Meta-Analysis Help and Writing Service UK 2026-2027

A meta-analysis is not a longer literature review, and the students who discover that late usually discover it in a viva.

Projectsdeal supplies bespoke, human-written model meta-analyses and reference material built to your own question, your programme's handbook and your marking rubric. Every model runs the full pipeline: a pre-specified protocol, a reproducible search with Boolean logic and controlled vocabulary, documented two-stage screening, risk-of-bias assessment matched to study design, a justified effect measure and model, heterogeneity investigated rather than merely reported, sensitivity analyses, and a certainty rating presented in a summary of findings table.

Meta-Analysis: The Statistical Half of Evidence Synthesis

A meta-analysis is not a longer literature review. It is a statistical procedure that combines the quantitative results of separate studies into a single pooled estimate, together with a measure of how much that estimate can be trusted and how far the underlying studies actually agree with one another. The synthesis is the arithmetic; the review that surrounds it — the protocol, the search, the screening, the appraisal — is what makes the arithmetic meaningful. Combine badly chosen studies well and you produce a precise answer to a question nobody asked. That is why every credible meta-analysis is embedded in a systematic review, and why the two phrases are so often used together.

Projectsdeal has supported UK students since 2001, with 115,000+ orders completed at an average 4.9/5 and 120+ PhD-qualified UK writers, including statisticians and evidence synthesis specialists who run these analyses in their own research. Our meta-analysis support covers the full pipeline: framing an answerable question, building and documenting a reproducible search, screening and extraction, risk-of-bias assessment matched to study design, model choice and pooling, heterogeneity investigation, small-study effects, sensitivity analysis and certainty rating — then writing the results up so a marker or reviewer can follow every decision. It sits alongside our systematic review service and our broader statistical analysis support.

Meta-Analysis, Systematic Review and Scoping Review Are Not the Same Thing

Students lose marks here before they have run a single calculation, because the three terms describe different commitments. A systematic review is a method for finding and appraising all the evidence on a tightly framed question; it may or may not end in a statistical pooling. A meta-analysis is that pooling, and it requires studies that are similar enough in population, intervention or exposure, comparator and outcome measurement that combining them is conceptually defensible. A scoping review does something different again: it maps the extent, range and nature of evidence on a broader topic, deliberately includes a wider range of designs and does not usually appraise quality as a basis for exclusion. Our dedicated guide to the scoping review method sets out that distinction in detail, and choosing between them is the first decision your methodology chapter must justify.

The practical test is simple and worth applying before you commit: if you would be uncomfortable putting the studies on one forest plot and reading a single diamond at the bottom as a meaningful average, do not pool them. Clinical or contextual heterogeneity that you cannot defend conceptually is not solved by choosing a random-effects model. In that situation a structured narrative synthesis is the honest answer, and saying so explicitly reads as methodological judgement rather than avoidance.

Question, Protocol and Registration

Meta-analysis begins with a question specific enough that inclusion decisions are almost mechanical. PICO remains the workhorse in intervention research: population, intervention, comparator and outcome, sometimes extended with study design and timeframe. For exposure questions the same structure appears as PECO, with exposure in place of intervention. Each element becomes an inclusion criterion, a search concept and a column in your extraction form, which is why a vague question produces chaos three months later. The single most useful discipline at this stage is naming the primary outcome and its measurement, because a review that treats every reported outcome as equally central ends up pooling incommensurable things.

The protocol is written before the search and specifies eligibility criteria, databases and search strategy, screening and extraction procedures, the risk-of-bias tools, the intended effect size, the model, the planned subgroup analyses and the planned sensitivity analyses. Pre-specification is what allows a reader to distinguish a planned subgroup finding from one discovered by looking. PROSPERO is the international register for systematic review protocols in health and social care and is the norm for reviews of health outcomes; student projects are sometimes ineligible or too compressed to register, in which case write the protocol anyway, date it, place it in an appendix and explain the position in your methodology. Registration or its absence is itself a reportable methodological fact.

The Search: Boolean Logic and Controlled Vocabulary

A meta-analysis stands or falls on whether the search would find the same studies if someone else ran it. The mechanics are unglamorous and entirely learnable. Each concept from your question becomes a block of synonyms combined with OR; the blocks are combined with AND; NOT is used sparingly because it discards more than students expect. Truncation catches word endings, wildcards catch spelling variants including transatlantic differences, phrase searching holds multi-word terms together and proximity or adjacency operators find terms near one another without demanding an exact phrase. Field tags let you restrict a term to the title and abstract when a broad free-text search would flood the results.

