Monday, November 1, 2004

previous entry | main | next entry | TrackBack (0)

Hey, network news producers!! Over here!!!

Joe Flint and Shailagh Murray have a great Wall Street Journal front-pager on the major networks' plans for reporting on the election Tuesday night:

This time around, the TV networks swear they aren't going to make the same mistake again. They say they have revamped the way they collect and analyze polling data, using more sophisticated equipment and better communications. To tone down their competitive instincts in "calling" states for either candidate, some are blocking their news desks from watching rivals' shows. All the networks are also striving to get their respective "decision desks" -- the units that make the calls -- to work more closely with the producers and reporters so information doesn't fall through the cracks.

"The real race is to get it right, not to be first," says NBC News Vice President Bill Wheatley, the executive in charge of the network's election coverage.

Still, always keeping the snafus of 2000 in mind, to get it right and be first is the goal this time around. Toward that end, perhaps the most significant change in tomorrow night's coverage will be the absence of Voter News Service, the now-defunct consortium formed by major news organizations in 1994 to handle exit polls, vote tabulations and projections. The consensus among the networks is that VNS could have done a better job.

For starters, VNS botched vote data in several counties around the country. Its exit-poll samples were seen as too small to be accurate and didn't take absentee ballots into consideration. Its vote tally in Florida was off when compared with the Associated Press's. And while Florida got all the attention, VNS data also led to bad calls in New Mexico and Washington.

In an 87-page report analyzing what went wrong with the 2000 election coverage, CBS News said the computers VNS used weren't sophisticated enough to compare voting data with historical information and were incapable of raising red flags where they were needed. "The old VNS was based on mainframe technology and was probably about 30 years old and wasn't very flexible to update," concurs Dan Merkle, director of ABC News's decision-desk unit.

Of course, part of the problem is that exit-poll data can be unreliable or overinterpreted, especially in a close race. News organizations, though, can't resist the urge to get an early read on a race. Calling elections based on exit-poll data first started in 1980, when NBC declared Ronald Reagan the winner in a landslide over President Carter at 8:15 p.m. Eastern time -- well before West Coast polls closed and much earlier than when CBS and ABC called the race for Mr. Reagan. After that, the competition to call races first began.

In those days, each network had its own exit-polling unit. In 1989, a cost-cutting effort led the networks and Associated Press to form Voter Research and Surveys, the joint venture that preceded VNS, to do the polling for them. Warren Mitofsky, a CBS executive, was tapped to head the venture but had left by the time VNS was started.

Now Mr. Mitofsky, a veteran pollster, is back and working for National Election Pool, a new consortium of Viacom Inc.'s CBS, News Corp.'s Fox News Channel, Walt Disney Co.'s ABC, General Electric Co.'s NBC, Time Warner Inc.'s CNN and the Associated Press. NEP hired Mr. Mitofsky and another polling veteran, Joe Lenski, who heads Edison Media Research.

The two main tasks of the old VNS -- collecting actual election results and conducting exit polls -- have now been separated. Edison Media Research and Mitofsky International are conducting exit polls and collecting returns from sample precincts -- early indicators that give media organizations an idea of which way a particular state or city is headed. The Associated Press is separately responsible for reporting vote counts as they are tallied at county election sites.

If the new system works the way it is supposed to, throughout Election Day, Edison and Mitofsky interviewers will speak to voters as they leave about 1,500 precincts, asking them whom they voted for and why. The "why" part will be analyzed later, but the "whos" will be tallied and shipped out to give news organizations a first look at where the race is headed.

As soon as the polls close, actual returns will flow from more than 3,000 sample precincts in the 50 states -- a much more accurate early indicator. The pollsters will fold into that data the results from pre-election-day telephone polls of early or absentee voters in 13 states -- 10 more states than in 2000, reflecting the increased prevalence of early voting and voting by mail.

Meanwhile, the AP will dispatch 5,000 stringers to county election sites, who will phone in official returns to 16 vote-collection centers. The first states to report will be Indiana and Kentucky, where polls close at 6 p.m. Eastern time. But in hotly contested states, or states with balloting problems, results could take many hours, even days, to dribble in.

One significant change from four years ago, says Mr. Mitofsky, is that the vote-counting database will automatically compare live AP data from the county election offices to increments that the news organization had previously reported. That should help to avoid debacles such as when Volusia County, Fla., reported a sudden vote surge for President Bush in 2000 -- a "red flag" indicating a glitch that would immediately jump out this year, Mr. Lenski says, because the patterns would be clearly laid out in the data, highlighting any aberrations. "In 2000, there was no way to notice that," Mr. Mitofksy says.

