Hypothesis Testing

Below is a sample data set that we will be using for today’s exercise. It lists the heights for
10 men and 12 women collected at Truman State University. The data will be entered in
the same way that it was in the SPSS introduction handout. In addition to the “Height”
and “Gender” columns, we will add another column “Hfemale”. Click on the variable view
tab and name the new column. Then return to the data view spreadsheet and enter the 12
heights for the females in this new column.

Male Female
69 65
70 61
65 67
72 65
76 70
70 62
70 63
66 60
68 66
73 66
65
64

Analyzing data
We commonly are interested in comparing the mean of the population to some known
standard. We do this by creating a hypothesis, sampling from the population, calculating a
test statistic from the sample, and then determining if the sample could reasonable come
from the population if the null hypothesis was correct. Below are several examples
illustrating various hypothesis tests.

2-tailed Hypothesis test

A researcher is interested in knowing if the average height for women at Truman State
University is equal to 62.5 inches. The appropriate hypothesis are

H o :  F  62.5 inches H A :  F  62.5 inches

The appropriate test for this hypothesis is the one sample t-test. To run this test, click on
Analyze-Compare Means-One Sample T-test.
This will bring up the One-Sample T Test window. To run the test we need to move
“hfemale” into the test variable and change the test value to 62.5 [the value of the null
hypothesis]. The standard hypothesis test will be run at a significance level of 5%, but if
you did wish to change this you can click the options button and change this setting.

SPSS will not indicate whether the null hypothesis has been rejected. That is up to the
researcher to determine from the presented results. Below is the output received from
SPSS for the female heights. We will use the P-value method to determine if the null
hypothesis should be rejected. The output contains the t value for the data and its degrees
of freedom. Additionally, it displays the test value [under Ho] and a column labeled Sig.(2-
tailed).
This “Sig” value is what we refer to as the P-value. It is the probability of seeing a test
score this extreme or more [in the direction of the alternative hypothesis] if the null
hypothesis is true. The smaller the number, the more rare our test score is under Ho and the
more likely that the null hypothesis isn’t true. Using the .05 significance level as our
cutoff, we find the P-value of .028 to be in our rejection region. We reject the null
hypothesis and conclude that there is sufficient evidence that the heights of women at
Truman are not equal to 62.5 inches.

One tailed hypothesis test (example 1)

A researcher believes that the female population at Truman State is taller than the average
height of the general population of the United States [62.5 inches].

H o :  F  62.5 inches H A :  F  62.5 inches

We conduct the analysis on SPSS exactly as it was done before. The only difference will
be in the interpretation of the computer output. The P-value for the test statistics is .028,
but we need to remember that our test is a one sided test and the P-value output by SPSS is
for a 2 sided test. As we are not interested in scores occurring in the lower tail, we must
divide the outputted P-value by 2. So the P-value for the test is .014, resulting in our
rejection of the null hypothesis. We would conclude that there is sufficient evidence that
Truman State females are taller than the general population.

IMPORTANT MISINTERPRETATIONS ARE POSSIBLE!!!

Unfortunately, because SPSS does not automatically conduct a one sided test, researchers
can make incorrect conclusions by naively dividing the P-value by 2 and making a
decision. A researcher must utilize common sense to realize which side of the hypothesis
their data lies upon. We may get a “rare” P-value but it might be in the opposite direction
of the alternative hypothesis.

One tailed hypothesis test (example 2)

Assume that the researcher believes that the female population at Truman is actually
shorter than the general population [62.5 inches].

H o :  F  62.5 inches H A :  F  62.5 inches

The hypothesis are setup as before, but now the alternative hypothesis states that women
are shorter than the general population. Running our SPSS analysis exactly as before we
once again obtain a P-value of .014, indicating that we have a rare sample if the average
height of women is truly 62.5 inches. But this time we CANNOT reject Ho and rule in
favor of HA. The P-value tells us that getting an average this far away from 62.5 is a rare
event, but looking at the data tells us its an even rarer event if the average height was
actually shorter [as the alternative hypothesis assumes!]. So we conclude that there is not
sufficient evidence that Truman females are shorter than the general population.
The true P-value for this test is actually obtained by dividing the Sig value by 2 and
subtracting that from 1. So P-value is .986. This indicates that we would almost always
expect test scores to be more extreme [in the direction of the alternative] if the null
hypothesis were true [this score is not rare at all!]. Basically the data shows almost no
evidence to support the alternative hypothesis.

Comparing 2 independent samples

One question of interest in this data set might be to determine if there is a statistical
difference in the heights of men and women. The appropriate hypotheses for this question
is

Ho : M  F  0 H A : M  F  0

The appropriate test for this hypothesis is the independent samples t-test. To run this test,
click on Analyze-Compare Means-Independent Samples t-test

We then move the height and gender to their appropriate fields and click on the Define
groups button.
We choose which groups we wish to compare [we can only choose 1 & 2 in this example]
and hit continue.

The output window will contain the results of the hypothesis test. The top row of the table
assumes that the variances of the two groups are equal. The bottom row assumes that the
variances are unequal and SPSS will calculate the correct degrees of freedom  , which we
did by hand. As before the table contains t-observed, the df, and the P-value for the test.
We can see that we have a P-value of .000 [which means its zero to 3 decimal places so it
could be .0004] so we reject the null hypothesis and conclude that there is sufficient
evidence that the average height of men and women are not the same.

If we check the calculation for df assuming unequal variances

2
 3.25 2 2.75 2 
  
  10 12 
 17.77
2 2
 3.25 2   2.75 2 
   
 10    12 
10  1 12  1
We see that the hand calculations match the SPSS output. Note also that we receive the
same conclusion regardless of which test we use, though in some cases they may lead to
different conclusions.
 Independent Samples Test

Lev ene's Test f or

Equality of
Variances t-t est f or Equality of Means

Sig. Mean St d. Error

F Sig. t df (2-tailed) Dif f erence Dif f erence
HEIGHT Equal
v ariances .085 .774 4.229 20 .000 5.40 1.28
assumed
Equal
v ariances not 4.162 17.765 .001 5.40 1.30
assumed

Comparing 2 dependent samples (Paired t-test)

If the test subjects are matched and paired by certain traits when assigned to the treatment
groups, we no longer have independent samples. Now each measurement in a treatment
group can be reasonably paired with a measurement in the other treatment group [and this
was done before the experiment was run!]. The data will be placed into 2 columns with
each row corresponding to a “paired” data set. For the river water quality example done in
class we have 6 locations with a measurement at the top and bottom at each location.

To test if there is a significant difference between top and bottom we run the paired
Samples t test in SPSS.
This will bring up the data window. We highlight the two variables representing our paired
data and then use the arrow key to select them.

Paired Sampl es Statistics

Std. Error
Mean N Std. Dev iation Mean
Pair RIVERTOP .44450 6 .14177 5. 79E-02
1 RIVERBOT .53617 6 .17133 6. 99E-02
 Paired Sampl es Test

Paired Dif f erenc es

95% Conf idence
Interv al of the
Std. Std. Error Dif f erence Sig.
Mean Dev iation Mean Lower Upper t df (2-tailed)
Pair 1 RIVERTOP - RIVERBOT-. 0917 .0607 2. 48E-02 -. 15536 -2.8E-02 -3.700 5 .014

We can see that the p value for the test is well below our cutoff of 0.05 and so we would
reject the null hypothesis and conclude there is significant evidence that the concentration
of pollutants differs from the top and bottom of the river.