Free-text searching alone is not enough in databases with controlled vocabulary. MEDLINE and PubMed index records using Medical Subject Headings, Embase uses Emtree, and CINAHL has its own subject headings, each with its own hierarchy. A competent strategy combines subject headings with free text for every concept, uses explosion where the narrower terms are genuinely wanted, and is rebuilt rather than copied when moving between databases because the vocabulary and syntax differ. Supplementary methods matter too: backward and forward citation searching, hand-searching key journals, checking trial registers and, where relevant, contacting authors for unpublished results.

Reporting is part of the method. State each database and platform, the date each search was run, all limits applied and why, and reproduce at least one full strategy verbatim in an appendix so it can be replicated line by line. Record the number of records retrieved from each source. Students who leave this to the end reconstruct it inaccurately from memory; students who keep a search log as they go write the appendix in twenty minutes. The same discipline underpins doctoral work through our PhD systematic review service.

Screening, Dual Review and Cohen’s Kappa

Screening proceeds in two stages: titles and abstracts against the eligibility criteria, then full texts, with a recorded reason for every full-text exclusion. Those reasons are not bureaucracy; they populate the PRISMA 2020 flow diagram, which accounts for every record from identification through screening and eligibility to inclusion, and a reader uses it to judge whether the review was conducted or merely reported. Deduplication before screening should also be documented, because duplicate counts affect every number downstream.

Best practice is dual independent screening, with two reviewers assessing records separately and a documented procedure for resolving disagreement, usually discussion first and a third reviewer where consensus fails. Agreement is quantified with Cohen’s kappa, which corrects raw percentage agreement for the agreement expected by chance, and it should be reported with the stage it refers to. Undergraduate and taught master’s students often work alone; the honest solution is to have a supervisor or peer double-screen a random sample, report agreement on that subset, and name solo screening as a limitation. Claiming dual screening you did not perform is a far worse error than acknowledging the constraint.

Extraction follows the same logic. A piloted extraction form fixes exactly what is taken from each study: bibliographic details, design, setting, sample characteristics, sample size per arm, the intervention or exposure and comparator, outcome definitions and time points, and the numerical results needed for pooling, which usually means means, standard deviations and group sizes for continuous outcomes or event counts and totals for binary ones. Dual extraction, or at minimum verification of a sample, catches the transcription errors that quietly poison a pooled estimate. Where a paper reports a standard error, confidence interval or test statistic instead of a standard deviation, the conversion must be documented rather than done silently.

Risk of Bias: Match the Tool to the Design

Quality appraisal in a meta-analysis is not a scoring exercise that produces a number to put in a table. It is a structured judgement about whether each study’s result is likely to be distorted, and it earns its keep by feeding into sensitivity analysis and into the certainty rating at the end. Using a single generic checklist for a mixed evidence base is one of the clearest signals of an inexperienced reviewer, because the mechanisms of bias in a randomised trial and in a retrospective cohort study are simply not the same.

Report appraisal visually as well as in prose — a traffic-light figure by domain and study, plus a summary bar chart, is the convention and it makes patterns legible at a glance. Then use it. If three of your eight studies are at high risk of bias, run the pooled analysis with and without them and report both. If the effect disappears when the weakest studies are removed, that is a finding, and saying so is exactly the kind of honesty that distinguishes a strong dissertation. AMSTAR 2 is relevant in a second way for students: it is the instrument by which your own review would be judged, so reading it early tells you what a reviewer will look for.

Choosing the Effect Size

The effect size is the common currency that makes pooling possible, and the choice follows the outcome type rather than personal preference. Continuous outcomes measured on the same instrument across studies can be pooled as a raw mean difference, which keeps the original units and is far easier to interpret clinically. Where studies use different instruments for the same construct, the standardised mean difference converts each to units of standard deviation; Hedges’ g applies a correction for the upward bias that affects Cohen’s d in small samples and is the safer default. Binary outcomes are pooled as risk ratios, odds ratios or risk differences, with ratio measures analysed on the log scale and back-transformed for presentation.

