of your sample. Usually, the larger the difference and the smaller the variability in
your sample, the greater the chance that the population mean differs significantly from the
hypothesized mean.
The 1-sample t-test also works well when the assumption of normality is violated, but only if the
underlying distribution is symmetric, unimodal, and continuous. If the values are highly skewed, it
might be appropriate to use a nonparametric procedure, such as a 1-sample sign test.
For 1-sample t, the hypotheses are:
Null hypothesis
H 0: = 0
Alternative hypothesis
Choose one:
H 1: 0
H 1: > 0
H 1: < 0
Calculate a range of values that is likely to include the difference between the population
means.
For example, you want to determine whether two grain dispensers are dispensing the same amount
of grain.
2-Sample t calculates a confidence interval and does a hypothesis test of the difference between two
population means when standard deviations are unknown and samples are drawn independently
from each other. This procedure is based on the t-distribution, and for small samples it works best if
the data were drawn from distributions that are normal or close to normal. You can have increasing
confidence in the results as the sample sizes increase.
To do a 2-sample t-test, the two populations must be independent; in other words, the observations
from the first sample must not have any bearing on the observations from the second sample. For
example, test scores of two separate groups of students are independent, but before-and-after
measurements on the same group of students are not independent, although both of these examples
have two samples. If you cannot support the assumption of sample independence, reconstruct your
experiment to use the paired t-test for dependent populations.
The 2-sample t-test also works well when the assumption of normality is violated, but only if the
underlying distribution is not highly skewed. With nonnormal and highly skewed distributions, it
might be more appropriate to use a nonparametric test.
For 2-sample t, the hypotheses are:
Null hypothesis
H0: 1 2 = 0
Alternative hypothesis
Choose one:
H1: 1 2 0
The difference between the population means (1 2) does not equal the
hypothesized difference (0).
H1: 1 2> 0
H1: 1 2 0
The difference between the population means (1 2) does not equal the
hypothesized difference (0).
H1: 1 2< 0
Determine whether the mean of the differences between two paired samples differs from 0.
Calculate a range of values that is likely to include the population mean of the differences.
Determine whether the mean of the differences between two paired samples differs from 0
(or a target value)
Calculate a range of values that is likely to include the population mean of the differences
For example, suppose managers at a fitness facility want to determine whether their weight-loss
program is effective. Because the "before" and "after" samples measure the same subjects, a paired
t-test is the most appropriate analysis.
The paired t-test calculates the difference within each before-and-after pair of measurements,
determines the mean of these changes, and reports whether this mean of the differences is
statistically significant.
A paired t-test can be more powerful than a 2-sample t-test because the latter includes additional
variation occurring from the independence of the observations. A paired t-test is not subject to this
variation because the paired observations are dependent. Also, a paired t-test does not require both
samples to have equal variance. Therefore, if you can logically address your research question with
a paired design, it may be advantageous to do so, in conjunction with a paired t-test, to get more
statistical power.
The paired t-test also works well when the assumption of normality is violated, but only if the
underlying distribution is symmetric, unimodal, and continuous. If the values are highly skewed, it
might be appropriate to use a nonparametric procedure, such as a 1-sample sign test.
For paired t, the hypotheses are:
Null hypothesis
H0: d = 0
Alternative hypothesis
Choose one:
H 1: d 0
The population mean of the differences (d) does not equal the
hypothesized mean of the differences (0).
H 1: d > 0
The population mean of the differences (d) is greater than the hypothesized
mean of the differences (0).
H 1: d < 0
The population mean of the differences (d) is less than the hypothesized
mean of the differences (0).