Chapter 26

Question 1

Power=1-beta. What is beta?

Answer 1

The probability of a Type II Error, usually .2

Question 2

Is it okay to make adjustments to alpha and beta pragmatically?

Answer 2

Yes, as long as it is clearly revealed.

Question 3

What does changing the alpha or beta accomplish?

Answer 3

It prevents wasteful, frivolous studies when the scientist finds that they cannot achieve a significance level and/or power that is acceptable.

Question 4

As sample size increases, what happens to…
power?
Type II Errors?
Ability to detect small effects/differences?
The margin or error of the confidence interval?

Answer 4

• increases
• decreases
• increases
• increases
• decreases

Question 5

Outliers (extreme
observations) can occur for a
variety of reasons. Some of
concern, some not.
• Explanations:

Answer 5

Random chance: its OK; it should happen.
• Biological diversity: it is important and interesting.
• Mistake: should be recognized and removed.
• Invalid assumption: be aware that you may have the
wrong model in mind. Maybe it’s not an outlier
under a lognormal model of variation.

Question 6

If you think that you would
remove outliers, then you
should probably have objective
criteria a priori.
• Because:

Answer 6

Humans have a tendency to see too many outliers.
• But,
• It can be legitimate to remove outliers if you are
concerned that they could lead to invalid results.
• Just don’t make a post hoc decision to remove an
outlier simply to make the results significant.

Question 7

Some Statistical Tests Are More Robust

• i.e.

Answer 7

Not greatly affected by
outliers or other differences in
distribution

Question 8

Some Statistical Tests Are More Robust

Examples:

Answer 8

Median vs Mean:
• The median will change very little if an outlier is
removed, but the mean will change more.
• Nonparametric Rank-Based Tests:
• This type of nonparametric test converts values of all
of the observations to their ranks.
• The outlier simply becomes the top or bottom
ranked observation, but all information about how
far it is from the rest of the observations is removed.