Here is how I would inspect schools:
* Collect together a bunch of metrics of each school. Eg. student test scores, parent satisfaction scores, number of police callouts to the school, number of leaks in the school roof. Also include metrics that aren't obviously good/bad: Number of acres of playgrounds, average tenure of staff, etc.
* Gather data of the success of past students, 30-50 years on. For example, employment rate, total earnings, percentage convicted, percentage in good health.
* Build a model to predict success metrics from the school metrics.
* To rate a school, go collect the school metrics, then run through the model to predict future success metrics. That is your rating.
Sure, such an approach has the correlation/causation problem. But this is self-correcting if schools try to optimize their scores as the models are rebuilt each year.
* Collect together a bunch of metrics of each school. Eg. student test scores, parent satisfaction scores, number of police callouts to the school, number of leaks in the school roof. Also include metrics that aren't obviously good/bad: Number of acres of playgrounds, average tenure of staff, etc.
* Gather data of the success of past students, 30-50 years on. For example, employment rate, total earnings, percentage convicted, percentage in good health.
* Build a model to predict success metrics from the school metrics.
* To rate a school, go collect the school metrics, then run through the model to predict future success metrics. That is your rating.
Sure, such an approach has the correlation/causation problem. But this is self-correcting if schools try to optimize their scores as the models are rebuilt each year.