Anda di halaman 1dari 2

Practice Exam 3

CHAPTER 13
1. What is inferential statistics? Someone once said, "The value of the inference depends on the data
generated as well as the sampling." Do you agree? Why?
Inferential statistics are statistical measures of a sample that can be generalized to the population with
a known degree of accuracy (confidence interval in classical view). Yes. Obviously, sample selection is the
cornerstone of inference, but getting the data in an unbiased fashion is the only way to get the measure of the
parameter about which one makes an inference.

2. Discuss what is meant by exploratory data analysis (EDA). Be sure to provide examples on how EDA could be
used to bring meaning to data.
Exploratory data analysis is an analytical approach to gain insight into the nature of the data and the
mechanism to bring meaning to the data at hand. EDA is loosely comprised of a set of data analysis and
visualization techniques to help the researcher understand the nature of the data. The techniques used can
include plots, bar charts, histograms, frequency tables and other types of data summarization and
visualization techniques.

3. What are Type I and Type II errors? Can these errors both decrease if you were to increase sample
size? Can an increase in sample size decrease one while having no effect on the other? How?
Type I error - incorrectly rejecting null hypothesis.
Type II error - accepting null hypothesis when it is really false.
Increasing sample size can decrease the chances of a Type II error with (Type I error) a fixed.
Reducing Type I error (setting α lower) with sample size (or experimental replications) fixed, increases the
chances of a Type II error. In practice, one sets α then calculates Type II error. If it is unacceptably large then
α is reset or sample size is increased. If Type II error is small and Type I error is too large (a mistake since α
should have been set low enough to begin with), then reset α , recalculate Type II errors, and decide if more
sample size is needed and/or cost effective.

4. Is there a relationship between the level (for example, nominal, ratio) of measurement and the choice
of a parametric versus nonparametric statistical tools? What impact does the resulting empirical level of data
have on the choice of a statistical technique to test a hypothesis?
Nominal data requires nonparametric tests, as when ordinal data does not fulfill the assumptions of
interval data. Most attitude scales like the Likert normally can be assumed to be interval in nature. Therefore
most ordinal (rating data, not ranks) can be tested with parametric methods. It is important to note here that
since ratio and interval data can be transformed to nominal categories, nonparametric tests can be used on this
transformed data.

5. Is it possible that a researcher who intends to gather interval data winds up with ordinal data instead?
What does it do to his or her plans for analysis if he or she winds up with mixed types?

Of course this is possible; for example, using a ten-point (natural zero) scale where on debriefing and
pretest one finds only four values used and the differences between these values are actually and/or
perceptually (by subjects) not equal. With mixed types, one may have to use nonparametric or special
techniques for analysis.

6. When choosing a statistical test for a contingency table, such as those given in Table 13.5, what are
the basic rules of thumb that one should follow? Suppose an analyst presented you (the decision maker) with a
contingency table and accompanied it with a test you are not familiar with. What would you do?
When choosing a test for a contingency table:
1. Determine the measurement level of each variable.
2. See if alternate explanations for measurement level are available (e.g., ordinal data is empirically
nominal since most respondents only used two categories).
3. What is the nature of the hypothesis? (Comparison with known population parameters, comparison with
another sample, etc.).
4. Are some of the variables dependent and others independent?
5. Are there at least five observations per cell?

If presented with a strange test:

1. Ask for an explicit explanation that point by point answers the specific hypothesis.
2. Ask to have the hypothesis re-tested with another test.
3. Ask for background research that shows comparative powers of both tests (i.e., is one test more
powerful--more likely to reject the null with a given sample size and the same data).

7. How does sample size affect the usefulness of the Z-test and the t-test used to test the differences
between groups?
If one increases sample size with a fixed α , it makes it easier to reject the null hypothesis and it decreases
Type II error (makes it easier to accept H0), there is a difference.

8. What rules of thumb could you suggest to a manager who received a long technical report based on
survey data and who was overwhelmed by the statistical figures in the report?

First, the manager should have taken a stronger hand in the study and asked for the report to
concentrate on the solution of his managerial problem(s). It may not be too late to request a summary report
which specifically answers his questions and only refers to the long technical report for support by the
statistical/data facts when necessary. If all else fails, hire a watchdog (internal or external) consultant
immediately--before paying the bill--to make sure that the report honestly approaches the research questions
and answers them one way or the other.

9. A vice president of marketing of AT&T once asked, "What's so magic about using a α level of .05? Why should
I believe you just because it is statistically significant at the .05 level?" Can you help him out?