@EvilChess I have made the same argument with no success. May as well bang the head against a wall. Off course a correspondence rating would be a far better data set to use than a blitz rating when a comparison is sought to OTB 90 games.
Have you plugged in the ratings to the formula and observed the resulting trends and predictions?
For example try 1078 blitz rating and 1169 classical (when the OP started the thread, classical was any game over 8 minutes, practically blitz) and view the predicted Fide rating. (should predict a Fide 1000) However, the result is much higher than the blitz rating! Try 2278 and 2369 and the prediction (should predict 2200) is much lower than the blitz rating! Makes sense right? LOL
+78 and +169 are from the OP's original premise on 1st post.
"Since there are many (fake) outliers, summarizing these data using the average would be misleading. We use the median instead:"
* A typical (median) user's FIDE rating tends to be 78 points lower than her Lichess Blitz rating
* A typical (median) user's FIDE rating tends to be 169 points lower than her Lichess Classical rating
OP
The formula makes the expected prediction for players rated 1900/2000. Ratings higher/lower and the predicted Fide rating becomes increasingly skewed towards each end of a plotted line.
If the formula is valid for only 1 class of players that fall within the median of +78 and +169, how can it possibly be valid for players that have a different skill set? ie: A blitz rating that is higher than their classical rating, which certainly applies to a great many players.