MRCPsych Paper B critical review is a reasoning task, not a formula-recognition contest. Start by identifying whether you are struggling with the arithmetic, the choice of measure, the study design or the interpretation. Then use MRCPsych Mentor, PassMRCPsych or another resource to practise that specific skill on unfamiliar data.
The Royal College of Psychiatrists' preparation page, checked on 24 September 2026, states that critical review accounts for one-third of Paper B. This article is published by iatroX and includes iatroX as a possible guided-reasoning tool, without treating any platform's explanation as a substitute for independently checked calculations.
Separate the failure mechanisms
An arithmetic error occurs when you select the correct calculation but perform it incorrectly. A denominator error occurs earlier: you divide by the wrong population. A measure-selection error occurs when the calculation does not answer the question. An interpretation error occurs when a correct number is given an unjustified meaning.
Study design and bias require a further distinction. You may calculate an association correctly from the supplied data while overlooking why the design does not support a causal conclusion. More repetitions of the calculation will not necessarily repair that problem.
Use an error log with these categories. "Statistics weak" hides the mechanism and encourages indiscriminate practice. "I used everyone tested as the denominator when the question asked about positive results" tells you what to rehearse next.
Dataset one: name the denominator before calculating
The following dataset is entirely fictional and created for this article. It represents a hypothetical test evaluated against a reference standard, not the performance of a real clinical test.
| Test result | Condition present | Condition absent | Total |
|---|---|---|---|
| Positive | 80 | 90 | 170 |
| Negative | 20 | 810 | 830 |
| Total | 100 | 900 | 1,000 |
Sensitivity asks what proportion of those with the condition test positive: 80 divided by 100, or 80%. Specificity asks what proportion of those without the condition test negative: 810 divided by 900, or 90%.
Positive predictive value asks a different question: among positive results, what proportion have the condition? That is 80 divided by 170, approximately 47.1%. Negative predictive value is 810 divided by 830, approximately 97.6%.
The arithmetic has been checked independently. The educational task is to explain why the denominator changes, not merely memorise where each figure sits in a familiar table. Before using a formula, translate the question into plain language: "among which group, how many have which characteristic?"
Dataset two: change the population without changing the test characteristics
Now use another original fictional population. There are 50 people with the condition and 950 without it. For this constructed exercise only, retain sensitivity of 80% and specificity of 90%. The resulting counts are 40 true positives, 10 false negatives, 95 false positives and 855 true negatives.
Positive predictive value becomes 40 divided by 135, approximately 29.6%. The lower value does not mean that the arithmetic changed or that the hypothetical test's specified sensitivity and specificity deteriorated. The constructed population has a different proportion of people with the condition.
This is a useful transfer exercise after a question-bank explanation. Change the data, hide the original answer and predict the direction of the effect before calculating. Do not assume that test characteristics remain constant across every real population; that constancy was stipulated to isolate the concept in this example.
Dataset three: absolute and relative effects
Consider an invented study with a stated follow-up of one year. Each group contains 200 participants. The outcome occurs in 20 participants in the intervention group and 30 in the comparison group. These are fictional event counts, not evidence supporting any treatment.
The risks are 10% and 15%. The absolute risk reduction is five percentage points, or 0.05 as a proportion. The risk ratio is 0.10 divided by 0.15, approximately 0.67. The relative risk reduction is approximately 33.3%. The reciprocal of the absolute risk reduction gives a number needed to treat of 20 over the specified year for the specified outcome, within this constructed example.
Each number answers a different question. Do not call five percentage points a 5% relative reduction. Do not remove the outcome or time period from the interpretation. Do not infer precision from the point estimate alone: this simplified dataset does not supply a confidence interval or enough design information for a complete appraisal.
A useful explanation should state these limits, not merely display the formula and celebrate a correct answer.
Add study design before drawing a conclusion
Imagine that the same event counts came from a non-randomised service comparison in which one group started with lower baseline risk. The calculations remain calculations of the observed data, but attributing the difference to the intervention would require additional justification.
Ask how participants entered the groups, whether the outcome was measured consistently, whether follow-up differed and what happened to missing data. These questions address the credibility of the inference rather than changing the numerator mechanically.
As an original exercise, create two short study descriptions with the same event table but different allocation and follow-up processes. Explain why your confidence in a causal interpretation differs. This prevents the table from becoming detached from the study that generated it.
How to inspect Mentor and PassMRCPsych explanations
MRCPsych Mentor and PassMRCPsych advertise examination preparation resources, as reviewed on 24 September 2026. Their public presence does not establish which explanation will resolve your particular difficulty, and this article does not report a paid-account comparison of their critical-review teaching.
Select a sample that matches your recurring error. After attempting it, inspect whether the explanation identifies the relevant population, states the calculation, interprets the answer and explains the main limitation. A formula alone may be enough for an arithmetic refresher but insufficient for a denominator or inference problem.
Then close the explanation and alter the original practice variables using your own fictional data. Do not copy or redistribute a provider's protected question. The learning objective can be practised independently without reproducing the vendor's stem or answer options.
A skills checklist for the next study session
| Skill | Demonstration to attempt | Common false reassurance |
|---|---|---|
| Identify the population | Name the denominator before calculating | Recognising the formula's appearance |
| Select the measure | Explain which question the measure answers | Calculating any available ratio correctly |
| Perform arithmetic | Recalculate using altered fictional data | Remembering the previous numerical answer |
| Interpret the result | Include the outcome, comparator and relevant time period | Repeating a percentage without context |
| Appraise the design | Identify a plausible limitation in the inference | Assuming a correct calculation proves causation |
| Transfer | Solve a new problem without notes | Completing the same item repeatedly |
Use the checklist to choose practice, not to manufacture an overall competence score. A small number of examples cannot validate mastery of the entire critical-review syllabus.
Protect critical review within mixed preparation
Stronger clinical topics can occupy most of a study session because they provide more immediate reassurance. Reserve a defined critical-review task before beginning the material you find easier. The task should specify a skill, such as interpreting a diagnostic table or distinguishing absolute from relative effects.
The College's preparation page, checked on 24 September 2026, also describes syllabus updates. Use the version applicable to your sitting rather than combining a future syllabus with assumptions from an older course. The official page is the appropriate starting point for that check.
Per iatroX product information, September 2026, the Socratic Tutor can ask targeted follow-up questions after an attempted item. It may help you articulate why a denominator or conclusion was wrong. Verify numerical results independently and return to an unaided attempt; neither conversational fluency nor a reference link establishes that the arithmetic is correct.
The resource decision
Keep Mentor or PassMRCPsych when the explanation resolves the mechanism and you can demonstrate transfer. Add selective teaching when you remain unable to explain the concept despite repeated questions. Consider iatroX for responsive questioning where it adds a useful step, not because AI assistance is automatically preferable.
The best next activity may be a fresh miniature dataset you construct and check, rather than another subscription. Critical review becomes more manageable when each mistake names a skill you can practise.
Frequently asked questions
How much of MRCPsych Paper B is critical review?
The College's preparation page, checked on 24 September 2026, describes critical review as one-third of Paper B. Use the current syllabus and instructions applicable to your sitting.
Should I memorise formulae before practising questions?
Know the relevant calculations, but connect each to the question it answers and the correct denominator. Formula recall alone does not demonstrate interpretation or appraisal skill.
Can I use AI to check my statistics answers?
AI can help explain a proposed setup, but numerical results and assumptions should be checked independently. An apparently confident explanation is not an independent verification of its own calculation.