"We kind of hold our breaths every election cycle to see if our methods are still holding up," said Chris Borick, director of the Institute of Public Opinion at Muhlenberg College in Allentown.
Nationwide, the proportion of cellphone-only households has doubled in less than four years, from 17.5 percent in the first half of 2008 to 34 percent at the end of 2011, according to the most recent government survey. Such people are harder to reach for various reasons.
They could be driving or crossing the street, for example, said Douglas Schwartz, director of the Quinnipiac University Poll, based in Connecticut. Or they may simply be ineligible to participate - perhaps because they are too young to vote, or because they have an area code from the state being polled but actually live in another.
Throw in the fact that willingness to participate in polls is down across the board, and it is increasingly tough to get a sample that represents the people who are likely to vote. So poll results are often "upweighted" or "downweighted" to match what the pollster predicts will happen on Election Day.
In other words, if the survey reaches, say, only half as many people as expected from Group X, then the responses from that group would count double in the overall results.
Therein lies the debate. How much to weight?
These statistical tweaks come into play especially for three groups that are more likely to live in cell-only households: young people, minorities, and the poor - who all are more likely to vote for Democrats but are also less likely to vote.
Voter turnout among African Americans and young people was higher than normal in 2008, a fact attributed to enthusiasm for the first-ever nomination of a black candidate. Will those groups turn out in big numbers this year, or will their participation be more similar to those in previous years?
In the two post-debate polls that varied so widely in Pennsylvania, pollsters made different calls on age and race, as well as geographic region.
In one of the surveys, conducted on behalf of The Inquirer by a bipartisan team from the polling firms Global Strategy Group and National Research, pollsters predicted that 18-to-29-year-olds would come in at 16 percent of the state's electorate.
In the other survey, conducted by the Harrisburg-based Republican firm of Susquehanna Polling & Research, pollsters weighted their results to reflect their prediction that 18-to-29-year-olds would make up just 9 percent of Pennsylvania voters. The raw numbers for that group were even smaller - less than 5 percent of the total, largely because of the cellphone problem, said Susquehanna president Jim Lee - so their responses had to be counted nearly twice as much.
So who is right?
In 2008, voters 18 to 29 made up 18 to 19 percent of Pennsylvania voters, according to various post-election surveys. But in 2010 - a midterm election, when younger people typically show less interest - turnout from that age group made up only 11 percent of the total.
So The Inquirer poll predicts this year's number for younger voters will come in somewhere in between, whereas Susquehanna has them turning out below the level in the midterms.
Overall, The Inquirer poll found that 50 percent of Pennsylvania voters were likely to vote for Obama after the first debate, compared with 42 percent for Romney. The Susquehanna poll found that Obama had just a 47-46 edge in the state.
Officials from both polls said the gap between their results was likely the result of their different turnout predictions in various categories.
"Everybody's got their own little formula, their own little mix of ingredients," said Susquehanna's Lee.
Then there is the question of race. Susquehanna estimates that blacks will make up 8 percent of the state's voters. The Inquirer poll is expecting they will account for 10 percent - about what they did in 2008. Pollsters arrived at that figure after an analysis of voting trends and voter enthusiasm, said Nick Gourevitch, director of research at Global Strategy Group.
"Some people might argue that 2012 isn't going to have the same level of enthusiasm as 2008," Gourevitch said. "Could it be 11? Could it be 9? Sure. Ultimately, this is our best guess based on a lot of different things."
The need for increased weighting of results is not solely the fault of the cellphone.
Even in those households with land lines, residents are increasingly less likely to pick up the phone, said Muhlenberg's Borick. Caller ID is widespread for those who want to screen out unwanted calls.
Finally, amid rising concerns about identity theft, some people are loath to divulge the kinds of personal information polls seek, Borick said.
"If you're calling somebody's house, you're almost expecting that you're going to get somebody's answering machine," he said. "I think we've just grown more cautious as a society."
Borick is so careful about predicting the makeup of the electorate that he weights for just two categories that remain fairly stable from election to election: gender and geographic region.
The safest gauge, he said, is to look at the combined results of many polls. But until the night of Nov. 6, no one knows for sure.
Contact Tom Avril at 215-854-2430 or firstname.lastname@example.org