Join us for a PyData Ann Arbor Meetup on Thursday, December 7, at 6 PM, hosted by TD Ameritrade and MIDAS.
Although ideal experimental design involves hypothesis testing with randomized controlled trials on concurrent populations to minimize selection bias and convoluting variables, it often arises that experiments cannot be run on variant populations simultaneously — for instance, when a stimulus necessarily impacts the entire population at a specific point in time (as in measuring a non-digital ad campaign’s effectiveness). Dealing with these situations is common in the social sciences, where the method of Interrupted Time Series Analysis is commonly used. In order to measure the effect size of a stimulus in situations such as these, we have to consider many convoluting factors caused by our populations being non-concurrent before deriving meaning from the experimental results.
In this Python-based tutorial, we will walk through a Monte Carlo-generated experiment measuring the lift induced by a simulated ad campaign. By the end of this tutorial, you will understand many complicating factors that could lead to faulty conclusions being drawn from the experimental results. However, you will also learn how best to design experiments and interpret results to mitigate these risks and take proper account of irreducible convoluting factors.
About: Drew Fustin is a former physicist and current data scientist in Chicago. He created and led the data science organization at SpotHero, focusing primarily on optimizing acquisition marketing spend and balancing supply and demand to generate inventory and rate recommendations. He’s also worked for GrubHub as the insights analyst, turning food facts into media content for the PR department and transforming data into actionable initiatives within the organization. He was also a data scientist with Digital H2O, a SaaS startup providing water intelligence for the oil/gas industry. He holds a PhD in physics from the University of Chicago, where he studied dark matter by looking for tiny bubbles in a chamber over a mile underground in a Canadian nickel mine.
PyData Ann Arbor is a group for amateurs, academics, and professionals currently exploring various data ecosystems. Specifically, we seek to engage with others around analysis, visualization, and management. We are primarily focused on how Python data tools can be used in innovative ways but also maintain a healthy interest in leveraging tools based in other languages such as R, Java/Scala, Rust, and Julia.
PyData Ann Arbor strives to be a welcoming and fully inclusive group and we observe the PyData Code of Conduct. PyData is organized by NumFOCUS.org, a 501(c)3 non-profit in the United States.
“use what you have learned to make something better and share with others”