sinjax

Hello. I am really interested in specifics of your model, I'm not sure how much you care to divulge as it might represent hours of your time and chunks of your income but, worth a shot.

I work as a research fellow for an EU project that is trying to improve such models using social media sources (i know, who isn't these days?). But tell me, what kind of model have you constructed here? Is it primarily heuristics or is it learnt from data? If so what data? What are your variables? The interactive model on your blog is interesting but it seems the few sliders available wouldn't be enough to predict so accurately? Do you have any papers i can read re: these details?

By the kind of model, I'm asking: are we talking a simple linear model? something more advanced? Trained NN, bayesian, SVM, or… what? Echoing WildTauntaun: "what is your statistical background?".

Finally (and most importantly), what's it all about? you know, everything, what's that all for?

edit: added MORE QUESTIONS!

Thanks for the technical answer, exactly what i wanted to know!

I am not surprised that you get such great results with a linear model, indeed i think most of the time people use non-linear madness to show off, practically speaking you can get 90% of the way there with most problems using OLS or similar.

What does surprise me is that those 3 variables can get you so far! I mean yes your classification task is small, but wow, 3% error with 3 variables, pretty interesting. I suppose the Gallup is itself a composite variable made up of many factors, but still, good to know!