Graphic Representation of Research Evidence Matrix Guide

The chart is an analytical instrument, not a bibliography laid out sideways, and the brief says so in the phrase that matters: it should organise your research *and allow you to analyse each one across certain key dimensions*. Analysis across dimensions means the columns are the argument. Once every study occupies the same row structure, patterns become visible that a reading list conceals — that all the positive findings come from one design, that the samples cluster in one age band, that the strongest studies are the oldest. Build the chart so those comparisons are possible, then read down each column rather than across each row. That reading is what the next assignment in the sequence will draw on, so a chart built carefully now saves the outline from being written out of memory later.

The supplied example gives you a workable column set and you are told you may add to it: date and author, sample, location, programme name, instrument and evaluation, type of study, findings, and level of evidence. Two of those do more work than the rest. *Type of study* is what lets you group by design, and it should carry a real label — randomised controlled trial, quasi-experimental, cohort, case study, qualitative — rather than the word "study". *Level of evidence* is a rating against a published hierarchy, and it is only meaningful if you say which hierarchy you used, because the numbering differs between schemes. The example's parenthetical note that an instrument's validity is questionable is worth imitating too, since a column entry that flags a weakness is more useful than one that merely records a fact.

Add columns where your own topic needs them, because the example's set is generic and the brief explicitly permits adaptation. If your research concerns an intervention, a column for dose or duration of exposure usually earns its place, since it often explains why findings disagree. If outcomes are measured differently across studies, a column separating the outcome from the instrument that measured it prevents you comparing things that are not comparable. A limitations column is almost always worth having. Resist adding columns you will fill with the same value for every row — a constant is not a dimension. Decide the columns before entering any data, because retrofitting a column across fifteen studies means rereading all fifteen. A column recording who funded the study is worth considering as well, since it is cheap to fill and occasionally explains a pattern in the findings column that nothing else accounts for.

Ordering the chart is a decision the brief hands to you — organise it in a logical way, by date, by programme or intervention, or by whatever your own research suggests — and different orders surface different patterns. Chronological ordering shows how the field moved and whether recent work supersedes older findings. Grouping by intervention shows where evidence converges and where it is thin. Ordering by level of evidence shows immediately whether your strongest claims rest on your weakest studies, which is the single most useful thing a chart of this kind can tell you. Pick one and say why in a sentence above the chart. If two orderings both look informative, build the chart in a spreadsheet so you can sort it either way rather than committing to one arrangement in a static table.

Filling the cells demands a discipline that is easy to lose halfway through. Keep entries comparable: if one row's sample says fifty girls aged seven to ten and the next says "children", the column has stopped functioning. Keep findings factual and short — what the study found, not what you think it means — because interpretation belongs in the paper the chart feeds, and a findings column full of judgement is harder to scan. Use consistent notation for anything absent, since a blank cell is ambiguous between not applicable and not reported, and those are very different facts about a study. Read your finished chart as a stranger would, checking that every cell would mean the same thing to someone who has not read the underlying paper. Where a study reports several outcomes, choose the one your research question is about and say so, rather than summarising all of them into a cell too small to hold them.

The mechanics are short: one to two pages, uploaded to the Dropbox. Two pages is a real constraint on a chart with eight or nine columns, so consider landscape orientation, abbreviate consistently with a key, and keep the number of studies proportionate to the space — a dozen well-completed rows read better than twenty-five truncated ones. Give the chart a title and a one-line note saying which evidence hierarchy the ratings use, since that is the entry a reader cannot interpret without you. Check that the chart is legible at print size rather than only on screen, because a table that requires zooming has failed the one job a graphic representation exists to do. Keep the underlying spreadsheet even after submitting, because the chart is a working document for the rest of the sequence and rebuilding it from the submitted page is wasted effort.