I assume that you are talking about something to do with opinion polling or some other thing like that and random sampling.
So, one way that this could happen would be if, for example, you were trying to find out how smart the average student is. If you asked "who would like to take an intelligence test" you would likely get a skewed sample because of the selection effect. Most likely, the students who chose to take the test would be the smartest students and your numbers would be higher than they ought to be.
I hope this is what you are asking.
The difficulty in your problem is that you neither say about your poulation size nor about your sample size.The criteria of poor practice is also undefined. Having defined the poor practise by the response make list of all the poor students.Now the best sample that could be chosen is the quality of maintaining the unbiasedness in your chosing the sample of a perticular size. The quality of unbiasedness means to give equal chance or probability to all the students with the characteristic of poor quality to be or not to be in the sample of the determined size.There are different types like simple random sampling, systematic sampling,stratifed sampling etc to avoid bias.
When the voluntary response is not availabe or not practicable, the determination of poor criteria has to be judged on the performance of the a questionair or questions designed to meet your standard of the quality of practice. Having evaluated the performance you have the list of all the students judged to be with a poor practice , and that list itself is the largest sample of poor practice . And if you want the sample of a particular size, go for a simple random sample or systematic random sample or a stratified random sample so that you can ensure the unbisedness of chosing any of the students among the list made for poor practice.
Voluntary response sample is one in which respondent to a survey or participant in a study elect themselves to become a part of the sample in response to a general appeal. In general the researcher does not have any control over the composition of the sample, and hence the chances of sample being biased are higher. However this method of voluntary response may be particularly unsuitable for some type of behavior. For example, consider a survey conducted to determine the extent and nature of leadership qualities and initiative among students in a school. If sample for this survey is drawn using voluntary response, it is quite likely that students that volunteer for the sample may have higher level of qualities like initiative and being more social, that are frequently associated with leadership effectiveness also. Thus this sample is likely to have a bias that will show higher level of average initiative and leadership abilities than the real average.