[Case Study] Berkson's Paradox

[AlperDuman]

1. Analyse des Berkson'schen Paradoxes

• Worum handelt es sich?

• Was sind die Ursachen?

• Wie kann man es erkennen und verhindern?

2. Datensatz mit R generieren, mit Grafiken usw. auswerten um daran das Paradox sichtbar zu machen

Gruppenmitglieder: Amir, Alper

[PhilippBach]

Dear @AlperDuman and Amir,

thanks for opening the issue, please first have a look on our already existing app for the collider bias, see https://colliderbias.herokuapp.com/ (we'll cover this in the next lecture) and here the repo https://github.com/DigitalCausalityLab/colliderbias

You can do something in this context, like adding a new example or improving the existing app, or add some extra explanations (like embedding the app in a blogpost or so).

So please, describe what exactly you want to do here in more detail. Thanks!

Possible real-data example

For example, in this youtube video, there's another example on books vs films https://www.youtube.com/watch?v=FUD8h9JpEVQ . There's even a real data set available which you could use (you could, for example, investigate whether you can detect the collider bias in the data set if you restrict attention to only famous books and movies, which you'd have to measure somehow like using bestseller's list and movie charts or so)

You can let me know if you'd like to try this. The next step would be to

• find the data source and see if it's avaialable
• find some data on the famousness of a movie and a book, i.e., hence find a subsample which is based on the selection mechanism
• do the visualization and
• interpret the results

I think this is quite explorative, but maybe you can find a nice example here 🕵️

So I'd recommend you to see if you can find the data source etc. and prepare the analysis. We can then discuss the current state in our first meeting in January. Thanks!

[AlperDuman]

Dear Philipp, thanks for the answer!

We originally planned on generating a data set to illustrate the topic at first, but when we get together as a group, we'll definitely consider building off of your feedback and incorporating some of these ideas into our project, thanks again for the detailed response!

Will there be an e-mail soon regarding the exact date and location of the next meeting in January?

Kind regards
Alper

[PhilippBach]

Hi @AlperDuman ,

thanks for your reply. I just sent a message on STiNE regarding the appointments in January. It would be great if you could concretize and prepare your data product idea for a discussion on January 11. I think we should spend the remaining time on developing the data product, so we should finalize the idea by next week.

Best,

Philipp

[PhilippBach]

Result: See hairdresser example (Fall 2022/2023)

• It's possible to extend this data product to a shiny app, see current state and call for extension