🔬 Research
Choice Based Conjoint Study Designer: Attribute and Level Audit, Price Level Spacing, Tasks and Concepts per Screen, Dual Response None, Design Efficiency Test, Sample Size Rule of Thumb, HB Estimation, and Simulator Plan
Design a choice based conjoint study that can answer a real pricing or packaging question: audit attributes and levels for overlap and ambiguity, space price levels across a believable range, choose tasks and concepts per screen, add a dual response none and holdout tasks, test design efficiency before fielding, size the sample with a stated rule of thumb, and plan HB estimation, quality checks, and a share simulator.
0Reviews
Prompt
Act as a market research methodologist who designs choice based conjoint (CBC) studies in Sawtooth Software (Discover and Lighthouse Studio) and Conjointly, and has had to explain away results from studies with overlapping attributes, a price range nobody would pay, and a sample too small for the segments the client wanted. Inputs: - The business decision the study must inform: [DecisionQuestion] - Draft attributes and levels, including price: [AttributeDraft] - Target respondents, screener criteria, and segments to compare: [Population] - Conjoint platform and panel source: [Platform] - Respondent budget and interview length limit in minutes: [SampleBudget] - Current or competitor prices in the market, if known: [PriceContext] - Output format: [Format] Generate: 1. A decision to design map: what each attribute lets the client decide, and any attribute that maps to no decision, flagged for removal. 2. An attribute and level audit of AttributeDraft: each attribute independent of the others, levels mutually exclusive and concrete (no "good" or "premium" without a definition), about 6 attributes or fewer for full profile tasks, and 2 to 5 levels each. Note the number of levels effect when one attribute has many more levels than the rest. 3. Price levels: evenly spaced across a range anchored to PriceContext, wide enough to show a slope and narrow enough to stay believable, with a note on whether price will be estimated per level or as a linear term. 4. Prohibitions: list any impossible combinations, keep them to the minimum, and explain the efficiency cost of each. 5. Task design: tasks per respondent, concepts per task, a dual response none question, and 2 fixed holdout tasks for validation, fitted to SampleBudget's interview length. 6. A design efficiency test before fielding: run the platform's test design with simulated respondents and review standard errors for each level, using the platform's guidance as the threshold. 7. Sample size: apply the Johnson and Orme rule of thumb n x t x a / c >= 500 (respondents, tasks, alternatives per task, largest number of levels on any attribute), show the floor, then raise it so each segment in Population has enough respondents, stated as a team choice. 8. Data quality: speeders, straightliners, and a fit statistic such as RLH from HB estimation compared with the chance level for the number of concepts. 9. An analysis plan: HB multinomial logit utilities, attribute importances with their caveats, a market simulator for share of preference, and willingness to pay reported only from simulations, never by dividing utilities. Constraints: - Never invent utilities, shares, or results; the output is a design. - Thresholds are labeled as guidelines or team choices. No em dashes.
Instructions
Replace every [bracket] with your details before running. Works on ChatGPT, Claude, and Gemini.
Generated Output
This image was generated using the prompt above.
Examples
Example Input
DecisionQuestion: should our dental clinic group launch an in house membership plan for patients without insurance, at what monthly fee, and what to include AttributeDraft: monthly fee ($19, $29, $39, $49); cleanings per year (1, 2); discount on other treatment (10%, 20%, 30%); emergency exam (not included, 1 per year included); commitment (cancel anytime, 12 month minimum); "premium service" (yes, no) Population: adults 25 to 64 without dental insurance within 20 miles of our 6 clinics; compare parents of children under 18 vs non parents Platform: Sawtooth Discover, online panel SampleBudget: up to 400 completes, 15 minute interview PriceContext: local plans we found range from $25 to $45 per month Format: design memo with tables
Example Output
CBC DESIGN MEMO: DENTAL MEMBERSHIP PLAN 1. DECISION MAP | Attribute | Decision it informs | | Monthly fee | price point | | Cleanings per year | core inclusion | | Treatment discount | discount tier | | Emergency exam | add or leave out | | Commitment | cancel policy | | "Premium service" | none: undefined, remove | 2. ATTRIBUTE AND LEVEL AUDIT - Remove "premium service". Respondents will each imagine something different. If it means same week appointments, test that as its own attribute: appointment wait (within 2 weeks, within 3 days). - Cleanings "1" vs "2": write as "1 cleaning and exam per year" and "2 cleanings and exams per year". - Final: 6 attributes. Fee has 4 levels, discount 3, the rest 2. A mild number of levels effect on fee is possible; accept it because 4 price points are needed to see the curve, and note it in the report. 3. PRICE LEVELS $19, $29, $39, $49: evenly spaced $10 steps around the $25 to $45 local range, with one level below and one above. Estimate per level first; if the part worths fall in a near straight line, refit price as a linear term for smoother simulations. 4. PROHIBITIONS None needed. Every combination can be offered. Avoid adding any, since each one lowers design efficiency. 5. TASK DESIGN - 10 random tasks + 2 fixed holdouts = 12 screens, fits about 8 minutes of a 15 minute interview. - 3 concepts per task. - Dual response none after each task: "Would you actually join this plan?" yes or no. - Holdouts: H1 $29, 2 cleanings, 20%, emergency included, cancel anytime vs two realistic alternatives; H2 mirrors your launch candidate. 6. DESIGN TEST Run Discover's design test with simulated respondents at n = 400. Review standard errors for every level and follow Sawtooth's guidance on acceptable size for main effects. If any level is weak, add tasks before adding respondents. 7. SAMPLE SIZE Rule of thumb: n x t x a / c >= 500 c = 4 (fee), t = 10, a = 3 n >= 500 x 4 / (10 x 3) = 66.7, so the floor is 67. That floor is for the total sample. You want to compare parents vs non parents, so plan at least 150 per segment. Team choice: 400 completes, quota of 200 parents and 200 non parents. 8. DATA QUALITY - Remove speeders under 40% of the median interview time. - Flag respondents who pick the same position in every task. - HB fit: RLH for chance with 3 concepts is 0.33. Review respondents near chance before removal and report how many were removed. 9. ANALYSIS PLAN - HB multinomial logit, utilities per respondent, zero centered. - Importances reported with the caveat that they depend on the level ranges tested. - Simulator: share of preference including the none option, test 3 to 4 plan scenarios against holdout hit rates first. - Willingness to pay: compute only by simulation (raise price until share matches the base plan), never by dividing utilities. - Segment results: parents vs non parents, with the n for each shown on every chart. No results are shown here; all utilities and shares come from fieldwork.