unless sample size is > 20,000

p here is the number of parameters, not the prob in the above formulas

The references show Riley, et al Part II twice. Need to add the reference for Part I here:

Minimum sample size for developing a multivariable prediction model: Part I - Continuous outcomes Richard D Riley 1 , Kym I E Snell 1 , Joie Ensor 1 , Danielle L Burke 1 , Frank E Harrell Jr 2 , Karel G M Moons 3 , Gary S Collins 4

PMID: 30347470 DOI: 10.1002/sim.7993

Here's a later edition of Clinical Prediction Models: A Practical Approach to Development, Validation, and Updating (Statistics for Biology and Health) 2nd ed. 2019 Edition

by Ewout W. Steyerberg (Author)

link to Amazon: https://www.google.com/url?sa=t&source=web&rct=j&opi=89978449&url=https://www.amazon.com/Clinical-Prediction-Models-Development-Validation/dp/3030163989&ved=2ahUKEwjSxoHgq5OGAxW3IUQIHWFdBtAQFnoECBsQAQ&usg=AOvVaw1rQeWGmpD-0CWnqdfmplNj

I'm not sure what this phrase means? How does this statement related to the use of the indispensable (IMHO) Hmisc::describe function, e.g. Section 8.2?? Or R Workflow Chp 9?

A model is basically a formula. The beauty of RMS are the tools for presenting and visualizing model results. Why focus on "descriptive statistics" ? I'm confused!

Suggest something like this: (graphical) visualizations of a (well-fitted)(validated) model provide the best (way)(tool) to (learn)(understand) the (underlying) data ....

this is a long sentence. Suggest replacing "as will auxiliary" with "as well as"

dropping the word "auxiliary" since most of the topics listed are too important

suggest replacing "her" with "them"

reword ?? "does not at all affect"???

I cringe every time I see "she" ... it seems as if "she" is always doing something wrong! suggest "they" (with or without changing investigator to investigators ...

widering? maybe "widening" or just "wider"? or even "increasing the width of"

"only" in rare ...

1991 The effects of transformations and preliminary tests for non-linearity in regression. Stat Med, 10, 697–709.

what happened to the datamethods discussion for this chapter?

typo: "fredom" >> "freedom"

? plot title? Predicted Fat Fraction based on Height and Abdominal Circumference given Age = 43 ??

Confused ... you list 5 methods? Maybe number them 1 to 5 ...

1. Fit simple

2. Scatterplot of ...

3. Stratify ...

4. Separately for ...

5. Fit flexible nonlinear ...

Ah now Method 5!

Method 4

Method 3 ?

this is method 2?

Just one reference? So maybe just

"Stone and Koo (1985)"

or

"Stone & Koo (1985)"

typo: "form" should be "from" (or maybe "using"

your link went to the desktop version.

Here's link to the server version (Yea Posit!) https://posit.co/products/open-source/rstudio-server/

Can't find this ?personal communication??: gus06per

this link is broken as of 5-05-2024

You've referenced Part II twice. I suspect the first reference in this pair should be to Part I : https://doi.org/10.1002/sim.7993

??

?insert "ed." or "edition" after "second"

typo .. "as" should be "has"

insert ":" Compute by treatment: the mean time unwell

In class, Frank says day should be year,not day

discussion in class very helpful for use of cancore option in transcan

Question: ?truncate? how would that var go in the formula? You said the prediction would be flat for wbc > 40000?

someone in class suggested define effective sample size in glossary!

and m is the sample size (number of observations?

Just to clarify, some ML 'packages' (e.g. AzureML, GoogleML) consider logistic regression to be one of the ML algorithms. FWIW, I've seen (high school) science fair projects use both of those and find that logistic is "the winner" ...

if the cutoff for dichotomizing one variable is unstable, shouldn't be surprised that a tree based on cutoffs by dichotomizing multiple variables is unstable ?

insert period after cutpoints

only appears

effect is present

or better: Do Not Forget ...

?what is not recommended? ?analyses on individual completed datasets?

nonmonotonically

duplicated look

biased

trying to understand this: maybe "the difference between the treatments with the greatest difference is badly biased"

just for readability:

"disease" or "no disease" when severity of disease is actually on a continuum

Pages for Part I are actually 1262--1275

in an object if you want

typo: Some of the functions ...

but also allow for shrinkage

add some simulated post data

When it increases

Suggested edit to emphasize you don't really suggest this approach?!

"Best way" to make model fit data well is to discard much of the data!!

maybe "Change from baseline will"

the idea is to assume

question: after the plan, or after the pivotal analysis, has been completed?

*Avoid * the tendency (or "urge"!)