#### Comment Preferences

• ##### To be pedantic...(66+ / 0-)

in statistics, 48.56 and 48.49 ARE a tie, but zra was wrong to quote numbers to the second decimal place.  In reality, based on the number of people polled, the plus/minus error in any one poll is probably 2 to 3 percent, and the actual percentages should never be quoted byond the "ones" place.

Therefore, the proper response would be "President Obama polled at 49% plus or minus 3%, or 46% to 54%.  Governor Romney polled at 48% plus or minus 3%, or 45% to 53%.  Because of the large overlap, this poll, and this poll alone are a tie."

Of course, he should then have added that "In thirty five polls from Ohio over the past two weeks, President Obama has led in thirty four and tied in this one, so it is reasonable to assume that president Obama is actually ahead, by an estimated three percent, the average of the differences."

That would have been mathematically and statistically correct, as well as honest.

The Scout Law (trustworthy, loyal, helpful...) is a GREAT liberal manifesto.

[ Parent ]

• ##### I agree with (and get) the statistical tie piece.(20+ / 0-)

But it seemed to me that if "49-48" and "49.56-48.49" are both within the same statistical MoE that makes them statistical ties, then it seems pretty moot for Ezra to take it to two decimal places to begin with.

Follow me on Twitter: @THTBAW. I crush trolls with truth, boredom with snark, and ignorance with links.

[ Parent ]

• ##### You're absolutely right.(5+ / 0-)

See my response below.  In my professional work, I am constantly dealing with numbers like 56.3, with quality assurance data that says that there's a plus or minus 10% accuracy to the number.  It's really 55, with no decimal place to be accurate.

The Scout Law (trustworthy, loyal, helpful...) is a GREAT liberal manifesto.

[ Parent ]

• ##### That's not pedantic(6+ / 0-)

that's just wrong. 48.56 does not equal 48.49. It would be a true statement that the two point estimates have overlapping confidence intervals, but the two estimates are not equal.

The important question is, what is the probability that the true parameter that the first statistic is estimating is greater than the parameter that the second statistic is estimating. If the two parameters were truly equal, you would expect half of the samples to have the first greater than the second. The probability that all of these samples would have the first greater than the second given that the two parameters are equal is very small.

Small varmints, if you will.

[ Parent ]

• ##### In statistics, they are equal, and wrong.(24+ / 0-)

By background, I have an engineering and a scientific degrees, as well as an advanced degree in geology.  I deal with statistically-based numbers all the time.  I live on computers, and in Excel spreadsheets, but...

I have my original computer hanging on a bookshelf next to my home computer.  All 10 inches of it.  A K&E Decilon slide rule.  It's the synthetic resin version of the classic log-log-duplex decitrig slide rule.  And I still know how to use it.

One thing that was pounded into us by our engineering professors was the concept of significant figures, mostly due to measurement limitations, but also due to statistical effects.  In a poll with a plus or minus 3% margin of error (95% confidence limit), reporting a number like 48.56 or 48.49 will get your butt failed on an engineering test.  The only significant figure is the 40%, and the units place (8 or 9) is an estimate.  Anything past the decimal place is garbage.

We used to know that when working with slide rules, where we could only use the first two to three digits to a number.  Computers allow us to routinely take numbers to an infinite number of decimal places.  In the numbers Ezra cites, the 0.56 or 0.49 are just garbage, that shouldn't be considered.  Even the 8 or 9 in front of the decimal aren't that certain, and are "close enough" that the numbers are equal, statistically.  That poll IS a tie, statistically, but the vast majority of other Ohio polls are not - Obama is the clear leader in them.  Ezra is just being an idiot in harping on this one.

The gift of the poll aggregators like Sam Wang and Nate Silver is that they cut through the chatter over individual polls and what decimal place to take them.  By pooling the daa, they come up with a much more powerful tool for determining who is ahead.  I find it much more informative than having the "talking heads" who know nothing of math or statistics blathering about how the race is tied.

The Scout Law (trustworthy, loyal, helpful...) is a GREAT liberal manifesto.

[ Parent ]

• ##### We must be the same vintage...(11+ / 0-)

I also still have my K&E Decilon slide rule form engineering school days! You are absolutely right about how slide rules prevented the silliness of five or six decimal places in an answer where there really should be only one or two at the most.

Just another faggity fag socialist fuckstick homosinner!

[ Parent ]

• ##### I had to whip mine out(9+ / 0-)

I thought that it was a log-log decitrig.  Wrong it's, a Pickett Model N 3-ES Power Log Exponential.  What a name!  Beats the crap out of TI SR-50.  BTW the SR-50 is long gone, but the Pickett still works.

“that our civil rights have no dependence on our religious opinions, any more than our opinions in physics or geometry.” Thomas Jefferson

[ Parent ]

• ##### In other words(9+ / 0-)

The numbers are very precise, but they are not very accurate.

Minority rights should never be subject to majority vote.

[ Parent ]

• ##### It's a false precision(5+ / 0-)

In a poll of ~1,000 people, one changed response results in a swing of ~0.2% in the margin.  Anything beyond the tenths of a percent isn't measuring anything meaningful.

I'm a statistician, and a statistical tie has a very specific meaning, and it doesn't have a thing to do with you think it means.

A tie is when two values are exactly equal (as in precisely the same). This usually only occurs with discrete data, and is important in doing something like the Mann-Whitney test.

Interestingly enough, the classic example used to teach the method of the Mann-Whitney test (the tortoise and the hare, cited in the wiki article) is exceedingly relevant to the current discussion.

Small varmints, if you will.

[ Parent ]

Recommended by:
louavul, Possiamo

I now remember why I hated Statistics!  Or as I liked to call it- sadistics!