Two rules prevent most trouble. First, decide direction before you begin, so that a positive value always means the same thing across every study, reversing signs where an outcome is scored in the opposite direction and documenting each reversal. Second, one effect size per study per analysis. Studies that report several outcomes, several time points or several intervention arms create statistical dependence, and putting them all on one plot double-counts the same participants. The defensible options are to select the pre-specified primary outcome, to average dependent effects within a study, or — if you have the technical grounding — to use a multilevel or robust variance estimation model that handles the dependency explicitly.

Fixed-Effect or Random-Effects: A Question About Assumptions

The two models answer different questions. A fixed-effect (or common-effect) model assumes every study estimates the same single true effect, and that the differences you see are sampling error alone; weights are simply the inverse of each study’s variance, so large precise studies dominate. A random-effects model assumes the true effect varies across studies and estimates the mean of a distribution of effects, adding the between-study variance to each weight, which flattens the weighting and widens the confidence interval. The choice should be made from the conceptual plausibility of a single common effect, stated in the protocol — not selected afterwards because it produced a tidier result.

In practice, most syntheses of complex interventions delivered in different settings and populations are better served by a random-effects model, because the assumption of one identical true effect is difficult to defend. Several estimators of between-study variance exist, the classic DerSimonian and Laird method being the most familiar and restricted maximum likelihood and Paule-Mandel common alternatives; with few studies, adjustments such as the Hartung-Knapp approach produce more appropriate confidence intervals and are increasingly recommended. Mantel-Haenszel methods are preferred for sparse binary data where inverse-variance weighting behaves badly. Report which estimator you used, because it is a decision a reader cannot infer from the forest plot.

Heterogeneity: Read It, Do Not Grade It

Heterogeneity is variation in true effects across studies, and investigating it is usually more interesting than the pooled estimate itself. Cochran’s Q tests the null hypothesis that all studies share a common effect; it has low power when studies are few and becomes over-sensitive when they are many, so a non-significant Q with eight small studies is weak reassurance. I-squared expresses the proportion of total variability attributable to genuine differences between studies rather than to chance. Tau-squared estimates the between-study variance on the scale of the effect measure, and tau itself is often the most interpretable summary because it is in the same units as the effect.

The most common error is treating I-squared as a fixed grading scale in which one number means low and another means high heterogeneity. It is not a measure of the amount of heterogeneity in absolute terms; it is a ratio that rises as the included studies become more precise, so a set of very large trials with clinically trivial differences can post a high value while a set of small studies with substantively different effects posts a low one. Interpret it alongside tau, alongside the confidence interval around I-squared itself, and alongside a visual reading of the forest plot. A prediction interval, showing the range in which the effect of a future study would be expected to fall, is often the single most informative statistic you can report, and it is frequently much wider than the confidence interval around the mean.

Subgroup Analysis and Meta-Regression

When heterogeneity is present, the next question is whether it is explicable. Subgroup analysis splits studies by a categorical characteristic — delivery setting, population age band, intervention intensity, risk-of-bias level, region — and formally tests whether effects differ between groups, which is not the same as noting that one subgroup reached significance and another did not. Meta-regression models the effect size against one or more study-level covariates, including continuous ones such as mean age, duration or baseline severity, and reports how much between-study variance the covariate explains.

Both are easy to abuse and markers know it. Pre-specify the small number of subgroups and covariates you will examine, and label anything else exploratory. Meta-regression needs a reasonable number of studies per covariate; a common working guide is roughly ten studies for each covariate examined, and with fewer than that the exercise is descriptive rather than inferential. Above all, remember that these are observational comparisons across studies, not randomised comparisons within them, so an association between a study-level characteristic and effect size is confounded by everything else that differs between those studies. The ecological fallacy is a real hazard here: a relationship at study level need not hold at participant level.

Publication Bias and Small-Study Effects

Studies with statistically significant, positive results are more likely to be published, published faster and published in English-language journals, so a synthesis built only on what is easy to find can overestimate an effect. The standard visual check is the funnel plot, which plots each study’s effect against a measure of its precision; in the absence of bias the scatter should be roughly symmetrical, with small imprecise studies spread widely at the base and large precise studies clustered at the top. Egger’s regression test provides a formal test of that asymmetry, and Begg’s rank correlation test is a lower-powered alternative.

The limitation is serious and must be stated. These methods have very low power when few studies are available, and the widely used working guide is that funnel plot asymmetry should not be tested with fewer than about ten studies. Most student meta-analyses include fewer, which means the correct action is to say that assessment was not appropriate and to explain why — not to produce a funnel plot with five points and interpret its shape. It is also important to remember that asymmetry has causes other than publication bias, including genuine differences between small and large studies, poorer methodological quality in smaller trials and true heterogeneity. Contour-enhanced funnel plots help distinguish these, and trim-and-fill offers a sensitivity assessment of what the estimate might look like under an assumed missing-study mechanism, but neither corrects the underlying problem.

