The Hong Kong Polytechnic University · FACULTY OF STATISTICS

RS517 Chap.7 Factorial and Multivariate Analysis

- one subject, every graph, every model, every mark
6 Chapters4-page Bible
Our own words - no uploaded lecturer files
Updated for this semester
Chapter 7 of 10 · RS517

Factorial and Multivariate Analysis

Interactions before averages

A factorial model estimates main effects and interactions. A main effect averages across the levels of another factor, while an interaction shows that one contrast changes across those levels. When profiles cross, an overall mean contrast may represent no setting well.

Repeated and adjusted models

Group-by-time interactions ask whether change profiles differ.

ANCOVA models a covariate, often baseline, and can improve precision when the role is prespecified. Post-treatment mediators should not be added casually because they change the estimand.

Multiple outcomes

MANOVA evaluates a group difference across a related outcome vector, and MANCOVA adds covariates.

A multivariate result does not identify the responsible outcome, so follow-up analysis needs a planned interpretation and control of multiplicity.

Read interactions before averaged effects

A factorial model can estimate the effect of each factor and whether one effect changes across the other. When an interaction is present, a main effect averages across levels that may have different or opposite contrasts.

Interpret the interaction estimate and simple effects with intervals, then decide whether an overall average remains useful.

The fact that one subgroup is significant and another is not does not itself prove interaction; the difference between effects needs its own direct test.

Respect repeated and baseline structure

Group-by-time models distinguish overall group differences, overall time change and whether change differs between groups. Within-group significance tests do not answer the between-group change question.

Baseline-adjusted follow-up analysis can gain precision when the relationship is modelled appropriately, whereas change scores incorporate baseline measurement error into the outcome.

Post-treatment variables such as adherence may be mediators, so routine adjustment can remove part of the treatment effect or introduce selection.

Use multivariate models for a joint question

MANOVA and MANCOVA model a vector of related outcomes and test a joint contrast.

A significant multivariate result does not identify the responsible outcome, make every follow-up significant or guarantee practical importance. Outcome selection needs a scientific rationale, adequate sample size and a plan for interpretation. Adding every collected measure can create an unstable covariance model and a vague hypothesis.

Follow-up analyses should preserve the joint question and control the claims actually pursued.

Revision drill

Sketch parallel, diverging and crossing profile plots. For each, write the interaction sentence before any main-effect sentence. Classify candidate covariates as baseline predictors, confounders, mediators or consequences of outcome.

Finish by defining a two-outcome vector whose joint interpretation is meaningful, and name what a significant omnibus result still leaves unanswered about the separate outcomes.

In this chapter

What this chapter covers

  • 01

    Main effects and interactions

  • 02

    Simple effects and profile plots

  • 03

    Between-person and within-person factors

  • 04

    Group-by-time interpretation

  • 05

    ANCOVA and covariate roles

  • 06

    MANOVA and MANCOVA

Worked example · free

Interpret crossing treatment profiles

Q [8 marks]. Practice note: the marks used here are not part of the university's published assessment scheme. An intervention improves an outcome by eight points in urban clinics and worsens it by two points in rural clinics. The overall effect is three points.
The crossing profiles show an interaction: the intervention improves the outcome in one setting and worsens it in the other.The crossing profiles show an interaction: the intervention improves the outcome in one setting and worsens it in the other.usual careinterventionsetting Asetting B
The crossing profiles show an interaction: the intervention improves the outcome in one setting and worsens it in the other.
  • 4Prioritise the treatment-by-setting interaction.
  • 4Report simple effects and limit causal explanation.
The overall three-point average hides setting-specific effects in opposite directions. Report the interaction and each simple effect with intervals. If setting was not randomised, the model describes effect modification by setting but does not establish which setting feature caused the difference.
Sia tip — When lines are not parallel, interpret the interaction before the averaged main effect.
Glossary

Key terms

Main Effect
The average contrast for one factor across levels of another factor.
Interaction
A model term showing that one contrast changes across another factor.
Simple Effect
The effect of one factor at a specific level of another factor.
Covariate
A measured predictor included to improve precision or address a defined adjustment purpose.
MANOVA
A multivariate procedure testing group differences across a vector of related outcomes.
FAQ

Factorial and Multivariate Analysis FAQ

Why should the interaction be read first?

An interaction shows that the effect of one factor changes across another. An averaged main effect can then conceal clinically important differences or even reverse direction across settings. Read the interaction estimate and simple effects before writing a general treatment sentence.

When is ANCOVA useful?

ANCOVA can improve precision and account for baseline outcome variation when the covariate is measured before treatment and the model fits the relationship. It is not a universal repair for confounding, and adding a post-treatment mediator can remove part of the effect being estimated.

What does a significant MANOVA mean?

It supports a group difference somewhere in the modelled outcome vector. It does not identify which outcome differs or make every follow-up significant. Interpret the joint question, then use planned follow-up analyses with suitable multiplicity control and original-scale estimates.

Study strategy

Exam move

Draw profile plots for parallel, diverging and crossing effects and write one sentence for each. For adjusted models, label every covariate as baseline predictor, confounder or post-treatment variable. Explain why each outcome belongs in a multivariate question before choosing MANOVA.

Working through Factorial and Multivariate Analysis in RS517? Sia is AskSia’s AI Statistics tutor — ask any RS517 Factorial and Multivariate Analysis question and get a clear, step-by-step explanation grounded in how RS517 is taught and assessed. Read this chapter free, then take your hardest questions to Sia.

A+Everything unlocked
Unlocks this Bible + all 24 of your subjects at The Hong Kong Polytechnic University, plus 1,000+ more Bibles across the AskSia library.
Sia - your RS517 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