From reader Alex Lee-MacDonald:
Exactly, actual Education experts like Prof Mark Priestley and Dr Marina Shapira argue that the big “decline” in Maths in PISA came in 2006 before the introduction of Curriculum for Excellence- see https://mrpriestley.wordpress.com/2019/12/06/what-do-the-pisa-results-tell-us-about-scottish-education/. Supposed declines since then are not statistically significant (maybe UK education systems should be better at teaching journalists statistical literacy!)
The declines between 2003 and 2006 actually seem pretty odd to me and I would probably guess they are due to sampling methods.
PISA is clearly seriously flawed and I was disappointed the Institute for Government gave it so much weight (though to be fair the actual report is more nuanced than the reporting of it suggests).
As mentioned in another comment above, it looks like England’s good performance in Maths in 2018 was likely due to sampling bias (and there is hard GCSE data to show that the sample was definitely biased). It appears England has been including too many schools with high attainment and too few schools with low attainment in its sample. https://www.theguardian.com/education/2021/apr/22/maths-scores-in-world-education-rankings-inflated-for-england-and-wales-study
I imagine this issue (and similar) is widespread in the PISA study (the actual study reported in the Guardian also criticised Scotland’s sampling methods but there was no hard data to show a definitely biased sample like there was for England and Wales). Given there’s no OECD post-PISA analysis (once exam results are known) to find out if samples were actually biased, it doesn’t look like OECD are interested in fixing this issue.