My first comment was going to be that I think the baseline scenario would be lognormal rather than normal… but the 2nd graph (with exponential x-axis) basically covers that perfectly. And we did indeed get my suspected triple peak, exactly where I was suspecting it
Perhaps unsurprisingly this would be my 2nd comment, although it shouldn’t make a huge difference in 2015vs 2019 or 2021 vs 2024 I feel like nominal introduces a bit of noise whereas admits/adjusted would reveal more of a real pattern across time. Definitely not going to change the big picture takeaway though.
If we didn’t pull anything 90-100 then the 100-110ish range is suppressed a bit by only getting 50%+ of the adjacency window that the other values have, right? This is another thing that doesn’t really affect the main point just want to make sure I’m understanding the process correctly.
The lowest hanging fruit going forward imo would be seeing the graphs for like 5% or 20%, but this is already a very cool dynamic to have quantified a bit more