Interesting, the census has followed in the footsteps of big tech and adopted differential privacy. I had the joy of working with differentially private data at

Interesting, the census has followed in the footsteps of big tech and adopted differential privacy. I had the joy of working with differentially private data at my last job. It was brutal. Typically when you do differentially privacy you also do thresholding (any group with less than N members is dropped unconditionally). This means intuitive things you'd want to do are now subtly broken. Want to take an average? You can't because the lower values are just thrown out. Same with standard deviation. Basically all downstream analysis needs to be REALLY aware that the data is anonymized. I don't know if the census's implementation will have these problems but if so then yeah, as the paper found lots of analysis will either need to change or be done differently.