eCollegeFinder has a new article on English as a Second Language, How to Educate ESL Students. It discusses two methods of education, "bilingual" (students only use their native language when they study subjects such as math and science) and "immersion" (English all the time; further divided into two methods depending on whether the students study English exclusively for a period at the beginning), then the pros and cons of each method are properly and clearly evaluated.
Here are my thoughts. Firstly, the bilingual method should be preferred if English is taught in a non-English-speaking country; it would be counter-productive if immersion were enforced, unless of course the students are outnumbered by English speaking natives in a small enclosed environment such as an American or British international school. My friend, president of a language school in China, reported that her students used to demand native English speakers who speak no Chinese as teachers many years ago, and slowly changed to bilingual teachers, as they found the latter to be more efficient for their learning. On the other hand, if the students are in an English-speaking country and surrounded by English all the time, the immersion method will definitely win. It's interesting that Texas, where I am now, is among the few that practice this "least popular" method. Some bilingual parents complain about the segregation in school and regret that they report in the school registration form before their kids go to elementary school that they speak xyz (a language other than English) at home. Unlike adults, young school children learn a foreign language better in immersion and should not be artificially isolated.
Secondly, the education science or pedagogy should learn from medical science. Strict control study should be a standard. Whether bilingual or immersion is better, and better under what condition, can be argued verbally based on personal experience and common sense, as I'm doing here. But nothing beats a well controlled study, perhaps followed for four to ten years, with statistical analysis. It may be impractical to implement a single- or double-blind study. Nevertheless, statistical numbers are more scientific than experiential observation.