I had a chance yesterday to listen to Mary Grace Crissey’s Informs podcast The Nitty-Gritty of Working with O.R. Providers. I thought she had some interesting insights about Operations Research and analytics in general and data issues in particular. I think that sometimes data issues are not addressed enough in OR classes, where the problems can be small and the data contrived to create certain situations. This can leave OR practioners in the dark when it comes to analyzing the dirty, incomplete data you often encounter in the real world. Some of her data highlights include:
- Before you can get better at something, you need to know how you did last year
- Getting data “straight” is no small task
- The quality of your data affects the quality of your model output
- You cannot do historical analysis on data that doesn’t exist
Numbers 1 and 3 echo a recent blog post of mine about modeling without enough data. It’s nice to see that very smart people have similar opinions about the issues I have found in my work. Number 4 is also interesting because I have run into several situations lately where there is a lot of discussion around the level of granularity necessary for data that is being collected. I plan to discuss this in more detail in an upcoming blog post but the basic idea that I have stressed in these situations is that you cannot analyze data at a lower granularity than you originally collected it at.
I also loved her statement
So much in our daily life has changed by the way we look at numbers today.
I feel that way about Operations Research too, that so much of my life and my work has been changed for the better by seeing the OR in it.
For those who have listened to the podcast, what parts of it resonated with you? Did you find parallels between her thoughts and your work in OR? And what challenges have you run into when deal with real-world data?