Hypothesis Testing Logic and Decision Methods

Scope Label

Core 9758. This branch develops the reasoning language behind one-tailed and two-tailed $Z$-tests for a single population mean.

Use it with the hub Hypothesis Testing and the procedural branch Z-Tests for a Population Mean.

A Test Measures Evidence Under $H_0$

A hypothesis test starts by temporarily assuming the null hypothesis is true.

It then asks:

Under that assumption, how unusual is the sample result?

This is the reason the sampling distribution is built under $H_0$. The test is not comparing two equally trusted claims. It is asking whether the sample gives enough evidence to reject the baseline claim.

Caption: A hypothesis test starts by assuming $H_0$, then measures how unusual the sample is under that assumption before making a decision.

Statistical Hypotheses

A statistical hypothesis is a claim about a population parameter.

For a population mean, the null hypothesis is usually

$$H_0:\mu ={\mu }_0.$$

The equality is essential because it fixes the reference distribution used for the test statistic.

The alternative hypothesis may be

$$H_1:\mu <{\mu }_0,H_1:\mu >{\mu }_0,H_1:\mu \ne {\mu }_0.$$

Choose $H_1$ from the wording and purpose of the investigation, not from the sample mean after it is observed.

Caption: The null hypothesis fixes the reference value, while the alternative hypothesis states the kind of departure being tested.

Translating Wording into $H_1$

Many testing errors happen before calculation begins.

Wording	Meaning	Alternative
has increased	definite increase	$H_1:\mu >{\mu }_0$
is greater than	definite increase	$H_1:\mu >{\mu }_0$
has decreased	definite decrease	$H_1:\mu <{\mu }_0$
is less than	definite decrease	$H_1:\mu <{\mu }_0$
has changed	either direction	$H_1:\mu \ne {\mu }_0$
is different	either direction	$H_1:\mu \ne {\mu }_0$
is affected	either direction	$H_1:\mu \ne {\mu }_0$
has overstated the mean	true mean is lower	$H_1:\mu <{\mu }_0$
has understated the mean	true mean is higher	$H_1:\mu >{\mu }_0$

For example, “the machine is not correctly calibrated” usually means the mean may be too high or too low, so the test is two-tailed.

Tail Direction

The form of $H_1$ determines where the rejection region lies.

Lower-tailed test

If

$$H_1:\mu <{\mu }_0,$$

then unusually small test-statistic values support $H_1$.

Upper-tailed test

If

$$H_1:\mu >{\mu }_0,$$

then unusually large test-statistic values support $H_1$.

Two-tailed test

If

$$H_1:\mu \ne {\mu }_0,$$

then unusually extreme values in either direction support $H_1$.

Caption: The alternative hypothesis decides which tail or tails contain the rejection region.

Significance Level

The significance level $\alpha$ is the chosen threshold for rejecting $H_0$.

It controls how extreme a result must be before it counts as sufficient evidence against $H_0$.

Typical levels are:

$$10\%,5\%,1\%.$$

A smaller significance level demands stronger evidence before rejecting $H_0$.

There is a more precise interpretation: $\alpha$ is the probability of rejecting $H_0$ when $H_0$ is actually true. In words, it is the allowed risk of a false rejection.

So a $5\%$ significance level means:

$$P(\text{reject }{H}_0\mid H_0\text{ is true})=\mathrm{0.05.}$$

This does not mean there is a $5\%$ probability that $H_0$ is true. The probability is calculated under the assumption that $H_0$ is true.

Critical-Region Method

The critical region is the set of test-statistic values that lead to rejection of $H_0$.

The critical-region method is:

1. decide the test direction from $H_1$;
2. use $\alpha$ to find the critical value or values;
3. compute the observed test statistic;
4. reject $H_0$ if the statistic lies in the critical region.

For a $Z$-test:

• an upper-tailed $5\%$ test uses the right tail;
• a lower-tailed $5\%$ test uses the left tail;
• a two-tailed $5\%$ test uses $2.5\%$ in each tail.

p-Value Method

The p-value is the probability, assuming $H_0$ is true, of obtaining a result at least as extreme as the observed result.

Decision rule:

$$\text{reject }{H}_0\text{if p-value}<\alpha .$$

A small p-value means the observed result would be unusual if $H_0$ were true. It does not mean that the p-value is the probability that $H_0$ is true.

Caption: The critical-region and p-value methods express the same testing logic in two different ways.

Critical Region Versus p-Value

The two methods are equivalent when used correctly.

Method	Question asked	Decision
Critical region	Did the statistic enter the rejection region?	Reject $H_0$ if yes
p-value	Is the observed result more extreme than $\alpha$ allows?	Reject $H_0$ if p-value $<\alpha$

Use whichever method the question asks for. If no method is specified, either is acceptable if the working is clear.

Mini-Examples: Wording Before Calculation

Overstated mean

Suppose a school claims that the mean score is at least $80$, and the question asks whether the school has overstated the mean.

The suspected direction is lower than the claim, so write

$$H_0:\mu =80,H_1:\mu <80.$$

Correctly calibrated

Suppose a machine is claimed to fill bottles with mean volume $500$ ml, and the question asks whether the machine is correctly calibrated.

Too low and too high both matter, so write

$$H_0:\mu =500,H_1:\mu \ne 500.$$

Do not change this to an upper-tailed test just because the observed sample mean happens to be above $500$.

Writing Conclusions

A testing conclusion must be cautious and contextual.

If $H_0$ is rejected:

There is sufficient evidence at the $\alpha$ level that …

If $H_0$ is not rejected:

There is insufficient evidence at the $\alpha$ level that …

Avoid saying:

• “$H_0$ is true”;
• “$H_0$ is accepted”;
• “the alternative has been proved”.

A hypothesis test makes a decision at a chosen level of evidence. It does not deliver absolute proof.

Common Pitfalls

• Choosing the tail direction from the observed sample mean instead of the question wording.
• Forgetting that $H_0$ must contain equality.
• Splitting $\alpha$ across two tails in a one-tailed test.
• Failing to split $\alpha$ across two tails in a two-tailed test.
• Treating the p-value as the probability that $H_0$ is true.
• Writing a conclusion without context.

Revision Checklist

• Can you explain why the test statistic is considered under $H_0$?
• Can you translate wording into $H_1$?
• Can you decide whether a test is lower-tailed, upper-tailed, or two-tailed?
• Can you use both critical-region and p-value methods?
• Can you write a proper conclusion without overclaiming?