Systematic Review and Meta-Analysis Starter Kit

The first day decides more than you’d think

Most reviews don’t fall apart at the statistics. They fall apart earlier, in small decisions nobody thinks to write down. Where the files live. What they’re named. Which version is the real one.

Six months in, the cracks show. You can’t reconstruct your own search. The numbers in the spreadsheet don’t match the PDF you pulled them from. A co-author saved over the extraction sheet, and there’s no way back.

That’s not bad luck. It’s what skipping setup costs you, paid back later with interest. The fix is dull, and it works: decide once how the project is organized, then protect that structure while the work piles up. Wagner and colleagues, writing in Research Synthesis Methods in 2026, make the larger point plainly. Data management in reviews has trailed behind every other part of the method, and at the scale reviews now run, it’s the first thing to break.7

Give the project a home, with fixed rooms

Start with one folder for the whole review, and a fixed set of subfolders inside it. The layout below isn’t sacred, but the logic under it is. Keep raw data apart from anything you change. Keep your code separate from your prose. Wilson and colleagues say it cleanly in their guide to good-enough computing: raw data and metadata go in their own place, and the files you generate during analysis go somewhere else.8 Project TIER builds the same split into a folder hierarchy you can hand to a reviewer and have it simply run.9

One rule protects all of it. Your raw data is read-only. Once a database export lands in 02_Search, nobody edits it. You copy from it, clean the copy, and leave the original alone. The same goes for the numbers you pull from each study. That single habit is what lets you, or a reviewer, trace any figure back to where it came from.8

Name files so you can find them in a year

A tidy folder falls apart the moment the files inside it are called Final, then Final_v2, then Final_REAL. You know the pattern. Here’s the cure, and it’s three small rules drawn from Broman and Woo’s work on data organization.1

Date first, in YYYY-MM-DD form, so the newest work never hides in the middle of the list.1 No spaces and no odd characters, because both trip up code and break links in shared repositories. Underscores separate the big pieces, hyphens join words inside a piece. And name the thing for what it actually is. A raw export gets the database and platform in its name, which is also what PRISMA-S asks you to report later.6

A data plan that survives the messy middle

Here’s a distinction worth holding onto, and it comes from that 2026 paper by Wagner and colleagues. Your raw data is what the database handed you. Your primary data is the cleaned, structured set you actually analyze. They are not the same thing, and the second should never quietly replace the first. Keep both.7

When you start extracting, the spreadsheet itself needs rules, or it turns into something only you can read. Broman and Woo lay out the ones that matter.1 One rectangle of data, one header row. Studies as rows, variables as columns. One thing per cell, so a mean, its standard deviation, and the sample size each get their own column instead of being crammed together. No blank cells either; write NA when a value is missing, so a reader can tell “missing” apart from “forgotten.” And never use colour or highlighting to carry meaning, because it vanishes the second you export to plain text. Make a separate column instead.

Two more habits round it out. Back up to more than one place, the old 3-2-1 rule, so a dead laptop doesn’t erase three months of extraction.8 And when you write analysis code, point it at files with relative paths, not C:UsersyouDesktop…. Absolute paths work on exactly one machine: yours. Relative paths travel, which is the whole point when a reviewer wants to rerun your analysis.9

Your first-day checklist

Work down this list before the first real search. Tick the boxes as you go.

• Draft the protocol against PRISMA-PPin down your question, eligibility criteria, and analysis plan before you search. The 17-item PRISMA-P checklist is the map.4
• Register it, before you screenPROSPERO for health-related reviews; OSF for everything else, including scoping reviews. Prospective registration is the point, not paperwork.
• Write down roles and a decision logWho screens, who extracts, who arbitrates. One running log for every methods decision and why you made it.
• Set up the reference managerZotero, EndNote, or Mendeley, shared with the team, ready for deduplication.
• Record the search as you run itExact database, platform, date, and the full strategy. Save the per-database counts before you deduplicate, because the PRISMA 2020 flow diagram needs them.56
• Choose screening and extraction toolsTwo reviewers screen the same records independently, then resolve conflicts. Splitting the list between you isn’t double screening.2

Where do I register? A quick read

The plumbing is universal. The vocabulary isn’t.

Folders, naming, backups, version control: a soil scientist and a health economist set them up the same way. What shifts between fields is the language at the edges, and it’s worth knowing which parts those are.

Your question framework, for one. PICO suits clinical trials. PECO fits exposure questions in epidemiology or toxicology. PCC works for scoping reviews, and SPIDER for qualitative work. Appraisal shifts too. Clinical reviews lean on risk-of-bias tools; many reviews in education, management, or ecology use quality or critical appraisal instead, because the studies aren’t trials. The standards bodies differ in name but agree on the spirit, from the Cochrane Handbook in health2 to the Campbell Standards in the social sciences.3 Registration is the other one. PROSPERO is health-gated; OSF and PROCEED take the rest.

Five day-one mistakes that cost you later

1. Not saving the raw search. You delete duplicates before recording how many each database returned. Now the top of your PRISMA diagram is a guess.6
2. Registering late. A registration date that lands after your search start reads as a warning sign to reviewers. Register first, then screen.
3. The solo screener. One person taking the first half of the abstracts and another the second half isn’t double screening. Same records, two readers, independently.2
4. Absolute paths in your code. A path that starts at your own drive breaks on every other machine. Use relative paths from the project folder.9
5. Editing the raw data. Overwrite the original numbers and the trail is gone. Raw stays read-only; you work on copies.8

Frequently asked