There are 3 grade levels and 9 subjects giving 27 possible groups. It seems entirely inappropriate to pick out 3 and highlight the fact that (on a different test of students studying a different curricula) the results went down.

Especially when the NCES Data Usage Agreement clearly states "Use the data in any dataset for statistical purposes only."

As an example 4th grade reading has on a scale from 0 to 500 been between 213-223 for 20 years with of course this year (215) not being the lowest point. This is a test where many jurisdictions have 10+ point swings from test to test. There really isn't that much signal in the noise here.