ACADEMIC WRITING GUIDE
How to Create a Data Extraction Form for a Systematic Review Dissertation (With Worked Examples)
A dissertation systematic review may have a carefully documented search and a defensible PRISMA flow diagram, yet still produce unreliable findings if information is extracted inconsistently from the included studies. One reviewer may record the final follow-up while another records an earlier measure. A paper might report several outcomes, but the spreadsheet captures only the most favorable one. Or two reports from the same underlying study may be mistakenly treated as independent evidence.
A systematic review data extraction form prevents many of these problems by specifying exactly which information reviewers will collect, where to find it, how to code uncertainty, and how differences will be resolved. The form is more than a table of author names and conclusions: it is the bridge between eligible studies and the claims made in a dissertation's Results chapter.
This guide explains how to develop a review-specific extraction form, pilot it, work through a fictional example, prevent common errors, and describe the process in a dissertation. It focuses on intervention-focused systematic reviews; qualitative evidence syntheses, diagnostic accuracy reviews, scoping reviews and other designs require adapted fields and their own methodological guidance. The core principles are consistent with Cochrane Handbook Chapter 5 and the PRISMA 2020 reporting guideline.
Teaching-example disclaimer: The study names, intervention scenarios, values and reviewer decisions below are invented. They illustrate a method; they are not genuine research results. Never paste invented extraction records into a real dissertation as evidence.
What is data extraction in a systematic review?
Data extraction, also called data collection, is the structured recording of information from each eligible study and its reports for a specified review question. Depending on the protocol, the information might include design, participant eligibility, setting, intervention and comparator details, outcome definitions, follow-up times, effect estimates, uncertainty intervals, missing data, funding, conflicts of interest and risk-of-bias-relevant details.
The central distinction is between what a paper says and what the review intends to synthesize. A journal article may present several analyses of the same sample. An extraction form should preserve relevant results while distinguishing the prespecified primary result from alternative analyses. It should make exclusions or non-use of otherwise available results traceable.
A systematic review's extraction form is not the same as a research survey administered to participants. It records information about already completed studies. It is also not the same as a risk-of-bias tool: extraction captures factual inputs, while appraisal uses specific criteria to judge potential distortion. Our risk-of-bias assessment guide explains that separate judgment process.
Start from the review question, not a generic spreadsheet
Before opening Excel or a review platform, identify the question, eligibility criteria, planned comparisons and the outcome data needed for synthesis. Consider this fictional review question:
Among first-year university students, does structured written formative feedback, compared with usual course feedback, improve completion of a first-semester module?
For this question, it would be useful to distinguish true module completion from intermediate quiz scores; to record when completion was measured; and to specify what “structured feedback” actually involved. Extracting every reported student satisfaction statistic may create busywork while leaving the primary outcome poorly defined.
If your project is still at the searching stage, see our guide to conducting a systematic literature search for a dissertation. A transparent PRISMA 2020 flow diagram records how reports were selected. Data extraction begins once the appropriate reports and study identities are established; it does not replace selection.
Make a field dictionary before entering any records
For every proposed variable, decide what it means, what format it accepts, how to identify its source location, and what to enter when it is not reported. This field dictionary is a practical way to reduce divergent interpretations among reviewers.
| Field | What to record | Rule that prevents an ambiguous entry |
|---|---|---|
| Study ID and report ID | A stable study identifier and a separate identifier for each publication or report | Link multiple reports of the same study instead of counting them as independent studies |
| Design and setting | Trial or other eligible design; country, institution and study setting | Copy the reported design; flag rather than guess missing design details |
| Population and eligibility | Recruitment criteria, sample characteristics and relevant exclusions | Distinguish randomized or analyzed populations where reported |
| Intervention and comparator | Who delivered each condition, frequency, duration and core components | Do not use generic labels such as “feedback intervention” alone |
| Outcome definition | What was measured, direction of benefit and instrument or data source | Do not mix course completion with self-reported engagement |
| Time point | Exact assessment timing and whether it matches the planned review window | Specify the rule for choosing among multiple follow-ups |
| Results and uncertainty | Relevant arm-level data, effect measure and confidence or other uncertainty information | Preserve the original units and label adjusted versus unadjusted analyses |
| Missing information | What is absent, unclear or awaiting confirmation | Use distinct codes for “not reported,” “not applicable” and “not yet checked” |
| Source location | Report identifier, page, table, supplement or registry entry | Every material data point should be traceable to its source |
| Review decisions | Discrepancies, decisions, reviewer initials and confirmation status | Keep a dated audit trail instead of silently overwriting entries |
This is a teaching template, not an official universal checklist. The actual fields must match your protocol, research design and synthesis approach. If your review investigates experiences rather than intervention effects, extraction may focus on qualitative findings, participant quotations and context using a method suited to qualitative synthesis.
