In other words, under about-as-perfect conditions as can be had, the correlation was—phew!— positive but still really weak.
- Maybe three, but certainly four observations of a teacher in a year are necessary to correlate consistently with performance—simply because the MET team found huge variations in teacher performance from one observation to another which connected to instructional content, not the abilities of the evaluators;
- Observers need not only to be trained but certified as to their ability to consistently rate the same lesson the same way (which doesn't seem possible without using a video library); and,
- The district has to employ a method to randomly check the accuracy of evaluations to maintain integrity.
In presenting these findings, the MET team of researchers makes a point of returning to its last set of findings, reminding us of the much more powerful finding from the last go-around, that the best measure next to value-added scores is to survey students about their own teacher's performance (and yes, the MET team shows these surveys can be designed fairly).