LEWISBURG — While most politicians understand that polling is an inexact science, candidates still lean heavily on poll results to capture snapshots of trends during a campaign.

With only two days left before Tuesday’s Pennsylvania primary election, the hotly contested Democratic race has Hillary Rodham Clinton ahead by anywhere from five to eight points, with about a 3 percent margin of error, according to most of the major pollsters, including Zogby, Quinnipiac and Rasmussen.

But how reliable are these polls?

“The results can be fairly reliable, if it’s a well-conducted poll,” said Scott Meike, assistant professor of political science at Bucknell University in Lewisburg.

But some of the reasons why a poll can be wrong include “making sampling errors,” Meike said.

“Every sample is just that, a sample. In a general sense, the greater the number of people in a sample, the smaller the margin of error. A small sample of only 500 voters, what’s called a ‘quick poll,’ might have a margin of error of 5 or 6. A 1,000-person sample would have a 3 percent margin.”

A second problem can occur when a pollster builds the sample.

“Every company does it differently,” he said. “Often, a sample is based on who has voted in the past. They have to figure out who is going to vote.

“Finally, some people change their minds about who they will vote for at the last second,” Meike said. “You can’t account for sudden changes of minds. That’s the inexact science aspect of this all.”

The shocking results in the New Hampshire Democratic primary earlier this year — when most polls predicted a win by Barack Obama over Clinton — prompted many to question the reliability of polls.

“But I think there were some credible explanations why the polls were so wrong,” Meike said. “The intensity had shifted as election day neared. The support for Clinton was vastly underrated. All this happened in the last few days prior to the vote.”

Meike recommended two Web sites for poll watchers: www.realclearpolitics.com and www.pollster.com

“They both provide good statistics and they don’t just follow one poll,” he said. “They take the average of a number of polls.”

Electioneering and voter trend analysis has come a long way since the first polls, in the 1930s, conducted by Gallup.

Before then, Meike said, politicians would judge how well they were doing by the size of the crowds they drew when they made appearances.