results coach available?

[Maguin, Eugene]

Michael,

Your problem sounds very similar to that of analyzing observational data
where each person's ongoing behavior is coded into a set of discrete
categories. I'm assuming that when you describe a person's last exposures to
a product, that exposure list is actually time ordered such that the first
item listed is the first viewed and the last is the last viewed. If this is
truly so, then methods developed for observational data might be applicable
to your problem. That said, it'd be helpful to understand what you want to
learn from your dataset. Without that there is simply thousands of things
you might focus on, none more important than the next.

Gene Maguin

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[ViAnn Beadle]

What are you trying to test or predict with this? Are you looking for
product associations here (AKA "market basket analysis").

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Pearmain, Michael
Sent: Friday, March 07, 2008 3:09 AM
To: [hidden email]
Subject: Path analysis help or ideas

Hi All,

I've recently been given a data file that contains transactional data, and i
was looking for some advice on the best way to analyse the data.

The data has a revenue value attached to it and has the individuals last
(upto)10 exposures to a set of five products before they bought.
Products may be:
APPLE =a
BANANA =b
GRAPES =g
ORANGE =o
PEAR =p

so an individuals stream may well something similar to
REVENUE - STREAM
23 - o,a,a,a,p,g,b
45 - o,o,o
24 - b,b,b,g,g,a,o,o,p,p

At the minute i've created lots of frequencies and ran an ANOVA on the last
exposure before the transaction, but ideally i'd like to look at the stream
the individual made, the only way i can see to do this (crude?) is to create
a huge set of nested loops (via python for elegance and ease) and run a
count on those Vs revenue to find the mean for each path.

At some later stage i might get the stream of those who don't make a
transaction, at which point i intent to build a Bayesian network, but until
that point does anyone have any suggestions on other types of analysis i
could do specifically towards the path?

=======
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

[Pearmain, Michael]

Hi Gene,

It's exactly as you have suggested, and indeed they are time ordered, the end goal of this research is to identify exposure patterns and link them with revenue, this way i can build some kind of ROI based on the exposure history and build what if scenarios after that event at any point on exposure path, giving some kind of indication of when maximum yeild for a particular product diminishes, whilst thinking about the number of products that have been seen at any exposure, so it takes the size of the sample into account

Does this make sense?

Mike




-----Original Message-----
From: SPSSX(r) Discussion on behalf of Gene Maguin
Sent: Fri 3/7/2008 10:09 AM
To: [hidden email]
Subject:      Re: Path analysis help or ideas
 
Michael,

Your problem sounds very similar to that of analyzing observational data
where each person's ongoing behavior is coded into a set of discrete
categories. I'm assuming that when you describe a person's last exposures to
a product, that exposure list is actually time ordered such that the first
item listed is the first viewed and the last is the last viewed. If this is
truly so, then methods developed for observational data might be applicable
to your problem. That said, it'd be helpful to understand what you want to
learn from your dataset. Without that there is simply thousands of things
you might focus on, none more important than the next.

Gene Maguin

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

====================To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD