TURF (Total Unduplicated Reach and Frequency) answers a portfolio question: if you can only offer k products, flavours or messages, which combination appeals to the most people? Picking the k individually most popular items often fails, because popular items tend to appeal to the same respondents. TURF counts each respondent once, however many items in the set they like.
Input: one yes/no column per item
Each candidate item is a column where a respondent is a yes (1, “1”, true) or a no (anything else). A typical source is a top-2-box recode of a purchase-interest scale, done beforehand in the Data Cleaner or Label Mapper. Analysis is unweighted: every row counts once.
Reach
A respondent is reached by a combination if they said yes to at least one item in it.
Reach % = 100 × (respondents with at least one yes in the set) / (all respondents)
Reach is unduplicated: a respondent who likes three items in the set still counts once. That is what makes TURF reward items that bring in new people rather than items that please the same people again.
Frequency
Frequency is the total number of yes answers to items in the set, summed over all respondents. Between two combinations with similar reach, higher frequency means the reached respondents have more options they like in the line-up.
Finding the best combination
Up to 20 candidate items: exhaustive search. Krosstabs scores every possible combination of k items (for 20 items and k = 5 that is 15,504 combinations) and ranks them by reach. The top row is the true optimum for your data.
More than 20 items: greedy search. The number of combinations grows too fast to score them all in a browser, so Krosstabs builds the set one item at a time: start with the single item of highest reach, then repeatedly add the item that raises reach the most, until the set has k items. Greedy search is fast and usually lands on or near the optimum, but it is not guaranteed to find it.
Ties. When two combinations reach the same number of people, the one with the higher frequency (more total picks among the people reached) ranks first. If that is equal too, they are ordered by name, so the result never depends on the order you picked the columns in. Greedy search breaks ties between candidate items the same way.
Forced items. If you mark items as must-include (for example a hero SKU that will ship regardless), only combinations containing them are considered, and greedy search starts from them.
Reading the results
- Look at how reach grows with k. When adding an item buys only a point or two of reach, the portfolio is close to saturated.
- Several combinations often sit within a point of each other. Treat them as equivalent and choose on cost, margin or brand fit.
- Reach is only as good as the yes/no definition. A loose threshold (any interest) inflates reach for every combination; a strict one (definitely would buy) is closer to real trial.
Privacy
TURF runs in a Web Worker inside your browser. Your respondent data never leaves your device.