Short version. A member of the San Jose teacher's association thinks one of two academic studies is wrong, because... we already have a consensus.
The one point raised in the article about multi-site employers is acknowledged by the article's favored Berkeley study: "some multi-site businesses report payroll and head counts separately for each of their locations, while others consolidate their data and provide information as if their business operated only at a single location."
Oh and by the way:
"This report was prepared at the request of the Office of the Mayor of Seattle."
"Naïve Bayes can only represent non-negative frequency counts of features; therefore it was not a candidate as accelerometer data has negative values. However, this could be mediated by simply scaling all the data to ensure positive values (i.e. multiplying all values times 100)."
I think there are two serious flaws here.
- Bayesian frequency counts aren't measured values - they're counts of measured values...
- Multiplying a dataset of positive and negative numbers isn't going to make it strictly positive (unless you multiply by 0). You'd have to add the minimum value to all values.
This lead me to look for the author, but sadly the author is some anonymous 'GuestBlogger' ...
The one point raised in the article about multi-site employers is acknowledged by the article's favored Berkeley study: "some multi-site businesses report payroll and head counts separately for each of their locations, while others consolidate their data and provide information as if their business operated only at a single location."
Oh and by the way: "This report was prepared at the request of the Office of the Mayor of Seattle."
http://politicalcalculations.blogspot.com/2017/06/the-most-s...