Login Create Account

MaxDiff (Best-Worst Scaling) Analyzer

Preference & Prioritization Conjoint Tool → TURF Tool →

Ask people to rate ten features on a 1–10 scale and they will call nine of them very important. Ask them to pick one Best and one Worst from a small set, and they have to trade off. This tool turns those forced choices into a ranked list of priorities on a 0–100 scale that sums to 100, flags which rank gaps are statistically real versus ties, and checks whether the average ranking hides opposing customer segments.

👨‍🏫 Professor Mode: Guided Learning Experience

New to MaxDiff? Enable Professor Mode for step-by-step guidance through reading a best-worst study, telling a real gap from a statistical dead heat, and checking whether the average ranking describes anyone!

OVERVIEW & WHAT THIS TOOL SOLVES

Why forced tradeoffs beat rating scales: On a 1–10 Likert scale (left), respondents rate all six features 8.2–8.9 (“everything matters”). Picking one ✓ Best and one ✗ Worst per screen (center) forces real tradeoffs and separates the list into a 0–100 priority budget (right) — revealing clear winners, a statistical tie in the middle, and a bottom item that looked popular on Likert.

Suppose a product team has budget to build two of six candidate features. If you run a standard 1–10 importance survey, almost every feature comes back between 8.2 and 8.9 because calling everything “very important” costs the respondent nothing. MaxDiff (Best-Worst Scaling) replaces rating scales with a scarce forced choice—“of these four, which is most important and which is least?”—turning those trade-offs into a 0–100 priority budget that sums to 100.

🎯 Questions It Answers

  • Which claims, messages, or product features sit at the top of customers' true priority list?
  • Which rank gaps are statistically real winners versus a statistical dead heat?
  • Does the overall average ranking hide two opposing customer segments with different priorities?

📂 What Data You Need

  • Long-format Best-Worst choice data: one row per item, per task set, per respondent.
  • Respondent ID, Task ID, Item label, plus binary Best (0/1) and Worst (0/1) columns.
  • Optional segment column to compare priority rankings across customer groups.

💡 What You'll Discover

  • 0–100 Priority Shares: Probability-rescaled priority scores summing to 100 against an equal-share benchmark.
  • Rank-Separation Verdicts: Confidence intervals and pairwise tests flagging real gaps vs. ties.
  • Heterogeneity & Counting Checks: Best-minus-Worst napkin counts and latent segment splits.
🔍 Key Nuances & Common Traps to Watch For Click to expand ▾
  • Why Best-Worst beats Likert ratings: A 1–10 rating is denominated in a private currency—generous raters live in the 8–10 band while strict raters never pass 6. Best-Worst records only ordinal trade-offs within each set, eliminating scale-use bias.
  • Relative priority, never absolute demand: Priority scores always sum to 100 across the tested list. If you run MaxDiff on ten features nobody wants, the least-hated item will still win the highest share.
  • When to use MaxDiff vs. Conjoint: Use MaxDiff to rank a list of standalone items (messages, claims, benefits, features). Use Conjoint when items are combined into multi-attribute product bundles with prices to simulate market share or Willingness-to-Pay.
📐 Under the Hood: Model Equations & Variable Definitions Click to expand ▾
1. Linked Best & Worst Multinomial Logit Choice Probabilities Best-Worst Logit
\[ P(\text{best} = b \mid S) = \frac{e^{u_b}}{\sum_{j \in S} e^{u_j}}, \qquad P(\text{worst} = w \mid b, S) = \frac{e^{-u_w}}{\sum_{j \in S \setminus \{b\}} e^{-u_j}} \]
💬 Read it in plain English: Picking the Best item follows a standard multinomial logit over all items in the shown set \(S\), while picking the Worst item is a multinomial logit over the remaining \(a - 1\) items using negated utilities (\(-u_w\)).
Variable & Symbol Directory (Every Term Defined) Hover or focus any role badge for details
b, w
Chosen Best & Worst Items YOUR DATA ⓘ
\(b\) is the item picked as Most Important; \(w\) is the item picked as Least Important in task set \(S\).
S
Task Choice Subset YOUR DATA ⓘ
The set of \(a\) items shown together on a single screen; \(S \setminus \{b\}\) is the set of remaining items after \(b\) is chosen as Best.
uj
Latent Item Utility (Log-Odds Scale) ESTIMATED ⓘ
Estimated priority utility of item \(j\) on a zero-centered log-odds scale (higher \(u_j\) = stronger priority).
2. 0–100 Probability-Rescaled Priority Share 0–100 Priority Scale
\[ \text{Score}_i = \frac{\exp(u_i)}{\exp(u_i) + a - 1} \times \frac{100}{\sum_{j=1}^{K} \frac{\exp(u_j)}{\exp(u_j) + a - 1}} \]
💬 Read it in plain English: Converts zero-centered log-odds utilities \(u_i\) into intuitive 0–100 priority shares that sum to 100 across all \(K\) items, where an average item sits near the equal-share benchmark \(100 / K\).
Variable & Symbol Directory (Every Term Defined) Hover or focus any role badge for details
Scorei
0–100 Priority Share OUTPUT ⓘ
Item \(i\)'s share of the 100-point priority budget (an item at 20 has twice the modeled relative priority weight of an item at 10).
a, K
Items per Screen & Total Items YOUR DATA ⓘ
\(a\) is the number of items displayed per task screen (typically 4 or 5); \(K\) is the total number of items in the study.

STEP 1: LOAD DATA

📚

Use a Case Study

Load a realistic MaxDiff dataset with a written brief.

Select a scenario above to load a preset MaxDiff dataset with context.

📤

Upload Your Own Data

Long-format best-worst data (CSV or TSV).

Drag & drop your MaxDiff file

Needs respondent, task, item, best and worst columns

No file loaded.