How to test which combination of product attributes generates better sales performance?

Question

I am looking for help in identifying what is the best statistical approach to address the following market research question:

(1) I have a dataset from an online website that helps users rent their vacation homes.

For each vacation home I have information on:

• Their attributes (predictor variables): location (city and distance from city center), # of rooms, guest rating (10 stars),price, etc.
• Their sales performance (dependent variables): conversion rate, # of nights sold, sales volume in $

(2) The website rents vacation homes mainly for US and Canada customers

Question: I would like to be able to identify which combinations of predictor variables are associated with better performance, in order to be more targeted in gathering new homes for the site. Because most of the website users (renters) originate from the US and Canada (and I have a sense that they have different preferences) I would be especially interested in understanding which combinations of attributes work best for each country?

I imagine that for Canadian customers you could have 3 combinations of attributes that would work best. For instance, one combination could be: price $50-$75/day, 2 rooms, far from the city center, etc...

Any help would be really appreciated.

Answer 1

There are a number of techniques you can use, but I always have a tendency to lean towards using Classification and Regression trees (rpart in R). They are simple yet robust as it can handle non-linear data and yields trees which are intuitive.

It's worth noting that the caret package mentioned above is a very powerful package to compare a number of models once you have the general structure of your data identified and think that is a great R library as well.

Answer 2

I would start with regressing (using OLS) the conversion rate and/or sales volume on the attributes separately for US and Canada (non-numeric attributes would have to be converted to binary flags first). Those variables that have highest absolute value of t-value (ratio of parameter value to its standard error) are the most important ones. One thing to watch out for is multi-collinearity (remove variables that have largest Variance Inflation Factor one by one until multi-collinearity is under control)

Ordinary Least Squares (OLS) regression comes with many tools including Excel where it is known as "Multiple Regression".

One could use more complex techniques as well such as decision trees etc. R package caret provides a convenient framework for that (but can have a steep learning curve if you are not familiar with R)