The purpose of focus groups is not to provide statistical representation (as in a probability sample), but to explore various cognitive framings of issues in a *social* condition. One of the main problems of conventional surveys is the they attempt to measure beahvior in settings that are very different from those in which that behavior actually occurs. Asking people how they would vote or what they would buy ignores the fact that voting and buying takes place in a social setting that frames the issue and influences the decisions people make. It is almost like studying intimate behavior by observing students gathered in an auditorium.
Focus groups try to avert that problem by re-creating social settings and observing how social interaction in a group affects people's decision. While it is not "representative" in a statistical sense (i.e. the probability of occurrence of the observed trait in the universe cannot be determined) - it is representative in the sense of likely responses to different sets of conditions.
Opinion polls and focus groups are based on very different models of human behavior and even science. Opinion polls assume that preferences governing behavior are given - the purpose of the inquiry is merely to determine their distribution in a population through statistical sampling (a method, btw, first developed in agriculture to study crops). Focus groups, otoh, assume that human behavior is malleable by social conditions, and they try to determine which buttons to push to achieve certain outcomes. The French sociologist Alain Touraine wrote extensively about it, he called this method "sociological intervention" and used it to study social movements. Stated differently, probability sampling is the science of bean counting, while focus group is the science of Machiavelian manipulation, or Herrenwissenschaft (science of the masters) as Horkheimer (or was it Adorno?) dubbed it.
Wojtek