Course · 3. Menu-based choice (MBC)

The menu model

Every time someone looks at the menu, each item gets a pull towards buying it. They buy the items whose pull is strong enough. The pull has four parts:

• The item's appeal at its usual price, the same for everyone.
• Prices: the item's own price and, by default, the other items' prices too.
• The person: a lasting taste for each item that stays with them on every menu. Someone who likes healthy food leans towards salad, soup and juice every time.
• The occasion: what they came in for that day, new on every menu. On a quick-lunch day they take a sandwich and crisps; a main replaces another main.

Items can be linked through either of the last two. Same-visit links come from both; links between one visit and the next come only from the person. Because each respondent's menus are kept in order, the model can tell them apart.

Technically this is a multivariate probit with a person effect (a panel model), fitted one pair of items at a time, which keeps it fast as the menu grows.

The comparison model

Chosŏn also fits the standard multivariate probit, which has the same item appeals, prices and same-visit links but treats every visit as a new person. The two agree on any one visit. They differ on anything that spans visits: who comes back, and how many different people a range reaches over a month.

Settings

• Cross-price effects (default on): every item's price can affect every item's buy rate. Off: each item responds to its own price only, which is steadier on small studies.
• Holdout (default on): keep each respondent's last menu out of the fit and check predictions on it.

Try it in the tutorial: model setup

Next: Reading the menu results

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