Quiz (Chapter 8)
Effect Size, Power, CIs, and Bonferroni
The Seven-Step Version of Hypothesis Testing
- (T/F) The last step of the 7-step version of hypothesis testing involves
determining whether Ho is true.
- (T/F) A result that's statistically significant can be completely
devoid of any practical significance.
- The "yardstick" that's used to measure practical significance is
- actual significance
- effect size
- Look at Excerpts 8.3, 8.4, 8.5, and 8.6. In these excerpts, what
4 letters were used to represent measures of effect size?
- (T/F) Effect size can be measured numerically.
- What three words do researchers use to describe the 3 different
"sizes" of effect size?
- What exactly would it mean if a researcher stated that his/her statistical
test had a power of .90?
- In a post hoc power analysis, is the effect size computed
by the researcher or is it chosen by him/her?
- If Ho is not rejected but statistical power is quite high,
would the chances of a Type II error be high or low?
The Nine-Step Version of Hypothesis Testing
- How many steps of the 6-step version of hypothesis testing appear
in the 9-step version of hypothesis testing?
- (T/F) In the 9-step version of hypothesis testing, the researcher
specifies ES after analyzing the sample data.
- (T/F) In the 9-step version of hypothesis testing, the sample
data are analyzed before a power level is specified.
- Of the 2 kinds of effect size measures ("standardized" or "raw"),
which one is better?
- (T/F) The needed sample size for a study cannot be determined unless
both ES & power are specified first.
- In Excerpt 8.9, if the n had been 50, the power would have been
___ (higher/lower) than .8.
- In Excerpt 8.10, which kind of effect size was used, standardized
- There is general agreement that power should be no lower than ___
- (T/F) If a researcher reaches a fail-to-reject decision at the end
of the 9-step version of hypothesis testing, a critic cannot
legitimately argue that a Type II error was made because the test was
Hypothesis Testing Using Confidence Intervals
- (T/F) When hypothesis testing is conducted via one or more
confidence intervals, there's no Ho or Ha.
- If a 95% confidence interval is used when doing hypothesis testing,
the level of significance = ___ .
- If a researcher has a sample with data on two variables, if r = +.40,
and if a 95% confidence interval around r extends from +.20 to +.57,
should Ho be rejected if Ho: r
= 0 (and Ha: r
- (Yes/No) In Excerpt 8.14, is the number .80 indicating statistical
- (T/F) If, in Excerpt 8.15, 40% of the people in the supervised
group had improved, the result might have been a nonsignificant difference
between the two
Adjusting for an Inflated Type I Error Rate
- (T/F) A Type I error is made when a researcher fails-to-reject
a null hypothesis that's really false.
- If you blindly select 1 card from a single well-shuffled deck of
cards, the probability that your card will be a "club" is .25 (i.e.,
1 out of 4). Instead of doing that, suppose you blindly select
a single card from each of 3 well-shuffled decks of cards. Here,
the probability that you'll end up with at least one club in
your set of 3 cards is:
- less than .25
- more than .25
- Do the terms "heightened probability of Type I error" & "inflated
Type I error risk" mean the same thing?
- Would the Bonferroni adjustment technique ever be used in a study
involving a single null hypothesis?
- (T/F) In Excerpt 8.19, the p-level of .017 would have been .025 if
there had been 2 comparisons (rather than 3 comparisons).
- When the Bonferroni adjustment technique is used, critical values
become ____ (more/less) demanding.
- Researchers sometimes use the "_____ modification" rather than the
Bonferroni adjustment technique.
A Few Cautions
- (T/F) The notion of an "effect size" (in the 7-step version of hypothesis
testing) is exactly the same as the notion of an "effect size" (in the
9-step version of hypothesis testing).
- (T/F) The numerical criteria for "small," "medium," and "large"
effects remain the same regardless of the statistical focus or the
kind of ES that's computed.
- (T/F) The 6-step version of hypothesis testing fails completely
to address the distinction between statistical significance and practical
- (Yes/No) If 21 null hypotheses are tested, each at a=.05,
and if Ho is true in each and every one of the 21
cases, would it be smart to bet that one or more Type I errors will
be made among the 21 tests.
Two Questions that are Supposed to be a Bit Challenging
- If all bivariate correlations are computed among 5 variables, and
if each r is tested to see of it is significantly different from 0,
what Bonferroni-corrected alpha level should be used if the researcher
desires to have a 5% chance of one or more Type I errors occurring?
- Suppose 2 researchers (Mary & Larry) each secretly pull a sample
from the same giant population. They both measure their subjects
on the same 2 variables (X & Y), they both test Ho: r
= 0, they both conduct a 2-tailed test, and they both set a=
.05. If Mary uses 5,000 subjects while Larry uses only 50, will
Mary's Type I Error rate be "inflated" as compared to Larry's?
Click here for answers.