Revision aid, not clinical advice. This page is educational material written for exam preparation. It is not a substitute for the current guideline or for clinical judgement, and must not be relied on for the care of any individual. Always check up-to-date national guidance and verify independently before acting on anything here.

AKT Statistics Guide · Part 2 · Chapter 7

Bias & Validity

The named biases the exam loves, plus validity vs reliability and how good studies fight bias — the “what’s wrong with this study?” chapter.

Where this shows up in the AKT

Naming the bias from a stem — selection, recall, measurement, attrition, lead-time, length-time.
Telling validity from reliability, and internal from external validity.
Knowing how randomisation, blinding, and ITT each reduce a different kind of bias.

The AKT is full of questions that essentially ask “what’s wrong with this study?”. Knowing the named biases and what they look like in a stem is high-yield revision.

The main types of bias

Selection bias

The people in the study are not representative of the population you want to generalise to. Example: recruiting only hospital inpatients for a study about a community-level intervention.

Recall bias

Participants’ memory of past exposures is influenced by whether they have the outcome. Example: mothers of children with congenital abnormalities recalling pregnancy exposures more thoroughly than mothers of healthy children. A classic problem in case-control studies.

Measurement (information / observer) bias

The way outcomes or exposures are measured differs between groups. Example: the investigator knows who got the drug and assesses their outcomes more favourably. Reduced by blinding.

Attrition bias

Participants who drop out differ systematically from those who stay. Example: in an antidepressant trial, patients who feel worse drop out, leaving only those who tolerated the drug well.

Publication bias

Positive studies are more likely to be published than negative ones (covered in detail with forest plots and funnel plots).

Lead-time bias

A screening test detects disease earlier, making survival look longer even if death occurs at the same time. A classic in screening questions.

Length-time bias

Screening preferentially detects slow-growing, less aggressive disease (because fast-growing tumours appear and kill between screens). This makes screening look more effective than it is.

The named biases at a glance

What each bias is and where it classically turns up.
Bias	What goes wrong	Classic setting
Selection	Sample not representative of the target population	Recruitment
Recall	Memory of exposure differs by outcome status	Case-control
Measurement	Outcomes assessed differently between groups	Unblinded studies
Attrition	Drop-outs differ systematically from completers	Long trials
Lead-time	Earlier diagnosis lengthens apparent survival, same death	Screening
Length-time	Slow-growing disease over-represented	Screening
Publication	Positive studies more likely to be published	Meta-analysis

Validity & reliability

Internal validity — does the study correctly answer its own question? (Free from bias and confounding within the study.)
External validity (generalisability) — do the results apply to other populations beyond the study?
Reliability — does the measurement give the same result on repeat testing? (Consistency.)
Validity — does the measurement actually measure what it claims to? (Accuracy.)

The dartboard analogy: reliable = arrows all land in the same spot; valid = arrows land on the bullseye. You can be reliable without being valid — that is consistently wrong.

Reducing bias

Randomisation — balances confounders between groups at the start.
Allocation concealment — recruiters do not know which group the next patient will be assigned to.
Blinding — single (patient), double (patient + investigator), or triple (patient + investigator + analyst).
ITT analysis — preserves randomisation.
Standardised outcome assessment — same tools, same training, same blinding for everyone.

The key insight is that these tackle different problems. Randomisation deals with confounders at the start; blinding deals with bias during follow-up. One does not substitute for the other.

The traps that cost marks

Trap — mixing up validity and reliability. Valid = right (on the bullseye). Reliable = consistent (same spot every time).

Trap — confusing internal and external validity. Internal = correct within the study. External = generalisable beyond it.

Trap — assuming blinding fixes everything. It reduces measurement bias but does nothing for selection bias.

Trap — treating “randomised” and “blinded” as synonyms. Randomisation handles confounders at the start; blinding handles bias during follow-up. They are not the same tool.

AKT-style question

A case-control study investigates whether anti-emetic use in pregnancy is associated with cleft palate. Mothers of affected babies recall their exposures more thoroughly than mothers of unaffected babies. Which bias is most likely?

Selection bias
Recall bias
Lead-time bias
Attrition bias

Reveal answer

B — recall bias. This is the classic problem in case-control studies: mothers of affected children scrutinise their pregnancy exposures far more than control mothers, which exaggerates the apparent association. Selection bias (A) is about who is recruited, not how they remember; lead-time bias (C) applies only to screening; and attrition bias (D) is about differential drop-out, which is not what is described here.

Lock it in

Know the named biases: selection, recall, measurement, attrition, lead-time, length-time, publication — and the classic setting for each.
Validity = right (bullseye); reliability = consistent (same spot). You can be reliable but invalid.
Internal validity = correct within the study; external validity = generalisable beyond it.
Randomisation fixes confounders at the start; blinding fixes measurement bias during follow-up — different jobs.

Source concepts: standard methodology on bias, confounding and validity (e.g. Centre for Evidence-Based Medicine; standard medical-statistics texts). A revision summary, not a clinical protocol.

Want more questions like this one?

Try the free 20-question AKT taster: sixteen clinical, two evidence-based practice and two admin questions, with the reasoning after each and a marked report. No card needed. The full AKT Pack has 2,000+ more.

Note: Independent revision resource for the MRCGP AKT and the MSRA. Educational content for exam preparation only — not clinical advice, and not affiliated with the RCGP. A study aid, not a substitute for the official curriculum or clinical judgement.

← Back to the AKT Statistics Guide

by Core Revision · corerevision.com