How to Extract and Chart Data from Studies for a Literature Review

That gap catches almost everyone, and it is not a reading problem. It is a format problem. Everything you know about those papers is currently stored as twenty-two separate impressions, one per document, in the order you happened to read them. To write a review you need the opposite shape: one field at a time, across all twenty-two, so you can see that six of them used the same measure and four of those found nothing.

Turning the first shape into the second is a distinct step with a name. In a systematic review it is called data extraction. In a scoping review it is called charting the data. It sits between the reading and the writing, it takes a couple of afternoons, and skipping it is the single most common reason a chapter comes back with the comment that it reads like a list of summaries.

Extraction and Charting: Two Words for One Habit

Both terms describe the same physical act. You open a paper, you decide in advance which facts you want from it, and you record those facts in fixed fields in a table rather than in prose. Do that for every included study and you end up with a grid where the rows are papers and the columns are questions.

The difference is what the table is for, and it follows from the kind of review you are writing. Data extraction, the systematic review term, pulls out results so they can be compared or pooled. The table is aiming at a finding: this intervention worked in seven studies and not in three. Charting, the scoping review term, pulls out characteristics so the shape of a field can be described. The table is aiming at a map: most of this research is quantitative, almost all of it is from high income countries, and nobody has studied adolescents at all.

That matters when you choose columns. A charting table often has more columns about the study and fewer about the outcome, because the gap in the literature is the finding. An extraction table goes deep on outcomes and effect sizes, because the result is the finding. If you are not yet sure which review you are doing, settle that first with scoping vs systematic vs narrative, because it changes the table you are about to build.

For a narrative review or a taught masters chapter, nobody will ask to see the table. Build it anyway. It is the fastest route to a chapter that compares rather than lists, and it costs less time than the rewrite you avoid.

Decide What Each Column Is For Before You Open a Paper

The instinct is to start extracting and let the columns emerge. It feels efficient and it costs you the most time of anything in this process, because by paper eleven you will have invented three new fields and will have to reopen the first ten to fill them in.

So set the columns first, and set them from your research question. Every column should earn its place by answering one of two things: what would I need to know to compare these studies fairly, or what would I need to quote to defend a claim in my chapter. A column that answers neither is a column you will fill in thirty times and never look at again.

Then pilot it. Take three studies that differ from each other as much as your included set allows, extract them fully, and see what breaks. Usually two things do. One field turns out to mean different things in different papers, and one thing you keep wanting to write down has no home. Fix both, and only then start the other twenty.

The Fields a Good Extraction Table Has

This is a solid default set. Trim it to your question rather than adopting all of it, and add anything your topic makes central.

The last two are the ones people leave out and later wish they had not. A page reference turns a five minute check into a ten second one. A note in your own words is where the synthesis actually starts, because it is the first time you have written anything comparative about the set.

Quantitative and Qualitative Studies Need Different Columns

The frame above works for both, but the outcome columns behave differently, and forcing one style onto the other is where mixed reviews go wrong.

For quantitative studies you are extracting values. Record the measure used, the comparison being made, the direction of the result, the magnitude, and the precision if it is reported. Copy numbers exactly as printed and note the unit. Resist the urge to round or to convert into a common scale while you are extracting, because that is analysis, and analysis done at the same moment as transcription is where quiet errors get in.

For qualitative studies you are extracting meaning, which does not fit in a cell as neatly. Record the themes exactly as the authors named them, one per line rather than merged into a sentence. Add one short illustrative quote per theme with its page number. Note the analytic approach, because thematic analysis and grounded theory produce different kinds of claim. The discipline here is to record their themes in their words first, and to do your own interpreting later in a separate column, so you can always see which is which.

If your review includes both, run two tabs with a shared set of descriptive columns and different outcome columns, rather than one table with a lot of empty cells. You will bring them back together at the synthesis stage, not the extraction stage.

Spreadsheet or Matrix: Where to Build the Thing

A spreadsheet wins for anything above roughly fifteen studies. You get filtering, sorting, and the ability to hide every column except design and outcome when you want to see whether the randomised trials disagree with the surveys. That one move, sorting by a column, is most of what the table is for.

A document table is fine for a smaller narrative review, and it has one real advantage: it holds long text without the cell clipping that makes spreadsheets awkward for qualitative work. If you go this way, keep the table in its own file rather than inside the draft chapter, so that editing the chapter never risks the record.

Either way, a few habits pay for themselves. Freeze the header row and the citation column. Use one row per study, never one row per finding, and put multiple findings on separate lines inside the cell instead. Add a column for the date you extracted it. Write not reported rather than leaving a cell blank, because a blank cell is ambiguous later and not reported is itself a finding you may want to count. And back the file up somewhere that is not only your laptop, alongside the rest of your organised sources.

