MaxDiff Analysis Software Compared: Standard vs. Anchored Best-Worst Scaling
MaxDiff (Maximum Difference Scaling) software generates balanced incomplete block designs where respondents pick the best and worst items from small subsets of 4–5 items, converting choices into a differentiated hierarchy of preferences. While dozens of survey tools offer basic best-worst counting, serious prioritisation requires Anchored MaxDiff, Hierarchical Bayesian item utilities, and TURF + Shapley portfolio reach.
Why Basic "Count-Based" MaxDiff Tools Fall Short
Many general survey tools (such as basic tiers of SurveyMonkey or Alchemer) score MaxDiff by subtracting "Times Chosen Worst" from "Times Chosen Best", or by fitting an aggregate logit model without anchoring. That creates three commercial blind spots:
Relative ranks without an absolute buy threshold (Unanchored MaxDiff). In unanchored MaxDiff, Feature A may rank #1 and Feature B may rank #2, yet customers might find every item on your list unimportant and unworthy of purchase. Anchored MaxDiff adds a per-respondent calibration question ("Would you actually consider / pay for the best item in this set?") that estimates an individual utility cutoff, separating true must-haves from items that are merely "least bad".
Ignoring preference polarization. If 50% of buyers love Feature A and 50% hate it, an aggregate count scores it in the middle of the pack. Hierarchical Bayes estimates utilities per respondent, preserving the polarized niche so k-means++ and Bayesian Latent Class segmentation can uncover it.
Missing TURF and Shapley portfolio reach. Roadmaps and marketing campaigns rarely ship a single claim or feature—they ship a bundle of 3 to 5. Without TURF (Total Unduplicated Reach and Frequency) and Shapley reach attribution, you risk picking your top 3 ranked items when all 3 appeal to the exact same 40% of customers.
Comparison of MaxDiff Platforms
Capability | Ordeen Choice Modelling | Sawtooth (Lighthouse / Discover) | Conjointly | General Survey Tools (Alchemer / QuestionPro / SurveyMonkey) |
|---|---|---|---|---|
Anchored MaxDiff | On by default (non-linear set-maximum logit threshold per respondent) | Supported (Dual-Response / Lattery anchoring in Lighthouse) | Supported | Rarely supported; usually unanchored relative scores |
Estimation Engine | Multi-chain Hierarchical Bayes with visible R-hat, ESS & divergence gate | HB (Gibbs/MH) or aggregate logit | Cloud HB | Simple Best-minus-Worst counts or aggregate logit |
TURF & Elbow Rule | Unified independent-appeal TURF with 95% intervals, ceiling line & statistical 2pp elbow stop | Supported | Supported | Manual export required |
Shapley Portfolio Attribution | Exact (n≤12) / sampled Shapley reach decomposition across items | Not standard in base reporting | Not standard | Not supported |
Segmentation | Signal-weighted k-means++ & Bayesian Latent Class with LOTO-CV & purity | Latent Class / k-means | Crosstabs | Basic demographic filters |
Respondent Warm-Up | Interactive item-recognition warm-up before tasks count | Manual text instructions | Manual instructions | None |
How Ordeen Runs MaxDiff End to End
In Ordeen, you paste your list of 10 to 30 features, claims or benefits (or let AI draft them from your brief). Ordeen generates a balanced incomplete block design, primes each respondent with a short item-familiarisation warm-up, and estimates individual item utilities and anchor thresholds with Hierarchical Bayes. In the results and simulator, you get:
Individual-level utilities and 95% credible intervals for every item.
The Unified TURF Card, showing the exact step-by-step item line-up that maximises unduplicated audience reach, complete with the recommended elbow stop (≥2pp gain and 95% CI excluding zero) and the category ceiling.
Portfolio-Reach Shapley, decomposing total reach fairly across items.
Item-pair interactions and full k-means / Latent Class segmentation.
Frequently Asked Questions
What is the difference between Standard MaxDiff and Anchored MaxDiff? Standard MaxDiff only ranks items relative to one another, so even the #1 item could be unappealing in absolute terms. Anchored MaxDiff adds a threshold question that estimates a per-respondent zero point, revealing which items clear the bar of genuine buyer appeal and enabling TURF reach analysis.
How many items can I test in a MaxDiff study? MaxDiff works best for lists of 10 to 30 items shown 4 or 5 at a time, with each item shown at least 3 to 5 times per respondent.
Does MaxDiff measure price sensitivity or Willingness to Pay? No. MaxDiff prioritises items on a flat list without price trade-offs. If you need to know how much buyers will pay for features or how share moves against competitors, use Choice-Based Conjoint (CBC), Adaptive CBC (ACBC), or Menu-Based Choice (MBC)—all included in the same Ordeen workspace.
Need to prioritise your product roadmap or claims? Book a 20-minute demo or email im@ordeen.com.
