search

uchicago-computation-workshop / 2019_spring_conf

MACSS 2019 Spring 2nd-year Thesis Lightning Talk Conference
1 stars 0 forks source link

Ling Dai #30

Open bensoltoff opened 5 years ago

rickecon commented 5 years ago
bensoltoff commented 5 years ago
dailing616 commented 5 years ago
  • Nice. Oleg Urminsky is advising two MACSS theses.
  • Your contribution to a thesis cannot just be in the data analysis part. You need to be involved in every step. I hope that is what you meant.
  • Is there any computational content to this thesis? I know it is a field experiment, but that is not enough for a computational exercise.

@rickecon To answer your questions, the experiment was ran by Prof. Urminsky and I didn't participate in running the field experiment, and there's not much I can do about that. However, I did propose several hypothesis testing in additional to Prof. Urminsky's original research plan, using the data they collected. (e.g. testing Calorie per USD Spent as an indicator of choice architecture was my original idea, and some other analyses of my own were not shown in the slides due to time constraints)

As for the computational content part, scraping and processing the receipts was in fact a lot of work, not in terms of computational power, but in terms of coding. Moreover, because the data were so dirty, Prof. Urminsky and Goswami were actually having a hard time processing those data, and that was the major reason I was initially brought into this project. Because we also included sales data from 2014 for the purpose of baselining, there were more than 100k receipts to process, and they actually came in more than one format. Also because there were many miscellaneous items that we couldn't directly lookup for calorie information, I also built ML predictive models to predict the calorie content of those items.

I hope that answers your questions. Thank you!

dailing616 commented 5 years ago
  • Was the signage in condition 2 for all products sold, or just the example products like shown on the slide?

@jmausolf Thanks for asking. It was just for the 3 items listed there, namely two kinds wraps and a salad. And I believe the specificity of its information is exactly why it wasn't as efficient as Signage 1.