Last fall, we witnessed a Presidential election that was closer than anyone could have imagined. Its closeness brought to light a series of flaws in election procedures and in reporting on the electoral process to the American people. Those flaws included errors made by the Voter News Service, of which I am Executive Director.

As dedicated professionals who have tabulated, analyzed and reported on thousands of elections, my staff and I have spent this post-election period working to understand precisely how those errors occurred and how to prevent them in the future.

The electoral process is a cornerstone of our nation. Reporting on the culmination of that process is a serious responsibility, and we owe the public an explanation of the mistakes we made last November. As individuals who have devoted much of our lives to educating the American people about elections, I assure you that we feel that sense of accountability very keenly.

At the outset, there is one matter that I would like to lay to rest. In reporting to our member news organizations and, indirectly, to the American public on Election Night, we have one paramount concern: reporting and analyzing the results of the election as accurately and quickly as possible. The notion that some kind of political bias enters into our work is, quite simply, without any foundation, and I am pleased, Mr. Chairman, that that appears to be common ground among those of us here today. I also appreciate the assurances you have given that this process will not offend the free speech principles that we both must clearly defend.

The Voter News Service (VNS) was created in 1993 and is owned by ABC News, The Associated Press, CBS News, CNN, Fox News, and NBC News. They are among the leading news organizations in the world and are committed to the highest standards of journalism.

The purpose of the Voter News Service is to collect, tabulate, and disseminate vote returns, exit poll data, and projections of presidential primaries and national and statewide election contests. That information is distributed to our six member organizations and to other subscribing news outlets. These news organizations take the data provided by VNS, conduct their own analysis and interpretation, and report it to the American people as they see fit.

Our organization makes possible the timely reporting and in-depth interpretation and analysis of election results that the American people have come to expect and rely upon. On Election Day 2000, our work involved more than 40,000 people who staffed nearly 28,000 individual precincts and went to some 4,600 counties to obtain the information that we needed.

In national elections, the final VNS National Vote totals, which are verified with official state canvases, become a record of the election that is widely published. The exit polls that we conduct are widely recognized as a critical source of information for understanding an election. They are used by students, scholars, officials, and journalists throughout the world. They tell the public, among other things: who voted and why, what issues mattered most to the voters, which candidates' policy positions were most effective, and which candidates' qualities attracted voters most.

In addition to providing information to analyze election results, on Election Night VNS projects the outcome of contests to its members and subscribers. VNS' projections are based on complex statistical analyses that take many factors into account including, among other things: the actual vote in sample precincts, tabulated vote at the county level, and the exit poll. All of this data is reviewed and interpreted by VNS analysts who add their own knowledge and experience before making a decision that it is possible to project the outcome in a given race. Projections are made by people - not by computers.

Since 1990, when the first joint polling and projection effort began, we have been involved in nearly 900 election contests across the nation. The methods that we use to project winners in those races have only been wrong once before. In other words, we have been right 99.8 percent of the time.

Despite our strong record of accuracy, we constantly strive to eliminate all errors.

In between elections, our staff evaluates the performance of our methods and models and considers how they might be improved. The models are based on the accepted statistical theory of sampling, the principles of which have not changed. Nevertheless, over the years, we have made improvements in the models and procedures. For example, we routinely research our sample precincts prior to an election, in order to take into account changes in precinct boundaries and demographic composition. This year, our decision screens for California and Washington were revised in light of a significant increase in absentee voting which has been evident in those states. In this regard, since 1996, we have conducted telephone polls of absentee voters in states where a high percentage of absentee ballots are cast, and we continue to work to improve the methods for polling absentee voters.

Unfortunately, when you make an error as glaring as calling Al Gore the winner in Florida at 7:52 p.m. on November 7^th, the number of times that you have been right seems less relevant. The plain fact is that, despite our best efforts, the Voter News Service let down its members, subscribers, and ultimately the American people, on Election Night 2000. We are determined never to let that happen again.

Toward this end, VNS has conducted an intensive internal investigation of what went wrong on Election Night - an inquiry that is still continuing. In addition, our members commissioned an independent review by the prestigious Research Triangle Institute. Several of our members have also conducted investigations of their own.

On Election Night our statistical models, based on our exit polls and actual vote from a number of sample precincts, showed Vice President Gore ahead - decisively it seemed - in Florida. Our decision team considered other variables, including absentee vote beyond that which already was accounted for in the models, and determined that the data clearly justified making a call, which we did shortly before 8:00 pm. The reality, however, is that the race was a virtual tie.

Based on all that we have learned since then, the error in Florida was due to the convergence of a number of the anomalies to which all polling and projections are subject, which in this case all pointed in the direction of a Gore victory. None of these factors alone would have caused the error, but, taken together, they did. These factors include:

· the exit poll, showing Gore ahead;

· the fact that the actual vote from the first sample precincts reporting indicated that the exit poll was actually understating the Gore vote;

· the fact that the model selected the 1998 gubernatorial election in Florida, rather than the 1996 Presidential race, as a basis for statistical comparison, when comparisons based on the latter would have prevented us from making the call; and

· a larger than expected absentee vote.

Later, after the Gore call in Florida had been retracted, we discovered problems in the tabulation of actual Florida votes that led to the race being called in favor of President Bush. In one case, Volusia County, VNS passed on incorrect numbers that were released by election officials, and this went undiscovered until after the Bush calls had been made. Moreover, we significantly underestimated the number of votes outstanding.

Based on this experience, and following the recommendations in the RTI and other reports, we are actively pursuing numerous improvements, including the following:

· using larger sample sizes for exit polls;

· developing new procedures to account for the effects of a growing absentee and early vote, including more extensive telephone polling of absentee voters;

· rewriting of the VNS projection and statistical models;

· working to improve the exit poll accuracy and response rate;

· completing work on the integration of The Associated Press's tabulated vote as a second source of information;

· developing more sophisticated quality control in the tabulated vote system and the rest of the VNS systems; and

· upgrading and modernizing the VNS technical capabilities and infrastructure.

We are taking these steps because, as journalists, we are deeply committed to the integrity and accuracy of our reporting. We are determined to do everything humanly possible to make sure that these mistakes will never be made again.

Our intention is to take all that we have learned and use it to improve our Election Night procedures - and thereby return to the American people the confidence that they will receive timely and accurate Election Night information.

Thank you, Mr. Chairman and Members of the Committee.