ideas for moving forward.

1) Remove fuel points and other problem transactions with large units.
What constitutes as “large”? I chose over “221” since there is a jump in the quantity to 400+ afterwards.

data sasuser.modified_transaction;
set sasuser.transaction_data;
if quantity > 221 then delete;

**Go back to the original and delete fuel points…try to identify other transactions and delete those as well.
**By doing this, will we have to go back and re-create our datasets? i.e. the total period counts data set (the data set in which we’ve done all of the model fitting?)

Yes, we’ll have to.

**How do we reconcile baskets without any quantity for products??

1b) remove cases with multiple campaigns at same time.
From the aggregated transaction table?
*I’m not exactly sure how to correctly do this…
**We must go back and make sure that the logic makes sense regarding the retain statement and seeing if it matches the campaign sequence. Start back with “” data set. Identify one or two customers to see whether the logic makes sense.

2) Test fixed effects vs. random effects approach using glimmix. (read the allison book)

General model:
y = µ + β +yz + α + e

Using Chapter 4 material found down below..

3) What does survival analysis book say about repeated measures.
The survival analysis book by Allison basically suggests using proc mixed in SAS to fit a random effects model or a simple fixed effects model. It also suggests employing “phreg”, but I’m not really sure if that applies here since we have not incorporated censoring of observations into our analysis. Could be something to look at if we were to try to go back to the aggregated transaction data set and censor these observations based off some arbitrarily chosen week number. (perhaps?) Maybe could treat campaign all the way up to the end of a time period as censoring…more on this later.

There is a bridge though between these count method...

Similar Essays