Conjoint Companion

This app accompanies the workflow introduced in Test-Retest Reliability in Metric Conjoint Experiments: A New Workflow to Evaluate Confidence in Model Results . It helps researchers prepare conjoint designs and evaluate response consistency in metric conjoint experiments.

What You Can Do

  • Generate full and fractional factorial designs for conjoint studies.
  • Inspect two-way interaction estimability for two-level fractional designs.
  • Inspect pairwise coverage and balance for N-level and mixed-level designs.
  • Upload conjoint response data and run the test-retest reliability workflow from the paper.
  • Review reliability tables, regression diagnostics, and plots.
  • Download sample data and analysis results.

How To Use The App

  1. Use the factorial design tabs if you still need to construct a conjoint design.
  2. Use the test-retest reliability tab when you have data from an initial and replication round.
  3. Start with the bundled demo data if you want to inspect the required structure first.
  4. Validate your uploaded data before running the analysis.

Data Handling

Uploads are processed only for the current session. The app accepts CSV and one-sheet XLSX files, applies a 5 MB upload limit, and rejects reliability datasets with more than 25,000 rows. Generated result files are stored in a session-specific temporary folder and removed when the session ends.

Privacy

Uploaded files and generated outputs are processed in a session-specific temporary folder. They are removed when the session ends, and reset clears the current session files manually. The app does not intentionally store uploaded datasets or track user behavior.

Paper, Data, And Authors

Please review your results carefully. This app supports the workflow described in the paper, but it does not replace substantive judgment about your conjoint design, measurement strategy, or model specification.

R Shiny app written by Jens Schüler.

Factorial Designs

Factorial designs help researchers reduce the number of profiles shown in conjoint studies while preserving the information needed to estimate the intended effects. This app provides a graphical interface to established R tools for creating full and fractional designs.

What The App Provides

  • A design table with generated profile combinations.
  • For two-level designs, an interaction estimability table that shows which two-way interactions work separately and which are confounded.
  • For N-level and mixed-level designs, an interaction coverage table that checks pairwise level-combination coverage and balance.
  • Access to design-generation functionality from FrF2 and DoE.base .

Two-Level Designs

  • Use this tab when all attributes have two levels, such as low/high or absent/present.
  • Choose the number of attributes, then select a full or fractional design.
  • For fractional designs, choose resolution III for main-effects-focused designs.
  • Choose resolution IV when main effects should be clear from two-way interactions, but some two-way interactions may be confounded with each other.
  • Choose resolution V when two-way interactions should be clear from main effects and from other two-way interactions, assuming higher-order interactions are negligible.
  • Use the interaction estimability table before interpreting two-way interactions.

N-Level And Mixed-Level Designs

  • Use this tab when at least one attribute has more than two levels.
  • Enter the number of levels for each attribute separated by commas, for example 4,4,4 or 2,4,4,3.
  • The app currently supports up to 7 attributes and up to 4 levels per attribute.
  • Full designs use all combinations. Fractional designs search mixed-level orthogonal arrays through DoE.base.
  • The app reports the full factorial size, the generated number of profiles, and whether a selected fractional design actually reduced the number of profiles.
  • The interaction coverage table reports pairwise level-combination coverage and balance. It is not an aliasing or confounding diagnostic.
  • The search for suitable fractional mixed-level designs can take time for some combinations, and some settings may still produce a full-factorial-sized design.

Method Notes

  • Full factorial designs are generated with fac.design().
  • Two-level fractional designs are generated with FrF2().
  • Mixed-level fractional designs use oa.design() and related orthogonal-array search utilities.
  • Classical Resolution III/IV/V applies to regular two-level fractional factorial designs. The N-level section uses an orthogonal-array criterion and pairwise coverage diagnostics instead.

Further Reading

Important: Always evaluate whether a generated design is suitable for your theoretical model before collecting data.

Test-Retest Reliability Workflow

This section implements the workflow from Test-Retest Reliability in Metric Conjoint Experiments: A New Workflow to Evaluate Confidence in Model Results . The workflow is designed to help researchers evaluate response consistency for manipulated conjoint attributes before interpreting the substantive model results.

Before You Upload

  • Use CSV or one-sheet XLSX files only.
  • Keep uploads below 5 MB.
  • Reliability datasets must contain no more than 25,000 rows.
  • Use long-format data with one row per respondent, round, and profile observation.
  • Use the bundled demo data if you want to inspect the required structure first.

Required Columns

  • respondent identifies the respondent.
  • round identifies the initial round as 1 and the replication round as 2.
  • profile identifies the conjoint profile.
  • dv contains the dependent variable. Analyze one dependent variable at a time.
  • att_1 , att_2 , ..., att_x contain the manipulated attributes. At least two attributes are required.

round , profile , dv , and all att_ columns must be numeric or cleanly coercible to numeric. Missing values should be left empty or coded as NA .

Workflow

  1. Upload your CSV or XLSX file.
  2. Click the validate button to check the file structure.
  3. Inspect the table or variable types if needed.
  4. Run the analysis to compute the reliability results.
  5. Download the Excel workbook or CSV archive if you need local copies of the result tables.
  6. Use reset to clear the current upload and generated session files.

What The Workflow Covers

  • Profile-level Pearson correlations.
  • ICC(3,k) reliability estimates.
  • Slope-difference checks between initial and replication rounds.
  • A pooled regression model with clustered standard errors.
  • Plots for response deviations, ICC summaries, and slope differences.

The workflow focuses on level-1 manipulated conjoint attributes and outcomes. Measured level-2 respondent variables are not part of this app's reliability workflow.