In my column two months ago, I said that those reading preelection polls should expect most of the undecided voters to vote for the challenger in races involving incumbents. In fact, frequently all of the undecideds will go to the incumbent. Since then, I have extended the analysis from the 24 incumbent polls done for our media clients to 71, using polls obtained from Gordon S. Black Research, CBS, Gallup Organization, Market Opinion Research and others.

The Incumbent Rule says: Analysis of poll results in races with incumbents should assume that undecideds won't split evenly; instead, a majority of undecideds usually go to the challenger. Therefore, the ''comfort'' threshold for the incumbent in a two-way race is about 50 percent.

I believe that the undecided voter is not literally undecided between candidates but is expressing doubt about the candidate known best. When an incumbent has served many years in office, pollsters should wonder what undecided voters are really telling them — they say they are undecided, but they may mean something else.

Another equally important point: Assuming most undecided voters will go to the challenger also means the point spread between stated preference for incumbent and challenger is meaningless. Point spreads can only hold up when undecideds split equally.

A recent example of point spread confusion comes from the Democratic mayoral primary election in Chicago. Just before the 1987 primary, WBBM-TV, the Sun-Times Gallup poll and the Tribune Poll found incumbent Mayor Harold Washington favored by from 51 to 55 percent of voters. His challenger, former Mayor Jayne Byrne, trailed with from 35 to 40 percent. In this case the undecided response was definitive, and the news stories said a majority of them would support Byrne. That most undecideds would go to Byrne was a major conclusion in stories published before that election and was confirmed when Washington ended with 53.6 percent of the vote.

But in on-air discussions of poll results reporters and anchors slipped into a point spread characterization. One number is easier than three: voting for Washington, voting for Byrne and undecided. Saying that most undecideds would go to the challenger at the same time the poll results are characterized in terms of point spread is a logical inconsistency. That's why a 16-or 19-point poll spread becomes a 7-point spread on election day.

In the 1987 Washington-Byrne race, the ultimate disposition of undecideds was fairly definitive, but what about the other cases?

In 59 of the 71 incumbent races we analyzed, the percentage gain over stated poll preference for the challenger exceeded the percentage gain over poll preference for the incumbent. In other words, most of the undecideds went to the challenger most of the time.

The undecided trend is important because it means polls can't always predict the winner. Incumbents who are poll leaders are likely to become election losers (or near losers) if they are leading with less than 50 percent, and this should be prominent in any news story. I found six cases where incumbents who were leading in late polls with about 50 percent or less ended up losing because undecided voters went for challengers. Other cases show incumbents with similar poll leads barely winning with 51 percent.

There is one important caveat to the Incumbent Rule: It has mixed (or no) application in presidential races. Apparently after several months of primary campaigning, a national convention and all the media attention, there is no difference in the voters' familiarity between incumbent and challenger. We examined 1980 and 1984 Gallup, ABC, CBS and NBC national polls provided by the Roper Center as well as four of our presidential polls in Illinois and Missouri. We found no pattern in ultimate disposition of undecided voters.

Traditionally, undecided response has been a nuisance response, eliminated through proportional or equal allocation, which is the classical treatment for non-response. At best, since no consistent pattern was observed in past presidential polls, the undecideds were assumed to be nonvoters, late deciders who would decide like early deciders, refusers or a combination of all three; all were allocated equally or proportionally among candidates.

But, the undecideds can make a difference in all other races. Some Illinois examples where the challenger got more of the undecideds than the incumbent:

• A 1986 CBS Illinois poll in early October had Democratic incumbent Sen. Alan J. Dixon at 63 percent; he got 2 points more by election day while 10 undecided percentage points went to his challenger.

• In 1984 Republican incumbent Sen. Charles H. Percy led with 45 percent in our final WBBM-TV poll in a story featuring the Incumbent Rule; he wound up with only 4 points more — from the 16 percent undecided — to lose the election 49 percent to 51 percent. Gallup also had Percy under 50 percent.

• In 1982 a final weekend poll by WCIA-TV (CBS, Champaign) had Gov. James R. Thompson down to 50 percent as he squeaked by for reelection, picking up none of the 6 percent undecided in that poll.

• Our earliest reading is from the 1976 Democratic primary for governor. Dan Walker at 46 percent in our WBBM-TV poll was leading Michael J. Howlett by 3 points in late February and wound up losing with 46 percent; he got no undecideds.

The overwhelming evidence is that an incumbent won't share the undecideds equally with the challenger. To suggest otherwise by emphasizing point spread or saying that an incumbent is ahead when his or her percentage is well under 50 percent is misleading.

Nick Panagakis is president of Market Shares Corporation, a marketing and public opinion research Firm headquartered in Mount Prospect. Panagakis conducts phone polls and exit polls for the news media in Chicago, St. Louis, and Milwaukee. He is a member of the National Council on Public Polls.