conjoint-design

What it does

Defines a Conjoint Design Expert that guides a user from attribute architecture to power calculations, then to treatment validation and estimation of effects (AMCE/AMIE). It includes concrete steps for structuring attributes (orthogonality, randomization at the respondent level, and D-optimal considerations), power logic (N_eff, SE formulas, MDE, Type S/M errors, and tool-based simulations), validation practices (DGP identification, realism checks, attention/processing validation, and pretests), and estimation framing (reference categories, SESOI, AMCE interpretation, and marginal means). It explicitly references a workflow example and enumerates the tools and packages to use for analysis and simulation (cjpowR, DeclareDesign, FindIt, cjbart, cjRank, CRTConjoint, Docker-based adaptive designs).

How it works

• Attribute Architecture: ensure attributes are independent, randomize order at the respondent level, consider D-optimal or constrained designs, monitor attribute density to manage fatigue, allow nested or restricted randomization under justified conditions, and set medium-level attribute descriptions.
• Statistical Power and Error Logic: compute effective N as Respondents × Tasks × Profiles; use the closed-form SE for AMCE with SE = sqrt(Var(Y) × L / N_eff); plan for larger samples when estimating interactions; reference typical AMCE benchmarks (0.02–0.10) and MDE guidance; guard against Type S/M errors; apply a rule of thumb for Low-N_eff; account for multiple testing with specified corrections; advise using cjpowR or DeclareDesign for power and simulations; and apply compromises when the respondent pool is fixed.
• Treatment Validation and Realism: define the DGP before attributes, distinguish experimental vs. mundane realism, audit attention and salience, ensure information availability and processing, beware Names-as-Cues, consider pretreatment mock vignettes, and plan a repeated task to estimate IRR (which enables bias-corrected AMCEs via projoint).
• Estimating Effects: specify a baseline level, define SESOI for AMCEs, frame AMCE as the change in probability of selection when an attribute changes from reference to the level of interest, discuss interpretation boundaries (not a direct share of preferences), and consider Marginal Means alongside AMCEs.

When to use it

Use when designing and analyzing conjoint experiments to ensure proper attribute construction, power planning, and reporting of AMCEs/MMs, with explicit tools and workflows for simulation, estimation, and robustness checks.

What it can touch

• Tools: cjpowR R package, associated Shiny app, DeclareDesign framework, FindIt, cjbart, cjRank, CRTConjoint, Docker containers for adaptive designs, and replication scaffolding via specific GitHub paths. It lists standard survey platforms and emphasizes analyzing with these packages.

Caveats

• The guidance references licensed or unlicensed tools and assumes access to specified packages and Docker images; no explicit licensing text beyond NOASSERTION is provided in the facts. It emphasizes statistical concepts, assumptions, and interpretations that require careful pre-specification and corrections for multiple testing. No explicit outcomes are promised.