"Incumbent leads by 8 points" declares the headline. The story goes on to describe how there would be a "comfortable" 8-point win for the incumbent if the election were held today. Another campaign season has opened, another round of TV and newspaper polls has started and perhaps another round of misleading poll stories.

If media polling suffers today, it is from a straw poll mind-set that polls must project likely outcomes. News directors and editors want to reduce the multitude of statistics produced in a poll to a single easy-to-understand horse race number on who is ahead and by how much.

Voter preference data should not be taken at face value. What the data mean is always far more important that what the data say. Rules of analysis are necessary, rules that are not as shortsighted as "8 points is comfortable" and "2 points is a close race."

Those who hear of an 8-point lead are led to believe that's what will happen on election day. In order for the 8-point lead to be sustained, a number of conditions must be in place: for instance, no change in voter recognition of the candidates, no new issues which may affect voter attitudes and, most importantly — based on simple arithmetic — the undecided vote will split evenly. The widely believed theory that undecided voters will divide equally or even proportionately does not seem to fit many elections. Races that include incumbents are subject most often to misleading analysis because very often the undecided vote does not split evenly. If an incumbent's 8-point lead means 53 percent to 45 percent with 2 percent undecided and it is two days before election day, then it may very well be a "comfortable" lead. If it is two months from election day and the challenger has not yet begun to campaign, then 53 percent is not so comfortable and any news reporter should recognize that a rather slim margin must be protected.

More to the point: If the 8-point lead is based on a 45 percent to 37 percent poll outcome, the incumbent could be in very serious trouble. In two-way races that include an incumbent, it is far more meaningful to analyze preference for the incumbent in relation to 50 percent and not the percentage voting for the challenger. Why? The majority of undecideds usually go to the challenger, especially lesser-known challengers who later become better known.

After having polled enough elections to develop an understanding of why elections turned out the way they did, we can still only theorize why this happens. It is reasonable to conclude that indecision must be an expression of doubt about the efficacy of the candidate the voters know best — the incumbent. After an incumbent has served many years in office, the key question should focus on the true disposition of the voters who haven't made up their minds months or just weeks before election day.

Analysis of poll results involving incumbents in two-way races should assume a threshold of 50 percent for incumbent vulnerability. All too often incumbents will wind up on election day with little more than the percentage of respondents stating they would vote for them in a poll.

Analyze past two-way races where incumbents have fallen below 50 percent in polls within two months of election day, and you will find that all were in trouble before or on election day: Percy, 1978; Bilandic, 1979; Carter, 1980; Carey in Cook County, 1980; Thompson, 1982; Missouri's Danforth, 1982; Percy again. 1984; Elrod in Cook County, 1986; Wisconsin's Earl and La Follette, 1986.

It is interesting that even candidates are telling reporters of this rule, but to no avail. In her two-year campaign to regain the mayor's office, Jane Byrne kept saying that as long as Mayor Harold Washington was slightly under or over 50 percent in the polls, the race was very close. The Chicago Tribune poll showed him at 52 percent and stressed how he would pick up very few undecideds. So did most other polls, but the media's stories stressed Washington's "significant edge" or "comfortable" margin. He ended 1 point better at 53 percent, which means he picked up just 1 point in undecideds. The same scenario was repeated a month later against two challengers in the general election: same poll results, same election outcome. Another example: In 1982 Illinois Gov. James R. Thompson expressed this rule when he said that he must be well above 50 percent in polls for confidence. This rule tells us what to expect in terms of final vote magnitude for the incumbent: The incumbent will end up close to the magnitude expressing a preference for him in polls conducted late in the campaign.

In 1978, despite flip-flops in voter sentiment in the last month of campaigning, the final WBBM-TV (CBS, Chicago) phone poll had U.S. Sen. Charles H. Percy 2 points away from the final election outcome. In 1979, when the low-profile incumbent mayor of Chicago, Michael Bilandic, had been in office just two years, the rule apparently did not apply; the final WBBM-TV Newspoll of 47 percent anticipated the exact Byrne winning margin of 2 points.

In 1982, a final weekend poll by WCIA-TV (CBS, Champaign) had Gov. Thompson down to 50 percent as he squeaked by for reelection, picking up none of the 6 percent undecided in that poll. In the 1983 Chicago mayoral Democratic primary, incumbent Byrne got no more than the 35 percent she was getting in late polls while the 16 percent undecided voters decided in favor of one of her two challengers.

In Illinois in 1984, incumbent President Reagan did 3 points better than stated preferences in a final WBBM-TV poll from an available 9 percent undecided. U.S. Sen. Percy, at 45 percent in that poll, wound up with 4 points more from the 16 percent undecided. Based on the same final poll, Cook County State's Atty. Richard M. Daley had 64 percent but picked up only 2 points on election day from the 16 percent undecided.

