In the aftermath of the 2000 election, and the mistakes made by NBC and the other television networks in calling the results of the election, NBC News has made important changes in how Election Night data is collected and analyzed.
The old election data consortium — Voter News Service — was dissolved and the partners (NBC, ABC, CBS, CNN, Fox and the Associated Press) formed the National Election Pool (NEP) to collect exit poll data and provide tools for data analysis.
Via NEP, NBC News will continue to use exit polls, precinct votes, county votes and statewide votes to project races, but this year’s computer models have been completely overhauled, modernized and fine-tuned so that analysts have much more detailed information to make judgments about how the election is likely to turn out in any given state.
The Associated Press will conduct the actual vote count with built-in mechanisms to lower the possibility of error. The AP has added a completely new quality control procedure to test the votes before they are entered into the system. This will help prevent incorrect county vote reports from inclusion in the NBC systems.
NBC News will not project a winner in a state until after the last scheduled poll closing time in that state. If the election appears to be close in any given state, an abundance of caution will be used before calling a race in that state.
The increase in absentee and early voting has required a thorough re-evaluation of election projection methods to ensure that absentee voters are properly accounted for. This year — via NEP — 13 absentee ballot polls have been conducted, compared with three in 2000.
The head of NBC News’ decision desk, the decision analysts and the quality control desk will be isolated from the calls of other networks. In addition, the entire decision desk area has been physically isolated.
Understanding Election Night calls
NBC News will be clearer about its nomenclature during the entire Election Night broadcast. In addition, NBC News will explain methodology for collecting data it presents. Here are some terms you will hear on TV and read on the Web on Election Night:
“Projected winner”: NBC has made a projection that a candidate will win the race, but the vote count is not complete. This call is made only after all the polls are scheduled to have closed in a state. Projections are based on one or more of the following types of information: exit polls, precinct models from scientifically selected sample precincts, county vote models using data from the Associated Press and/or analysis of the actual raw vote.
“Winner”: A candidate who has clearly won the race, beyond the normal margin for a recount. “Winner” will not be used unless returns make the outcome a virtual certainty. Note that neither of these terms refers to the “official” winner, since most states take weeks to certify a winner in an election.
“Too early to call”: There is not enough data in the NBC News decision computer systems to allow analysts to make a call.
“Too close to call”: While there is data in the systems, the numbers are too close to allow analysts to make a call.
Understanding Election Night results on MSNBC.com
As part of NBC News, MSNBC.com receives data from NEP and directly from the Associated Press. NEP gathers two sets of data on Election Day: Exit polls, conducted by NEP, and projection information which is used by NBC to “call,” or project, a race once all the polls have closed in that state. The Associated Press collects unofficial vote tallies, as reported by state governments after the polls close.
For exit polls, voters leaving the polling place in selected voting districts are handed a questionnaire with both demographic (sex, race, age) and attitudinal questions (Do you approve or disapprove of the way Congress is handling its job?). They are also asked to indicate how they voted in the presidential, Senate, House and gubernatorial races, and for selected ballot propositions. Those responses are collected and then transmitted to the NEP data center in Atlanta, where the results are tabulated and reported.
NEP conducts exit polls in every state except Oregon, which votes entirely by mail. There is an absentee ballot poll for Oregon.
For vote tallies, results are gathered from election officials in each state as the vote is reported. These are “unofficial” returns because the tallies have not been rechecked and certified by state election authorities.
MSNBC displays the election results two ways — in summary tables and in breakdowns by candidate. Summary tables are used to show overall results for president, Senate, House and gubernatorial elections. This is the summary table for Senate:
The Current column shows the current composition of the Senate. The Holdover column shows the seats not at stake in this election. Seats won shows how many seats have been won by each party. And the New column shows the projected post-election composition of the Senate. At the beginning of the night before any results are reported, the numbers in this column reflect the seats that are not at stake in this election.
The row called TBD shows seats that are at stake but for which results have not yet been reported. As the night goes on, seats shift out of TBD and into Democrat, Republican or other parties. For the House of Representatives, the projected post-election composition also has a margin of error expressed as plus or minus a number of seats.
The House and Governor summary tables are similar to the Senate table.
How to read our exit polls
Two types of exit polls are being conducted on Election Day. One is a national poll, with surveys being conducted in key precincts nationwide. The other type is a series of state polls — one for every state except Oregon. Exit poll results are displayed on national pages and state-by-state pages.
On each survey, the answers to both demographic and attitudinal questions are displayed in the left-most “% of total” column. In the example below, 51 percent of the respondents were men and 49 percent were women. This column will normally total 100 percent, within some rounding error. In sum, the left-hand column should be read down.
Respondents in the state exit polls are asked which candidate they voted for in the presidential, and if relevant, Senate and gubernatorial races. In some states they are also asked about selected ballot initiatives. Respondents in the national exit poll are asked the party affiliation of the congressional candidate they voted for in their districts. The totals of the demographic and attitudinal questions are then further broken down, or “cross tabulated,” by candidate in the “% of category” column.
In the national exit poll example to the left, 41 percent of the male respondents voted for “candidate #1” in their district and 58 percent voted for “candidate #2”. Of the women, 52 percent voted for “candidate #1” and 46 percent voted for “candidate #2”. These rows will normally total 100 percent, unless a significant vote was captured by third-party candidates. (See outlined row.)
From these results, inferences can be drawn about the electorate as a whole. For example, “candidate #1” seems to be more popular with female voters than male voters. How accurate this inference is depends primarily on a) how well the respondents reflect the attitudes of the electorate at large, and b) how large the sample is. Proper random sampling techniques are required to ensure the former; for the latter, the larger the sample size, the greater the confidence in the data.
For most of the state exit polls on these pages, the uncertainty is plus or minus 4 percent, which means that the “true” values for the entire electorate are almost certain to be within 4 percentage points of the results tallied for the sample. (“Almost certain” is generally taken to be 95 percent certain, which means that if the poll were repeated 100 times, in 95 cases the true values would be within 4 percentage points of the sample values.) The national exit polls, because of a larger sample size, have a lower uncertainty of plus or minus 2 percentage points.