Chapter 9

advertisement

Chapter 9

Fundamentals of Hypothesis Testing:

One-Sample Tests

9.1: Hypothesis Testing Methodology

• Confidence Intervals were our first Inference

• Hypothesis Tests are our second Inference

• “Methodology” implies a series of steps:

1. Develop hypotheses

2. Determine decision rule

3. Calculate test statistic

4. Compare results from 2 and 3: make a decision

5. Write conclusion

Step 1: Develop hypotheses

• You will need to develop 2 hypotheses:

1. Null hypothesis

2. Alternative hypothesis

– Hypotheses concern the population parameter in question (ie “µ” or “π” or other)

The Null Hypothesis

• A theory or idea about the population parameter .

• Always contains some sort of equality.

• Very often described as the hypothesis of “no difference” or

“status quo.”

• H

0

: µ = 368

The Alternative Hypothesis

• An idea about a population parameter that is the opposite of the idea in the null hypothesis

• NEVER contains any sort of equality!

• H

1

: µ

368 (sometimes use H a

:)

Hypotheses

• Null and alternative hypotheses are mutually exclusive and collectively exhaustive.

• Our sample either contains enough information to reject the NULL hypothesis

OR the sample does not contain enough information to reject the null hypothesis.

“Proof”

• There is no proof.

• There is only supporting information.

Step 2: Decision Rule or

“Rejection Region”

• Always says something like “we shall reject H

0 for some extreme value of the test statistic.”

• The “Rejection Region” is the range of the test statistic that is extreme enough—so extreme that the test statistic probably would not occur IF the null hypothesis is true.

• Figure 9.1 shows the rejection region for a hypothesis test of the mean.

• The “critical value” is looked up based on the error rate that you are comfortable with.

Step 3: Calculating the Test

Statistic

• The test statistic depends on the sampling distribution in use. This depends on the parameter.

• This will be determined the same way it was in chapter 8.

Step 4 & 5: Decision and

Conclusion

• The decision is always either (1) reject

H

0 or (2) fail to reject H

0

. This is determined by evaluating the decision rule in step 2.

• The conclusion always says “At α =

0.05, there is ( in )sufficient information to say H

1

Alpha

• α is the probability of committing a Type I error: erroneously rejecting a true Null

Hypothesis.

• α is called “The Level of Significance”

• α is determined before the sample results are examined.

• α determines the critical value and rejection region(s).

• α is set at an acceptably low level.

Beta

• Beta is the probability of committing a Type II error: erroneously failing to reject a false null hypothesis.

• Beta depends on several factors and it cannot be arbitrarily set. Beta can be indirectly influenced.

Compliments of Alpha and Beta

• (1-α) is called the confidence coefficient.

This is what we used in Chapter 8.

• (1-beta) is called the Power of the test.

Power is the chance of rejecting a null hypothesis that ought to be rejected, ie a false null. Bigger is better. Power cannot be set directly.

9.2: z Test of Hypothesis for the mean

• Use this test ONLY for the mean and only when σ is known.

• There are two approaches:

– critical value approach

– “p” value approach

Critical Value Approach

Remember your methodology (steps):

1. Create hypotheses

2. Create decision rule

– depends on α

– depends on distribution

3. Calculate test statistic

4. State the result

5. State the managerial conclusion

Hypotheses

• The discussion in 7.2 assumes a twotail test because the sample mean might be extremely large or extremely small.

– Either one would make you think the null hypothesis is wrong.

Rejection Region

• The standard approach requires that the value of α be divided evenly between the tail areas.

• These tail areas are called the

“rejection region.”

Conclusion

• See Step 6 on pages 308-309.

“p-value” approach

• Rewrite the decision rule to say, “we will reject the null hypothesis if the ‘p-value’ is less than the value of α .”

• “p-value” definition, page 309.

• “p-value” is called the observed level of significance.

• Excel--most statistical software--does a good job of this (that’s why it’s a popular approach).

Estimation and Hypothesis Testing

• The two inferences are closely related.

• Estimation answers the question “what is it?”

• Hypothesis Testing answers the question “is it ______ than some number?”

• See page 312.

9.3: One-Tail Tests

• The rejection region is one single area.

• Sometimes called a directional test.

• Mechanics:

– see the text example

• Problem identification:

– hypothesis test or interval estimation?

– One-tail or Two-tail?

– If One-tail, which is the null?

Text Example

• Page 314, the “milk problem.”

• Are we buying “watered-down milk” ?

– Watered-down milk freezes at a colder temperature than normal milk.

– What are the null and alternative hypotheses?

– Hint: what do you want to conclude?

– Hint: what is the hypothesis of action?

– Hint: what is the hypothesis of status quo?

Mechanics of the One-tailed test

• Different hypotheses.

• Different decision rule/rejection region.

• Different “p-value” or observed significance of observed level of significance.

Consider Problem 9.44, page 317

Reading only the context, not the steps (a, b, etc.), can you tell that the problem calls for a hypothesis test?

– Knowing that a test of hypothesis is called for, can you determine that a one-tail test is appropriate?

• Knowing that a one-tail test is to be used, can you set up the hypotheses?

9.4: t test of Hypothesis for the

Mean (σ Unknown)

• When σ is unknown, the distribution for x-bar is a “t” distribution with n-1 degrees of freedom.

• Use sample standard deviation “s” to estimate σ.

• This test is more commonly used than the z test.

Assumptions

• Random Sample.

• You must assume that the underlying population, i.e. the underlying random variable x is distributed normally.

• This test is very “robust” in that it does not lose power for small violations of the above assumption.

Methodology

You still need your 5 step Hypothesis Test

Methodology.

– The critical value approach is the same as that for the z test.

– The p-value method does not work as well when done by hand because of limitations in the “t” table.

– One- and Two-tail tests are possible.

• You could add another step to check assumptions.

EXAMPLE

• 9.54, page 323

9.5: z Test of Hypothesis for the Proportion

• For the nominal variable—variable values are categories and you tend to describe the data set in terms of proportions.

• Both one- and two-tail tests are possible.

• Problem 9.72 on page 329 is a good example.

Assumptions

• The number of observations of interest (successes) and the number of uninteresting observations

(failures) are both at least 5.

Download