The Conceptual Regression EquationPosted: January 24, 2012
Honestly, I am not sure if this will work but I want to try it. I want us to think about creating a conceptual Regression Equation for each of our possible client companies. The assignment for Thursday’s class is at the bottom of htis post. But first let me quickly review how we talked about multiple regression in class.
Multiple Regression Review.
Remember the purpose of a regression equation is to determine how independent variables contribute to a dependent variable. Generally they take the form of
Y = a = B1X1 + B2X2 + B3X3 + ….
Y = dependent variable — the thing we want to predict
a = the constant (throw this out — we don’t need it)
X = the independent variable — the things we use to predict the dependent
B = the factor loading — basically the correlation coefficient — it can vary between -1 and +1. It tells us how much it contributes to the dependent variable.
Remember our extended example in class. We wanted to predict the amount that someone tips a waitperson. Y = amount of tip. We thought about all the possible factors that contributed to it — the independent variables. Quality of service. Quality of food. Total amount of the bill. Those are the Xs.
So now we understand the basics of regression.
A Conceptual Multiple Regression Equation.
Focus on the main purpose of the multiple regression equation as prediction. We want to use somethings to predict something. And if you really want to get manipulative, you can see a multiple regression equation as a recipe. We can manipulate the ingredients (independent variables) to determine the final product (the dependent variable).
So now I want to take the most abstract approach to multiple regression equations and see if we can use it to help determine what we want to do with our companies.
You have been thinking about and researching your companies. Now we need to determine if we can actually use the this approach to help our client companies deliver value to their constituencies. Wow, that sounded very vague, abstract, and downright hard. Here is what I mean.
For a company, what is it that they want to accomplish. For instance, let me try an example. What does GoogleTV want to do?
Here are a couple of links about Google TV.
Ed Cotton: Is There A TV Revolution Happening? @PSFK
What’s Really Next for Apple in Television – NYTimes.com
In the simplest terms, I believe the true value of GoogleTV is going to be predicting what TV program you want to watch. So in the most simple conceptual terms, think of the independent variable (Y) as program recommendation (at the end of this extended example I will tell you about the problem with this example but just swing with it for a minute). What can it use to make the best possible prediction? In other words what are the the independent variables.
X1 = previous viewing. I think that would have a high loading.
X2 = time of day — I like to watch certain things at certain times of day.
X3 = my mood. Do I need the TV to lift me up or calm me down.
X4 = my other interests (am I a sports fan?)
So I could put together a conceptual algorithm that allows a GoogleTV (or AppleTV or connectedTV) to make accurate suggestions. And then, of course, I would add a feedback variable to the equation to measure its value — did I actually choose to watch it and how much did I watch.
So that would be my conceptual algorithm for predicting what TV programs to suggest. [And on a technical note this isn’t a good example because the TV Program isn’t a variable at all — it is nominal data — categories — so you can’t calculate correlation. But hang with it for just a few more minutes.]
Is this a valuable algorithm? I would say yes. Helping people get the program they want to watch is good for them — no duh. And I would argue that it is good for the program producer — in my mind they are still implementing some form of advertising so getting them a program they like will encourage them to watch more.
So here is the assignment. Can you come up with a conceptual multiple regression equation — heck, let’s just call it an algorithm — for your company. Is there something they want to predict? Is there something they want to show them? Is there something they want them to do? And are there things that might help motivate them to do that? Are their ingredients that we can draw from social media to get the user there?
OK, this is an example of one of those digressions that you think sounds brilliant on the way in and by the time you are at the end you wonder if it makes any sense. So tell me on Thursday if this works. And if we can’t at least conceptualize the dependent variables for the companies, then I believe that ReallyGetsMe won’t work for them.
We have to leave class on Thursday with our companies in hand.