Saturday, September 02, 2006

Cool Link of the Day

Here's a really cool website that I just found via Fire Dog Lake.

It's called Pollster.Com, and it features graphs detailing the latest polling trends in many different races. They compile all recent polls for a particular contest so you can get a consensus view. It's a very handy tool, and I'm sure I'll be visiting this site many times in the future.

4 comments:

harriet said...

great poll, if these numbers keep trending down for Joe he'll be in single digits along with his Republican running mate by October!

imajoebob said...

This is pretty. Pretty useless. It falls under the category of "Lies, damned lies, and statistics.

First, what the heck is that "Avg Last 5" all about? It's as useful as the total of the last 5 polls. It simply pollutes the latest data with old information.

Next is the "Polling Trend." This looks like a simplistic average, or maybe moving average. In terms of Public Opinion, this is not a very sophisticated trend analysis. And the format for a trend line, more often than not, is a striaght line or hyperbolic.

And Confidence Intervals? This is (again) a useless graph, because we don't know the Confidence Level. Which is better: leading 50-46 with an error of ±3% or 48-47 with an error of ±5%? Answer: neither. What's better is a poll with a higher Confidence Level. We don't know if the CL of these polls is 90%, 95%, 99%, or 50%. A small margin of error and a low CL is problematic. Not knowing the CL is useless.

There doesn't appear to be any consideration for Time in this graph, except as a marker for the individual data sets. Is there any weight given to recent information? Lamont's early anonymity makes his average (especially 10-set average) much lower. On the other hand, it also makes his trend line incredibly dramatic. Not knowing the time-factor makes it all useless.

What I do find useful, though it's not due to this graph, is the recent volatility in Lieberman's numbers, and the relative flat results for Lamont. I can infer (the data can imply, I can infer) that Lieberman's support is tenuous. A minor event may cause a notable change, a major event may show an extreme change. I might also infer from lamont's numbers that he has plateaued. He needs to do something (or something has to happen) to boost his numbers more than a point or two.

But the best use of this site, other than to see a collection of the different polls in one place, is to learn how easy it is to be deceived by (or deceive with) statistics. After all, 4 out of 5 people believe statistics are true. Who can argue with numbers like that?

Anonymous said...

I just wonder what that "I" next to Lieberman's name stands for?

CT Bob said...

imajoebob - one of the things I like about the Pollster graphs IS the trending lines. Yes, it does show Ned's numbers increasing dramatically, and it does show a timeline along the bottom of the chart.

However, I do agree that numbers and statistics CAN be inaccurate. I was actually horrified when the 2nd to last Q-poll before the primary showed Ned leading by an astounding 13 points. We all knew that was WAY over the top!

Anyway, I look at this data like the movie review website "Rotten Tomatoes" (http://www.rottentomatoes.com/), where they compile ALL the reviews for a particular film and then give it a consensus rating based on them.

Pollster is sort of like that. (Ned would get a fresh tomato, and Joe deserves a rotten tomato.)