Sensitivity Analysis and Rating the Certainty of Evidence

Sensitivity analysis asks whether the conclusion survives reasonable alternative decisions. The standard set includes rerunning the pooled analysis under the alternative model, excluding studies at high risk of bias, excluding studies whose data required conversion or estimation, excluding unpublished or non-peer-reviewed sources, and leave-one-out analysis to identify a single dominant study. Report these as a compact table rather than in prose, and say plainly whether the conclusion held. A meta-analysis whose direction reverses when one study is removed is a fragile meta-analysis, and the write-up must acknowledge that rather than bury it.

The final step is certainty rating. GRADE assesses the body of evidence for each outcome rather than individual studies, starting from a level determined by design and moving down for risk of bias, inconsistency, indirectness, imprecision and publication bias, and moving up for a large magnitude of effect, a dose-response gradient, or plausible residual confounding that would only reduce the observed effect. The result is a rating of high, moderate, low or very low certainty, presented in a summary of findings table with the absolute and relative effects alongside it. That table is what a practitioner or policymaker actually reads, and producing one lifts a dissertation noticeably. Health and social care students commonly bring this into a public health dissertation or a nursing dissertation, where the summary of findings table is often the most cited page of the whole document.

Software: What Runs, and What Markers Expect to See

Software choice is a practical matter, but it has methodological consequences because packages differ in their defaults, and defaults become decisions if nobody examines them. What matters for your write-up is that you name the software and version, name the package, name the estimator and state the model, so that another researcher could reproduce the analysis exactly. A results section that says only that a meta-analysis was performed has omitted the information a reader needs most.

Whatever you use, keep the analysis reproducible: a script or command log, a clean data file with one row per effect size and documented variable definitions, and a record of every conversion applied to reported statistics. Students already working in R or Stata for other parts of a project usually find it far quicker to stay in one environment than to move the analysis into a graphical package for the sake of familiarity.

Writing It Up: What Separates a First from a 2:2

A meta-analysis is reported in a recognisable order, and marks are lost less often in the statistics than in the reporting. Methods must be written so the study could be repeated: eligibility criteria, information sources with dates, the full strategy for at least one database, the selection and extraction process including how many reviewers were involved, the risk-of-bias tools, the effect measure, the model and estimator, and the planned heterogeneity, subgroup, bias and sensitivity analyses. Results follow the same sequence: the flow diagram, a study characteristics table, appraisal findings, the forest plot and pooled estimate with its confidence interval, heterogeneity statistics with a prediction interval, then the planned secondary analyses.

Two habits reliably lift a mark. The first is discussing the pooled estimate and the heterogeneity together rather than separately, because the average of a widely dispersed set of effects is a weak guide to what a practitioner should expect. The second is a limitations section that names the specific constraints of your review — the number of studies, the languages searched, reliance on published data, solo screening, an inability to assess small-study effects — rather than a generic paragraph. Statistical writing of this kind is also where our methodology chapter service and quantitative analysis support most often make the difference.

The six failures we correct most often

How Our Meta-Analysis Support Works

Send us your research question or topic, your programme’s dissertation handbook and marking rubric, your word count and referencing style, your supervisor’s feedback so far, and any work already done — a protocol, a partial search, a screening spreadsheet or an extraction table. We match the project to a writer with genuine evidence synthesis and statistical experience in your field, because the difference between a competent review and a publishable one is almost entirely a matter of judgement calls made by someone who has made them before.

Design and protocol stage

We test whether your question is poolable at all, which is the most valuable hour in the whole project. If it is not, we say so and set out the defensible alternative, whether that is a narrower question, a structured narrative synthesis or a scoping approach. If it is, you receive a full protocol with eligibility criteria, search strategy, appraisal tools, effect measure, model and pre-specified secondary analyses.

Analysis and reporting

Model outputs include the study characteristics table, appraisal tables and traffic-light figure, forest and funnel plots, heterogeneity statistics with a prediction interval, subgroup or meta-regression output where the number of studies supports it, sensitivity analyses and a summary of findings table with certainty ratings, plus the annotated code or command log so every figure is reproducible. Ordering is online 24x7 with WhatsApp support on +447447882377, and every order carries free unlimited revisions, on-time delivery, a money-back guarantee, GDPR-compliant confidentiality and instalment options on larger projects.

