Explanation - Prediction trade off in Political Science research?

[bhargavvader]

Thanks for the paper! I'm always excited to read papers using ML in clever ways in the social sciences. I particularly liked your method of using donation/contributions data and information to try and map to ideology - it certainly makes sense that this information would highly influence ideology, and it is always interesting to see possibly exactly how much?

On that note, a common problem in the ML model selection trade off is between the predictive power of the model and the ability for it to explain behaviour. We can see early on in your paper that your focus is more on prediction, which might also influence your model choices (SVR, Random Forest), which are models which often predict behaviour better than they explain it.

You do mention in the final bit of your paper that a future direction of research might be to use more sophisticated learning methods (such as kernel regularised least squares) on such data-sets. Could you maybe discuss this in more detail, especially with what possible information you would expect to glean? Do you have any ongoing work in this particular sphere?

[w4rner]

On politics & ML, would encourage y'all to scope out the beautiful Bhargav chatting Trump tweets: https://www.youtube.com/watch?v=E5YppPJCCq0

[bhargavvader]

Awww @w4rner , thanks for the plug <3