University College London · FACULTY OF STATISTICS

BENV0139 Chap.4 Inference and Hypothesis Tests

- one subject, every graph, every model, every mark
7 Chapters4-page Bible
Our own words - no uploaded lecturer files
Updated for this semester
Chapter 4 of 8 · BENV0139

Inference and Hypothesis Tests

Inference and Hypothesis Tests explains how evidence is evaluated against a null model without turning p-values into probabilities that a claim is true. A test begins with explicit null and alternative hypotheses, a pre-specified tail and an error threshold. Type I errors reject a true null; Type II errors miss a specified alternative.

Standard error describes uncertainty in an estimated mean, while a two-sample t statistic compares an observed mean difference with its estimated sampling variation. Independent and paired designs require different calculations. A small p-value can accompany an unimportant effect, and a large p-value does not prove equivalence.

Confidence intervals show both magnitude and uncertainty, while power depends on effect size, spread, sample size and alpha. The chapter closes every test with a heritage consequence and the assumptions that could change it.

In this chapter

What this chapter covers

  • 01

    Null and alternative hypotheses

  • 02

    P-value interpretation

  • 03

    Type I error

  • 04

    Type II error

  • 05

    Standard error

  • 06

    Independent and paired t tests

  • 07

    Confidence intervals, effect size, power and practical significance

Worked example · free

Calculate a pooled two-sample statistic

Q [8 marks]. Samples have means 52 N and 47 N, sizes 12 and 10, and SDs 6 N and 5 N. Calculate the pooled t statistic. The marks are AskSia practice weighting, not a published UCL marking scheme.
The decision matrix separates correct conclusions from a Type I false positive and a Type II missed effect.The decision matrix separates correct conclusions from a Type I false positive and a Type II missed effect. test decisiondo not rejectreject H0H0 trueH0 false correctType Ifalse positive Type IImissed effectcorrecttruth
The decision matrix separates correct conclusions from a Type I false positive and a Type II missed effect.
  • 2State the null of equal population means and the independent-group assumption. The marks are original practice weighting.
  • 2Calculate pooled S = sqrt[((11)(36) + (9)(25))/20] = 5.57 N.
  • 2Calculate SE = 5.57 sqrt(1/12 + 1/10) = 2.38 N and t = 5/2.38 = 2.10.
  • 2Use 20 degrees of freedom, then report the five-newton effect and assumptions alongside the p-value.
The pooled statistic is about 2.10 with 20 degrees of freedom. The practice weighting is not a UCL marking scheme.
Sia tip — Never write the significance label without the estimated difference, its uncertainty and a sentence about practical importance.
Glossary

Key terms

Null hypothesis
The parameter claim used to derive the reference distribution for a test statistic.
Type I error
Rejecting the null hypothesis when it is true.
Statistical power
The probability that a planned test rejects the null for a specified alternative effect.
FAQ

Inference and Hypothesis Tests FAQ

What does a p-value mean?

It is the probability, under the null model, of obtaining the observed result or one at least as extreme according to the test statistic. It is not the probability that the null is true and does not measure the practical importance of the effect.

Does non-significance prove no difference?

No. The data may be compatible with zero and with effects that matter. Inspect the estimate and confidence interval, consider precision and power, and use an equivalence design only when the goal is to support a practically negligible difference.

When should I use a paired test?

Use a paired method when observations are naturally linked, such as before and after measurements on the same object or matched sites. Analyse within-pair differences because an independent-groups calculation discards the link and uses the wrong uncertainty structure.

Study strategy

Exam move

Use a fixed hypothesis-test template: research claim, null and alternative, one-sided or two-sided direction, assumptions, effect estimate, standard error, statistic, degrees of freedom, p-value or critical comparison, confidence interval and practical conclusion. Practise switching between independent and paired examples until the observational link determines the test automatically.

Draw a two-by-two error table from memory and attach a heritage cost to each error, such as adopting a harmful treatment or missing a protective measure. Rework p-value statements that incorrectly say the null is probably true or the result happened by chance. For every significance result, write a companion sentence about magnitude. For every non-significant result, list the effects still compatible with the interval.

Study power prospectively by specifying the effect worth detecting. Finish with assumptions about independence, measurement quality and distributional approximation. That final audit makes the inference useful rather than ceremonial. Recalculate the pooled example on blank paper, showing pooled spread, standard error, signed difference, statistic and degrees of freedom with units.

Then change the design so each fibre is measured before and after treatment and explain why the analysis moves to paired differences. For every result, place zero and a pre-defined practically important effect on an interval line; write what benefit and harm remain compatible instead of relying on a binary label.

Practise correcting five forbidden claims: the null is probably true, the result occurred by chance, non-significant means identical, significant means important and p equals the replication probability. Create a multiple-testing scenario and decide which outcomes were primary before seeing results. Plan power around a named conservation threshold, anticipated spread and independent sample size.

End with a decision paragraph that reports effect direction and units, explains the relevant Type I and Type II consequences, names the strongest assumption and gives a conditional recommendation. If variance equality is doubtful, state why the uncertainty method should change without altering the observed mean difference.

Working through Inference and Hypothesis Tests in BENV0139? Sia is AskSia’s AI Statistics tutor — ask any BENV0139 Inference and Hypothesis Tests question and get a clear, step-by-step explanation grounded in how BENV0139 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.

A+Everything unlocked
Unlocks this Bible + all 32 of your subjects at University College London, plus 1,000+ more Bibles across the AskSia library.
Sia - your BENV0139 tutor, unlimited, worked the way the exam marks it
The full 4-page Bible + practice bank with worked solutions
Chrome extension - sync your LMS so Sia knows your deadlines
Bilingual EN / Chinese on every Bible and every Sia answer
$0.99 Trial
30-day money-back · cancel in one tap · how it works