Scope, Turnaround and Typical Deliverables

Price depends on the number of studies, the complexity of the outcome structure, whether searching and screening are included, level of study and deadline. The cheapest useful intervention is almost always the earliest: an hour spent testing whether a question can be pooled prevents months spent discovering it cannot. If your analysis is already run and the problem is interpretation and write-up, send the dataset and the output — rebuilding a results and discussion chapter around an existing analysis is one of our most frequent requests, and it sits naturally alongside our wider literature review writing and research methods support.

How It Works — 3 Steps, Open 24x7

Join 115,000+ UK students since 2001 • ✅ Subject specialists • ✅ No hidden fees • ✅ Money-back guarantee

Written by Subject Specialists

Every meta-analysis brief is matched to a named UK academic who holds a degree in that discipline and has marked or taught at this level. That matters more than any general writing skill: a specialist already knows the standard theories, the seminal texts, the methods your module expects you to apply and the difference between what earns a 2:1 and what earns a first in this subject. They write to your brief, your module handbook and your marking rubric, and they explain their reasoning in the work so the structure is transferable to your next assignment.

Our Guarantees, In Writing

What UK Students Say

Frequently Asked Questions

1. What is the difference between a meta-analysis and a systematic review?
A systematic review is a method for finding, selecting and appraising all the evidence relevant to a tightly framed question. A meta-analysis is the statistical pooling of results from those studies into a single estimate. A systematic review may end in a meta-analysis, or it may conclude that pooling is inappropriate and present a structured narrative synthesis instead. You cannot credibly run a meta-analysis without the systematic review machinery around it, because the arithmetic is only as meaningful as the study selection behind it.

2. How is a meta-analysis different from a scoping review?
They answer fundamentally different questions. A meta-analysis estimates a pooled effect from studies similar enough in population, intervention, comparator and outcome for averaging to make conceptual sense. A scoping review maps the extent, range and nature of evidence on a broader topic, deliberately includes a wider variety of designs and does not usually appraise quality as a basis for exclusion. If you would be uncomfortable putting your studies on one forest plot and reading the diamond as a meaningful average, you do not have a meta-analysis.

3. How many studies do I need for a meta-analysis?
There is no minimum that makes a meta-analysis automatically valid, and two studies can technically be pooled, but the number affects everything you can do afterwards. With very few studies the between-study variance is poorly estimated, subgroup analysis is uninformative and funnel plot asymmetry cannot sensibly be assessed, which conventionally requires around ten studies. Meta-regression needs roughly ten studies per covariate to be more than descriptive. If your search yields four studies, pool them if they are genuinely combinable, but report the limitations honestly.

4. What is PROSPERO and do I have to register my review?
PROSPERO is the international prospective register of systematic review protocols in health and social care, designed to reduce duplication and to make it visible when a review's methods change after the fact. Registration is standard practice for systematic reviews of health outcomes, but many student projects are ineligible or too compressed to register within the timetable. If you cannot register, write the protocol anyway, date it, place it in an appendix and explain the position in your methodology. Whether or not you registered is itself a reportable methodological fact.

5. What is PRISMA and what do I have to report?
PRISMA is the reporting standard for systematic reviews and meta-analyses, and the 2020 statement sets out the items a complete report should contain. In practice it means reporting eligibility criteria, every information source with the date searched, at least one full search strategy verbatim, the selection and extraction process including how many reviewers were involved, the risk-of-bias tools, the effect measure, the model, and all heterogeneity, subgroup, publication bias and sensitivity analyses. The flow diagram accounting for every record from identification to inclusion is the item markers check first.

6. Which risk of bias tool should I use?
Match the tool to the study design rather than using one generic checklist. RoB 2 is the standard for randomised controlled trials. ROBINS-I assesses non-randomised studies of interventions against a hypothetical target trial. The Newcastle-Ottawa Scale is widely used for cohort and case-control studies, and is best reported by domain rather than as a single total score. AMSTAR 2 appraises systematic reviews themselves, and QUADAS-2 covers diagnostic test accuracy studies. Whichever you use, the appraisal must feed into sensitivity analysis rather than sitting in a table.

