The author uses a bad graph to convince the reader that there is no correlation, even when (by his own admission, see the comments) one is present.
The year to year correlation is 0.3. The correlation across percentile groups is much higher because that increases the sample size and thereby reduces statistical noise.
The conclusion we can draw here is that measuring the performance of a single teacher based on a single class will yield a very large confidence interval. That's not the same as "bullshit".
Comments
The author uses a bad graph to convince the reader that there is no correlation, even when (by his own admission, see the comments) one is present.
The year to year correlation is 0.3. The correlation across percentile groups is much higher because that increases the sample size and thereby reduces statistical noise.
The conclusion we can draw here is that measuring the performance of a single teacher based on a single class will yield a very large confidence interval. That's not the same as "bullshit".