Learn · Start here

Your first analysis in Logit

This walkthrough goes from a file of responses to calibrated measures, a Wright map, and a shareable report. If you have your own data, use it; the shape below is all Logit needs.

1 · Shape your data

One row per respondent, one column per item, cells holding the scored responses — 0/1 for right/wrong, or 0/1/2/… for rating scales. An ID column is fine (you’ll tell Logit which one it is); empty cells are treated as missing and handled by the model rather than dropped. Logit accepts CSV, TSV, Excel (.xlsx/.xls), and SPSS (.sav) — .sav files keep their variable names and labels.

id,   q1, q2, q3, q4, q5
P001,  1,  1,  0,  1,  0
P002,  1,  0,  0,  0,  0
P003,  1,  1,  1,  1,  1

2 · Create a project and upload

Sign in, create a project, and upload the file to it. Logit previews the parsed table so you can confirm the columns and spot scoring surprises (a “9 = missing” code, a reverse-keyed item) before any model runs. A project holds datasets and any number of analyses over them, so re-running with different choices never clutters your original data.

3 · Choose a model

Three choices, driven by your response format: dichotomous for 0/1 scoring, Rating Scale (RSM) for polytomous items that share one response scale (the common Likert survey case), Partial Credit (PCM) when each item’s steps deserve their own structure. If you’re unsure between RSM and PCM, the category structure guide has the trade-off; running both and comparing is cheap.

4 · Read the results, in this order

Wright map first. Are the items where the people are? Any gaps? Does the item ordering, top to bottom, match what your theory says should be hard and easy? That last check is the fastest validity test you’ll ever run — the Wright map guide covers how to read it.

Fit second. Sort the item table by infit and inspect anything outside roughly 0.5–1.5, using the fit guide to tell noise from redundancy. Click a flagged item to see its ICC — the observed proportions overlaid on the model curve usually make the problem obvious.

Then the summary numbers. Person reliability and separation tell you how many levels of the trait the instrument distinguishes in this sample; for rating scales, the category structure table tells you whether every response option is earning its place.

5 · Export the report

The PDF report packages the summary statistics, Wright map, item and person tables, fit flags, category structure, and score-to-measure table — the set a methods section or a committee needs. Charts export individually as SVG/PNG for papers and slides.

What to try next

Re-run the same dataset under a different model and compare; add a grouping column and run a DIF analysis to check whether items behave the same across groups; or follow the Liking for Science walkthrough to replicate a published textbook calibration and see every diagnostic on data whose answers are known.