Justin Holman is CEO of Aftermarket Analytics, where he leads efforts to develop cutting edge sales forecasting and inventory optimization technology for the Automotive Aftermarket. Prior to joining Aftermarket Analytics, Justin managed corporate consulting for the Strategy & Analytics division at MapInfo Corporation, leading major projects for retail clients including The Home Depot, Darden Restaurants, Bridgestone-Firestone, Sainsbury’s and New York & Company. Before that, Justin served as Vice President of Software Development at LogicTools, now part of IBM's supply chain application software group. Justin holds a B.A. from Claremont McKenna College, a Ph.D. from the University of Oregon and an Executive Management certificate from Northwestern University's Kellogg School of Management.
3 Comments
Greeetings, Earthlings!
Good luck on your test, I recommend reviewing the material relating to the topics of discussion this semester.
8 more weeks after this week, one is Spring Break and another is finals… Much taco-espresso smoothies.
Hello, Professor Holdman.
I’ve got a quick question, when is it acceptable to have a negative coefficient? I believe it was that if you could boost your Rsquared number, while having a low P score, it would be acceptable to have a negative coefficient. I’m probably making this harder that what it is, but I want this to come out perfect.
Thanks,
Rique Lucero
Hi Rique,
Sorry I didn’t reply to this question in time for exam 2. There’s nothing wrong with having a negative coefficient. But, it should make sense in the context of the problem you’re trying to solve. The real estate data example with a negative coefficient for number of bedrooms was a result of multicollinearity. But the reason it isn’t appropriate to utilize such a model is because it doesn’t make sense to reduce a selling price estimate for every additional bedroom, not because the coefficient is negative. Make sense?
Best,
JH