Beruflich Dokumente
Kultur Dokumente
by Christoph Strehl
SUMMARY ........................................................................................................................................................... 3
CONCEPT OF HYPOTHESIS TESTING ......................................................................................................... 3
GENERAL PROCEDURE OF HYPOTHESIS TESTING............................................................................... 4
DEFINE THE BUSINESS OBJECTIVE ....................................................................................................................... 4
STATE THE HYPOTHESIS ...................................................................................................................................... 4
CHOOSE AN ALPHA-VALUE.................................................................................................................................. 4
CALCULATE THE TEST-STATISTIC ........................................................................................................................ 5
DECIDE WHETHER TO ACCEPT OR REJECT THE NULL-HYPOTHESIS ...................................................................... 5
INTERPRET THE DECISION .................................................................................................................................... 5
OVERVIEW OF TEST TYPES........................................................................................................................... 6
PARAMETRIC TESTS ............................................................................................................................................ 6
1-sample T-test .............................................................................................................................................. 6
1-sample Z-Test ............................................................................................................................................. 6
2-sample T-test .............................................................................................................................................. 6
2-sample F-test .............................................................................................................................................. 6
NON-PARAMETRIC TESTS .................................................................................................................................... 6
Chi square-test............................................................................................................................................... 6
Sign Test ........................................................................................................................................................ 6
Wilcoxon Signed Rank test ............................................................................................................................ 7
Mann Whitney U-test ..................................................................................................................................... 7
Kolgomorov-Smirnoff test.............................................................................................................................. 7
CONCLUSION...................................................................................................................................................... 7
BIBLIOGRAPHY ................................................................................................................................................. 7
Summary
“Too few researchers understand what statistical hypothesis testing does and doesn't do, and
consequently their results are misinterpreted. Even more commonly, researchers understand
elements of statistical hypothesis testing, but the concept is not integrated into their research.
For example, the influence of sample size on statistical significance may be acknowledged by
a researcher, but this insight is not conveyed when interpreting results in a study with several
thousand subjects.”1
A perfectly conducted market research study is worthless if you don’t know how to interpret
its results. Hypothesis tests can help you to make quantitative and objective conclusions based
on your survey results. So it is important to understand how to choose among different
Hypothesis tests and how to interpret them.
This report concentrates on the general concept of Hypotheses testing and will not go into
details of statistical reasoning or specific tests. Non-statisticians will be able to understand the
basic concepts as well as the power and the caveats of hypotheses tests.
So basically, in Hypothesis testing the researcher states two hypotheses – the null and the
alternative hypothesis - , computes the test and finally decides in favour of one of those
hypothesises.
When deciding to accept or to reject a particular hypothesis, one can never be sure of making
the correct decision. Two types of errors are possible: Accepting a false hypothesis and
rejecting a true hypothesis. With H0 denoting the null hypothesis, the following table lists the
four possible constellations:4
H0 is true H0 is false
Accept H0 No Error Type II error
Reject H0 Type I error No Error
If for example a company the produces soda wanted to analyze a survey and publish an
advertisement that says that more than 75% prefer their soda instead of the competitor’s
product, their Null-hypothesis would be “Less than or equal to 75% prefer our product” and
the alternative hypothesis would be “More than 75% prefer our product”. Of course the
sample data may not support the alternative hypothesis and they have to accept the null-
hypothesis, but this is the only way to be able to gain statistical difference for the hypothesis
the company supports.
Choose an alpha-value
As stated above, there are two types of error that can be made. The Type-I-error is the error if
you reject the Null-Hypothesis although it is in fact true. This error can be limited.
