dagblog - Comments for "Jack and the Statstalk"

Sorry I missed this Donal.I

Richard Day — Mon, 13 Sep 2010 07:19:43 +0000

Sorry I missed this Donal.

I always laugh when some cop in civil dress flashes a badge for 1/10 of a second and threatens someone with 'obstruction of justice' on TV. How am I supposed to know he did not win the 'badge' at a carnival side show?

And 'surveys', well I have been yelling about them for two years but not from a strictly scientific perspective. I just know that Rassmussen has shown the same damn results for approval ratings of President Obama for two frickin years. 45% approve and 55% think he should be shot.

And, at the same time, Gallop's figures change all the time. And they never add up to 100%. Because you and I know that at least 3% of our population does not have the faintest idea who is president.

This is a nice follow up on this Lott guy. Is he related to the ex gay and fanatically dishonest senator?

Real fine work here Donal.

I fought a losing battle over

acanuck — Thu, 09 Sep 2010 19:39:43 +0000

In reply to Quite right. Good

I fought a losing battle over at TPM, trying to persuade them to change how they reported the reliability of polls. You can't just say, "The margin of error is X%." Typically, it needs to be, "The margin of error is X%, 19 times out of 20."

Big difference. It means you should expect every 20th poll to be off by more than the stated margin. In an election cycle, with hundreds of polls being taken, that makes for lots and lots of really wrong poll results. Yet individual, small-sample polls showing minuscule movement within the margin of error consistently get hyped as showing "trends." They simply can't be used that way, but they are.

Now that you mention it, I do

Donal — Thu, 09 Sep 2010 15:05:21 +0000

In reply to I was referring to the study

Now that you mention it, I do recalling hearing about Wakefield, but the articles I read highlighted the Geiers' errors.

Quite right. Good

DF — Thu, 09 Sep 2010 14:30:48 +0000

In reply to One more thing that should be

Quite right. Good statistical analysis doesn't tell you what's right or wrong. It tells you what the probability that you've arrived at the wrong conclusion is. That can't stop people from excitedly jumping to the wrong conclusion.

I was referring to the study

DF — Thu, 09 Sep 2010 14:29:14 +0000

In reply to I didn't mention the Geiers

I was referring to the study and ensuing scandal involving this guy. The entire history of this episode informs us that in spite of a mountain of statistics proving that there was no evidence of the connection between ASD and the MMR vaccine, a bunch of people wanted to believe this guy. You can't blame the numbers for that.

One more thing that should be

Atheist — Thu, 09 Sep 2010 14:28:02 +0000

In reply to Jack and the Statstalk

One more thing that should be said, although it should also be obvious to anyone who thinks about it: assuming all of the statistics are performed flawlessly, you'd still expect about 5% of all scientific studies that show 95% confidence to be wrong (and 1% of all studies that show 99% confidence, etc.). The problem, of course, is in figuring out which 5%.

One of the founders of statistics liked to use 99.9% certainty (and even then, 1 out of every 1000 such results would be due to chance), but scientists quickly found out that such certainty typically required far too many experiments.

I didn't mention the Geiers

Donal — Thu, 09 Sep 2010 14:10:37 +0000

In reply to You forgot the part about how

I didn't mention the Geiers by name, but I think I covered it. My impression is that thiomersal was removed as a precaution because it contained mercury. The removal spurred parents' interest in Geier's research, which was eventually found to be severely flawed.

You forgot the part about how

DF — Thu, 09 Sep 2010 13:52:45 +0000

In reply to Jack and the Statstalk

You forgot the part about how the whole anti-vax movement was spawned by one guy and one completely fraudulent piece of work that has been thoroughly discredited, but continues to be cited.

Seriously, it's easy to blame the numbers, but it's the people that are the problem. Statistics isn't that hard, but not even rigorous statistical analysis can prevent people from making false claims.

Neil Young got it right, "Red

quinn esq — Wed, 08 Sep 2010 03:22:17 +0000

In reply to Jack and the Statstalk

Neil Young got it right, "Red means run, son... numbers add up to nuthin'."

Tough to argue with that.

The mere fact that these

cmaukonen — Wed, 08 Sep 2010 03:16:42 +0000

In reply to Jack and the Statstalk

The mere fact that these scientific studies are performed by, analyzed by and published by humans indicates that they are biased by default. There is no such thing as an unbiased scientific study.