27/10/2016

Inferences on Population
Proportions

Use Calculation from Sample to

Estimate Population Parameter
(select)
Population Sample

(describes) (calculate)

(estimate)
Parameter Statistic
p? pˆ  63%

Statistic Parameter
 Describes a  Describes a
sample. population.
 Always known  Usually unknown
 Changes upon  Is fixed
repeated sampling.
 Examples:  Examples:

x , s 2 , s, pˆ  , 2 , , p
 27/10/2016

A Statistic is a Random Variable

 Upon repeated sampling of the same
population, the value of a statistic changes
 variable.
 While we don’t know what the next sample
will yield, we do know the overall pattern
over many, many samplings  random.
 The distribution of possible values of a
statistic for repeated samples of the same
size from a population is called the
sampling distribution of the statistic.

Proportion
 We are interested in the distribution of
y
pˆ 
n
 Note, p̂ is cY where c = 1/n is a constant and
Y is a binomial random variable.

 If Y is normally distributed, cY will also be

normally distributed.

If Y is normal cY is normal

90 92 94 96 98 100 102 104 106 108 110 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

y cy
0.5y
For example: If Y is n(µ=100,σ2=4),
then (0.5)Y is n(µ=50, σ2=1)
 27/10/2016

Distribution of Sample Proportions

A normal curve can be used to
approximate the distribution of sample
proportions if:

The size of the sample or number of

repetitions is relatively large (say 30 or
more).
While the sample size is relatively small
compared to the population size (say <
10%)

Sampling Distribution for p̂

The sampling distribution of p̂ based on
large n is approximately normal.

Sampling Distribution for p̂

To completely define the normal distribution of

p̂
We need the mean (expected value) and variance.
 27/10/2016

Sampling Distribution of p̂

1-α

p(1  p) p p (1  p)
p  z / 2 p  z / 2
n n
p(1  p)
So, at most, p̂ will be z n
/2away from p, (1-
α)100% of the time. We call this (1-α)100% the level
of confidence.

Repeated Sampling of Size n

95% 95% of the time

our estimate
will be within
p (1  p )
1.96
n
of the truth.

Standard Error
 We don’t know the value of p, so we will
use p̂

 When we use p̂ , we have an estimate of

the standard deviation for the sampling
distribution of p̂ .

 We call this estimate the standard error.

pˆ (1  pˆ )
n
 27/10/2016

Confidence Interval for p

pˆ (1  pˆ )
(1   )100%CI for p : pˆ  z / 2
n
Example:

pˆ (1  pˆ )
95%CI for p : pˆ  1.96
n

Confidence Interval Based on

Normal Distribution
Pt Est  ( Z Value) (Standard Error)

pˆ (1  pˆ )
pˆ  z  / 2
n
Standard error is our estimate of the standard
deviation for the distribution of the point
estimate.

Effect of Sample Size on a

Confidence Interval

 At a given level of confidence, as one

increases the sample size what happens to
the width of the confidence interval?

 So choose your sample size before you

estimate a proportion with a confidence
interval
 27/10/2016

Sample Size
pˆ (1  pˆ )
Maximum Error  e  z/2
n
(solving for n)

z2 / 2 pˆ (1  pˆ )
n
e2

Sample Size
 If you can’t guess a value for p̂ :

z2 / 2 (.5)(.5) z2 / 2

n 2
 2
e 4e
 What sample size would you recommend
to estimate the proportion of exeedences
with a 95% confidence interval and
maximum error of 2%?

Research Update!
 Recent research shows that we get better
coverage for (1-α)100% CI’s on
p-hat if we alter the CI formula.

pˆ (1  pˆ )
 95%CI: pˆ  z / 2
n4
where x2
pˆ  (Agresti/Coull)
n4
 27/10/2016

Estimation Procedures: An Example

Basic Logic
 In estimation procedures, statistics
calculated from random samples are
used to estimate the value of
population parameters.
 Example:
 If we know 42% of a random sample
drawn from a city are Republicans, we
can estimate the percentage of all city
residents who are Republicans.

Basic Logic
 Information from
POPULATION
samples is used to
estimate
information about SAMPLE
the population.
 Statistics are used
to estimate PARAMETER
parameters.

STATISTIC
 27/10/2016

Basic Logic
 Sampling Distribution
is the link between POPULATION
sample and
population.
 The value of the
parameters are
SAMPLING DISTRIBUTION
unknown but
characteristics of the
S.D. are defined by
theorems.
SAMPLE

Two Estimation Procedures

 A point estimate is a sample statistic
used to estimate a population value.
 A newspaper story reports that 74% of a
sample of randomly selected Americans
support capital punishment.
 Confidence intervals consist of a
range of values.
 ”between 71% and 77% of Americans
approve of capital punishment.”

