The place you will start is the Elections Board or Registrar for each of the counties represented in the Congressional district. The Board will provide to you voter information, which includes critical data such as addresses, how often a person votes, in what primary (Democratic or Republican), and general descriptions of the voter such as year of birth.
Whether it is a primary or general election, looking at the voting patterns for each voter over ten years will give you a clue as to what party the voter leans. The objective of obtaining voter data is to predict who will vote in the election and what party they are most likely to vote. All the votes, including municipal, county, off-election years, and even special option tax votes, will indicate how likely someone is to vote.
For example, Joe Voter has voted in a Republican primary six times in the past ten years. His voting record indicates Joe has never missed voting in a general election and votes in all of the municipal and county elections. Sue Voter, in the past ten years, has split her vote between Republican and Democrat primaries. She seldom votes in municipal election or county elections, but votes in every general election. In analyzing the voting patterns of Joe and Sue, their voting history suggests Joe will vote Republican, and Sue is leaning towards being unaffiliated with either party; she is more Independent. Their history indicates they are regular voters, and unless some unforeseen catastrophe occurs, they will show up on election day to cast a vote!
How do you know who to include or exclude? Looking at the general voting statistics for each county in the district, you select the ten-year average percentage of votes cast by each party. If the district average party voting percentage is 54% Democrat and 46% Republican over the ten years, then you can expect unless there are outliers, the party voting pattern will be similar. For each voting precinct in the county, you will verify the median trend with the actual for each precinct. Once you have collated all of this data, you conduct the poll by the percentage of the party of voters and the size of the voters in the population. It sounds complicated, but it isn't!
For example, if Voting County 1 has 10,000 predicted voters and it votes 35% Democrat and 65% Republican, your polling population to draw from your sample size should represent 3500 Democrats and 6500 Republicans. You do not need to contact all of them. For most political polls, you want to reach between 5% - 10% of the total, keeping in mind the median or average percentage of voters by party affiliation. The most efficient way is to use the phone. However, that has become increasingly difficult as many voters have unlisted cell phones. Polling organizations will have to canvas some precincts door-to-door to make personal contact. Since the Elections Board data included addresses with the voters's names, many candidates collate this data with the phone survey to get a better picture of the potential voter. You also, to the extent you can, make sure your poll is demographically representative by age, race, and gender, if possible. Sophisticated polling can take aggregated data and slice the data into small meaningful units.
Modern polling only differs from the election polling of thirty years ago by the use of technology. The methodology remains consistent. However, polling does encounter glitches, and predictions sometimes are less than accurate. This is why many candidates use rolling polls or poll the electorate every few weeks up to election day. The key to winning an election is getting your supporters to turn out and vote, while discouraging your opponent's supporters from voting.