I really like the coefficient divide by 4 rule at 31:30 for logit, I hadn't thought of that before. Then again, it seems that works well exactly in situations where LPM would work well anyway (events where most predicted probs are close to 0.5). Also lots of folks seem to report odds ratios , seems hard to apply a rule like this to those. Something I didn't see mentioned that may be worth discussion is the issue of interpreting logit coefficients across models. Social scientists often run a reg without controls then with controls and assess how the coefficients of interest move move. This doesn't really work for logit coeffs since the metric of the latent variable is arbitrary (see e.g. www3.nd.edu/~rwilliam/stats/Oglm.pdf, www3.nd.edu/~rwilliam/stats3/L04.pdf).
![⭐ฮิตในTikTok!! ( หมูเด้ง MooDeng ) Ver. แดนซ์โจ๊ะๆ ReMix BY [ ดีเจกิต รีมิกซ์ ]](http://i.ytimg.com/vi/3krsyfWUeHU/mqdefault.jpg)
amazing, simply amazing. thank you Sir!
Thank you Prof. King.
I really like the coefficient divide by 4 rule at 31:30 for logit, I hadn't thought of that before. Then again, it seems that works well exactly in situations where LPM would work well anyway (events where most predicted probs are close to 0.5). Also lots of folks seem to report odds ratios
, seems hard to apply a rule like this to those.
Something I didn't see mentioned that may be worth discussion is the issue of interpreting logit coefficients across models. Social scientists often run a reg without controls then with controls and assess how the coefficients of interest move move. This doesn't really work for logit coeffs since the metric of the latent variable is arbitrary (see e.g. www3.nd.edu/~rwilliam/stats/Oglm.pdf, www3.nd.edu/~rwilliam/stats3/L04.pdf).