[thelist] Sales and Visits

Luther, Ron Ron.Luther at hp.com
Tue Jun 1 15:39:45 CDT 2010

Fred Jones shared a chart at:

>>I have here a chart http://docs.google.com/View?id=dcpssp27_0gpcvjjck

And asked:

>>Of course this is a fairly small sample, but still, 
>>[the apparent lack of correlation] seems surprising, no?

Hi Fred,

No.  Not really.  Your model is using 7 datapoints and *1* variable! 

It would be a freaking miracle [1] if you got a decent fit from that!   ;-)

Seriously.  What I think your graph is actually telling you is that there 
are *more* factors (and/or more important factors) to explain sales 
volume than merely 'number of website visits'.  Looked at in those 
terms, I don't find it to be a very surprising result at all.

I think that's one area {and you never heard me say this!} where econometrics 
is a tad ahead of simple statistics.  The econometrician crowd look for likely 
explanatory factors, then search for data to represent those factors and build 
their model - and test - and rebuild - and retest - and refine.

>From a purely mathematical point of view I would suspect that an 8th degree 
polynomial based on number of visits might get you a decent fit to your data 
... but that wouldn't really explain anything or have predictive value now would 

So I would probably recommend:
(a) Going through the logs to see if you can determine how many visits, on 
average, were made by buyers prior to purchase.
(b) Phoning up and talking to a few {i.e. as many as you can} of the buyers 
to get a handle on the logic of their buying process and what (potentially 
modelable) factors went into their decision making.
(c) While talking to the purchasers be sure to ask if they had any trouble 
navigating your site and making their purchase.  E.g. Did they make multiple 
visits prior to purchase because your site was hard to use?  Try to remember that 
large customers may have a different process than smaller customers  and 
be sensitive to those kinds of differences.
(d) Here's a tricky part - also talk to some visitors who did _not_ purchase 
from you and see if you can get a handle on their decision making process.

Research can be frustrating,  It can also be fun.  Unfortunately, it often 
ends up being pretty complicated.  Articles and textbooks are pretty 
misleading in this regard - those are written long after the fact and the 
dead ends and errors are swept away in order to make for a better [and a 
heckuva lot shorter!] story.


[1] And, dreadfully sorry, but most likely a misleading one at that!

More information about the thelist mailing list