Chapter 3

[dashaasienga]

@katcorr

Hope you had a great weekend!

Given our wrap-up of the application section of the thesis (#30), I've completed a draft of Chapter 3 (see https://github.com/dashaasienga/Statistics-Senior-Honors-Thesis/blob/main/Thesis/index/_book/Dasha-Asienga_StatThesis.pdf).

It was definitely a challenge trying to be concise and selective with what to include since I had done quite a bit! I'd love your comments on feedback for how we could improve the chapter, but no rush with this on your end since I will probably get back to editing that during Spring break.

For this week, I'd love to continue following the plan (#25) and finalize ideas for the simulation, revise Chapter 4.1, and get to a decent place with the simulation code by the end of the week. I will probably post another issue regarding this some time tomorrow!

See you Tuesday.

[katcorr]

Excellent work on Chapter 3, Dasha! You conducted such a thorough exploratory analysis, and you condensed the results already a lot. I tried to add in edits to make even more concise in this draft, and identify sections that could be cut in Chapter 3. But overall, the writing is really well organized.

One thought on something that might be added (or not, up to you): a paragraph (perhaps early on in this chapter, like in section 3.1) discussing how even though we're using the variable "recidivate" as a "gold standard truth" of whether someone reoffended or not here . . . it is, in fact, itself likely biased by societal factors (e.g., police officers less likely to arrest white people / more likely to arrest Black people for same offense; judges less likely to convict white people/more likely to convict Black people for same offense). While perhaps tangential to the theoretical underpinnings of the Seldonian algorithm and your work in comparing the Seldonian algorithm to logistic regression in this thesis -- I think this context is a very important thing to keep in mind more generally when trying to apply computer science-driven "solutions" to addressing algorithmic bias.

I hope you're enjoying Spring break! No need to work on updating this chapter this week; resuming with thesis work next week will be fine :)

[dashaasienga]

@katcorr

I hope you had a great Spring break! I definitely did, and I'm ready to put in the work for the final stretch!

Thanks for the feedback on Chapter 3! I will review it throughout the week and return a revised draft before next week.

I agree with your suggestion. I think it'll be important to underscore that some of these fairness guarantees rely on the assumption of the 'objectivity' of the response variable, especially when looking at error rate discrepancies. The discrimination statistic is calculated with this assumption, so I definitely agree that adding that information will be beneficial for setting up the proper context.

See you tomorrow :)

[dashaasienga]

@katcorr

Hope you're doing well! See https://github.com/dashaasienga/Statistics-Senior-Honors-Thesis/blob/main/Thesis/index/_book/Dasha-Asienga_StatThesis.pdf for the revised Chapter 3, including your suggested edits. They were super helpful, and I think the draft looks much cleaner now! I'd love to hear your thoughts on it!

[dashaasienga]

See #32 for updates.