Having spent more than a year and a half struggling on my own as an independant IT-consultant specializing in web analytics (I was reseller for this product), it never ceased to amaze me how poorly the innocent and brainwashed folks understood this wonderful world of measuring things.
So much magical hocus-pocus really. Ironically enough, if only you could truly understand what it all 'really' meant, you would fork in alot more profit than you ever thought was possible.
For awhile folks believed that just by collecting tons and tons of irrelevant information from megabytes of web logs, they could somehow make far-reaching conclusions based on magnificent graphs that were made on the fly.
Nothing could be further from the truth. It is all very personal really, a correlation of trends and how these reflect your ongoing business ups-and-downs.
You can collect data, generate a series of standard and often complicated graphs, see how (you think) the oscillating curves match your expectations, ad infinitum.
Sorry sir, mister CEO sir, but this is completely backwards!
The correct methodology is to decide 'first' how your business success can best be measured, meaning that the relevant metrics are defined 'before' the analysis is to take place. Using these carefully thought-out metrics, one can then properly filter through and collect the correct data which provides the most meaningful information.
For example, what is more important: total number of hits or page visits? Do you know the difference? Is it more important to have visitors meandering around your site or should they be spending more time in certain areas? Are visitors migrating to the correct regions of your web site, and how long is it expected for them to get there? How do you define a visitor? Is it more important to have more visitors hitting landing page or are you more concerned with the conversion rates? What is a conversion rate?
You might not care at all about the quality of the visitors or how long they stay, only the bare numbers that came and went during a given period. For example, when deciding how effective a given advertisement campaign was.
Then again, it may be more important that (potential) customers reacting to a given special offering enter via the correct landing page and progress in an orderly fashion from point A to point B and create an order by hitting the ok-button. So why is there such an increase in the fall-out rate between steps 3 and 4 resulting in much fewer conversions than hoped for?
Those are just a bunch of questions not all of which have one correct answer.
LTNS
>>Sorry sir, mister CEO sir, but this is completely backwards!
So what about backward propagation networks, where you look at the output and try to discover which input parameters which would have been significant.
Both sides of the coin are important in statistical analysis.