Tech Economics Newsletter
What's new from economist Andrew Chamberlain, Ph.D.
Summer is here, and I'm back with a quick update from the world of tech, economics and data science.
Where Americans Are Moving for Jobs: I've been trying to do something interesting with Glassdoor's real-time job applications data for years. Last month we finally did it -- here is my new "metro movers" study. Favorite takeaway: Geographically mobile tech workers may be fueling inequality among cities.
Smarter A/B Testing With Bayes' Rule: Traditional A/B testing usually assumes we know nothing before the test -- a null hypothesis of "no effect". But in most business problems we usually know a lot, from real-world experience or previous tests. There's a better way: Bayesian A/B testing. Here's a nice example walking through the math of doing this in practice.
Why CEOs Should Be Skeptical of AI: What can AI really do today? And where does it fall short? Here is a smart take from the brilliant Susan Athey on why business leaders should be aware of the limitations of what ML models can realistically do in today's business world.
Build A 5-Minute Recommendation Engine: Here's a nice example I ran into this month walking through how to build a simple recommendation engine in Python. Best part: It only takes about 5 minutes. A nice tutorial for teaching young analysts the basics of today's widely used collaborative filtering models.
Your Very Own Stack Overflow: Most of us who write code today spend a lot of time on Stack Overflow, looking for solutions to problems others have already solved. Now there's an awesome private version for teams, starting at around $10 per month -- a nice solution for creating searchable knowledge for data science teams.
Cheddar TV is a Real Thing: By streaming to Hulu, Facebook and other online platforms, Cheddar is rapidly rising as a major player in broadcast media. Here's me last Friday talking about the latest jobs report with the folks at Cheddar's business network.
Extracting Secrets from Neural Network Models: Some ML models today are trained on "secret" data. For example, sentences types in a text messaging app might be used in a model to suggest the next word you might type. Here's a worrying paper about how it's often possible to extract secrets from training data that may be inadvertently revealed by "deep learning" models. Be careful out there my friends!
Thanks again for staying in touch -- and best of luck in the coming month.
Andrew Chamberlain, Ph.D.