1. by including only the frat houses you have biased the student population. If you want to know how many a student consumes you need to get a cross section of all students. If you only want to know how much a frat student drinks then you need to change your question.

2. having law enforcement administer the test will eliminate many students because they may be under-age or afraid that the law enforcement officers will use the survey to target the next party.

A reduces randomness in the survey

B causes a skew in results (reduces randomness by surveying a population within the population)

C causes dishonest answers

D causes a skew in results (reduces randomness by surveying a population within the population)

Edit: Making sure your survey proportions are identical to the population proportions is an invalid and incorrect way of conducting a truly random survey folks. Random means random. A survey that is not random is not a reputable survey.

It may reduce randomness a bit, but it maintains the makeup of the sample group and would be far more representative of the whole than B or D. C would just be bad.