Overview (CONJOINT command)

CONJOINT analyzes score or rank data from full-concept conjoint studies. A plan file that is generated by ORTHOPLAN or entered by the user describes the set of full concepts that are scored or ranked in terms of preference. A variety of continuous and discrete models is available to estimate utilities for each individual subject and for the group. Simulation estimates for concepts that are not rated can also be computed.

Options

Data Input. You can analyze data recorded as rankings of an ordered set of profiles (or cards) as the profile numbers arranged in rank order, or as preference scores of an ordered set of profiles.

Model Specification. You can specify how each factor is expected to be related to the scores or ranks.

Display Output. The output can include the analysis of the experimental data, results of simulation data, or both.

Writing an External File. An data file containing utility estimates and associated statistics for each subject can be written for use in further analyses or graphs.

Basic Specification

  • The basic specification is CONJOINT, a PLAN or DATA subcommand, and a SEQUENCE, RANK, or SCORE subcommand to describe the type of data.
  • CONJOINT requires two files: a plan file and a data file. If only the PLAN subcommand or the DATA subcommand—but not both—is specified, CONJOINT will read the file that is specified on the PLAN or DATA subcommand and use the active dataset as the other file.
  • By default, estimates are computed by using the DISCRETE model for all variables in the plan file (except those named STATUS_ and CARD_). Output includes Kendall’s tau and Pearson’s product-moment correlation coefficients measuring the relationship between predicted scores and actual scores. Significance levels for one-tailed tests are displayed.

Subcommand Order

  • Subcommands can appear in any order.

Syntax Rules

  • Multiple FACTORS subcommands are all executed. For all other subcommands, only the last occurrence is executed.

Operations

  • Both the plan and data files can be external IBM® SPSS® Statistics data files. In this case, CONJOINT can be used before an active dataset is defined.
  • The variable STATUS_ in the plan file must equal 0 for experimental profiles, 1 for holdout profiles, and 2 for simulation profiles. Holdout profiles are judged by the subjects but are not used when CONJOINT estimates utilities. Instead, these profiles are used as a check on the validity of the estimated utilities. Simulation profiles are factor-level combinations that are not rated by the subjects but are estimated by CONJOINT based on the ratings of the experimental profiles. If there is no STATUS_ variable, all profiles in the plan file are assumed to be experimental profiles.
  • All variables in the plan file except STATUS_ and CARD_ are used by CONJOINT as factors.
  • In addition to the estimates for each individual subject, average estimates for each split-file group that is identified in the data file are computed. The plan file cannot have a split-file structure.
  • Factors are tested for orthogonality by CONJOINT. If all of the factors are not orthogonal, a matrix of Cramér’s V statistics is displayed to describe the non-orthogonality.
  • When SEQUENCE or RANK data are used, CONJOINT internally reverses the ranking scale so that the computed coefficients are positive.
  • The plan file cannot be sorted or modified in any way after the data are collected, because the sequence of profiles in the plan file must match the sequence of values in the data file in a one-to-one correspondence. (CONJOINT uses the order of profiles as they appear in the plan file, not the value of CARD_, to determine profile order.) If RANK or SCORE is the data-recording method, the first response from the first subject in the data file is the rank or score of the first profile in the plan file. If SEQUENCE is the data-recording method, the first response from the first subject in the data file is the profile number (determined by the order of profiles in the plan file) of the most preferred profile.

Limitations

  • Factors must be numeric.
  • The plan file cannot contain missing values or case weights. In the active dataset, profiles with missing values on the SUBJECT variable are grouped together and averaged at the end. If any preference data (the ranks, scores, or profile numbers) are missing, that subject is skipped.
  • Factors must have at least two levels. The maximum number of levels for each factor is 99. Note that ORTHOPLAN will only produce plans with factors with 9 or fewer levels for each factor.