Q1 guidance

Hi all,

The project is pretty structured now, so I don't expect many issues at this stage. Kindly ensure your group has done the minimum steps required. The textbook examples and data are a good way to first practice implementation of any procedure on SPSS.

Spent some time a while back looking over one group's attempts with building a demand model using secondary data.

Some pointers:

1. Know the dimensions of your data matrices: How many observations/rows? How many columns in the X variables?

More # observations (or rows in your Y and X data matrices) is good. More # columns in your X matrices is not so good.

So use the secondary data sent to setup as many rows as you can get info for.

The group I looked at was planning to pick up the top 15 brand-models for one year and had collected some 20 X variables (brand dummies, size dummies and so on). Obviously the model gave nonsense results.

A good rule of thumb is that the # X rows should be at least 3x or 4x the # X columns! Else, inference is severely impaired.

2. Reduce the number of X variables. I suggested using a size proxy (like car length - available from brand/model/specs/dimesnsions at sites like carazoo.com) instead of using dummy variables for different size categories. This reduces the number of size related variables to 1 ratio-measured variable.

3. Be creative in the Xs you choose.
carazoo.com specs have plenty of info. Is mileage important? Is engine capacity (in litres) important enough to be included as an X? These are calls you have to take.

In general, its better to collect more data than less. One can always exclude variables that are there but not important. But how do you include variables that are not there?

End of the day, you want to have identified the major variables that account for a clear majority of Y variation. Look at R-square to judge fit and all.

4. Try out different specs - try using a quadratic term for some variable you believe follows a diminishing return or ideal point pattern. Try using the log-log specification. Compare model fit using adjusted R-square. And select the best fitting and most intuitive demand model.

5. See if the demand model fits into the recommendations: For a given size of car type, for a given price range that you have in mind, you can compute predicted sales. Now this may or may not help in the recommendation you make to the client. But do check and see if it helps.

Hope that helped.

Sudhir