Thanks for the feedback.

]]>1. Your phrasing “The probability of that in a Poisson process with mean λw is” is a bit misleading because it comes after making the statement about P(W > w), yet the following probability shown is the one for F(w).

2. There’s an extra parenthesis when you first show the derivative.

3. The derivative is F'(w), not F'(x).

4. Some important steps were skipped, for instance it’s not clear where the extra factorials come from when you derive.

Overall, the proof is nice and that’s what I was looking for, but the bad presentation made me look for it elsewhere.

]]>I forgot about that article. That was almost 9 years ago. Where did the time go?

Sure, I imagine the issue could arise with linear data. Fit a highly flexible non-linear model to linear data and you can get high r squared values despite the model being wrong. That’s an example of overfitting the data. Adjusted R squared in this case wouldn’t fix the fact you fit the wrong model.

]]>I was referring to the article ‘Is R-squared Useless?” On the University of Virginia blog it attributes authorship to you. In it the arguments issued by Cosma Shalizi against the r2 metric were reviewed by you. In particular, I was asking you about the demonstration of how r2 can be arbitrarily high even with an incorrect model. You used non-linear data to prove the point. But my question is if the same issue can result using linear data? Secondly, I wanted to know in your opinion how well can adjusted r2 values compensate for these alleged failings.

Link to article in question

https://library.virginia.edu/data/articles/is-r-squared-useless

I’m not sure I follow your question. I’m also not sure what article you’re talking about.

]]>Yeah, that was me. Happy to try and answer any questions you might have.

]]>Thank you!

Kyle

Thanks for the comment and nice words! I have to admit, in the 10 years since I wrote this up, I’ve developed a less-rigid approach to mixed-effect modeling. I no longer subscribe to the recipes advocated above where we first fit an “unconditional means model” and so forth. I think you should use subject matter expertise, your research question, and the structure of your data to guide your modeling. Propose a sensible model that could reasonably be answered by the amount of data you have. Try different models as needed. Be transparent and report all your models, or at least report that you tried different models. Be modest about your results. Accept the uncertainty. I hope this helps.

]]>