Unfortunately, there proves to be a “trade-off” between the probabilities of both error-types:
Lowering the probability for a type-I-error leads to an increase in the probability of a type-II-
error and vice versa.5
The alpha-value, denoted by α, is the probability of making a type-I-error. The most common
choice for an alpha is 5%. That means if one were to hypothetically perform a test 100 times,
one would make the wrong conclusion in about 5 of these tests. Arriving at the wrong
conclusion in this context means that one has concluded that the results are not due to chance,
when in fact they were.6
But nevertheless the choice of the “right” alpha is a subjective choice that very much depends
on the specific business problem. For example an airline company that tries to make
inferences about if the tires of their planes have to be replaced or not will much likely choose
a very low alpha, because making a mistake in that decision would be disastrous for both the
company and the customers. On the other side a company who wants to publish a brand-
awareness result may choose a very high alpha because they don’t care about making a type-I-
error.
As an output, almost every statistical tool will return you a so called p-value that can be easily
used to decide in favour of one of the two hypotheses. That decision is described in the next
step.
In order to decide for one of the two hypotheses one has to compare the calculated p-value to
the prespecified alpha: If the p-value is lower than alpha, the Null-Hypothesis has to be
rejected in favour of the alternative hypothesis.
There is one exception to that rule. If the test is a two-sided test, the p-value has to be divided
by two before comparing it to the alpha.
It gets more tricky if the Null-Hypothesis has to be accepted. In this case the only statement
that can be drawn from that is that there is not enough statistical significance to reject the
Null-Hypothesis. IT DOES NOT mean that there is a statistical significance that the Null-
Hypothesis is true, which is a commonly made mistake.
Parametric Tests
1-sample T-test
A very commonly used test to make inferences about the mean of a population from a sample
mean. Questions that can be answered with this test might be:
• Is the average money that people spend in a fast food restaurant greater than 4$?
• Is the awareness of my product less than 50%?
1-sample Z-Test
This test is very similar to the 1-sample T-Test, but it requires knowledge about the standard
deviation in the population in order to be computed. Because of this restriction which is
seldom fulfilled this test is not frequently used.
2-sample T-test
This test compares two samples and tries to find out differences about both populations.
Questions for this test might be:
2-sample F-test
Test to determine whether the two variances are equal.
• Does the deviation of the satisfaction of my customers in two different areas differ?
Non-parametric tests
Chi square-test
This test makes inferences about the relationship of two variables. A possible question is:
Sign Test
A test that can be applied for data with a poor quality. That is data where only positive or
negative changes were measured.
Kolgomorov-Smirnoff test
This test can make inferences about the population distribution of a sample.
• Is the money someone spends per visit in a fast food restaurant normally distributed?
Conclusion
-There are many different tests out there that can be executed. A good market researcher
doesn’t have to know all of them nor know how to compute them. But he has to understand
the general procedure of hypothesis-testing in order to be able to correctly interpret results
from practical market research.
Bibliography
1
Thompson, Bruce (1994). The concept of statistical significance testing. Practical
Assessment, Research & Evaluation, 4(5). [online] Retrieved March 1, 2005 from
http://PAREonline.net/getvn.asp?v=4&n=5
2
David M. Lane (2003), HyperStat Online Textbook. [online] Retrieved March 1, 2005 from
http://davidmlane.com/hyperstat/A29136.html
3
William M.K. Trochim (2002), Hypotheses [online], Retrieved March 1, 2005 from
http://www.socialresearchmethods.net/kb/hypothes.htm
4
Missong, Martin(2004), MMC-class „Applied Data Analysis“, Lesson 4 „Statistical
Hypothesis Testing“
5
Missong, Martin(2004), MMC-class „Applied Data Analysis“, Lesson 4 „Statistical
Hypothesis Testing“
6
Ed Colet (2002), THE (IN)SIGNIFICANCE OF STATISTICAL SIGNIFICANCE [online],
Retrieved March 1, 2005 from http://www.tgc.com/dsstar/00/1003/102239.html
7
iSixSigma (2003), p-value [online], Retrieved March 1, 2005 from
http://www.isixsigma.com/dictionary/P-Value-301.htm
8
StatSoft Inc (2003), Nonparametric Statistics [online], Retrieved March 1, 2005 from
http://www.statsoft.com/textbook/stnonpar.html