README

🛠️ The One-Stop Conjoint Shop: projoint

Conjoint survey designs are spreading across the social sciences due to their unusual capacity to estimate many quantities of interest from a single randomized experiment. Projoint is general-purpose software for the design, implementation, and analysis of conjoint surveys. Its easy-to-use components include a drag-and-drop web-based application for survey design and an R package for analysis.

We created this software for our AJPS article (“Correcting Measurement Error Bias in Conjoint Survey Experiments”), which offers easier and more powerful ways of designing and analyzing conjoint experiments – including everything from writing Qualtrics survey code for you to preparing figures for your paper or report. It also includes easy measurement error corrections, which are essential because conjoint’s tendency to mirror complicated real-world choices generates substantial measurement error and, without corrections, considerable bias.

We intend for Projoint software to be the easiest and fastest way to design, field, and analyze conjoint surveys.^ⓘ Take the 6 steps below and you should be all set.

🚀 Six Easy Steps

1. Install the development version from GitHub

Open R (or install R if you do not have it), and run the following command in your coding environment.

devtools::install_github("yhoriuchi/projoint")

2. Design your survey

Online surveys are frequently written with an online software called Qualtrics. Using our web tool, called the Projoint Survey Designer, you don’t need to learn how to write a survey in Qualtrics.

Use the Projoint Survey Designer and export surveys formatted for Qualtrics.
Follow the step-by-step guide to learn how to set up your Qualtrics survey.

3. Field your survey

Using the .QSF file export from the Projoint Survey Designer, load your survey into Qualtrics.
- Log into your Qualtrics account.
- Click “Create a new project”
- Under “From scratch” select “Survey” and then “Get started”
- Enter a name and under “How do you want to start your survey” select “Import a QSF file”
- Click “Choose file” and select your .QSF file.
- Click “Create project”
You are free to field your Qualtrics survey through online vendors.
When you are done fielding your survey, you will now need to export your data from Qualtrics to R.
- Click “Download Data”.
- Choose CSV format.
- Critically, select “Use choice text” rather than coded values.

4. Read the results into R and wrangle your data into structured form

Load your survey responses into R:

library(projoint)
dat <- read_Qualtrics("your_file.csv")

Prepare the data for analysis:

dat <- reshape_projoint(
  .dataframe = dat,
  .outcomes = c(paste0("choice", 1:8), "choice1_repeated_flipped")
)

Follow the step-by-step guide to learn how to read and reshape data for conjoint analysis.

5. Analyze and visualize important Quantities of Interest

Estimate Marginal Means (MMs) or Average Marginal Component Effects (AMCEs) with correction for measurement error:

Note: The following example illustrates a profile_level analysis. This approach is common in social science and useful as an initial diagnostic, but we encourage researchers to consider choice_level analysis (the default for .structure). See Choice-Level Analysis. Detailed steps of analysis appear in the step-by-step guide.

output <- projoint(out1_arranged, .structure = "profile_level")
print(output)
summary(output)

Visualize your results easily:

plot(output)

Estimate additional quantities of interest and explore subgroup comparisons using choice-level analysis.
Follow the step-by-step guide to learn how to:
- Estimate and correct marginal means (MMs) or average marginal component effects (AMCEs), including predicting IRR if necessary.
- Visualize the marginal means (MMs) or average marginal component effects (AMCEs).

6. Explore additional quantities and compare subgroups

Go beyond standard profile-level summaries.
Use choice-level analysis to:
- Directly compare trade-offs (e.g., low housing cost vs. low crime).
- Collapse multiple levels (e.g., city vs. suburban preferences).
- Estimate subgroup differences (e.g., Democrats vs. Republicans).
See detailed examples in the Explore and Compare Further vignette.

📦 Additional Info

Upcoming Features

Weighted estimation for features and respondents
Support for non-binary outcomes (ratings, rankings)

Comments, questions, or suggestions?

Please check issues already filed at our GitHub repository. If you don’t see an answer to your question, click on the green “new issue” button.

✨ Thank you for using projoint! ✨