Ah, so did the polling method here just consist of blatantly making up numbers? Reminds me of Strategic Vision.
Crappy partisan polling often plays around heavily with that order of questions that prime for certain answers, because people are more likely to give answers that are superficially "consistent" with their previous answer so as not to appear inconsistent, even if they wouldn't make the same choice if the question came on its own. If we assume that the questions were asked in the same wording and order as on the link*, then they first asked whether respondents were pro-life or pro-choice, then asked whether they were satisfied with "their current congressman" without naming him, and then asked about Thompson vs. Marcy without including the words "Democrat" or "Republican". All of these would prime an exaggeratedly Republican response in a district like this.
*Of course, this assumption is just a guess, but not an unreasonable one given the result and source.
