Research & Evidence

Search for a method

Search titles, questions, territories and MSC identifiers.

    ← All methods
    MSC-H-004Customer scienceScientific territory

    Customer and choice science: behavior, value and heterogeneity

    This territory models choice, retention, future value and segments without confusing prediction, causality and description. Observation time and heterogeneity are explicit.

    Scientific editorial team: Marketing Science Center

    Direct answer

    Estimate a probability, expected value or segmentation with out-of-sample validation.

    This territory models choice, retention, future value and segments without confusing prediction, causality and description. Observation time and heterogeneity are explicit.

    Model or deliverableQuestion → design → estimand → uncertainty → decision boundary

    Fader, Hardie & Lee, 2005McFadden, 1974

    ESTIMAND

    Estimation profile

    Unit of analysis
    Decision × population × horizon
    Exact estimand
    A defensible chain from question to evidence
    Model or deliverable
    Question → design → estimand → uncertainty → decision boundary

    01—10

    Territory pathway

    01 · MSC-P-026Logit vs Probit: how do you choose for purchase probability?→02 · MSC-P-028How do you estimate CLV with BG/NBD and Gamma-Gamma?→03 · MSC-P-029Which customers have the highest probability of churn?→04 · MSC-P-030How do you analyze retention with a survival model?→05 · MSC-P-031How do you build a useful customer segmentation?→06 · MSC-P-032How do you test segmentation stability?→07 · MSC-P-043What is a subscriber worth with only six months of retention data?→08 · MSC-P-044What is a customer who buys at most once a season worth?→09 · MSC-P-045What are a firm’s current and future customers worth?→10 · MSC-P-046How much will a customer spend per purchase? The Gamma-Gamma model→

    01—08

    Verifiable analysis framework

    1. 01
      Question

      Customer and choice science: behavior, value and heterogeneity

    2. 02
      Estimand

      A defensible chain from question to evidence · Decision × population × horizon

    3. 03
      Data

      population · unit_of_analysis · decision_horizon · outcome · assumptions

    4. 04
      Model or deliverable

      Question → design → estimand → uncertainty → decision boundary

    5. 05
      Declared calculation

      Question → design → estimand → uncertainty → decision boundary → A defensible chain from question to evidence

    6. 06
      Uncertainty checks

      Unique decision intent · Explicit identification assumptions · Reproducible evidence trail

    7. 07
      Method-specific validation

      Check data, estimate stability and interpretation limits. Unique decision intent · Explicit identification assumptions · Reproducible evidence trail

    8. 08
      Limitations

      Present a predictive score as proof that an action will change behavior.

    Required variables

    • population
    • unit_of_analysis
    • decision_horizon
    • outcome
    • assumptions

    Checks

    • Unique decision intent
    • Explicit identification assumptions
    • Reproducible evidence trail
    Possible decision

    Estimate a probability, expected value or segmentation with out-of-sample validation.

    Unsupported decision

    Present a predictive score as proof that an action will change behavior.

    Scientific sources

    2 sources

    1. Fader, Hardie & Lee, 2005
    2. McFadden, 1974

    Read next

    MSC-P-026Logit vs Probit: how do you choose for purchase probability?→MSC-P-028How do you estimate CLV with BG/NBD and Gamma-Gamma?→MSC-P-029Which customers have the highest probability of churn?→MSC-P-030How do you analyze retention with a survival model?→MSC-P-031How do you build a useful customer segmentation?→MSC-P-032How do you test segmentation stability?→