An extraction table, or literature matrix, is one row per included source and one column per thing you decided to pull from all of them. It is the one artefact in a review that pays back most of the work put into it, because the synthesis can only say things the table can show. Choose the columns badly and you will re-read forty papers; choose them well and the argument writes itself from the table. This page is about choosing them, and about the arithmetic of filling them in.
The columns decide the argument
If you want to say that the effect differs by setting, setting has to be a column. If you want to say the older studies used a different measure, the measure and the year have to be columns. Working backwards from the claims you expect to make is the fastest way to design the table, and it is why the design happens before reading rather than during it.
A default set worth starting from
Citation, year, country or setting, design, sample and its size, what was measured and how, the main finding, the limitation that matters, and your own note on how it bears on your question. Ten columns is a comfortable maximum for a narrative chapter; a systematic review with an appraisal tool will carry more. Your own question should add two or three and remove one or two of these.
Fill it consistently or not at all
A half-filled matrix is worse than none, because it looks complete and is not. If a field is genuinely not reported, record that explicitly rather than leaving the cell empty: not reported is a finding, and in a table of thirty studies a column of them is often the most quotable thing in the chapter.
What it costs, honestly
Cells times minutes. The free worksheet on this site multiplies your sources by your columns, adds whatever share you are double-extracting, and turns it into hours and weeks at the pace you can actually work. People consistently underestimate this by a factor of two or three, and discovering it in advance changes either the column count or the timetable rather than the quality.
Questions people ask about extraction table
What columns should a literature matrix have?
Work backwards from the claims you expect to make: anything you want to compare across studies has to be a column. Citation, year, setting, design, sample, measure, finding and limitation is a reasonable starting set to adapt.
Should I extract into a spreadsheet?
It works, and it is what most people do. Its weaknesses are version control and the fact that it holds no reason for an exclusion, which is exactly the gap the product on this site fills. Whatever you use, decide the columns before reading.
What if a study does not report something?
Record not reported rather than leaving the cell blank. A column full of them is a real finding about the field and often the most quotable sentence in the chapter.