Reputation
Next tag badge:
97/100 score
28/20 answers
Badges
1 52 128
Newest
 Nice Answer
Impact
~1.7m people reached

13h
comment Any necessary EDA before logistic?
That doesn't change how the calculations are done. The best way to keep the meaning of independent and dependent variables is to do a loess nonparametric smoother relating continuous x to binary y. But looking at this may possibly bias model specification.
19h
comment Should cohort studies in diagnosis research be age-matched?
I don't agree with that way of looking at it. See my expanded answer.
19h
revised Should cohort studies in diagnosis research be age-matched?
answered OP follow-up question
19h
answered Should cohort studies in diagnosis research be age-matched?
1d
answered Different results using Brier score and Logarithmic scoring rule
Feb
8
comment Evaluate fit of classification model in large sample
First clarify if you are analyzing the rawest form of the variables, vs. categorizing any of the raw data into fewer categories.
Feb
7
comment Any necessary EDA before logistic?
You are showing the distribution of $X|Y$ whereas we want $Y|X$ e.g. $Prob(Y=1|X)$.
Feb
7
comment Any necessary EDA before logistic?
More to @RustyStatistician but just a bit to you.
Feb
6
comment Is my understanding of regularized logistic regression correct?
I guess you could have a non-monotonic one but then predictors would often have to be transformed non-monotonically. But there is seldom need for a non-monotonic function to transform $X\beta$ to $P$. The sunflower example, to me, is an excellent example of why I like regression splines to relax the linearity assumption (and monotonicity assumption) in $X$.
Feb
6
comment Any necessary EDA before logistic?
I can see some value in that but also danger if the plot makes the statistician change the model, resulting in inaccurate statement of the true number of degrees of freedom present and inflation of type I error, lack of coverage of confidence intervals. Moreover, box plots seem to be interchanging what is considered the dependent variable and what is the independent variable.
Feb
6
revised Is my understanding of regularized logistic regression correct?
edited tags
Feb
6
answered Is my understanding of regularized logistic regression correct?
Feb
6
answered Multiple binary logit regressions vs multinomial logit regressions?
Feb
6
revised Multiple binary logit regressions vs multinomial logit regressions?
edited tags
Feb
5
comment Any necessary EDA before logistic?
How do the box plots help you?
Feb
5
comment Evaluate fit of classification model in large sample
Categorizing independent (or dependent) variables will result in massive lack of fit that needs no statistical test to detect. Worse still you say you are doing classification but I'll bet that is not the goal at all. The goal is or should be risk estimation.
Feb
5
comment logistic regression predictive modeling
@nootodis the optimum order of doing things is not fully known but it is reasonable to unbiasedly assess the smooth calibration curve after knowing that the model is competitive in terms of predictive discrimination.
Feb
4
comment logistic regression predictive modeling
loess is a good way to get a nonparametric calibration curve. Bootstrapping can be done with or without penalization; that is a separate issue.
Feb
4
comment logistic regression predictive modeling
The last step is biased. I suggest you correct for the bias using the Efrong-Gong optimism bootstrap (see R rms package).
Feb
4
revised logistic regression predictive modeling
edited tags