The question is fundamentally about technical fields, e.g., EE; however, any other experience related to top-ten schools is welcome, especially MIT.
I've heard that the acceptance to graduate schools in top-ten schools is often committee-based. In particular, applicants declare their research interests in their SoP, but after the acceptance, they will attend the seminars of faculty members to become familiar with their active research. According to these seminars and any possible lab rotations (like what is provided in Caltech), the student may switch his field from what he already declared in SoP to another.
In contrary, here is an experience declared by a person (at UCB) saying:
What you might not know is that admissions are done by area (at least here at Berkeley). The department can admit N students. Area 1 gets n1 slots, area 2 gets n2 slots, etc, and each area decides which students to admit. Since there are some areas with lots of students applying, it is very hard to get into those areas. Some areas have very few students applying, and thus it may be "easier" to get in. It really works against you though, if you apply to area 2, thinking that it will get you in, and then try to switch to area 1, which is more popular. In the most popular areas, it can be really hard to find an adviser, get funding, etc. That is why they do the admissions by area, they actually admit only as many students in each area as they feel they can support (both financially and with advisers). A few always end up switching, but they do try to keep it balanced.
Is this really the common approach to reviewing SoP (which contradicts the first paragraph especially the philosophy of