Although the poll analysis suggested Fianna Fail's Eamon Scanlon would win the third seat, Harkin says this was not sustained by the poll figures and "represented a serious misrepresentation of the outcome".

The poll was inaccurate "by a factor of 50%" on the allocation of the third seat and "this cast doubt on the method of interpretation of polls conducted on behalf of TG4, " she added.

The poll was taken before Harkin announced she would not be contesting the general election, so she was included in it. The former maths teacher says that the moment she looked at the figures, she knew there was a problem.

"The fundamental flaw in the analysis is that they attributed number twos from [front-runners] John Perry and Jimmy Devins to Scanlon, when only a very tiny fraction of these votes would come into play [because they would be close to the quota and would have little surplus to distribute]. It makes no sense to ask people to indicate preferences and then not to count them as they would be counted in an election, " Harkin said.

Describing it as the "sloppiest work I've ever seen", she added that, on the assumption that the same "flawed analysis" was used in the five other recent polls, "it also calls into question their validity". Harkin's analysis of the poll finding, which includes only a small number of transfers from Perry and Devins, showed her, not Scanlon, taking the final seat.

Harkin has accused TG4 of "going to ground" on the issue. "Politicians are accountable, as we should be, but it's not only politicians that should be accountable, " she said.

TG4, which has no plans to carry out further polls, referred queries on the subject to TNS/MRBI.

Damien Loscher, managing director of TNS/MRBI . . . one of the most respected polling companies in the state . . . said Harkin's issue was not with the poll findings but their interpretation. He said the poll results were "not wrong", but Harkin was "entirely right to question the interpretation".

Loscher said there were two ways of looking at transfer patterns in constituency polls, neither perfect. One was to assess the transfer of candidates' surpluses and the votes of those eliminated, but this was not statistically reliable as the numbers were too small. The second method . . . the one used in the recent polls . . . is to look at overall transfer patterns.

Loscher said constituency polls were "always controversial" and "very tricky things". Because the polls were done on a constituency basis, the budget was smaller, meaning a smaller sample. But the level of analysis of constituency polls was much higher, he said, noting that, in national polls, transfers were never analysed. Loscher stressed that once it goes beyond the first and second counts, it's "very much open to interpretation" and the data was not robust enough to go to third-, fourth- and fifth-preference votes.

Loscher said the company did intend to do such polls in the future but said: "we have a job to do on educating people", to increase understanding of what they can achieve and, how the analysis, beyond first-preference votes, was open to interpretation. He also said the industry was looking at increasing sample size.