Course · 4. MaxDiff

Your MaxDiff data

Chosŏn reads one row per item shown in a set:

• Each set shows at least 3 items, with no item twice, exactly one best and one worst, not the same item.
• Sets may have different sizes, and respondents may see different numbers of sets.
• set numbers the sets in the order shown. With a holdout, each respondent's last set is kept back to check the fit.

Designing the survey

• Sets of 4 or 5 items. More than 6 per set gets hard to answer well.
• Each item about 3 times per respondent. With 16 items in sets of 4, that is 12 sets. Each person's own scores come from these few showings, so fewer makes them vague.
• Balance. Show every item about equally often, and each pair of items together about equally often, in different positions. The Design the survey card on a MaxDiff project makes such sets: one version per respondent, as a CSV for your survey tool.
• Every item must link to every other through sets shown together, directly or in a chain. Otherwise the two groups cannot be compared, and the data check stops with an error.
• Aim for 200 respondents or more (the tutorial has 300). The ranges narrow as respondents are added.

The data check

Before fitting, Chosŏn checks the format and warns about weak spots: items shown fewer than twice per respondent, items shown much less often than the rest, too few sets for a holdout, fewer than 100 respondents. Counts per item (shown, picked best, picked worst) are listed under the check.

Size limits

Free plan: up to 300 respondents, 20 sets, 6 items per set and 30 items. Premium: up to 1,000 respondents, 24 sets and 40 items. MaxDiff fits run on our servers' CPUs on every plan; the tutorial (300 respondents, 12 sets of 4, 16 items) takes about 40 seconds.

Try it in the tutorial: the data

Next: The model

Tutorial links open the example project in the app; it is free and needs a free account.