In 1986, a WTMJ-TV (NBC, Milwaukee) poll had Wisconsin incumbent U.S. Sen. Robert Kasten at 49 percent in October polls, but he picked up only 2 points on election day from the 12 percent undecided. Incumbent Gov. Tony Earl was at 44 percent in the final poll with 12 percent undecided and moved up 1 point to 45 percent on election day. Incumbent Atty. Gen. Bronson La Follette had 41 percent in a mid-October poll with 26 percent undecided and picked up only 5 points by election day.

In other words, there is overwhelming evidence suggesting that an incumbent won't share the undecided vote equally with the challenger and that emphasizing point spreads in news reports of polls is misleading.

Then there is the rule of straight-ticket voting. It mainly applies to Illinois races but it may also apply to other states. It recognizes the exogenous effect of straight ticket voting produced by effective political organization. (Exogenous means outside factors other than voter preference for a particular candidate.)

In races for several offices in general elections, in cases where high straight-ticket voting is expected, analysis of leading/trailing candidates within the same party is needed because trailing candidates can exceed stated poll preferences on election day. In such cases, trailing candidates can receive additional support from those casting a party vote.

A good example of a trailing candidate who was the chief beneficiary of a strong straight-ticket vote was Adlai E. Stevenson III in his race for governor in 1982. The straight Democratic vote in Chicago was strong that year for a variety of reasons, partly because of an unprecedented Democratic "Punch 10" media campaign but most importantly because of newly registered black voters sending Harold Washington a message that they would be there if he declared for mayor.

It is evident that Stevenson was the chief beneficiary of this straight-ticket vote: He got less votes in more city wards than all other Democrats, including University of Illinois trustee candidates. Jerry Cosentino was runner-up to Stevenson in the number of wards he trailed the ticket, and he was the only other Democrat who got more votes than stated poll preferences for him. The exogenous effect of straight-ticket voting was present but was not as noticeable in the 1978 Illinois elections.

In 1986, when voters had to split their ballot to avoid electing Lyndon LaRouche backers on the Democratic ticket and make a conscious effort to vote for Stevenson on the new solidarity party ticket, he could not match his 1982 performance.

When there is a wide gap in preference for candidates of the same party, stated preference data should be adjusted based on past straight-ticket voting to see how preferences could change to anticipate the overriding effect of voting for party and not person. In the 1984 U.S. Senate race in Illinois, such analysis of potential straight-ticket voting in WBBM-TV Newspolls showed before the election that trailing Democratic candidate Paul Simon could beat Republican Percy on election day.

Of course, this pitfall doesn't exist in primary election polls which, contrary to conventional wisdom, makes them easier to do.

Voter preference data must be interpreted in light of other survey findings. For example, the magnitude of those voting for an incumbent must be compared with his level of job approval and favorable opinion. If approval or opinion do not support stated voting preference for an incumbent, a decline can be expected, particularly if he or she has a strong name recognition advantage. In 1979, despite a strong 16-point first wave lead by Chicago Mayor Michael Bilandic, the magnitude of voter preference for him was far above the level approving the job he was doing in office. Early preference was mainly due to his unknown challenger. This was a key element in an early WBBM-TV poll story that anticipated Jane Byrne taking the lead in later polls.

All polls should obtain diagnostic or "reason why" data. A fundamental element should be the extent to which all of the candidates are known to voters. Name recognition must be monitored during the course of the campaign. The trend in relative familiarity of challengers will be a factor in the success of their campaigns at the time each poll is conducted and should be part of any news story.

How do you determine the effect of disparate recognition between the candidates? Re-tabulating the preference data by those aware of both candidates gives a good indication of how the election might look after the lesser-known candidate becomes better known. One example of this comes from a February 1976 poll for Illinois governor. Among all voters — 39 percent preferred Democrat Michael J. Howlett, 39 percent for Republican Thompson; among those aware of both, 60 percent for Thompson. Thompson got 63 percent in November 1976.

Other critical diagnostic data depend on the circumstances of each election. For example, if a candidate is the target of an investigative probe, how many voters are familiar with it and how will it affect their votes? If a party organization is supposed to affect the outcome, how many voters have been contacted by party workers and how are they voting?

Most polls ask a decades-old voting intention question which attaches party labels to the candidates and asks respondents to assume the election was held today. The option of indecision is withheld. The theory is that to offer this option might be an open invitation to respondents to be uncooperative and choose this option to hide their true intentions; this is probably true.

When some respondents are initially undecided, a "which candidate do you lean to" question is frequently asked.

There are some variations. A futuristic twist is added by some who ask respondents to assume the election was held "tomorrow." If it's an early poll, some ask voters whom they "would like to see win." But in all cases, indecision is discouraged.

I believe survey respondents are typically cooperative. If asked to state a preference without being given an equal or no preference option they are likely to state a preference, regardless of their conviction. We have noticed in other polls that preference under these conditions can overstate in favor of a better-known incumbent; e.g., Chicago Mayor Byrne in 1983; Wisconsin Gov. Earl in 1986.

The true level of undecided voters should be a legitimate survey finding and this response should not be discouraged.

