MSCI581. Workshop 3. Conjoint analysis
Ivan Svetunkov
Centre for Marketing Analytics and Forecasting, Lancaster University
Abstract
This time we will see how to carry out the conjoint analysis and then analyse the collected data. You will learn how to set up the questions for the conjoint analysis, how to prepare the cards and how to construct models for individuals and for a group of respondents.
Contents
Basics of Conjoint Analysis Conjoint Analysis for a sample
2 3
1
Figure 1: An example of the table for the conjoint analysis
Figure 2: An example of the regression model for your data
The first part of workshop will be done in Excel it will allow us collecting the data, and after that we will switch to R.
Basics of Conjoint Analysis
In this part of the workshop we will decide, what product to analyse, what atributes to include and what should be the levels of those attributes. This is done in class together with your tutors. After doing that, we will create 9 products with orthogonal attributes and you will need to rate them on the scale from 0 to 100.
Having done that you will need to open Excel and create a table with the attributes, options and your rating for the products, something like this see Figure 1.
After doing that, you will need to create a new table with the dummy variables instead of attributes and options.
Question 1. How many dummy variables will you need for each attribute? How many dummy variables will you have in your model?
The model that we want to construct is a basic linear one with rating as the response variable and all the others as the explanatory ones.
After that, please, use Data Analysis Regression and construct the regression with the dummy variables, specifying rating in the Y Range. You should have something like this see Figure 2, but note that the numbers will be different.
Question 2. What are the most important options of the attributes for you? Is any of the variables significant in your model? Why?
BASICS OF CONJOINT ANALYSIS
2
Based on that output, please, create a table with the relative importance of the attributes, similar to the one that we had in the lecture.
Question 3. What is the most important attribute of the product for you? Conjoint Analysis for a sample
Now that you have done the basic analysis based on your specific preferences, we will move to a more complex one, based on the sample. In order to do that, please, go to Moodle and find Data collection for the Workshop 3 form. Fill it in using your ratings for the products and submit. After that your tutors will collect the results and provide you with the data in csv format. This is the sample of responses of your classmates. The sample will contain the ratings in the first column, the attributes in the 2 4 columns and the id of the respondent in the last column. Import this data in MS Excel and transform the attributes and id into factors.
Hint. You should use factor function in R for that.
We will use greybox package for our purposes and the alm function, which allows constructing linear regression models.
Using your data, construct the regression of Rating from all the attributes. The command in R should have the following form please, substitute Attribute1, Attribute2 and Attribute3 with the respective names of variables in the imported data and TableExample with the name of your imported data frame:
almModel1 almRatingAttribute1Attribute2Attribute3, TableExample In order to see the summary statistics of the model, please use the code:
summaryalmModel1
Question 4. What can you say about the constructed model? Which of the options of attributes is the most important?
The parameters of this model can be extracted using the function coef: coefalmModel1
If you want to calculate the relative importance, then you would need to create several variables for attribute parameters. This can be done using the functions coef, diff and range in R, like this assuming that we have 2 options per attribute, 3 attributes and an intercept. Otherwise the values in the square brackets will be different:
Based on that we can calculate the relative importance of attributes for our sample:
Question 5. Which of the attributes is the most important for the respondents in your sample? Exercise 1. Build a new model, including now the id of respondents, and repeat the analysis.
Hint. This is a crude way of taking the potential differences in perception of the respondents into account. The better option would be to do a mixed effects model, but this is out of the scope of our course.
Question 6. Do the results change in the new model in comparison with the old one?
CONJOINT ANALYSIS FOR A SAMPLE
Attribute1Range diffrangec0,coeftest2:3 Attribute2Range diffrangec0,coeftest4:5 Attribute3Range diffrangec0,coeftest6:7
Attribute1Range sumAttribute1Range,Attribute2Range,Attribute3Range Attribute2Range sumAttribute1Range,Attribute2Range,Attribute3Range Attribute3Range sumAttribute1Range,Attribute2Range,Attribute3Range
3