What One Good Row Looks Like

Here is the same fictional study recorded twice. The first version is what most first attempts look like.

Everything in there is true and almost none of it is usable. It cannot be sorted, it cannot be compared with the next study, and in three weeks the phrase quite effective will not tell you whether this was a large effect or a small one. It is also a summary, which means you have done the work of reading without producing anything the chapter can stand on.

That took about eight minutes and it is worth noticing what it bought. The comparison group and the missing adjustment are now visible, so the result cannot quietly become causal in your chapter. The number is there, so you can compare it. The self-selection problem is written down at the moment you spotted it rather than three weeks later when you cannot remember which study it applied to. And that last line already contains a comparative claim, which is a sentence of your review written early.

Extract and Appraise in the Same Pass

Running quality assessment as a separate project a fortnight later means opening every paper twice. Since you are already reading the methods section closely to fill in design and sample, that is the natural moment to form a judgement about whether the study should be believed.

So bolt a few appraisal columns onto the same sheet rather than starting a new one, and let the appraisal checklist tell you which columns those should be. Keep the judgement as a short sentence rather than a score, because a total out of ten hides the thing you actually need, which is what specifically is weak about this study.

One more check belongs in the same pass, and it takes seconds: confirm the paper has not been retracted before you spend eight minutes extracting from it. Better to find out now than after it is cited in a submitted chapter.

How the Table Turns Into the Chapter

This is the payoff, and it is worth being concrete about, because the table can otherwise feel like admin.

Start by reading down a column instead of across a row. Reading across gives you one study, which you already knew. Reading down the outcome column across all twenty-two gives you the pattern, and the pattern is the thing you have to say. Sort by design and see whether the stronger studies agree with the weaker ones. Sort by country and see whether the finding travels. Sort by year and see whether the field changed its mind.

Each of those sorts produces a sentence, and the sentences have a recognisable shape. Most studies report a positive effect, but the three using registry data rather than self report show a smaller one. Every study in this set was conducted in a high income country, so how the finding transfers is unknown. Those are synthesis sentences, and you got them from sorting a spreadsheet rather than from staring at a blank page.

Groups of related sentences then become your themes, which is exactly the input that structuring a review by theme needs, and the paragraph craft that turns three rows into one argued paragraph is covered in how to synthesize sources. The extraction table is the bridge between them. A synthesis matrix organises what you will say. An extraction table records what the studies said, which is what you argue from.

If you are writing a systematic or scoping review, the table also earns its keep in the write up itself. The methods section needs a description of what you extracted and who did it, which belongs alongside your search strategy and eligibility criteria in the same methodology section. A trimmed version of the table usually appears as the study characteristics table in the results, and the counts you have been keeping feed the PRISMA flow diagram.

Five Ways This Goes Wrong

These are the ones that cost real time, roughly in order of how often they turn up.

1. Extracting into prose instead of a table. Paragraph notes per paper feel like progress and cannot be sorted, filtered or compared. If your notes are twenty-two headings in one document, you have summarised, not extracted. Converting them is an afternoon, and it is an afternoon that will save you a week.

2. Fields that drift between studies. Sample size means participants in row three and schools in row eleven. Outcome holds a percentage in one row and a theme in the next. The table looks complete and cannot be compared. Write a one line definition for each column at the top of the sheet and keep it visible while you work.

3. Extracting everything you can see. Thirty columns because each seemed like it might be useful. This triples the time per paper and buys nothing, since a column you never sort by is a column you did not need. Every field should trace back to your research question or to a comparison you know you will make.

4. No link back to quality. Without an appraisal column, a well controlled trial and a thirty person convenience survey sit in the table looking identical, and eventually get cited in the same sentence as though they carry the same weight. That is the error a marker notices immediately.

5. Assuming a single extractor never slips. Formal systematic reviews use two independent extractors for good reason: transcription errors are common and invisible. Most students do not have a second person, and the practical substitute is to re-extract a sample of your own rows a few days later without looking at the first attempt, then check for disagreements. Finding two or three is normal, and it tells you which columns are ambiguous. If you had help from anyone, say so in the methods.

Where This Sits in the Process

The chain is fairly rigid, and each step assumes the one before it is done. You choose a review type, set your eligibility criteria, build a search strategy, screen what comes back and report the counts, appraise what survived, and then extract from the studies you kept. Writing starts after that, and it starts much faster because the table has already done the comparing.

One reassurance about the effort. Extraction is the last step where you are still gathering rather than composing, and it is the one that most reliably converts into finished text. Twenty studies at eight minutes each is under three hours of work that you would otherwise spend twice, once now in a vaguer form and once again when you cannot remember which paper had the odd sample.