7. What is Cohen's kappa used for in a meta-analysis?
It quantifies agreement between two independent reviewers at the screening stage, correcting the raw percentage agreement for the agreement you would expect by chance alone. Best practice is dual independent screening of titles, abstracts and full texts, with a documented process for resolving disagreement. Taught students often have to screen alone, in which case the honest solution is to have a supervisor or peer double-screen a random sample, report kappa for that subset and name solo screening as a limitation.

8. How do I choose the right effect size for my meta-analysis?
Let the outcome type decide. Continuous outcomes measured on the same instrument across studies pool as a mean difference, which keeps the original units. Different instruments measuring the same construct pool as a standardised mean difference, with Hedges' g preferred because it corrects the small-sample bias affecting Cohen's d. Binary outcomes pool as risk ratios, odds ratios or risk differences, with ratio measures analysed on the log scale. Time-to-event data pool as hazard ratios and correlations pool after Fisher's z transformation.

9. Should I use a fixed-effect or random-effects model?
Decide from the assumption you can defend, and state it in the protocol rather than choosing after seeing the output. A fixed-effect model assumes all studies estimate one identical true effect and that differences are sampling error alone. A random-effects model assumes the true effect varies across studies and estimates the mean of that distribution, which is usually more plausible for complex interventions delivered in different settings and populations. Report the between-study variance estimator you used, since a forest plot does not reveal it.

10. How should I interpret I-squared?
Not as a fixed grading scale. I-squared expresses the proportion of total variability attributable to genuine differences between studies rather than chance, so it is a ratio rather than an absolute amount of heterogeneity, and it rises as the included studies become more precise. A set of very large trials with trivial differences can post a high value while small studies with substantively different effects post a low one. Read it alongside tau, alongside the confidence interval around I-squared itself, and alongside the shape of the forest plot.

11. What is a prediction interval and why should I report one?
A prediction interval shows the range within which the true effect of a future study in a new setting would be expected to fall, whereas the confidence interval only describes uncertainty about the average effect. When heterogeneity is present the prediction interval is often much wider than the confidence interval, and sometimes crosses the line of no effect when the pooled estimate does not. That contrast is frequently the single most informative thing a meta-analysis can tell a practitioner, and reporting it signals genuine statistical maturity.

12. How do I test for publication bias with only a few studies?
Usually you do not, and saying so is the correct answer. Funnel plot asymmetry tests including Egger's regression have very low power with few studies, and the widely used working guide is that they should not be applied with fewer than about ten. With five studies, producing a funnel plot and interpreting its shape is worse than declining to assess. Explain in your limitations that assessment of small-study effects was not appropriate given the number of included studies, and note that asymmetry has causes other than publication bias in any case.

13. What sensitivity analyses should a meta-analysis include?
Rerun the pooled analysis under the alternative model, exclude studies at high risk of bias, exclude studies whose data required conversion or estimation, exclude unpublished or non-peer-reviewed sources, and run a leave-one-out analysis to identify whether one study is driving the result. Present them as a compact table rather than in prose and state plainly whether the conclusion held. If the direction reverses when a single study is removed, the finding is fragile and the discussion must acknowledge that.

14. What is GRADE and how do I apply it?
GRADE rates the certainty of the body of evidence for each outcome rather than the quality of individual studies. You start from a level determined by study design, then consider downgrading for risk of bias, inconsistency, indirectness, imprecision and publication bias, and upgrading for a large effect, a dose-response gradient or plausible residual confounding that would only have reduced the observed effect. The result is a rating of high, moderate, low or very low certainty, presented in a summary of findings table alongside the absolute and relative effects.

15. What software should I use to run a meta-analysis?
R with the metafor package offers the most flexibility, including multilevel models, meta-regression and robust variance estimation, while the meta package gives readable syntax for standard pooling and plots. RevMan is built around Cochrane review structure and suits health students following those conventions. Stata has a mature meta suite with strong graphics. Point-and-click options lower the barrier for a first analysis but are harder to document reproducibly. Whatever you choose, report the software, version, package, model and estimator.

16. How long does a meta-analysis take and what does help cost?
Feasibility work and question framing take one to two days, a protocol two to four, a search and screening pack three to five, extraction and appraisal three to six, and the statistical synthesis three to six, with a complete model review typically ten to twenty-one days. Cost depends on the number of studies, the complexity of the outcome structure, whether searching and screening are included, level of study and deadline. The most economical intervention is the earliest, because an hour spent testing whether a question is poolable prevents months of misdirected work.