Scribd
Upload
24 views

Testing A Sample Mean: Large Samples: Using The Normal Distribution

1) The document discusses using statistical tests like the z-test and t-test to analyze sample data and test hypotheses about population parameters like means and proportions. 2) As an example, it analyzes a sample of 30 orders that averaged 450 units to test if there has been a fall in demand from the known population mean of 500 units. The z-score is calculated to be -1.37, which does not provide enough evidence to reject the null hypothesis that demand has not fallen. 3) It also discusses testing a claim about the proportion of cars needing repair by using a sample of 50 cars to calculate a z-score and compare it to the critical value, in this case not rejecting the

Uploaded by

bo
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
24 views

Testing A Sample Mean: Large Samples: Using The Normal Distribution

1) The document discusses using statistical tests like the z-test and t-test to analyze sample data and test hypotheses about population parameters like means and proportions. 2) As an example, it analyzes a sample of 30 orders that averaged 450 units to test if there has been a fall in demand from the known population mean of 500 units. The z-score is calculated to be -1.37, which does not provide enough evidence to reject the null hypothesis that demand has not fallen. 3) It also discusses testing a claim about the proportion of cars needing repair by using a sample of 50 cars to calculate a z-score and compare it to the critical value, in this case not rejecting the

Uploaded by

bo
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

Large samples: using the Normal distribution

Testing a sample mean

A firm knows that the average size of order received is 500 units, with s.d. 200. The last 30
orders have averaged 450 units. Does this indicate a fall in demand?

The hypothesis is

H0:  = 500 demand has not fallen


H1:  < 500 demand has fallen.

If H1 is true, the firm might want to take action, e.g. layoffs. But we don't know which is
true hence we could take the wrong action, e.g. layoff workers, reduce investment when
demand has not fallen, or continue as before when demand has fallen.

Using the 5% criterion (5% is the significance level of the test), the critical value of the
standard Normal distribution is 1.64. That is, 1.64 standard errors from the mean cuts off the
bottom 5% of the distribution. If our sample evidence yields a z-score of below –1.64, then
the null is rejected.

The z-score is

x 450  500


z   1.37
2
s n 200 2 30

Hence we do not reject H0 on this basis. In fact, -1.37 cuts off 8.53% in the tail of the
distribution, implying that there is an 8.53% chance of getting such sample evidence if the
null is true. Only if the probability is less than 5% do we reject.

The probability 8.53%, associated with the calculated test statistic, is known as the prob-
value (or sometimes P-value) of the test. Hence the null is only rejected if the prob-value is
less than the chosen significance level.

One and two tail tests

If H1 were that demand had not changed, we would have a two tail test (H1:   500) and the
critical value would be 1.96. Otherwise the test is identical and the null rejected if the test
statistic falls outside the range –1.96 to +1.96.

Testing a sample proportion

A car manufacturer claims that no more than 10% of its cars need repair in first three years.
This is doubted by a consumers’ association. A sample of 50 three year old cars finds that
eight needed attention. Does this contradict the maker's claim?

Sample data: p=0.16, n=50.


H0:  = 0.10
H1:  0.10

Choosing the 5% significance level, the critical value is z*= 1.96.

The test statistic is

z
p   
0.16  0.10
 1.41
  1    0.1  0.9
n 50

Since z < z* we do not reject H0 with 95% confidence.

The effect of sample size

Suppose the data were p=0.11, n=3500. Then z = 1.972, and H0 is rejected. However, the
extra 1% is not very large, and H0 is rejected simply because of large sample size. Hence
ideally we should alter the significance level with sample size. Hence it is important to look
at the effect size as well as the significance.

Testing two sample means for equality

For the hypothesis

H0: µ1- µ2= 0


H1: 1-2  0

the test statistic is

( x1  x2 )  ( 1   2 )
z
s12 s22

n1 n2

Testing two proportions

For the hypothesis

H0: 1- 2 = 0
H1: 1- 2  0

the test statistic is

( p1  p2 )  ( 1   2 )
z
P1  P  / n1  P1  P  / n2

2
where P = (n1p1+n2p2)/(n1+n2). We must have a common value P because it is presumed by
the null hypothesis.

Small samples: using the t distribution

Testing the sample mean

The local tax bill in Tory boroughs is known to be £280. A sample of 10 Labour boroughs
reveals the average bill to be £300, with s.d. 38.58. Do Labour authorities charge more than
Tory ones?

H0: µ=280
H1: µ>280

Significance level of the test: 5%

Critical value of the test: t* = 1.833 ( = 9)

Test statistic:

x  300  280
t   1. 64
s2 / n 38.582 / 10

t < t* hence cannot reject H0.

Testing two sample means for equality

( x1  x2 )  ( 1   2 )
t
S 2 n1  S 2 n2

( n1  1) s12  ( n2  1) s22
where S² =
n1  n2  2

You might also like