I found this a really interesting statement, and am trying to source the evidence. It may be my research skills, but there doesn't look to be much checking into the disparity between predicted grades and achieved grades out there in state vs independent schools.
But I am not a researcher!
From a Sutton Trust report from 2020: PQA: Reforming University Admissions - The Sutton Trust " those from working class backgrounds were, in general, more likely to be overpredicted: 34% compared to 29% of those from middle-class backgrounds. While previous data has shown this pattern to differ between disadvantaged pupils at the low and high ends of the attainment spectrum, the sample size available in this survey was not sufficient for such analysis."
But of course working class/middle class is hardly an exact match for state/independent!
It does also say that "Students from state schools were more likely to be underpredicted than those from private schools (32% vs 26% of private school students)", which confirms half of @Philandbill 's statement, but now I'm still left trying to find out which type of school is more likely to overpredict! As the Sutton Trust says that about 75% of students are overpredicted, and only 9% underpredicted, the overpredicting seems more of a concern (not least to the poor pupils who miss their offers)
UCAS is apparently going to pilot a new scheme to help teachers hone their predicting skills, and show schools how their own predictions compare with achieved grades (though I'm sure most schools must monitor this themselves anyway, surely?): UCAS pilots new reports to help teachers strengthen grade predictions and support student choice | UCAS It says that "Part of this research highlights how predicted grades can differ by student context. For example, disadvantaged applicants (defined using the Income Deprivation Affecting Children Index score) achieve further below UCAS predicted grades than advantaged applicants"...
but again, I'm not finding reference to school type, only to socio-economic groupings, which is less than helpful.
Meanwhile, other research from Bristol/Cambridge using UCAS and other data (but only from the National Schools Database of state schools, unfortunately) Tracing the origins of teacher gender bias - ScienceDirect is muddying the waters by bringing in female/male disparities! Gender gap in predicted grades raises questions over their use in university admissions | University of Cambridge, finding that "Depending on the subject, girls were between 3% and 11% likelier than boys to have their grades overpredicted, and between 15% and 41% less likely to have them underpredicted."