In an insane society, the sane man would appear insane

[ Parent ]

• ##### Right On DaytonMike(9+ / 0-)

Thanks for mentioning "slide rule" accuracy. I actually was in college in the mid-70s when the transition from slide rules tro calculators occurred. I didn't even own a calculator until my sophmore year. Our physics, math & engineering profs beat into our heads that just because you can get 5 or 6 digits behind the decimal, that doesn't make it real. It's the quality of the input data that helps dictate signifcance. Even today I see engineers and finance types in my company demand levels of accuracy way beyond the actual input data (6 BOPD for January and 7 BOPD for February must mean you averaged 6.47465763 BOPD over that period, right).

[ Parent ]

• ##### Right on!(9+ / 0-)

We found that the best use of our calculators was to enter the value 416437 E 09, hit inverse, and hit inverse again, and get 4.136437 E 14.  Turn it upside down before the two inverses, and you get two messages to my alma mater for all three degrees!

Truue story - As a geologist, I make use of our state's located well log files.  I digitized their locations to generate maps of drift thickness and bedrock topography, and I used the Ohio Coordinate system (locations in feet east and north of an arbitrary datum in Kentucky), as I live near the intersection of two different UTM grids.  Students at a neighboring college used my ideas and methods to do a similar study elsewhere, and I was asked to beon their thesis committee.  An appendix to their thesis showed the locations of all wells, in feet, to six decimal places. Mind you, a well is anywhere from 8 inches to a foot in diameter.  I asked them which side of the well they digitized, and whether it was on the inside or outside of the well casing.  They looked confused.

I then pointed out they were marking well locations on a USGS 1:24,000 scale map (1 inch = 2,000 feet), and were using a 600 dpi digitizing table, so none of their locations were better than a little over a yard.  In addition, if they digitized the same dot 10 times, the error would be more like 10 feet.

The final version of their appendix showed locations to the foot, but with no decimal places.  I let that false accuracy go.

The Scout Law (trustworthy, loyal, helpful...) is a GREAT liberal manifesto.

[ Parent ]

• ##### Cartographer here ...(3+ / 0-)
Recommended by:
DaytonMike, TampaCPA, ichibon

Thanks for keeping it real.

I was part of that same transition, in college science classes late '60s early '70s. Because electronic calculators were expensive still, they were not allowed, in the interest of fairness and equity. We all used slide rules.

• ##### Hm...(1+ / 0-)
Recommended by:
DaytonMike

I wonder how many people are trying to turn their computer monitors upside down to see the secret messages from inverting 416437 E 09.

Join the 48ForEastAfrica Blogathon for the famine in east Africa: Donate to Oxfam America

[ Parent ]

• ##### I wonder how many people here are following(2+ / 0-)
Recommended by:
Sam Sara, DaytonMike

this conversation in anguish because it is very interesting even though, like me, they understand so little of it.

However, I will take away more knowledge than I started with so carry on!

• ##### My bad for not turning my monitor upside down...(1+ / 0-)
Recommended by:
fumie

it should have been 431437 09, which (upside down) is

GO LEhIGh

inverted, and inverted again

hI LEhIG.h

Engineers (now Mountain Hawks) are so easily amused!

The Scout Law (trustworthy, loyal, helpful...) is a GREAT liberal manifesto.

[ Parent ]

• ##### Sorry, that's just wrong. (2+ / 0-)
Recommended by:
TheChocolateChips, Garrett

Significant digits or not, if one more person in the sample prefers Obama to Romney, that is not a tie.  Period.

In a sample of 999 people, that is a very slight advantage (53% to 47%)  but it is an advantage.

In practice, you might informally ignore such slight differences, but there is no such thing as a "statistical tie" except for an exact tie, with identical numbers.

• ##### Thank you ...(3+ / 0-)
Recommended by:
DaytonMike, TampaCPA, fumie

Old school, here. I still have a slide rule somewhere, was never very proficient at it, but did learn and understand the concept of significant figures, and how to use scientific notation, etc.

It's annoying when people try to read degrees of accuracy and precision into findings (statistics, polls, maps ...) that are not there to be had.

• ##### Recommended for the slide rule(2+ / 0-)
Recommended by:
DaytonMike, fumie

Sold mine at a garage sale 25 years ago and now wish I hadn't.

There has never been a protracted war from which a country has benefited. The supreme art of war is to subdue the enemy without fighting. - Sun Tzu

[ Parent ]

• ##### Round 1975 or so(2+ / 0-)
Recommended by:
DaytonMike, OHeyeO

I was walking though the UC Berkeley book store and they were unloading their stock of slide rules. We're talking really nice K&E's or Picketts for \$5-\$10. Sadly, I was in a hurry and didn't stop. By the time I thought to go back, they were all gone.

• ##### sadly Ezra gets to spew that bullshit on a nationa(10+ / 0-)

media web site.  it's pretty embarassing.

makes you wonder what else he gets wrong if he doesn't even understand significant figures.

of course, I work with engineers at work who keep using 4 decimal places for numbers when the shouldn't, so can't expect much from a media hack.

big badda boom : GRB 090423

[ Parent ]

• ##### Nothing. He was talking about just one poll.(1+ / 0-)
Recommended by:
jds1978

In that one poll, there is a statistical tie. I'm sure if asked about all the polls, he would be equally precise.

• ##### They are young engineers...(0+ / 0-)

who didn't have Lehigh's John Leibig hounding them.  Of course, he taught surveying where we had better be precise, to two decimal places while taping distances, and where the error of closure had better be tiny, but not zero.  If it was zero, he assumed w cooked the numbers, because nobody was perfect.

The Scout Law (trustworthy, loyal, helpful...) is a GREAT liberal manifesto.

[ Parent ]

Subscribe or Donate to support Daily Kos.