week 8 - effect size
TRANSCRIPT
-
7/27/2019 Week 8 - Effect Size
1/16
Effect SizeCohens d & r2
-
7/27/2019 Week 8 - Effect Size
2/16
Statistical versus
Practical Significance With Z-tests and t-tests, we are evaluating the
statistical significance of our result. Can we
reject the null hypothesis? Practical significance addresses a different
question: Assuming the result is statistically
significant, is it practically significant?
Is the effect big enough to matter?
Effect Size is a measure of the magnitude of the
effect. How big an effect was it?
-
7/27/2019 Week 8 - Effect Size
3/16
Measures of Effect Size:
Cohens d, and r2
Cohen's d = standardized
mean difference betweengroups
One sample t-test:
OR
Two-sample t-test
Cohens d: Small effect = .20
Med. effect = .50
Large effect = .80
r2 = proportion of variancein the DV that is
accounted for by the IV
r2 = t2 / (t2 + df)
r2 : Small effect = .01
Med. effect = .09
Large effect = .25
sXd /)( 1
d (X1 X2) /s2p
-
7/27/2019 Week 8 - Effect Size
4/16
Subjective Life
Expectancy Accuracy of Subj Life Expectancies: A subjective
life expectancy is an individuals estimate of what
age he or she expects to live to. One interestingquestion is how subjective life expectancies
compare with actual life expectancies as
represented by actuarial predictions. Financial
experts worry that many people underestimate
how long they will live and in doing so fail to plan
or save enough money to carry them throughout
their lives.
-
7/27/2019 Week 8 - Effect Size
5/16
Subjective Life
Expectancy In a study of this issue, Robbins, (1988) asked 49
women to indicate their expected age in response
to the question: Approximately how long do youexpect to live? These estimates were then
compared with actuarial predictions from the
National Center for Health Statistics. Results
showed that mean subjective life expectancy for
females in the samples was 77.2 years with a
standard deviation of 3.4 years. The actuarial
prediction was a mean of 78.2 years. Do females
significantly underestimate their lifespans?
-
7/27/2019 Week 8 - Effect Size
6/16
Descriptive Statistics &
Hypotheses H0: SubLE = 78.2
H1: SubLE < 78.2
Sample N = 49
= 78.2 years
Sample Mean = 77.2
years
s = 3.4 years
-
7/27/2019 Week 8 - Effect Size
7/16
Expectations Sample distribution of the Mean if we did this experiment over and
over again, each time drawing 49 females, what values for mean lifeexpectancy would I expect?
Mean of means = = 78.2
SE = 3.4/(49) = 0.49 Approx normal (N > 30)
Decision rule: p = .05 but now fatness of tails change w/ samplesize
Crit t (48) estimated by being more conservative
Crit t (40) = 2.021
Calculate obtained and find critical test stats:
Obt t (48) = (77.2 78.2) / 0.49 = -2.04
-
7/27/2019 Week 8 - Effect Size
8/16
Substantive Conclusion On average, women significantly underestimate their life
expectancy compared to actuarial predictions. Thus, the financial
experts may be correct they may not be planning ahead
sufficiently
-
7/27/2019 Week 8 - Effect Size
9/16
Cohens d Cohen's d is commonly used measure of effect size
Used to indicate the standardized difference between 2 means.
The difference between two means divided by a standard deviationfor the data
Cohens d: Small effect = .20
mid effect = .50
large effect = .80
sXd /
-
7/27/2019 Week 8 - Effect Size
10/16
How Cohens d Relates to the
Example
d = | | / s = |77.2 78.2| / 3.4 = .29
Here, we rejected H0 and said women significantly underestimate
their life spans. But the effect size is relatively small, so we might
have profited from a relatively large sample of 49.
So, it does have an effect, but how much does it actually tell us about
how much life span estimates are affected?
Effect size not influenced by n
Used before conducting research to see how much power (i.e., n) we
need, based on similar research effect size(s)
X
-
7/27/2019 Week 8 - Effect Size
11/16
Graduates Optimism
Example A psychologist has prepared an Optimism Test that is
administered yearly to graduating college seniors. The test
measures how each graduating class feels about its future thehigher the score, the more optimistic the class. Administratorsare concerned that this years class might be significantly lessoptimistic about the job market than the class 5 years ago due to
the economy. Five years ago, the graduating class had a mean
score of 12.3. A randomly-selected sample of 9 seniors from this
years class was selected and tested. Their scores are:
-
7/27/2019 Week 8 - Effect Size
12/16
XX
10X
-
7/27/2019 Week 8 - Effect Size
13/16
Graduates Optimism
Example IV & DV
H0 & H1
What do you expect under null
Stats
Conclusion
-
7/27/2019 Week 8 - Effect Size
14/16
Graduates Optimism
Example DV = optimism scores IV = surveyed before or now
H0: Now = 12.3 (Before) H1: Now < 12.3
Expect under null
Mean of means = = 12.3
SE = 1.14
Approaches normal (above N=30)
Decision rule based on =.05 critical t(8) = 2.306
Stats obtained t(8) = (10-12.3)/1.14 = -2.02
Conclusion
Substantive: Based on this sample, there is insufficient evidence tosuggest this years class is any more or less optimistic than the class from5 years ago.
-
7/27/2019 Week 8 - Effect Size
15/16
Effect size in t-test Cohens d
d = |mean diff|/s = |10-12.3|/3.43 = 0.67
Thus, even though unable to reject H0, effect size suggests that
there is something going on. That there has been a decrease in
optimism in graduating seniors compared to 5 years ago. We
just had very low power to detect the effect
-
7/27/2019 Week 8 - Effect Size
16/16
Power Is the probability that the test will reject a false null hypothesis
(that it will not make a type II error).
As power increases, the chances of a Type II error decrease.
Increases chance of Type I error.
It can also be used to calculate the minimum
effect size that is likely to be detected in a study using a given sample
size. sample size required to accept the outcome of a statistical test with a
particular level of confidence.