Hierarchical Bayes Conjoint

Hierarchical Bayes conjoint analysis estimates individual-level part-worth utilities from conjoint choice or rating data by combining each respondent's answers with the distribution of preferences across the sample.

계층적 베이즈 컨조인트 분석은 각 응답자의 응답과 표본 전체의 선호 분포를 결합하여 컨조인트 선택 또는 평정 자료로부터 개인 수준의 부분가치 효용을 추정하는 방법이다.

How do I estimate each respondent's preferences for product attributes from a conjoint survey?

컨조인트 조사에서 응답자별 속성 선호를 어떻게 추정하나요?

Hierarchical Bayes estimation fits a choice model in which each respondent has their own attribute utilities drawn from a population distribution, so individual estimates borrow strength from the whole sample. It fits choice-based conjoint studies where each person answers too few tasks for separate models but individual-level preferences are needed for segmentation or market simulation.

When to use Hierarchical Bayes Conjoint

• Choice-based conjoint data with several tasks per respondent
• Individual-level utilities needed for segmentation or market share simulation
• Calculating attribute importance and willingness to pay by respondent

When to use something else

• Only average preferences for the whole sample are needed: use a multinomial logit model
• Preferences are thought to form a few discrete segments: use latent class conjoint analysis

Assumptions and requirements

• A well-designed experimental choice design with balanced attribute levels
• Enough choice tasks per respondent
• Convergence of the Markov chain Monte Carlo sampler
• Respondents answer attentively and consistently

What the analysis reports

• Individual-level part-worth utilities
• Mean utilities and their distribution across respondents
• Relative attribute importance
• Root likelihood or hit rate on holdout tasks
• Market share simulations

How to report it

Hierarchical Bayes estimation showed that [attribute] was the most important attribute (mean relative importance = [value]%), with a holdout hit rate of [value]%.

Related methods

• Discrete Choice Experiment (DCE)
• Multinomial Logit Model
• Latent Class Conjoint Analysis

Run Hierarchical Bayes Conjoint on Quorant