A study matrix table for a systematic review dissertation needs a working structure that you can populate, check and update without losing the source of an entry. This guide takes you from an empty workbook to a checked extraction matrix, using a fictional nursing review to demonstrate the decisions.
The finished output is a usable extraction system: named sheets, defined fields, traceable cells and a change log. For choosing which evidence deserves weight in an academic argument, use the separate nursing evidence-table guide.
Set up the workbook before extracting your first study
Create five sheets: Read me, Studies, Reports, Extraction and Changes. These are suggested working labels, rather than a compulsory university format. A small dissertation may combine sheets, provided the study identifiers and source locations remain clear.
| Sheet | What to record | What it produces |
|---|---|---|
| Read me | Review question, extraction rules, missing-value codes and version date | Instructions another reviewer can follow |
| Studies | One identifier for each underlying study | A study register |
| Reports | Article, supplement or other report linked to a study identifier | A source register |
| Extraction | Relevant characteristics and findings with source locations | The working matrix |
| Changes | Field changes, reasons, affected rows and checks | An extraction audit trail |
Keep the original extraction file separate from the table formatted for your dissertation. Narrowing a column for a Word appendix should not remove information from the working workbook. Save a clearly dated version before making a structural change.
Give studies and reports different identifiers
Cochrane’s data-collection guidance distinguishes the underlying study from its reports and explains why linked publications must be recognised (Li, Higgins and Deeks, 2024). In your workbook, a study could be S01, while its main article and supplement are S01-R01 and S01-R02. The labels themselves are your choice.
For example, one fictional trial might report an initial outcome in one article and later follow-up elsewhere. Two articles would not automatically mean two independent trials. Record the connection in the Reports sheet before using the findings.
Do not join reports solely because their titles look similar. Check the available study details and record uncertainty. If the connection remains unresolved, flag it for discussion with your supervisor rather than silently combining the records.
Create a data dictionary with rules for each column
A column heading alone rarely explains how a cell should be completed. Alongside each field, record its definition, permitted entries, source location and any decision rule. This prevents the meaning of a column changing halfway through extraction.
| Field | Entry rule | Example entry |
|---|---|---|
| Study_ID | Use the identifier in the Studies sheet | S01 |
| Setting | Record the setting as described; flag any later grouping separately | Adult medical ward |
| Outcome_definition | Preserve the reported definition and time period | Definition recorded from methods, page 4 |
| Finding | Enter the result relevant to the review question | Relevant estimate or qualitative finding |
| Source_location | Identify the report and page, table or section | S01-R01, Table 2 |
| Check_status | Use agreed status labels consistently | Unchecked; checked; query open |
Use separate fields when two concepts need separate checks. For instance, enrolled participants and participants included in an analysis should not share an unexplained “sample” cell. Similarly, a measurement instrument and the time at which it was used require different entries.
Build a blank matrix around your review question
Consider a fictional review about nurses’ experiences of discharge communication with adult patients. A first-pass matrix might contain Study_ID, country, service setting, participant group, study design, data-collection approach, relevant finding, source location and reviewer query.
Those fields suit an experience question. An intervention review would need a different structure, potentially including comparison groups, outcome definitions and assessment times. Do not copy every column from this example into an unrelated review.
Write one sentence beside each proposed column explaining its job. “Setting helps us examine whether discharge communication differs between inpatient and community services” is a useful justification. “Other dissertations included it” is not enough.
Choose the row unit deliberately
A single row per study is convenient for a characteristics table. However, a study with several relevant findings may need more than one extraction row. Decide whether each row represents a study, an outcome at a particular time, or an individual qualitative finding.
If using several rows per study, repeat the study identifier and add a finding identifier. Avoid merged cells in the working data: they can make filtering and sorting confusing. The presentation table can look different once extraction is complete.
Keep a note explaining how you count included studies. Ten extraction rows from one study remain one underlying study. This distinction becomes particularly useful when writing the review’s study-characteristics section.
Pilot the matrix using contrasting reports
Before extracting everything, select a small set of reports that expose different practical difficulties. For example, choose one with a clear results table and another with findings embedded in narrative text. The purpose is to test the form, not to select only easy papers.
- Complete the first report using the draft dictionary.
- Record every field that needs interpretation.
- Try the second report without changing the rules silently.
- Revise ambiguous definitions and record why.
- Return to earlier entries affected by the revision.
A useful pilot question is: could you explain how you completed the same field in both reports? If not, the extraction instruction probably needs more work. A polished spreadsheet cannot compensate for an undefined field.
Use explicit codes for incomplete information
Choose distinct labels for “not reported”, “not applicable” and “unclear”. Define them on the Read me sheet. An empty cell gives no indication whether you forgot the entry or searched and could not find it.
For example, use NR only after checking the relevant available report sections. Use unclear when information exists but cannot be interpreted confidently. Record the problem in a query field so you know what needs resolution before synthesis.
Do not convert missing reporting into a claim that a procedure did not happen. The matrix should record what can be established from the available sources, together with the limits of that information.
Keep extraction, checking and judgement visible
Record the finding reported by the authors separately from your own note about its relevance or limitations. A reader should be able to distinguish an extracted entry from the reviewer’s interpretation.
Agree the checking arrangement with your supervisor and follow the review protocol. If the dissertation uses one student extractor with selected checks, describe that actual arrangement. Do not claim independent duplicate extraction unless it occurred.
Where a second checker is involved, retain the query and its resolution. “Corrected after checking Table 3” is more informative than replacing a number without a record. Your method should describe how disagreements were handled.
Run a row-by-row quality check
- Every study identifier links to a registered study.
- Every report identifier links to an available source.
- Each relevant finding has a source location.
- Units, denominators and assessment times remain visible where needed.
- Missing entries use the agreed codes.
- Open queries are resolved or explicitly retained.
- Changes to field definitions have been applied to earlier rows.
Filtering the Check_status column helps reveal unfinished rows. Sorting can reveal inconsistent spelling or categories, but save a version first and sort the complete table so cells stay attached to the correct record.
Turn the working matrix into an appendix table
Once checked, create a presentation copy. Retain the fields the reader needs to understand the included evidence and use consistent abbreviations with a key. Very wide extraction details may remain in the working file or a separate appendix, depending on your brief.
Explain the extraction procedure in the methodology chapter: what was recorded, how the form was piloted, who checked it and how queries were resolved. Do not rely on the appendix to communicate every methodological decision.
The matrix supports later analysis; it does not write the synthesis. For the next stage, use the evidence-table guide to examine how differences between studies affect the conclusions you can draw.
Frequently asked questions
Should the matrix be made in Word or a spreadsheet?
Use the format your programme accepts. A spreadsheet is convenient for filtering and checking working entries; Word may suit the final presentation. Preserve the complete working record when creating a shorter display table.
Can one study occupy several rows?
Yes, if the row unit is clearly defined and the study identifier is repeated. Distinguish the number of findings or outcomes from the number of studies.
What should I do when a source corrects an earlier result?
Record the correction, link the relevant source and update affected entries with a dated explanation. Review any downstream table or statement that used the previous value.
Conclusion
A usable study matrix has consistent fields, stable identifiers, source locations and a visible checking process. Finish with a workbook you can explain and verify, then produce the appendix from that checked record.
Reference
Li, T., Higgins, J.P. and Deeks, J.J. (2024) ‘Chapter 5: Collecting data’, in Cochrane Handbook for Systematic Reviews of Interventions, version 6.5. Cochrane Handbook, Chapter 5.
