The technique used to calculate APT/ATT (the underlying metrics for timeliness) are documented here (you have to click through a few folds)[1].
They don't appear to be simulated or theoretical: they're based on riders' swipes into the system. They do make certain ridership assumptions, however, because of the limited data (e.g. that customers use the same "corridor" during weekdays, and that the next swipe-in corresponds roughly to the last exit during weekdays). Those assumptions don't strike me as unreasonable, given the MTA's inability to collect exit data.
Edit: The page goes into detail on those restrictions as well. Interestingly, it sounds like OMNY gives them more precise timestamps, and thus has allowed them to improve the quality of the their metrics over the past ~2 years. The fact that they haven't swung substantially during OMNY's rollout indicates that there wasn't significant error in the previous data.
One of the challenges of "poor-er" people...e.g. those on local stops in outer-boroughs is the curse of express trains running past their stops. They are expected to switch again and come back to the stop where they should have stopped anyway. Subway metrics show the trains are operational, but they arent -- it is people adding 20-30+min to a commute because only half the station is operational.
Meanwhile the metrics (and exit counts) will show regular operation. Sharing dashboards like these suggest a working system when it isnt actually working that well.
That's a great point, and I agree: the subway doesn't do a uniformly good job of serving the city's population. There's a reason the wealthy areas are around the lines that they're around.
That being said: that's been true for 100 years. The claim (which I agree with!) that the subway experience is worse when poor doesn't have the temporal aspect that your original claim did (that subway performance has gotten worse recently.)
They don't appear to be simulated or theoretical: they're based on riders' swipes into the system. They do make certain ridership assumptions, however, because of the limited data (e.g. that customers use the same "corridor" during weekdays, and that the next swipe-in corresponds roughly to the last exit during weekdays). Those assumptions don't strike me as unreasonable, given the MTA's inability to collect exit data.
Edit: The page goes into detail on those restrictions as well. Interestingly, it sounds like OMNY gives them more precise timestamps, and thus has allowed them to improve the quality of the their metrics over the past ~2 years. The fact that they haven't swung substantially during OMNY's rollout indicates that there wasn't significant error in the previous data.
[1]: http://dashboard.mta.info/Help