FL-Gravis Marketing: GOP convention has helped Obama in Florida

[Tender Branson]

Gravis Marketing and Capitol Correspondent conducted a telephone survey of 1,288 likely or somewhat likely voters in the state of Florida on the afternoon and early evening of September 2, 2012.

The questions covered preference for a given presidential candidate, the Florida U.S. Senate Race between Connie Mack and Bill Nelson, and Governor Rick Scott’s performance rating.

Overall, Romney holds a 48.0% to a 46.7% lead over Obama, with a margin of error of about 2.7%.

How is the Republican National Convention Helping Romney?

The Republican National Convention does not appear to have helped Romney, with Romney at the same 48% take as he was on our August 20th poll. Over this time frame, Obama has gained 1.7% from his 45% August 20th take.

http://www.scribd.com/doc/104872042/Gravis-Florida-September-2-2012-Presidential-Poll

[MorningInAmerica]

You'd think from the title of the thread that Obama is leading in Florida. He is not, per the survey;

Romney: 48%
Obama: 46.7%

Rick Scott's approval rating:
Approve: 34% (35%)
Disapprove: 40% (38%)

Senate Race:
Nelson - 43% (46%)
Mack - 42% (38%)

[5280]

Slight lead over Obama is more appropriate title.

[Eraserhead]

There's nothing in the thread title that indicates that Obama is leading, it's fine. The poll is probably garbage anyway, it's still kind of funny though.

[MorningInAmerica]

Polling seems split as to whether Gary Johnson takes more from Obama or Romney. But at least for Gravis, Johnson takes slightly more from Obama.

If the election were held today, would you vote for Obama, Romney, Libertarian Gary Johnson, or someone else/unsure?

•Mitt Romney 47.8%
•Barack Obama 46.1%
•Gary Johnson 2.3%

Also, Party ID on this poll is 44/42/14%, or D+2.

[Associate Justice PiT]

     Giving results to a tenth of a percentage point is a telltale sign of poor polling.

[MorningInAmerica]

    Giving results to a tenth of a percentage point is a telltale sign of poor polling.

What exactly is the logic behind this? A question for everyone who keeps assuming decimal points mean poor polling - you do realize that every pollster, be they PPP, Quinnipiac, Rasmussen, Gravis, etc, obtains results to the decimal point, right? Just most of them only publish the rounded numbers. It's simple math. I'd question a pollster who only ever happens to wind up with whole numbers (simply because the mathematical odds of that are slim). Why is this difficult to understand? Or am I missing something?

[Niemeyerite]

     Giving results to a tenth of a percentage point is a telltale sign of poor polling.

What exactly is the logic behind this? A question for everyone who keeps assuming decimal points mean poor polling - you do realize that every pollster, be they PPP, Quinnipiac, Rasmussen, Gravis, etc, obtains results to the decimal point, right? Just most of them only publish the rounded numbers. It's simple math. I'd question a pollster who only ever happens to wind up with whole numbers (simply because the mathematical odds of that are slim). Why is this difficult to understand? Or am I missing something?

That's been the rule here for years. If a poll releases their numbers with decimals, they tend to be hackish/bad/useless polls.

[Sbane]

     Giving results to a tenth of a percentage point is a telltale sign of poor polling.

What exactly is the logic behind this? A question for everyone who keeps assuming decimal points mean poor polling - you do realize that every pollster, be they PPP, Quinnipiac, Rasmussen, Gravis, etc, obtains results to the decimal point, right? Just most of them only publish the rounded numbers. It's simple math. I'd question a pollster who only ever happens to wind up with whole numbers (simply because the mathematical odds of that are slim). Why is this difficult to understand? Or am I missing something?

It has to do with the margin of error obviously. You can round up or down when publishing numbers because with the margin of error they are basically the same thing. Any polling company publishing the numbers with decimals is trying to gain credibility by trying to prove how accurate they are. Which would meant that they have no credibility to begin with and are not taken seriously.

[MorningInAmerica]

    Giving results to a tenth of a percentage point is a telltale sign of poor polling.

What exactly is the logic behind this? A question for everyone who keeps assuming decimal points mean poor polling - you do realize that every pollster, be they PPP, Quinnipiac, Rasmussen, Gravis, etc, obtains results to the decimal point, right? Just most of them only publish the rounded numbers. It's simple math. I'd question a pollster who only ever happens to wind up with whole numbers (simply because the mathematical odds of that are slim). Why is this difficult to understand? Or am I missing something?

It has to do with the margin of error obviously. You can round up or down when publishing numbers because with the margin of error they are basically the same thing. Any polling company publishing the numbers with decimals is trying to gain credibility by trying to prove how accurate they are. Which would meant that they have no credibility to begin with and are not taken seriously.

It seems arbitrary. All pollsters obtain results with decimal points. If it bugs someone, round it yourself. Sometimes it can mean the difference between someone leading by 1 point and 0.1 point. For example, if a pollster found Obama leading Romney 48.5% - 48.4%, Rasmussen or PPP would report that as 49-48%, and not the actual 0.1 point lead that it is.  I understand it's all within the margin of error, but still, why not let your reader round however he or she wants? It can provide a more accurate picture.

[Sbane]

Again, it just usually means they don't have much credibility. Theoretically speaking, there is nothing wrong with reporting decimals.

[TJ in Oregon]

The underlying issue is the number of significant digits in the result. In general, you only report a result with a precision on the same order of magnitude as your error, ie. if the result has an error of ±3 points, they would report down to the 1% digit. This convention is not quite practiced universally and some people prefer to include one extra digit so you know if a result is 44.6 ±3 or 45.2 ± 3 instead of saying just 45 ± 3, but it looks weird.

In polling, I think it becomes a bit trivial to include extra digits because even the statistical uncertainty in the measurement has additional errors built in from various biases that make polling results fairly unreliable in general.

When a polling company does this it makes it look like they missed the 10th grade lesson on sig figs or at the very least are using a non-standard form of notation.