At the risk of prolonging a derail:
It is important to distinguish between research on one hand, and service improvement/audit on the other. The former is done to gain generalisable knowledge, and requires ethical approval and informed consent. The latter is done for local purposes, and does not require ethical approval or informed consent.
Intentionality is therefore key here, as it often is in ethical issues. I have two scars, of about the same size, and both caused by edged weapons. One was made by a surgeon, and the other by a member of a Glasgow street gang. One of these had altruistic intentions: the other, not so much.
RCTs are tricky in educational settings, since there are so many uncontrollable variables. One solution is to recruit volunteers, who can then be allocated to different 'equipoised'* arms of a study. This obviously raises a sampling issue, regarding who volunteers, and stratified sampling or some other sampling approach may be necessary. Small incentives may also be helpful. A cross-over design may be helpful to address equipoise. In this, the study cohort is divided into, say, Groups A and B, and Group A is exposed to Condition 1 for half the time, then to Condition 2, while Group B experiences the opposite.
Observational studies analysed by methods such as multi-level modelling (Wikipaedia is helpful) are useful in education, where many variables can be treated separately.
It may emerge that an audit of service change suggests a generalisable conclusion, even though it was intended for a local purpose only. There might then be an ethical desire to publish it, even though it wasn't originally badged as research, and did not receive prior ethical approval. In these circumstances, it may be appropriate to seek retrospective approval from an ethics committee, to confirm that publication does not breach Helsinki principles.
https://www.who.int/bulletin/archives/79(4)373.pdf
Ethically-approved studies need not require informed consent either. Under particular circumstances, as long as participants are anonymised and can suffer no possible harms, a study can be carried out without consent (on blood samples, for instance, obtained for other purposes). In the extreme, a
deception study can receive ethical approval, where the expressed aim is not the true aim, but it is hard to persuade ethics committees of the necessity for this.
Changes to teaching practices in schools generally come under the heading of 'service improvement', and parents may or may not subscribe to these. There may be a body of research evidence in support of the change. There may be the opportunity to withdraw a child from one setting and send them to another (frying pan/fire), but this is not always a realistic option. The best strategy for individual parents, perhaps, is involvement through School Boards etc.
As an aside, as someone whose first degree was in physics, and who worked in lab research for many years before moving into social sciences research, I would not speak slightly of social sciences research in general. Indeed, often the problems are more difficult and require more sophisticated means of study design and analysis than those of the physical sciences.
*equipoise, roughly speaking, is the requirement that one experimental condition should not be
known to be better than the other.