Analyze responses from a conjoint survey
To setup a conjoint study from scratch we need to determine the attributes and attributes levels that should be included. Once that has been done we would typically need to generate a fractional factorial design of conjoint profiles. This is a subset of all possible profiles that could be generated for the set of attributes and levels that were selected (see Design > DOE > Design of Experiments).
Once data is available from respondents their evaluations are analyzed to determine Part Worths (PW) and Importance Weights (IW).
To estimate a model select respondents ratings (or rankings) as the
Profile evaluations and select the profile
Attributes. Press the
Estimate button or
CMD-enter on mac) to generate results.
A respondent was presented with 18 product profiles for a carpet cleaning product described on five attributes in a conjoint study. To access the
carpet dataset go to Data > Manage, select
Examples from the
Load data of type dropdown, and press the
Load button. Then select the
Based on the attributes, 108 possible profiles could be created (i.e., 3x3x3x2x2 = 108). The respondent was given a set of 18 and was asked to rank the profiles from most preferred (rank 1) to least preferred (rank 18). The first five columns represent the five attributes and the last column is the respondent’s ranking.
The VIF scores displayed below indicate that the attributes are perfectly orthogonal. In the fractional factorial design profiles were deliberately selected such that all attributes are uncorrelated.
Multicollinearity diagnostics: design brand price seal money_back VIF 1 1 1 1 1 Rsq 0 0 0 0 0
We can confirm this results in three steps: (1) Create a new dataset with all 36 profiles in Data > Transform using
Expand grid, (2) Predict the utility for each of the profiles by selecting the newly created dataset in the Multivariate > Conjoint > Predict tab and storing the prediction in a new variable
predict_ca, (3) Sort the new dataset on
predict_ca in the Data > View tab. These three steps are shown in the screen shots below
If profile evaluations are available for multiple respondents and a respondent id variable is included in the dataset we can estimate conjoint results at the individual level by selecting the respondent id from the
By dropdown. We can then save the Partworths and/or Importance weights for all respondents to a new dataset in Radiant and use that for segmentation using Multivariate > K-clustering.
Add code to Report > Rmd to (re)create the analysis by clicking the icon on the bottom left of your screen or by pressing
ALT-enter on your keyboard.
If a plot was created it can be customized using
ggplot2 commands or with
patchwork. See example below and Data > Visualize for details.
plot(result, plots = c("pw", "iw"), custom = TRUE) %>% wrap_plots(plot_list, ncol = 2) + plot_annotation(title = "Conjoint Analysis")
For an overview of related R-functions used by Radiant to estimate a conjoint model see Multivariate > Conjoint
The key functions used in the
conjoint tool are
lm from the
stats package and
vif from the