Good illustration, but I couldn't stop wondering (and being sceptical) - in relation to the probit model - of the statement early on that since we CAN'T observe y* (the latent variable), we can simply ASSUME it has a normal distribution. What on earth is the theoretical justification for this? Presumably a similar assumption is implied by the logistic regression's standard functional form (i.e. a linear relationship between the log of the odds ratio (the dependent variable) and the explanatory variables, though the video is completely silent about what the theoretical justificationbof that implied assumption might be...
A really great explanation! Thank you! Some minor mistake in writing: at 14:19 you wrote F_Unlikely = exp(-0.82) but you meant 0.44/(1+0.44) instead. Same for F_somewhat Thanks!
Honestly, this is exactly what I needed for a refresher for stats work at work. I especially appreciate that you took the time to explain the motivation of IQR. Thank you so much, Dr. Harbaugh
This video sucks. It assumes that the mean doesn't change when a new data point is added. I'm looking for the effect on both mean and standard deviation when the new value doesn't give the same mean and standard deviation as the original sample.
dear can you help me i work my study contain likert scale for both dependent and independent variables then how can i analysis using ordinary regression analysis was it possible to transform thier response into other means or sum then use ORA
Wonderful video, thank you! One question, how would we be able to state the effect of an independent variable on the outcome? For instance, how can I express whether or not public school attendance is significantly associated with an increase likelihood of applying to grad school?
I believe a p-value of less than 0.05 for the predictor is considered significant. I'm not sure how to get p-values in ordinal regression though in R. There may be a different metric for measuring variable significance in ordinal regression.
Thank you so much! This is extremely helpful. I do have one question - at 14:20 shouldn't the probability equations be F(unlikely) = 0.44/(1+0.44) = 0.306 and F(somewhat) = 3.58/(1+3.58) = 0.782 instead of exp()?
Thanks for this useful video. Can you please tell me how we should deal with these outliers (either mild or extreme ones)? How we should modify it? there is a method called winsorization in which we replace the outlier by the 95th percentile value. For my dataset, when I do that it has some contradictions with the IQR rule.
Thanks! this is very helpful! I'm wondering if there is a way to assess the effect of a predictor X on the thresholds (intercepts) estimation. Are the thresholds assumed to be constant and pre-determined?
Thanks for this easy but informativ tutorial! It would be nice if you can do a tutorial on hierarchical ordinal regression! & is there a heuristic on how much categories there sould be at maximum?
How did you calculate the probability values for the ordinal probit regression model? For the instance with the specific student, how did you calculate the probability of them answering "somewhat" as 0.449?
Thanks for a helpful video. I'm new to ordinal regression and your's was the first video I've watched. The point where I was lost was the animation of bell curve movement with changes in GPA. How can the bell curve move if the value of boundaries is already defined? I can see how we'll get different Y* value with changes in GPA. However, I can't see how bell curve itself will move around on a likert scale!
The y* value comes from the equation, and the y* is the location of the center (mean) of the bell curve. So, say x increases, then y* increases, and thus the center of the bell curve increases along the axis. I hope this helps.