As the information flows in, Messrs. Mitofsky and Lenski will plug it into their computer system and pass it through models to produce calculations. The pollsters will review the results and make judgments. But in a major change from the VNS system, the news outlets will be able to see precinct-level data in real time, giving them access to the early numbers that they couldn't see in previous years. "They may make different judgments than we make about the very same results," Mr. Mitofsky says.

I have a humble request for the nets -- show us how the sausage is made. In other words, instead of hiding the data from the exit polls from us, explain as the returns come in what the polls say and compare and contrast them to the incoing returns.

[Won't that be kind of... dull?--ed. It would still be much more interesting than Tim Russert and his f@#$ing midget whiteboard, or Dan Rather and his nonsensical similes.]

UPDATE: Some network should really hire myster pollster Mark Blumenthal to explain how the sausage is made -- go read his infomative post on the merit of exit polls.

posted by Dan on 11.01.04 at 05:49 PM


Can someone please explain to me what practical difference it makes whether the election is declared won at 11 p.m., 2 a.m. or 8 a.m.? Perhaps it makes a difference to someone shorting the yen, but the whole exit poll-breathless extrapolation et al reminds me unfortunately of the poor souls who call sports talk shows in the 14th week of the NFL season to discuss all the possible combinations and permuations of who will play whom in the first playoff round and who gets a bye. Too much free time and too much computing power have brought us Everests of meaningless academic research, here they exist to fill media whitespace and occupy the reserve army of the unemployed punditry (a lumpen group of the worse sort).

posted by: gene on 11.01.04 at 05:49 PM [permalink]

Dan, I hate midgets, too.

posted by: jonk on 11.01.04 at 05:49 PM [permalink]

Also, we already now what Russert is going to write on the white board Ohio Ohio Ohio. (Maybe a brief mention of FL and PA too.)

posted by: Joel B. on 11.01.04 at 05:49 PM [permalink]

I vote for a media blackout. b/c of the obnoxious factor.

posted by: jason on 11.01.04 at 05:49 PM [permalink]

I agree with Dan. Sunshine is the best disinfectant. Take the mystery out of it, and let everyone have fun.

We ain't going to get rid of sausage.

posted by: Hal on 11.01.04 at 05:49 PM [permalink]

I wonder how--if at all--the exit pollers are going to factor in the millions who voted early and who won't be walking out of the polling stations?

Since early voting is a new phenomenon, there's no track record to compare against. Are the majority of early voters Republican or Democrat? Do they break as evenly as other voters in the same precincts? How do we know?

It strikes me as seriously dangerous (in terms of once again getting it wrong) for the media to not be dealing with early voters in predicting winners.

posted by: John on 11.01.04 at 05:49 PM [permalink]

Transparency is the key. I personally just want to hear a concession speech tomorrow night from somebody. I don't want any of this month-long drawn-out process. Ideally, networks' polls support the results eventually reported, and then the system looks good and no one asks questions.

It's only when it doesn't work that we begin to question it all. I personally think Kerry is much stronger than polls show. Young voters, new voters, minority voters I think will all turn out in record numbers. Then the networks will have an easier race to call.

posted by: theDamascus on 11.01.04 at 05:49 PM [permalink]

One thing I've not seen is a more sophisticated approach to looking at the results. Here in Australia, the ABC's election guru Antony Green has built up a model of the electorate down to the polling booth level. As results come in from a given polling booth, he's able to compare them against the same booth from previous elections, producing much more reliable predictions from a much smaller set of results. Is there any reason that something like this wouldn't work in the US? (For instance, if the results are not given in that detail).

posted by: anthony baxter on 11.01.04 at 05:49 PM [permalink]

> "The old VNS was based on mainframe
> technology and was probably about 30
> years old and wasn't very flexible to
> update," concurs Dan Merkle, director
> of ABC News's decision-desk unit.

Um, if you absolutely, positively have to crunch some critical numbers with 100.000000% accuracy, a mainfame-based system is EXACTLY what you use. Perferably a DECSystem-10/20, but and IBM 390 will do.

What you ABSOLUTELY don't want is some cobbled-together MS-Windows-based dreck coded by 20somethings who don't understand the basics of high reliability systems and who calls any technology older than the day he started kindergarten "archaic".

That single snippet leads me to question the reliability of the entire process.


posted by: Cranky Observer on 11.01.04 at 05:49 PM [permalink]

Post a Comment:


Email Address:



Remember your info?