Yup, and to add context to the error, here is a Twitter reply from Matt Chingos, who directs the Center on Education Data and Policy at the Urban Institute:
So, not fun story: This morning, Adam Looney told me that he found that 33% of student debt is held by the top income quartile, not 49% as reported in an @urbaninstitute blog post from 2018. He was right and we are correcting the post this afternoon.
It turns out that using aweights with Stata's collapse command produces the wrong answer for sums (pweights are the correct option for a survey dataset like SCF).
May you all learn from our mistake and never confuse weight types!
Thanks to Adam Looney and @JHWeissmann for helping us identify and correct this error.