MaxDiff Sample Size Requirements
How Many Respondents Does a MaxDiff Study Need?
MaxDiff is less sample-hungry than conjoint analysis because each choice task is cognitively simpler. A respondent picks the best and worst from 4-5 items rather than evaluating multi-attribute product profiles. This simplicity means you get cleaner data per respondent, which translates to lower sample requirements.
The practical answer: 200-300 respondents covers most MaxDiff studies. But the exact number depends on your analysis goals, number of items, and whether you need segment-level comparisons.
Sample Size by Analysis Type
| Analysis Goal | Minimum | Recommended | Notes |
|---|---|---|---|
| Aggregate ranking (top-line priorities) | 100 | 200 | Count analysis or aggregate logit |
| Segment-level comparison | 150 per segment | 200-300 per segment | Separate HB estimation per segment |
| Individual-level scores (HB) | 200 | 300-500 | Enables clustering, individual predictions |
| Latent class segmentation | 300 | 500+ | More classes require more data |
Aggregate Ranking
If you just need to know which 5 features out of 20 are most important across your entire audience, 100-200 respondents will give you a stable ranking. Count analysis (best count minus worst count for each item) produces a reliable rank order at this sample size. The ranking from 150 respondents will look very similar to the ranking from 500 respondents at the aggregate level.
Segment Comparisons
This is where sample requirements jump. If you want to compare priorities between enterprise buyers and SMB buyers, each segment needs its own viable sample. With 150 per segment, you'll get directional differences. With 200-300 per segment, the differences become statistically testable.
Plan your segments before fielding. Three segments of 200 each means 600 total respondents, not 200 split three ways. This is the same principle that catches people off guard in conjoint sample planning.
Individual-Level Modeling
Hierarchical Bayesian (HB) estimation produces utility scores for every individual respondent, not just group averages. This enables clustering (finding natural preference-based segments), individual-level predictions, and detailed cross-tabulation. HB works with 200+ respondents, and the estimates stabilize around 300-500.
Below 200 respondents, HB can still run, but the population-level priors heavily influence individual estimates. The individual scores start looking more like "group average with noise" than genuine individual preferences.
Factors That Affect Sample Needs
Number of Items
More items require more sets per respondent, which means each respondent provides more data points. A 30-item study with 18 sets gives you more information per person than a 12-item study with 9 sets. This partially offsets the increased complexity, but larger item lists still benefit from larger samples (300+ vs. 200).
Items Per Set
Showing 5 items per set instead of 4 extracts slightly more information per task. This can modestly reduce sample requirements for a fixed number of items. The difference is small enough that it shouldn't drive your set size decision, but it's worth noting: a study with 20 items shown 5-per-set needs slightly fewer respondents than the same study shown 4-per-set.
Number of Sets Per Respondent
More sets per respondent means more data per person. If each item appears 4 times instead of 3, you're collecting 33% more data per respondent, which either improves precision at the same sample size or lets you hit the same precision with fewer people. The limit is respondent fatigue: past 18-20 sets, data quality degrades.
Anchored MaxDiff
Adding an anchoring question (e.g., "Would you actually want any of these features?") doesn't materially change sample size requirements. The anchoring data is analyzed separately from the MaxDiff scores, and the MaxDiff estimation itself remains the same.
MaxDiff vs Conjoint: Sample Comparison
| Method | Minimum Sample | Recommended Sample | Why |
|---|---|---|---|
| MaxDiff | 100 | 200-300 | Simpler tasks, more data per choice |
| CBC Conjoint | 200 | 300-500 | Complex profiles, more parameters to estimate |
| ACBC Conjoint | 100-150 | 200-300 | Adaptive design collects more per person |
MaxDiff's sample advantage comes from task simplicity. Choosing best/worst from 4 items is faster and more reliable than evaluating 4-attribute product profiles. Each MaxDiff task also produces two data points (best and worst) while a CBC task produces one (chosen profile). This double-data-point structure is part of why MaxDiff works well with moderate samples.
Right-Sizing Without Overspending
Start With Your Segments
Define segments first, then multiply. If you need to compare 3 segments, that's 200 x 3 = 600, not 200 total. If budget only allows 400, drop to 2 segments or accept that 3-segment comparisons will be directional only.
Add Tasks Before Adding Respondents
Going from 12 to 15 sets per respondent adds about 60 seconds of survey time but gives you 25% more data per person. That's cheaper than recruiting 25% more respondents. Don't exceed 18-20 sets.
Use Count Analysis for Quick Results
If your deadline is tight and you have 100-150 respondents, run count analysis instead of HB. You'll get a reliable aggregate ranking. Save HB for when you have 300+ and need segment or individual-level insights.
Consider the Cost-Precision Curve
Precision improves with the square root of sample size. Doubling from 200 to 400 improves precision by ~41%, not 100%. For most MaxDiff studies, the practical sweet spot is 250-350: large enough for HB estimation and basic segment comparisons, small enough to keep panel costs reasonable.
Frequently Asked Questions
Can I run MaxDiff with 50 respondents?
You can produce an aggregate ranking, and it'll be directionally correct for the top and bottom items. But the middle of the ranking will be unstable, and you won't be able to make segment comparisons or run HB estimation. For exploratory research or internal prioritization exercises, 50 is workable. For client-facing or product-decision research, aim for 200+.
How does MaxDiff sample size compare to a ranking question?
A ranking question (drag items from most to least important) requires similar sample sizes for aggregate results but produces ordinal data, not ratio-scaled data. MaxDiff's advantage isn't sample efficiency; it's data quality. At the same sample size, MaxDiff produces more discriminating, more reliable, and more analytically flexible output.
Does the number of items change the sample size I need?
Modestly. A 30-item study benefits from 300+ respondents because the item space is larger and individual items get fewer appearances per set. A 12-item study can produce stable results with 150-200. The relationship isn't dramatic, but it's worth planning for.
Related Guides
- MaxDiff Analysis: Complete Guide -- Full methodology overview
- How to Design a MaxDiff Survey -- Design decisions that affect sample needs
- Conjoint Analysis Sample Size Requirements -- Comparison method sample planning
- Sample Size Determination -- General principles for research sample sizing
- MaxDiff Sample Size Calculator -- Calculate your required sample
Plan your MaxDiff study -- try Quali-Fi free for 14 days.