The point being, if they're strongly correlated, the "why" can already be inferred (assuming adoption of this particular theoretical causal chain).
I agree the movement toward testing is likely to continue unabated. The point was the futility of additional questions; when a strong correlation is already known, the responses can be predicted with a high degree of accuracy, so the additional information is not really "additional". Yes, social scientist prefer multi-item measures and avoiding a single source bias (particularly when publishing theory), but in the end, if the correlation is maintained, not much new is really learned.
(And one of the reasons for multiple election polls is that the results change over time, and leading up to an election, that's relevant. Additionally, the entire electorate isn't polled each time, so the "cost" to the system is lower, relatively speaking.)
Let's make some simplifying assumptions: the survey identifies two causes of poor performance, a lack of academic rigor, and a poor classroom environment (such as a teacher that's mean and unresponsive to requests for help or clarification). Those two categories are weighted equally on the survey - a teacher who gets 100% on the survey is good in both categories, a teacher who gets 0% is bad in both categories, and a teacher has more than one way to score 50%.
So, if both of those factors affect student performance on standardized tests, then there will be a strong correlation between the overall survey scores and the test scores. But analyzing the details of the survey results can offer actionable guidance that the test scores can't - the survey does provide useful information for how mediocre teachers can improve, even when it matches the test scores in predictive power for future test scores.
I agree the movement toward testing is likely to continue unabated. The point was the futility of additional questions; when a strong correlation is already known, the responses can be predicted with a high degree of accuracy, so the additional information is not really "additional". Yes, social scientist prefer multi-item measures and avoiding a single source bias (particularly when publishing theory), but in the end, if the correlation is maintained, not much new is really learned.
(And one of the reasons for multiple election polls is that the results change over time, and leading up to an election, that's relevant. Additionally, the entire electorate isn't polled each time, so the "cost" to the system is lower, relatively speaking.)