A few points for those who, like me, wasted part of an otherwise fine Sunday morning reading the RJ poll on the Titus-Porter race.
1. It is well and good as cocktail party or bar fodder to compare this poll with Titus' internal poll of two weeks ago that showed a very different result but such a comparison has no real meaning as a statistically valid statement.
Certainly not as implied by the RJ story's quoted experts that this provides evidence the race has changed in any way. That could have happened but this poll compared with that poll cannot be the basis for such a claim.
(While on the subject, the RJ should never have reported either campaign's internal poll as news, unless they were able to see the entire survey results and how the sample was weighted. Without knowing that, there's almost no way to assess the validity of the horse-race result and certainly no way to compare it with other horse-race results).
The only statistically valid comparison is with this same poll's last last result which appears to have been in June -- and which had the same +3 margin for Porter.
2. A survey with a sample of 236 (according to the RJ website) for a Congressional race is not that much of a poll. 300 is the minimum sample that would ordinarily be considered usable for random-sampling method of a universe that large. This small sample size must be the primary reason for the whopping +/-6 margin of error, meaning a 12-point "fork" of statistically likely outcomes, ranging from Porter +9 to Titus +3. In short, this survey would have been unlikely to pass a grad course on statistical methods.
3. In general, an incumbent is expected to lose a large portion of the late deciders, because they've had a long time (in this case 6 years in office) to get to know him. That's why an incumbent below 50% in the final stages of a campaign is thought to be in trouble. To be a three-term incumbent at 43% is really the significant news here. (It is true, as the story points out, that Titus is actually better known to respondents of this poll than the incumbent, but the general rule of thumb remains an incumbent below 50% is in trouble.)
By the way, on the topic of name recognition and favorability, its not reported in the story but really significant: Porter, the 3-term incumbent, was not recognized by 21% and viewed neutrally by another 5% of voters. (Compared with only 14% +3% for Titus). That's an open invitation to the challenger to, ahem, introduce him to a quarter of the electorate.
4. That many undecideds this close to the election is crying out for additional questions to see which way they are leaning.
5. Most survey polling does not use a pure random sample method; it weights the results to try to account for what might appear to the researcher to be wide disparities between the sample and the universe (in this case, those who will actually vote). It appears from the fine print at the bottom of the poll that the corrections were made in the statewide raw data to account for expected turnout by county but not by party affiliation, sex, age or other factors. Since we don't know the composition of the sample with respect to those factors, it's really a crap shoot to know how well it approximates any given statistical model of the electorate.
This is not to be harsh on the RJ (always a worthy endeavor), other than to say they should spend more to ask more questions, but to say this is really not the news that it is made out to be by the academic experts quoted in the RJ story. Then again, I guess that's why they are quoted in the story.