Maths

Common mistakes in hypothesis testing

5 mistakes learners make with hypothesis testing, each one named and explained.

All common mistakes

Compare the probability of what you saw against the significance level, and never claim certainty.

Certainty claimed from a test

What it looks like

The conclusion says the alternative hypothesis is accepted with certainty.

Why it happens

A clear result feels conclusive, and hedged wording sounds weaker than the evidence.

A worked example

Question. A significant result at the 5 per cent level

A common answer. the coin is definitely biased

The answer. there is evidence that the coin is biased

Why. A 5 per cent test still leaves a real chance of a false positive.

Put the two side by side

Definitely: a claim no test can support.
There is evidence: what the test actually shows.

Hypothesis tests never prove. Write evidence, sufficient evidence, or reject.

The critical region put in the wrong tail

What it looks like

The region is written as greater than or equal when the alternative hypothesis was less than.

Why it happens

The inequality has to be copied from H1, and it is one symbol.

A worked example

Question. H1: p < 0.4

A common answer. X >= 7

The answer. X <= 7

Why. A one-tailed test puts the region in the tail the alternative hypothesis points at.

Put the two side by side

H1: p < 0.4 puts the region in the lower tail.
H1: p > 0.4 would put it in the upper tail.

Copy the direction straight from H1. They always match.

The critical region taken one value too far

What it looks like

A value whose cumulative probability exceeds the significance level is included in the region.

Why it happens

The two cumulative probabilities straddle the level and choosing between them is the whole question.

A worked example

Question. P(X <= 7) = 0.0301 and P(X <= 8) = 0.0611 at the 5 per cent level

A common answer. X <= 8

The answer. X <= 7

Why. 0.0611 is above 0.05, so including 8 would make the test more than 5 per cent.

Put the two side by side

X <= 7: 3.01 per cent, inside the 5 per cent.
X <= 8: 6.11 per cent, outside it.

Take the largest region whose probability is still below the level. Never the first one above it.

The probability compared the wrong way

What it looks like

A probability below the significance level leads to not rejecting the null hypothesis.

Why it happens

Two small numbers and a comparison, with no intuition to say which way it should go.

A worked example

Question. P(X >= 15) = 0.0207 at the 5 per cent level

A common answer. do not reject H0

The answer. reject H0

Why. 0.0207 is smaller than 0.05, so what was observed is unusually unlikely under H0.

Put the two side by side

Probability above the level: the result is unremarkable, so keep H0.
Probability below the level: the result is surprising, so reject H0.

Small probability means strong evidence. Small p, reject.

The wrong hypothesis rejected

What it looks like

The conclusion talks about rejecting the alternative hypothesis.

Why it happens

Two hypotheses, and only the convention says which one is on trial.

A worked example

Question. H0: p = 0.5, H1: p > 0.5, significant result

A common answer. reject H1

The answer. reject H0

Why. The null hypothesis is the one being tested; the alternative is never rejected.

Put the two side by side

H0 is on trial.
H1 is what you conclude if H0 fails.

Only H0 can be rejected. If your conclusion rejects H1, the hypotheses have swapped.

WAJD spots these patterns in your child's answers and names the one behind their wrong answers, instead of just marking them wrong.

More maths topics