Transcript Document
Chapter 11
Statistical Inference: OneSample Confidence Interval
I Criticisms of Null Hypothesis Significance
Testing
Does not indicate whether the effect is large or
small
1
Answers the wrong question: Prob(D|H0). The
correct question concerns Prob(H0|D).
Is a trivial exercise; all null hypotheses are false.
Turns a continuum of uncertainty into a reject-donot reject decision.
II Confidence Interval for
A confidence interval for is a segment on the
real number line such that that has a high
probability of lying on the segment.
2
f (t ) /2 = .05/2
/2 = .05/2
1 = .95
t
t.05/2,
0
t.05/2,
Figure 1. Sampling distribution of t. If one t statistic is randomly
sampled from this population of t’s, the probability is .95 that the
obtained t will come from the interval from –t.05/2, to t.05/2, .
3
1. From Figure 1, the following probability statement
follows:
Prob(t.05/ 2, t t.05/ 2, ) 1
2. Replacing t with ( X ) / (φ / n ) and using
some algebra gives the following 100(1 – )%
two-sided confidence interval for
t / 2, φ
t / 2, φ
Pr ob(X
X
) 1
1 44 2 4 n43
1 44 2 4n 43
L1
L2
4
3. L1 and L2 denote, respectively, the lower and
upper endpoints of the open confidence interval
for .
4. A researcher can be 100(1 – )% confident that
is greater than L1 and less than L2.
5. The probability (1 – ) is called the confidence
coefficient and is usually equal to (1 – .05 ) = .95.
5
6. The assumptions associated with a confidence
interval are the same as those for a one-sample t
statistic.
A. Computational Example: Two-Sided Interval
1. Consider the following hypotheses for the
Idle-On-In College registration example:
H0: = 0
H1: ≠ 0
6
2. A two-sided 100(1 – .05) = 95% confidence
interval for , where X = 2.90, φ = 0.3013,
and t.05/ 2,26 2.056, is
X
t .05/ 2,26φ
n
2.90
2.056(0.3013)
t .05/ 2,26φ
X
)
n
2.90
27
2.056(0.3013)
27
2.78 3.02
7
3. The dean can be 100(1 – .05) = 95% confident
that is greater than 2.78 and less than 3.02.
4. The dean can be even more confident that lies
in the interval from L1 to L2 by computing a
100(1 – .01) = 99% confidence interval.
8
5. A two-sided 100(1 – .01) = 99% confidence
interval for , where t.01/2, 26 = 2.779, is given by
X
t .01/ 2,26φ
n
2.90
2.779(0.3013)
t .01/2,26φ
X
)
n
2.90
27
2.779(0.3013)
27
2.74 3.06
9
6. Graphs of the two confidence intervals
L1 = 2.78
2.6
2.7
2.8
L2 = 3.02
2.9
3.0
3.1
3.2
95% confidence interval for
L1 = 2.74
2.6
2.7
2.8
L2 = 3.06
2.9
3.0
3.1
3.2
99% confidence interval for
10
7. As the dean’s confidence that she has captured
increases, so does the size of the interval from L1
to L2.
B. More On the Interpretation of Confidence
Intervals
C. Computational Example: One-Sided Interval
1. Suppose that one-tailed hypotheses, H0: ≥ 0
and H1: < 0, reflect the dean’s hunch about
the new registration procedure.
11
2. A one-sided 100(1 – .05) = 95% confidence
interval for , where X = 2.90, φ = 0.3013,
and t.05,26 1.706, is
t .05,26φ
X
)
n
2.90
1.706(0.3013)
27
3.00
12
3. Comparison of one- and two-sided confidence
intervals
L2 = 3.00
2.6
2.7
2.8
2.9
3.0
3.1
3.2
One-sided 95% confidence interval for
L1 = 2.78
2.6
2.7
2.8
L2 = 3.02
2.9
3.0
3.1
3.2
Two-sided 95% confidence interval for
13
D. Advantages of Confidence Interval Estimation
Over Hypothesis Testing
1. Hypothesis testing is not very informative. A
confidence interval narrows the range of possible
values for .
2. Confidence intervals can be used to test all null
hypotheses such as H0: = 0. Any 0 that lies
outside of the confidence interval corresponds to a
rejectable null hypothesis.
14
3. A sample mean and confidence interval provide
an estimate of the population parameter and a
range of values—the error variation—qualifying
the estimate.
4. A 100(1 – )% confident interval for contains
all of the values of 0 for which the null
hypothesis would not be rejected.
15
III Practical Significance
A. Estimator of Cohen’s d
1. Hedges’s g for the registration example
| X 0 | |12.90 3.10 |
g
0.66
φ
0.3013
2. Interpretation of g
g = 0.2 is a small effect
g = 0.5 is a medium effect
g = 0.8 is a large effect
16
3. Computation of g from t statistics in research
reports
g
t
n
4. For the registration example, t = 3.449 and n = 27
g
t
n
3.449
0.66
27
17