How to design the data extraction form: a practical workflow
Step 1: Map every field to a question or planned analysis
Create a brief crosswalk from protocol to extraction form. For instance, if the main comparison concerns module completion at the end of the semester, the form needs an explicit outcome definition, assessment time, denominator or analysis population when available, and the data required for the selected effect measure. If intervention intensity is an exploratory subgroup, define intensity consistently rather than inventing categories after reading which studies look favorable.
Ask of every field: Which decision, display, appraisal or synthesis will use this information? Some metadata fields are retained simply to ensure traceability, but fields without a purpose can become a maintenance burden.
Step 2: Separate study identity from report identity
One trial may have a protocol, registry entry, primary publication and later follow-up paper. These are multiple reports of one underlying study, not automatically multiple studies. Create a study-level record and attach every applicable report identifier and extraction source. This matters when a synthesis table counts included studies or when later reports provide missing follow-up details.
Conversely, two reports by the same authors are not necessarily the same study. Check sample dates, sites, trial registration, recruitment and participant overlap before merging them. Where identity remains uncertain, record the uncertainty and resolve it explicitly.
Step 3: Write rules for outcomes and time points
Before extracting results, define which outcomes and follow-up windows count for the planned comparison. If the protocol prioritizes end-of-semester completion, a later alumni questionnaire should not quietly replace it because its effect looks stronger. Record all prespecified or otherwise relevant results, then mark which one meets the chosen synthesis rule.
Where a study reports several instruments measuring related but non-identical constructs, preserve their names, directions and units. The extraction form should not force unlike outcomes into one column labeled “performance.” The rationale for any conversion or grouping belongs in the methods and synthesis plan.
Step 4: Capture exact reported values and provenance
For numerical outcomes, collect the values necessary for the planned analysis, including units, group labels, relevant denominators, sample sizes, effect estimates and uncertainty information as applicable. Preserve whether a reported estimate is adjusted, unadjusted, intention-to-treat or based on a narrower analytic sample.
Do not infer missing standard deviations, participant counts or confidence intervals merely because the synthesis spreadsheet has an empty field. If a value can legitimately be derived, record the transformation, assumptions, original data and a reproducible calculation separately. For the reporting stage, consult Cochrane guidance on preparing study results for synthesis.
Step 5: Build in missing-data and discrepancy codes
Use distinct entries such as NR (not reported), NA (not applicable), UNCLEAR (reported but ambiguous), and PENDING (awaiting source verification). Keep a short key in the form. Do not treat NR as zero: no reported effect estimate is very different from an effect estimate of zero.
Store the original extracted entry and the final reconciled decision where feasible. Reviewers should be able to explain why a cell was changed. That is especially important if data from a supplement contradicts the main article or if reported denominators change across tables.
Step 6: Pilot the draft before full extraction
A pilot is a small, deliberately varied trial of the form on eligible reports. Select reports that challenge your field definitions: perhaps a multi-arm trial, a paper with several follow-up periods and a report with incomplete outcome data. Have the planned extractors apply the instructions, compare entries and revise fields that produce inconsistent interpretation.
Do not claim that piloting is a one-time box to tick. Significant form revisions may require rechecking previously extracted studies to maintain consistency. Cochrane emphasizes development and piloting of data collection forms, and Cochrane Handbook guidance explains why piloting helps identify process problems early. Report the pilot procedure that actually occurred rather than implying that all dissertations performed the same number of pilot extractions.
