I'm a data architect with a PhD in the area of automation for machine learning architectures. I hold data science and machine learning certifications. I have to be anonymous because I'm a contractor for the DoD so please don't try to figure out who I am or I will have to delete my account. Understand about all that regarding correlation not equal causation, other features in datasets having impacts, etc. But the deltas between implementation impacts are too high to be explainable by other causes, particularly when you factor in dates of policy changes and subsequent impacts.