Otherwise, preference can be distorted. Journalists have a tendency to be far too direct and literal when drafting or evaluating questionnaire copy. That's why so much attention is given nuances like "today" v. "tomorrow."

As polltakers we prefer an indirect question. We ask voters to "suppose you had 10 votes to cast" between candidates.

When more of the 10 are given to one candidate over other(s), then it counts as a vote for that candidate. How many given determines whether that voter is strong or soft on a candidate. If the votes are divided equally, then that voter is undecided. The final result shows the votes for Candidate A, Candidate B, and the undecided votes. There are advantages to this method:

• Unlike the standard poll voting question, it permits equal preference between candidates and a more valid measure of the undecided level so that you can make the kind of judgments discussed above.

• It allows data separation by strong or soft voters for each candidate and undecideds for analysis of all other survey findings. The separation of the electorate into five groups, the strongs and softs for each candidate and the undecideds, is a critical element in the analysis. The three groups not firmly committed are the persuadable voters. They are identified according to their demographics, their areas of residence, the issues which concern them, their perceptions of the candidates and all other survey findings. Analysis of data by strong voters identifies the more compelling issues and candidate appeals. In other words, a characterization of voters is possible.

• It provides a measure of preference intensity and can foretell the likelihood of future change. In 1984, the ratio of strong to soft in all waves of Illinois and Missouri polls suggested that Reagan's strong lead over Mondale would hold on election day. In 1982 in Missouri, a low ratio of strong to soft among incumbent Sen. John Danforth voters in an early poll led to a story that challenger Harriett Woods could catch up by election day despite his strong early-lead. She came within 2 percentage points of victory.

In multiple candidate races, it identifies which two candidates between whom most voters are undecided. In the 1983 Chicago Democratic mayoral primary, we could tell that the larger group of undecided voters, black voters, were deciding between Jane Byrne and Harold Washington. This meant that Richard M. Daley, son of the late mayor who was trailing badly in this race, would pick up very few undecideds. Anti-Byrne sentiment in the black community had put Daley ahead of Byrne in two-way matchups the year before.

Note that we haven't talked about sample size. While size is important, it is really secondary to how the poll was done and how it was interpreted. How is always more important than how many. How the sample is selected is more important than how many when only 20 percent are expected to vote.

Pseudo-polls are now a problem for the industry. They deceive with the promise of large numbers. In 1984, one Chicago radio station conducted a phone-in poll asking whether people sided with Mayor Washington or Democratic Chairman Edward Vrdolyak during the so-called Council Wars: 29,000 listeners responded, siding with Vrdolyak 4 to 1. A black radio station responded with its own phone-in poll, and Washington won 4 to 1. This anomaly was obviously a function of station audience demographics. (The real winner at 50 cents a call was AT&T.)

In 1972, one noted pollster theorized that primary polling is only "a straw in the wind" since candidates are lesser known and since it's impossible to predict who will vote in these low-turnout elections. This view ignores that diagnostic data (such as name recognition) should figure in the analysis and that those likely to vote can be identified. If respondents are properly screened for voting likelihood, a primary poll will be easier because straight-ticket voting is not a factor.

Poll results should not be accepted uncritically. A new circumstance affecting likely-to-vote measures arose this year in the Chicago mayoral elections.

The effect that turnout would have in the 1987 Chicago mayoral Democratic primary election outcome was given wide play by the media and by Mayor Washington in his campaign. Proven voting likelihood questions failed when they indicated that 95 percent of blacks would vote, making blacks the majority of likely voters in some polls. This placed Harold Washington well above the 53 percent he ended up getting. Ninety-five percent turnout sounds more like a legislative body than a constituency. Prudent poll users reweighted their data to bring them in line with realistic expectations based on history. One TV station didn't and had to apologize later for faulty assumptions. The real problem was uncritical acceptance. The same station was victim to the same kind of problem in the same election — only this time in a poll conducted by another pollster. An error produced a bias favoring challenger Jane Byrne. After her 1-point lead was contradicted by other media polls (Byrne didn't seem to believe it either), it came out that blacks made up only 36 percent of the sample.

How do you learn more about polls? The book is still being written by practitioners as each new circumstance arises. There is a clutter of published faulty theories, too many to address here.

Theorizing is easy. The best guide to knowing what to believe is to accept only what has been validated with actual election experience. The subject of polling is technical and should only be practiced by Professionals. This sounds like "do not attempt without professional supervision, "but it's true. Polls need competent analysis before the real meaning of their results becomes a valid news story.

Nick Panagakis is president of Market Shares Corporation, a marketing and public opinion research firm headquartered in Mount Prospect. Panagakis is best known for pre-election and exit polls conducted for the news media in Chicago, St. Louis and Milwaukee. His clients include WBBM-TV in Chicago (1974-1985), the Chicago Tribune as of 1987. KMOX-TV in St. Louis and WTMJ in Milwaukee.

August & September 1987/Illinois Issues/77