When creating a poll to judge the popularity of particular Congressional candidates, it's important first to determine which candidates will be part of the poll and which districts they represent. Knowing the boundaries of the district will keep the pollster from samplings outside of the requisite voting area.
Generally, the wider the net the pollster casts across the entire district and the greater number of voters sampled, the more accurate the poll will be. This includes sampling voters from different neighborhoods within the district, different ethnics groups, different age groups, different sexes, and different religions, among other factors. By oversampling, say, young Latinas, the pollster may get a very different reaction than if all sexes, ages, and ethnicities are sampled. Naturally, the makeup of the district will have an effect on whether certain groups are more heavily sampled than others.
Two primary sampling methods are commonly used. The first is called a "probability sample." In this method, every person in the defined population (in this case, the Congressional district) has an equal chance of being polled. One major advantage to probability sampling is that the pollster can fairly adequately determine how representative the polling sample correlates to the overall population. In other words, pollsters can determine accuracy within a certain margin of error.
The second method is called "non-probability sampling." In this method, the population sample a) does not give everyone an equal chance of being selected; b) is not selected randomly; and c) is not known to the pollster beforehand. Rather, participants are selected based on other means, such as volunteering, or "opting in," to the polling survey. Unfortunately, this method might wind up oversampling groups who are, for example, more politically active, and statistic models must be used to estimate margin of error, since there is no simple way to calculate it using this method.
Therefore, using a probability sample will typically give the most representative response of the population as a whole. Examples of probability samples include "random-digit dialing" (RDD) and "registration-based sampling" (RBS). The former involves calling all numbers with proper area codes and exchanges within the district, while the latter includes contacting registered voters culled from voter lists. RDD is more expensive, but it can often give a more accurate depiction of "likely" voters than the RBS system can.
Data from the polls commonly include a number of important data points, including a) the number of respondents polled; b) the total population they are meant to represent; c) the percentage of responses giving one candidate or another particular favorability ratings; and d) the margin of error showing how likely the polling group is to represent the population at large. Data could also be broken down into more specific sub-categories, such as age, race, religion, or sex.