%202.png)
Advanced Systematic Review and Meta-Analysis
Protocol-standard evidence synthesis: reproducible searching, dual screening, risk of bias, honest pooling decisions and a formal certainty judgement.
Free. Works on Claude, ChatGPT, Gemini or any assistant that accepts a skill file.
What this skill does
The method, encoded.
A systematic review is a study whose participants are studies, and it fails the way any study fails: decisions made after seeing the results, a search nobody could repeat, screening by one person whose judgement was never checked, and a pooled number produced because the software would produce one.
This skill installs the machinery that separates a review at doctoral and publication standard from a well-organised reading list. It starts with a protocol fixed and dated before searching, so that later choices are pre-specified rather than retrofitted. It builds searches that can be re-run, not merely described. It screens in two independent passes and reports the agreement before reconciliation, because low agreement is diagnostic of underspecified criteria rather than embarrassing. It assesses risk of bias by domain, against the designs actually included, and per outcome rather than per study.
Its central judgement is whether pooling is legitimate at all. Combining studies that measure different constructs produces a precise wrong answer that is more persuasive than any study it came from, and this skill applies comparability criteria before any estimate is computed. Where pooling is appropriate, it covers effect size conversion, model choice by assumption rather than by result, prediction intervals, heterogeneity read as a finding about the literature, reporting bias with its real limits stated, and sensitivity analysis. Where the evidence is qualitative, it distinguishes meta-ethnography, thematic synthesis and framework synthesis by their aims and their differing epistemological commitments.
Every conclusion ends with a certainty rating, because precision and certainty are not the same thing.
Best used for
- Doctoral review chapters that must meet publication standard
- Estimating the magnitude of an effect across a comparable literature
- Demonstrating that a literature cannot legitimately be pooled
- Synthesising qualitative studies with a named and justified approach
- Evidence submissions to guideline or policy bodies
Typical inputs
What you give it.
Structured review question with population, exposure, comparator and outcomes, Database access and full-text availability, A second independent screener, Reporting guideline and registration requirements for the field, Effect size data extractable from included papers, Statistical software for pooling and sensitivity analysis
Typical outputs
What you get back.
Registered protocol with pre-specified analyses, Reproducible search strategies per database with dates and record counts, Screening agreement statistic and full-text exclusion list with reasons, Reconciled flow diagram and study-to-report mapping, Duplicate extraction table with per-cell locators, Domain-level risk-of-bias judgements per outcome, Pooled estimates with model justification, prediction intervals and heterogeneity, Reporting-bias and sensitivity analyses, Qualitative synthesis where pooling is inappropriate, Summary-of-findings table with certainty ratings
Method coverage
What the skill works through.
- Why the protocol comes before the search
- Building a search someone else can re-run
- Dual independent screening and what the agreement statistic tells you
- Making the flow diagram numbers reconcile
- Risk of bias by domain, by outcome, never by total score
- The pooling decision: when combining studies produces a precise wrong answer
- Effect size extraction, conversion and dependency
- Fixed versus random effects, and what each model assumes
- Heterogeneity as a finding about the literature
- Publication bias assessment and its limits
- Sensitivity analysis that could change the conclusion
- Qualitative evidence synthesis: choosing the approach by its aim
- Rating the certainty of the evidence
- Academic integrity and AI use
Download
Free skill. One file.
Enter your email once. Every skill you download after that takes a single click.
How to install
Add the skill file and the five kernel protocols to a Claude Project, a ChatGPT Project, a Gemini Gem, or paste them at the top of any assistant conversation. Then give it your real research material, not a description of it.
Download skillQuestions
Common questions.
When can studies be combined in a meta-analysis?
When they address the same question, in populations similar enough that a common effect is a meaningful idea, with comparable interventions or exposures, comparable comparators, and outcomes that measure the same construct even if on different scales. Test those substantively before looking at any statistic. Heterogeneity measures diagnose inconsistency between results; they cannot tell you that two studies measured different things.
What is the difference between fixed-effect and random-effects meta-analysis?
A fixed-effect model assumes there is one true effect and that all variation between studies is sampling error, which is defensible only for near-replicates. A random-effects model assumes true effects vary across studies and estimates the mean of that distribution, which is a different quantity. Choose from the substantive question and the design of the set, not from which produces a narrower interval, and report a prediction interval alongside any random-effects result.
Is high heterogeneity a problem?
It is a finding. It says the effect depends on something, and the useful work is identifying what through pre-specified subgroup analysis or meta-regression. What is never acceptable is removing studies until the heterogeneity index falls, or presenting a pooled estimate from a highly inconsistent set as though it described a real population.
Do I need two people to screen?
For a review at publication or doctoral standard, yes, and the agreement before reconciliation should be reported. Single screening is possible but changes what the review can claim and must be declared as a limitation. Low agreement usually means the eligibility criteria are underspecified, in which case the fix is to amend the criteria and re-screen rather than arbitrate case by case.
How do I synthesise qualitative studies?
Choose the approach by its aim. Meta-ethnography translates studies' concepts into one another to reach an interpretation beyond any single study. Thematic synthesis codes findings across a larger set to answer a defined question. Framework synthesis maps findings onto an a priori framework. They carry different commitments about whether findings from different traditions can legitimately be combined, and that question is genuinely contested, so state the position you take and why.
Can AI run a systematic review for me?
No, and this skill does not do so. It can structure a protocol, check search logic, interrogate a pooling decision, verify that flow numbers reconcile and critique a certainty rating. It never generates study records, effect sizes or citations, and any AI involvement in screening or extraction must be disclosed in the methods with a statement of what a human verified. Institutional AI policies and declaration requirements vary and must be followed.
The skill chain
Works well with.
Research where people already are.
Analyse it where you already work.
Yazi helps researchers conduct surveys, AI interviews and longitudinal research directly through WhatsApp.
