I recently did some channel attribution consulting work. The idea is that you have a bunch of ad spend over time on a bunch of channels, and you have to try to figure out what amount of money buys a conversion on average for each channel, and then you can try to infer your organic growth rate based on the error you see when you back-forecast your actual conversion data based on your model and your spend.
Multivariate regression analysis is basically as complicated as statistical analysis gets, conceptually speaking. But it’s all deterministic except for picking which fields to assign as dependent versus independent. Once you pick which fields are which, SPSS (Or PSPP) does the rest. So this made me wonder if GPT-3 could pick the fields. On its face, it seems like a perfect problem for a GPT to solve. If you can just conceptually understand what the field names mean, and then decide which fields are unlike the others, and then which of those two groups makes more sense to be looking for (or which group of things affects the other group of things), then you can answer the question.
Conceptually it’s a lot, but once you understand the concept of what you’re trying to do, it’s easy. But it’s always going to be a non-deterministic problem; it’s always going to depend on what the field names mean and how the concepts relate to each other. In short, it seems like a potentially excellent use case for GPT.
Let’s Try It
The bold portion is the prompt.
A user would like to perform a multivariate regression analysis on a table of data. Based on the following field names, create two comma separated lists. The first list will be the dependent variables. The second list will be the independent variables.
Field Names: Facebook Ad Spend, Google Ad Spend, Reddit Ad Spend, Conversions
Dependent Variables: Conversions
Independent Variables: Facebook Ad Spend, Google Ad Spend, Reddit Ad Spend
Analysis
As you can see, GPT-3 had no trouble with this task. This opens up a whole world of fascinating potential research opportunities!