By Thomas Lang
Before next Tuesday’s New Hampshire primary, it’s worth taking one last look at how the networks handled the results from Iowa this past Monday night.
That’s because the Iowa Democratic caucus marked the debut of the National Election Pool (NEP), a new polling operation that involves the Associated Press, Fox News Channel, CNN, ABC, CBS, and NBC. The system replaces one that no one was happy with — the one that led to CBS’s Dan Rather miscalling the 2000 Presidential election.
And yet in Iowa, each network presented its viewers with sharply varying degrees of information. Were they really reading from the same script?
Well … yes.
As the 8’oclock hour wore on, Fox News, as well as MSNBC, chose to use specific numbers from the entrance polls, while CNN, CBS, and ABC ranked each candidate by standing, without providing any numbers. Although none of the networks seriously miscalled the Iowa caucus, a second look at Fox News’ broadcast, as well as a look at the process as a whole, shows that there is still ample opportunity for the networks to misread the data and air a premature conclusion.
The old system, the Voter News Service (VNS), also a project of the same six media outlets, was abandoned after the 2002 midterm election when it failed to report the exit poll and final vote results in a timely and accurate manner. Problems within the VNS system led more than one network to fumble its prognostications on election night, 2000.
To replace VNS, the networks created the NEP, which acts as a clearinghouse for both entrance/exit polling and final vote results. Edison/Mitofsky, an outside organization contracted by the NEP, conducted the Iowa entrance poll which merely reflected the “initial preference” of caucus-goers as they entered the intimate meetings.
Edison/Mitofsky interviewed randomly selected caucus-goers as they entered the polls at 50 precincts. The company then manipulated these raw numbers using eight to twelve statistical models to offer the networks a whole menu of numbers, based on various considerations such as past voting history of the precinct and demographic considerations. This information was then made available to all the networks at the same time.
Here’s an example: Imagine a Democrat running versus a Republican for elected office in a state that contains only two precincts. Precinct A always votes Republican. Precinct B always votes Democratic. If Precinct A managed to call in its initial vote two hours before Precinct B, then the raw data would show the Republican way ahead of the Democrat. However, if the data is filtered through statistical models that consider past voting tendencies, the manipulated data would produce a more accurate picture of the race.
Even though all the networks received the same information through the NEP on Monday night, it was still left to each network to either report specific numbers spit out by one model or another, or to cautiously characterize the race using no numbers at all.
On Monday night Fox News led off its dedicated caucus coverage at 8:00 p.m. EST with its own figures, drawn from one or more of the statistical models, giving the results as Kerry 29 percent, Edwards 22 percent, Dean 21 percent, Gephardt 16 percent. At 8:40 PM, Fox News gave its final report from the entrance polls, reporting Kerry 31 percent, Edwards 24 percent, Dean 21 percent, and Gephardt 15 percent. Six minutes later, it shifted to preliminary numbers compiled by AP that reported Kerry at 36 percent, Edwards 34 percent, and Dean 18 percent (it did not report a number for Gephardt).
These numbers were remarkably close to the final polling results, with Kerry winning 38 percent of the county delegates, Sen. John Edwards 32 percent, Gov. Howard Dean 18 percent, and Rep. Richard Gephardt 11 percent (the county delegates will choose the state’s delegates to the Democratic Convention).
MSNBC also reported specific numbers, but apparently chose its numbers from a statistical model (or models) that more closely resembled the final vote count.
Fox’s reports turned out to be both accurate and inaccurate. On one hand, John Kerry finished in first place, with Edwards in second, Dean in third, and finally Gephardt in fourth. On the other hand, the early entrance numbers Fox chose to provide its viewers missed the mark, at first by as many as 10 points, then later by as many as 8 points. Had Fox chosen to withhold the numbers and merely describe the developing race through the relative placement of the candidates, Edwards’s eventual strong finish and Dean’s weak finish would not have appeared so volatile. When contacted about their editorial process, Fox declined to comment.
Dan Merkle, decision desk director at ABC, said the network typically keeps numbers from statistical models off the air and waits to report “tabulation of the [actual] vote.”
Similarly, CBS first interrupted regular programming at 8:24 p.m. to present, “a characterization based on entrance polling estimates,” according to Sandy Genelious, a spokeswoman for the network. CNN also followed the procedure of reporting the entrance polling without providing specific numbers. All the networks were careful to note that these projected results were preliminary and subject to change.
By 9:24 p.m., CBS felt confident to call the race a victory for Sen. John Kerry. CBS based this decision on earlier-than-expected reports from the Iowa State Democratic party that corroborated with the Edison/Mitofsky entrance polls. The rest of the networks followed CBS’s lead in due time, and coverage quickly switched to analyzing the actual results.
So, in the end, in its first trial the NEP system functioned with few hiccups. However, it’s clear the system leaves open the possibility — and the temptation — for a network to rush to judgment based on any one of several statistical models, which may or may not bear a resemblance to the final results.
That’s the danger for the future.