Last week after my Twitter Predictive dump, I ended up chatting with a few people about the whole thing. I'm super glad to be able to start having these conversations even if it only helps me realize how little I know about what I'm trying to do :)
One of the conversations that came up was talk about regression analysis and email trends.
Still Learning
First it's worth noting that the Markov chain I've been toying with has very close ties with Bayesian inference model which is in fact are a type of regression...at least according to Wikipedia. Yes, snicker if you must, but I'm catching up on a lot of my Math on Wikipedia while I try and locate the right resources for what I'm attempting. (I did get a Math minor at A&M, but unfortunately none of the courses I remember covered this type of modeling and it's far outside the realm of any books I kept.) I did locate some courses on iTunesU that I'll checkout this week, but if you have any resources, let's have'em.
No Email Wednesdays
Now when it comes to email, apparently Wednesday is the worst day to send it. It doesn't get opened on Wednesday. On top of that each day has it's own twists and turns that correlate to open rates.
Now one way to look at this that each day should be part of any predictive model having to do with email. (And while we're at it twitter...or any other communication based activity.) We can help determine what tomorrow's activity will look like by factoring in what day tomorrow will be.
In this case, the day of the week becomes one of many environmental variables to include in the model. I'm not against this, and I'm sure a mature model does include many environment variables to help predictions. (Think weather forecasting again.)
However, what I'm most interested in right now are behavioral variables. What are the actions I can look at today to determine tomorrow? It may not be that much of a difference but it is in my head.
In fact, in my head, by monitoring series of behaviors, we can actually prove a weekly pattern. Instead of Wednesday being the worst day for email because it's Wednesday, it's just part of an Active, Active, Dormant, Active.... behavioral pattern. (I guess after seeing this over and over, perhaps a behavioral pattern becomes so accepted that it becomes an environmental variable?)
Like I said, there may be no difference there at all, but I thought I'd share my thoughts on it anyways. Keep'em coming.
Tuesday, March 9, 2010
Subscribe to:
Posts (Atom)