Documentation & Code review

[jaeyk]

[Averysaurus]

Risk of recit repository

line 7: are -1 values in tibble missing data? was that explained anywhere I missed?

[Averysaurus]

Risk of recit repository

line 12: suggestion : maybe explain AIC scores a bit.. Lowest score is best, yeah?

[Averysaurus]

Risk of recit repository

line 14 & 16 : suggestion : could we explain, in a non-technical way, what the chart illustrates?

[Averysaurus]

Risk of recit repository

Summary: this is brilliant statistical coding. All code that I dod not mention runs clean copy/pasted into R-studio.

If we can attach a down to earth narrative to these data, functions, and outputs, workshop will be a breakthrough A++ IMO.

~Avery

[Averysaurus]

pt. 2, Risk of violent recit

line 3: Data import has verbose warning script. dim() output is == on local environment.

[Averysaurus]

Line 9 in part 1 & Line 8 part 2:

Good place to explain or re-iterate narrative about what the recidivism score is.

[Averysaurus]

pt. 2, Risk of violent recit

Line 12: a lot is going on in this part! 👍 As elegant as the code is, it may help to include an explanation of what (with categorical DV) and (with continuous DV) are.

[Averysaurus]

Algorithm Accuracy Analysis

line 7: include narrative explanation of what these organized decile scores are expressing?

line 8: you are employing a cox hazard model here? Is there a simple way to describe it? what is Surv()? What is survival rate in this context? Why so much variance with Native Amer. category?

line 10-12: awesome visuals, excellent opportunity to distill findings into simple, "one sentence" type statement?

line 13: awesome that you went into python here. This transition may pose challenges if people are following in workshop. I think it is really cool myself. 👍

line 13: may help us to explain what a false positive /negative implies in this circumstance.

Again, super brilliant stuff. Imagining a workshop: the code speaks for itself, explaining a bit about the statistics and implications of outputs, "what it means" from a model -> reality standpoint would be valuable narrative threads to support the body of work here. +1000 overall. :)

[Averysaurus]

01_compas_analysis_revisited-checkpoint.ipynb

Intro: could we add links to the PBS video clip and background articles as prep material, something we circulate in our registration email? It may help us to have participants queued in on context before we begin.. It's possible participants enroll and not prepare, this narrative intro is key to everything else making sense to a beginner's eyes.

Line 5: change ae to ea in 'two_yaers_violent' ?

Line 28: It may help to add narrative explaining differences in these stratified outputs. What is being expressed here in the decile scores?

Line 30: in the other notebooks you evaluated different modeling methods with an AIC score. Could that be applied here? If so, should it?

Line 37: concordance is advanced stat terminology? Would there be a concise way to explain concordance and the difference of values reported?

Line 47: It would be awesome to explain concepts of overprediction and underprediction here.

Line 54: change "two year" -> "two years"

All together awesome stuff +1. Again pushing accessibility, explanation of stat methods, particularly the surv() functions, adding ylim(.25) to the output, a summary at the end of notebook is all I can suggest on my end.

[jaeyk]

@Averysaurus I think these comments are all great and related to accessibility issues. Can you add some multimedia references (e.g., YouTube video clips) to the R markdown files? Also, please feel free to edit those markdown files if some parts are too dense, etc.

[Averysaurus]

Part 3: notes, interpretation challenge.

what are hazard ratios?
Hazard is the instantaneous potential for change in an outcome per unit of time at a specific time, t, relative to the size of the population (where the event has not occurred yet), at time t. Hazard Ratios compare hazard between exposure status,(in this case race or other identity category).

ln 84 - 93: we should explain what the x-axis is, what those numeric values mean.

ln 102: we may want to restructure hazard ratio visualizations to look more like this:
typical HR vis

also Python in an r-markown. Can we make it work simply?