You are on page 1of 23

Statistics for

Business and Economics


8th Edition

Chapter 8

Estimation: Additional Topics


Comparisons

n Many interesting questions involve comparisons:


n Do pre-school programs increase student test scores

(after relative to before)?


n What is the difference in labor market earnings

between men and women?


n How do immigrants, relative to natives, fare in the

economy?
Comparisons

n Many interesting questions involve comparisons:


n Do pre-school programs increase student test scores

(after relative to before)? (dependent samples)


n What is the difference in labor market earnings

between men and women? (independent samples)


n How do immigrants, relative to natives, fare in the

economy? (independent samples)


Dependent Samples

n Assume that we have repeated measures of the same


individuals (e.g. test scores of kids before and after a
pre-school program).
n We want to measure the average change in outcomes in
the population.
n Assume that both populations, before (X) and after (Y),
are normally distributed
n The population parameter of interest is the difference in
population means: E(Y)-E(X)
n Estimator: the difference in sample means (before vs.
after).
Difference in Sample Means
The difference for individual i:
di = yi - xi

The point estimate for n


the population mean
difference is d :
åd i
d= i=1
n
n is the number of individuals in the sample
Confidence Interval for
Mean Difference
If the standard deviation is known, the
confidence interval for the difference
between two population means, μd , is
sd
d ± za/2
n
Where
n = the sample size
(number of individuals in the sample)
Example
nYou have data on test scores for ten preschool students, before
and after the program.
Student Before After

1 60 65
2 52 58
3 41 38
4 72 78
5 52 65
6 24 37
7 16 34
8 72 62
9 44 44
10 36 45
Example
n The first step is to calculate the difference for each student.

Student Before After Difference

1 60 65 5
2 52 58 6
3 41 38 -3
4 72 78 6
5 52 65 13
6 24 37 13
7 16 34 18
8 72 62 -10
9 44 44 0
10 36 45 9
Example
n The next step is to calculate the mean difference in the sample.

Student Before After Difference

1 60 65 5
2 52 58 6
3 41 38 -3
4 72 78 6
5 52 65 13
6 24 37 13
7 16 34 18
8 72 62 -10
9 44 44 0
10 36 45 9
Mean 5.7
Example
nEquivalently, we could calculate the mean before, the mean after,
and then take the difference in these means.
Student Before After Difference

1 60 65 5
2 52 58 6
3 41 38 -3
4 72 78 6
5 52 65 13
6 24 37 13
7 16 34 18
8 72 62 -10
9 44 44 0
10 36 45 9
Mean 46.9 52.6 5.7
Confidence Interval
n The mean difference in the sample equals 5.7.
n Assume that the standard deviation for the difference is known to be 8

n The standard error equals


2
8
= 2.53
10
n The 95 percent confidence interval is thus given by:

5.7 ± 1.96 ´ 2.53 = [0.74,10.66]


n 95 percent of confidence intervals constructed in this way will include
the true mean difference in the population.
Difference Between Two Means:
Independent Samples

Goal: Form a confidence interval


for the difference between two
population means, μx – μy
n Different data sources: The sample selected from one
population has no effect on the sample selected from the
other population. (e.g., earnings for men and women).
n The point estimate is the difference between the two
sample means:

x–y
Normal with σx2 and σy2 known

When σx and σy are known and both populations are


normal, the standard error of the estimator equals:

var( X - Y )

var( X ) + var( Y )
2 2
σx σy
+
nx ny
Confidence interval

Then, the confidence interval is given by:

σ 2
σ 2y
(x - y) ± z α/2 +
x
nx ny
Example
n A researcher wants to measure the difference in income between
male-headed and female-headed households in the population.
n Data from the Survey of Consumer Finances:
group Sample size Mean income (1,000s) Sd of income (1,000s)
Male-headed 4,592 106.8 439.9
Female-headed 1,423 37.8 70.2
n 95 percent confidence interval for difference in income:
2
σ 2x σ y
(x - y) ± z α/2 +
nx ny

439.92 70.22
(106.8 - 37.8) ± 1.96 +
4592 1423
69 ± 13.2
Two Population Proportions

Goal: Form a confidence interval for the


difference between two population
proportions, Px – Py

Assumptions:
Both sample sizes are large and
independent

The point estimate for


the difference is
pˆ x - pˆ y
Two Population Proportions

n The standard error in this case is calculated


as:
var(p̂ x - p̂ y )

var(p̂ x ) + var( p̂ y )

Px (1 - Px ) Py (1 - Py )
+
nx ny
Confidence Interval for
Two Population Proportions

The confidence limits for Px – Py are:

pˆ x (1- pˆ x ) pˆ y (1- pˆ y )
(pˆ x - pˆ y ) ± za / 2 +
nx ny
Support for Stricter Gun Laws
Confidence interval calculation
n A recent survey shows that 64 percent support
stricter gun control.
n Among women, 72 percent (560/775) support gun
control.
n Among men, 56 percent (420/751) support gun
control.
n Form a 90 percent confidence interval for the
difference in proportions.
Example:
Two Population Proportions
420
Men: p! ! =
751
= 0.56

560
Women: p! " =
775
= 0.72

p̂ x (1 - p̂ x ) p̂ y (1 - p̂ y )
+
nx ny

0.56(0.44) 0.72(0.28)
= + = 0.024
751 775

For 90% confidence, Za/2 = 1.645


Example:
Two Population Proportions

The confidence limits are:

p" ! (1 − p" ! ) p" " (1 − p" " )


("p! − p" " ) ± Zα/2 +
n! n"
= (0.56 − 0.72) ± 1.645(0.024)

so the confidence interval is

-0.199 < Px – Py < -0.121


Summary

n Many interesting questions involve comparisons.


n We have developed confidence interval estimates for
both independent and dependent samples
n We also constructed confidence interval estimates for
sample proportions.

You might also like