Step 7: Extract independently when required and reconcile transparently
For full intervention systematic reviews, guidance commonly recommends independent extraction by more than one reviewer for critical data, with a planned procedure to resolve differences. The Cochrane Handbook describes duplicate extraction as a safeguard against error. A dissertation completed by one student may face practical limits; those limits do not automatically turn a single-person extraction into a two-reviewer process.
Discuss feasibility and methodological expectations with your supervisor. If a verifier checks only selected fields or reports, describe precisely what was verified and identify the limitation. Do not invent a second reviewer, independence, or consensus meeting.
Step 8: Lock the final form and maintain an audit trail
When the pilot resolves its design questions, assign the form a version and record the date it began being used. If the form changes later, document the reason and which existing entries were revisited. Keep source documents, extraction instructions and reconciled records securely, following copyright, data-access and institutional requirements.
The purpose is not bureaucracy. It is to prevent one table being prepared under one definition and another table under a different, undocumented definition.
Worked example: extract a fictional education study correctly
Imagine an entirely fictional report called Study Cedar, which evaluates scheduled instructor feedback against routine course comments in a first-year university module. The report describes an end-of-semester module-completion outcome and a mid-semester student-confidence questionnaire. A supplement clarifies that some participants withdrew before the questionnaire but not before the university's official completion record was finalized.
The extraction task is not to decide that feedback worked. It is to capture the relevant study features and avoid confusing the two outcomes.
| Extraction item | Record for Study Cedar | Source or resolution note |
|---|---|---|
| Study identity | Cedar (fictional) | Internal teaching ID only |
| Design | Parallel-group intervention study, as described in fictional report | Verify design and allocation procedures before classifying as randomized |
| Participants | First-year students enrolled in one module | Record actual inclusion/exclusion criteria if using real research |
| Intervention | Scheduled written feedback against assessment criteria | Capture timing, format, delivery and fidelity details |
| Comparator | Routine course comments | Do not assume comparator means “no feedback” |
| Primary review outcome | Module completion at end of semester | Eligible for the fictional review question |
| Additional reported outcome | Mid-semester student confidence | Separate construct; not interchangeable with completion |
| Outcome data | Not supplied in this teaching example | Do not invent an effect size or favorable direction |
| Attrition issue | Questionnaire withdrawals mentioned in supplement | Distinguish outcome-specific missingness; verify relevant denominators |
| Extraction status | Provisional pending full source checks | Record reviewer, date, and eventual reconciliation |
Notice that this example intentionally does not manufacture participant counts or effect estimates. A high-quality extraction record can still show what information must be located before synthesis. In a real project, the reviewer must obtain eligible numerical results and their source locations, or clearly document that they were not available.
A disagreement worth recording
Suppose Reviewer A records student confidence as the primary outcome because the abstract emphasizes it. Reviewer B records module completion because the review protocol prespecified that outcome. The correct resolution is not to pick the reviewer with greater confidence. The team should compare the protocol with the full study report, retain both outcomes in appropriate fields and document which is eligible for the primary synthesis.
Similarly, if a later paper reports a follow-up using the same participants, link it to the existing study record and decide whether its time point fits the synthesis. Neither a second publication nor an additional measurement is automatically a second independent study.
What the worked example teaches
The extraction form creates a boundary between reported information, reviewer interpretation, and review eligibility decisions. That boundary makes subsequent synthesis more transparent. If the review later uses structured narrative synthesis rather than meta-analysis, the evidence matrix already records comparable outcomes and important differences instead of forcing the writer to reconstruct them from scattered PDFs.