Constructing Confidence
Intervals For Means
 Set the alpha (probability that the interval
will be wrong).
 Setting alpha equal to 0.05, a 95% confidence
level, means the researcher is willing to be
wrong 5% of the time.
 Find the Z score associated with alpha.
 If alpha is equal to 0.05, we would place half
(0.025) of this probability in the lower tail and
half in the upper tail of the distribution.
 Substitute values into equation.
 27/10/2016

Confidence Intervals For Means

 For a random sample of 178
households, average TV viewing was
6 hours/day with s = 3. Alpha = .05.
N=178.
 c.i. = 6.0 ±1.96(3/√177)
 c.i. = 6.0 ±1.96(3/13.30)
 c.i. = 6.0 ±1.96(.23)
 c.i. = 6.0 ± .44

Confidence Intervals For Means

 We can estimate that households in this
community average 6.0±.44 hours of TV
watching each day.
 Another way to state the interval:
 5.56≤μ≤6.44
 We estimate that the population mean is greater
than or equal to 5.56 and less than or equal to
6.44.
 This interval has a .05 chance of being
wrong.

Confidence Intervals For Means

 Even if the statistic is as much as
±1.96 standard deviations from the
mean of the sampling distribution the
confidence interval will still include
the value of μ.
 Only rarely (5 times out of 100) will
the interval not include μ.
 27/10/2016

Other confidence levels

Confidence Alpha Alpha/2 Z score
level
90% .10 .05 +/- 1.65

95% .05 .025 +/- 1.96

99% .01 .0050 +/- 2.58

99.9% .001 .0005 +/- 3.29

Constructing Confidence Intervals

For Proportions
 Procedures:
 Set alpha.
 Find the associated Z score.
 Substitute the sample information into
Formula

Confidence Intervals For

Proportions
 If 42% of a random sample of 764 from a
Midwestern city are Republicans, what % of
the entire city are Republicans?
 Don’t forget to change the % to a
proportion.
 c.i. = .42 ±1.96 (√.25/764)
 c.i. = .42 ±1.96 (√.00033)
 c.i. = .42 ±1.96 (.018)
 c.i. = .42 ±.04
 27/10/2016

Confidence Intervals For

Proportions
 Changing back to %s, we can estimate that
42% ± 4% of city residents are
Republicans.
 Another way to state the interval:
 38%≤Pu≤ 46%
 We estimate the population value is greater than
or equal to 38% and less than or equal to 46%.
 This interval has a .05 chance of being
wrong.

SUMMARY
 In this situation, identify the
following:
 Population
 Sample
 Statistic
 Parameter

SUMMARY
 Population = All residents of the
city.
 Sample = the 764 people selected
for the sample and interviewed.
 Statistic = Ps = .42 (or 42%)
 Parameter = unknown. The % of all
residents of the city who are
Republicans.
 27/10/2016

Estimating Population Parameters:

Another Example

Estimating Population Proportion

Example: When surveying 500 people selected at random
from the general population we get 200 responses of “yes”
when asked if they like broccoli. Estimate the proportion of
the population that likes broccoli.

p’ = x/n = 200/500 = 0.40

q’ = 1 - p’ = 0.60
np’ = 200 > 25 and nq’ = 300 > 25  normal aprox. is
OK
At the 95% confidence level we have  /2 = 0.025 and z/2 = 1.96

The margin for error is E = z/2(.4*.6/500) = 1.96(.022) = 0.043

P(p’- 0.043 < p < p’+ 0.043) = P(.357 < p < .443) = 0.95

Estimating Population Proportion

There are two user controlled factors that determine the
margin of error

E = z/2(p’q’/n) (1)
1. The confidence level 1 - 
The smaller , the greater z/2 and the greater E

2. The sample size n

The larger a value of n, the smaller E
If the experimenter wants to fix the width of the confidence
interval (by setting E to a pre-determined constant) and set
the confidence level (by selecting a particular ), then we
can use equation (1) above to determine the sample size
needed to achieve this level of precision.
 27/10/2016

Estimating Population Proportion

E = z/2(p’q’/n) (1)

Set E and  to desired values  E and z/2 are

constant. Solve equation (1) for n

n = z/22(p’q’)/E2 (2)

In equation (2) we have not yet taken a sample from the

population, so we cannot be sure what the proportion of
successes might be. The value of p’ that we use in this
equation may be an estimate that we make based upon
some prior knowledge, or, we may chose p’ = q’ = 0.5,
which maximizes n for particular choice of  and E