A simple extraction workbook structure
Many dissertation teams use a spreadsheet, while others use dedicated review software. The choice of software is less important than clear definitions, traceable entries, appropriate access controls and consistent review decisions. A practical workbook can separate identity and decision history from outcome-level information rather than squeezing everything into one oversized row.
| Worksheet | Purpose | Example columns |
|---|---|---|
| Study register | Track unique underlying studies and their linked reports | Study ID, report IDs, registry ID, status, notes |
| Study characteristics | Record design, participants, context and interventions | Study ID, setting, eligibility, intervention, comparator |
| Outcome-level extraction | Preserve distinct outcomes, time points and analyses | Study ID, outcome, time, groups, estimate, uncertainty, source location |
| Missing-data queries | Record questions and attempts to clarify reporting | Study ID, missing field, query source, date, resolution |
| Verification log | Track discrepancies and final reconciled entries | Field ID, original value, verifier decision, rationale, date |
| Field dictionary | Define required variables and allowed codes | Field, definition, format, allowed values, instructions |
If you are working alone, do not pretend the verification log records independent adjudication. It can document self-checks, supervisor review or a second extractor only where those activities actually happened. Make it easy for another reader to retrace how each summary table was built.
Should you copy the form into your dissertation appendix?
Often, a blank or suitably redacted extraction form and field dictionary are useful appendices, particularly when transparency is central to your methods assessment. Check institutional requirements and confidentiality limits. Explain in the Methods chapter where the protocol, extraction form and completed data records can be found, if they are available. Our guide to writing and formatting dissertation appendices discusses presenting supporting materials clearly.
How to write the data extraction section of a dissertation
A strong Methods paragraph answers questions that a skeptical reader would ask: who extracted the data, what was captured, whether the form was piloted, how multiple reports were handled, what happened when results conflicted, and whether study authors were contacted for clarification.
PRISMA 2020 asks authors to report the data collection process under item 9, sought outcome data under item 10a, and other collected variables and assumptions under item 10b. These are reporting requirements, not evidence that a particular extraction process was performed well. A completed checklist cannot substitute for reliable data.
Sample Methods wording — only if the described procedures actually occurred
The review team developed a standardized extraction form from the prespecified review question, eligibility criteria and planned synthesis. Fields covered study identification, design, setting, participants, intervention and comparator characteristics, outcome definitions, time points, eligible results, source locations and missing information. The draft form was piloted on selected reports representing different reporting patterns and revised before full extraction. Each eligible study was assigned a stable study ID, and multiple reports were linked to the same underlying study. The review team recorded discrepancies and documented their resolution according to the protocol. Reviewer roles and any departures from the protocol are described below.
This paragraph is a model of structure, not text to copy unchanged. Replace “review team,” piloting, discrepancy resolution and protocol claims with what you truly did. For instance, if one researcher extracted all records and a supervisor checked only a sample, state those procedures precisely rather than claiming independent duplicate extraction.
Report outcome selection and assumptions
Specify which outcomes were sought, how different instruments or time points were handled, and what qualified as the primary result for synthesis. Report any post-protocol adjustments and why they were necessary. If relevant numerical information was missing, describe what was sought and whether investigators were contacted. Do not automatically treat unavailable values as evidence of no effect.
Once the data are extracted, a separate synthesis plan determines whether pooling is justifiable and how uncertainty and bias will affect interpretation. Our guides to risk-of-bias judgments, narrative synthesis and GRADE certainty-of-evidence reporting explain subsequent but distinct stages.
Common data extraction mistakes and how to avoid them
Treating abstracts as complete data sources
An abstract may omit essential denominators, outcome definitions, analytic approaches or adverse results. Use full reports and relevant supplements where available, and record uncertainties. Some fields may be confirmed from a registry or other linked report, but keep the provenance visible.
Double-counting the same participants
Multiple papers may reuse the same cohort or report different stages of one trial. Create a single study-level identity and link all reports. Document any uncertainty about whether participant samples overlap.
Extracting only statistically significant findings
Selecting the strongest or most favorable results creates misleading syntheses. Follow protocol rules for all eligible outcomes and explain any selection of a particular estimate, time point or model.
Confusing missing values with zero values
A blank outcome field, an NR code and a measured value of zero mean different things. Define missingness codes explicitly and train extractors to use them consistently.
Changing field definitions without revisiting old records
If “end of study” changes from final course assessment to the last available follow-up halfway through extraction, the earlier records may no longer be comparable. Version the change and recheck affected entries.
Letting automated extraction replace verification
Review software, spreadsheets and AI-assisted systems can speed up administration, but they can misread PDFs, confuse arms or omit qualifiers. The JBI Manual's standardized extraction guidance and methodological reviews underscore the importance of a defined tool and verification procedures. Document any automation and independently inspect critical data against the original report. Do not treat generated text as an authoritative study result.
Frequently asked questions
Is a data extraction form compulsory for a dissertation?
Requirements depend on the type of dissertation and institution. A formal systematic review needs a transparent, reproducible collection process; a structured form is standard practice. An ordinary narrative literature review may not require the same protocol-level extraction procedures.
Can I use Excel for systematic review data extraction?
Yes, if your review and institution permit it. A well-designed workbook with version control, a field dictionary, source links and discrepancy tracking can support the process. Dedicated review software may provide stronger collaboration and audit features, but no tool automatically makes extraction methodologically sound.
How many reviewers should extract data?
Follow the applicable method and protocol. Cochrane systematic-review guidance recommends independent extraction of key data by more than one person to reduce errors. If resources constrain your dissertation, describe the actual review process and discuss the limitation honestly instead of implying that independent duplicate extraction occurred.
How many studies should I pilot the form on?
There is no universally sufficient number for every dissertation. Choose enough varied reports to uncover important ambiguity in the fields, test the planned reviewer workflow and revise before full use. Follow any review-specific methodological or institutional requirements.
What if a study reports multiple outcomes or follow-ups?
Record each relevant result separately, including the construct, instrument, timing and source. Apply prespecified rules to select the result for each synthesis. Do not choose the one with the most favorable conclusion after examining outcomes.
Is data extraction the same as critical appraisal?
No. Extraction records what studies report; critical appraisal and risk-of-bias assessments evaluate whether the evidence may be distorted or otherwise limited. They can share source information but should not be conflated.
Does PRISMA validate my data extraction form?
No. PRISMA is a reporting guideline. It helps readers understand how you collected data and what you sought; it does not certify that the form, reviewer decisions or extracted values are correct.
Final checklist before you write the Results chapter
The form's fields map to the question, protocol, outcomes and planned synthesis.
Every underlying study has one stable study identifier, with multiple reports linked appropriately.
Each field has a clear definition, permitted entries and a source-location rule.
Outcome time points, scales, analyses and denominators are distinguished where relevant.
Missing information and uncertainty are coded consistently; missing is never silently treated as zero.
A pilot was completed and documented if it was actually performed.
Reviewer roles, independent extraction and verification are described truthfully.
Important discrepancies, protocol changes and reconciliations have an audit trail.
Any derived data preserve their source inputs, assumptions and calculations.
Study-level results can be traced to the original eligible reports.
The Methods section addresses PRISMA 2020 items 9, 10a and 10b where applicable.
Appendix and sharing decisions respect institutional requirements and data-access rules.
Final takeaway
A useful extraction form is not the one with the most columns. It is the one that consistently captures the exact information needed to answer the review question, keeps different studies and reports straight, and lets readers trace every result back to its source. Define fields before extracting, pilot the form, record decisions and avoid quietly filling gaps with assumptions. That preparation makes a dissertation's later analysis clearer, more reproducible and easier to defend.
If you need human academic help organizing your review's evidence matrix, checking the clarity of your Methods section or presenting results transparently, explore Acadelyra's Research Support and Dissertation & Thesis Support services. Acadelyra's role is to support sound research practice and academic integrity; the actual source extraction, methodological decisions and submitted scholarship must remain your own.
Authoritative references and further reading
Li, T., Higgins, J. P. T., & Deeks, J. J. (2024). Chapter 5: Collecting data. In Cochrane Handbook for Systematic Reviews of Interventions (version 6.5). Cochrane.
Page, M. J., et al. (2021). The PRISMA 2020 statement: An updated guideline for reporting systematic reviews. BMJ, 372, n71. Open-access full text.
JBI. JBI Manual for Evidence Synthesis, standardized data extraction guidance.
Development, testing and use of data extraction forms in systematic reviews: A review of methodological guidance (2020). Open-access full text.
Cochrane. Cochrane Handbook, Chapter 9: Summarizing study characteristics and preparing for synthesis.