Estimating Population Proportion

Returning to our previous example, suppose we choose  = 0.05,
E = 0.025, and have no prior knowledge about p’

The from equation (2) on the previous slide we obtain

n = z/22(p’q’)/E2
Where

z/2 = 1.96 and

n= (1.96)2 (0.25)/0.0252
n = 1536.4  1537

Estimating the population mean,  known

Assume:
•Sample size, n > 30
•Population standard deviation is known

Then from the Central Limit Theorem, we know that the

sampling distribution for samples of size n
•Has mean x’ = , the mean of the original population
•Has x’ = /n
 27/10/2016

Estimating the population mean,  known

Let x’ be the mean of the sample of size n, then we

have for a confidence interval 1 -  given by

P(x’ – z/2/n <  < x’ + z/2/n) = 1 - 

Let the margin of error E = z/2/n

Then with a probability of 1 -  the population

mean lies between x’ – E and x’ + E

Estimating the population mean,  known

x mean from sample of size n

E ’
If the mean were 1 the probability of
getting a sample  x’ would be /2

1 /2

E
If the mean were 2 the
probability of getting a
sample  x’ would be /2

/2 2

Estimating Population mean with  unknown

If the population standard deviation is unknown, the

sample will have to provide both an estimate on the
population mean and standard deviation.

Estimation of the population mean will be similar to how it

was done when  is known, but the sample standard
deviation will be used instead. Student’s t-distribution will
be used to determine the margin of error, and the confidence
interval will be somewhat wider than it would be for the
same sample size if  were known.
 27/10/2016

Estimating Population mean with  unknown

Step 1. For a sample of size n, n  30, calculate the sample
mean x’, and sample standard deviation s
Recall: the sample variance s2 =(fixi2 – (fixi)2)/(n – 1)

Step 2. Convert to a standard t-score

t = (x’ - )/(s/n)
Where  is the (unknown) mean of both the original
population and the sampling distribution
Step 3. Select a confidence level 1 - , and determine t/2

Step 4. Find the margin of error E = t/2 (s/n)

Then P(x’ – E    x’ + E) = 1 - 

Estimating Population mean with  unknown

Finding t/2

Choose  = 0.05 , and assume n = 30 – From table A-3

Degrees of Area in One Tail

freedom 0.005 0.01 0.025 0.05
……………………………………………………….
29 2.756 2.462 2.045 1.699

Then t/2 = 2.045 for n –1 = 29 degrees of freedom

Estimating Population mean with  unknown

Using the TI calculator to find confidence interval for a

statistic with t-distribution
Let n = 106, x’ = 98.2, s = 0.62
Construct 99% Confidence Interval
Step 1: Select STAT > TESTS scroll down to 8: TInterval
Step 2: Select Stats if x’, n, and s are known
Select Data if these values are to be calculated from a
list
Step 3: (Stats) Use the arrow key to move to each prompt and
enter the values given above. Then Calculate <enter>
Answer: TInterval (98.081, 98.319)
 27/10/2016

Estimating Population Variance

Requirements:

1. The sample is a simple random sample

2. The population must have normally distributed
values

The sample variance has 2 distribution

2 = (n – 1) s2/2
Where
n = sample size
s2 = sample variance
2 = population variance (to be determined)

Estimating Population Standard Deviation

Confidence Interval for the Population Variance:

(n-1)s2/2R < 2 < (n-1)s2/2L

/2 /2

0 2
2L 2R

2 distribution is skewed right and always positive

Estimating Population Standard Deviation

Example: Given the following data, find the 95% confidence
interval for the population standard deviation

n = 41, x’ = 67200, s = 18277

P{(n-1)s2/2R < 2 < (n-1)s2/2L} = 0.95

First find 2R and 2L when each tail of the distribution
contains 2.5% of the area under the curve
From Table A-4 for the Chi-Square Distribution

Degrees of Area to the right of the Critical Value

Freedom 0.995 0.99 0.975 0.95 0.10 0.05 0.025
………………………………………………………………………………………….
40 20.707 22.164 24.443 26.509 51.805 55.758 59.342
 27/10/2016

Estimating Population Standard Deviation

From the previous slide we have 2R = 59.342 and 2L = 24.433
And therefore:

(n-1)s2/2R = 40 (18277)2/59.342 = 2.2516 x 108

and
(n-1)s2/2L = 40 (18277)2/24.433 = 5.4688 x 108

Thus we have:
P(2.2516 x 108 < 2 < 5.4688 x 108) = 0.95
and for the standard deviation
P(15,005 <  < 23385) = 0.95