Margins of Error - Torgo χ


2012-11-03 (Saturday)

Dear Log,

So, there's chatter on the news about whoever's numbers, in whatever place, dropping from 49% to 48%, ohnoes!...

It drives me batty, because look in the corner of the screen and it'll say "±4%". Not 0.4%, oh my no. That tiny little footnote says 4%, and you know how the large print giveth and the small print giveth away!

Now, this is what a no-margin-of-error 49% looks like, zooming in on the range 40%-60% (which I'll let stand, even though I generally squint unhappily at a graph whose Y axis starts at anything other than 0)...

49% exactly (60%)
(40%)

Now, this is what a no-margin-of-error 48% looks like:

48% exactly (60%)
(40%)

Yes, it is White gaining a whole extra point on Black, and that's how they draw it.

But when they say a one point drop with a little "±4%" in the corner of the screen, that means that they're really saying that things went from this:

49% ±4% (60%)
(40%)

to this:

48% ±4% (60%)
(40%)

Suddenly it's kinda harder to make gripping news chatter about that. Because the data visualization just got kinda... confusing. Namely, got real. And if you say the numbers instead of confusingly showing them, well, you might as well be just reading lists of lotto numbers.

(Meta-details: I did an even gradient effect between 49-4 and 49+4 instead of a chromatic bell curve, and I pretended that there was precision on those numbers: 49.00 ± 4.00%. And I presumed that ±4% was with a p=0.00 instead of a p value of our pal 0.05%, for example. Also, your display's color correction management might tweak this all anyway.)

So, ya see, look for the "± x %" in the corner of whatever graph. It's the difference between this being news, or the opposite of news: something that leaves you less informed than before you read it.

* * *

A somewhat unrelated gripe of mine: When we tally votes or responses, we end up with integers for one response or another.  Sometimes we turn them into percentages, sometimes as fractions. Regardless, people report the one tally versus the other, and mention the difference as the lead. It is my imperial, autocratic, and kommissarial opinion that we should report the swing instead.

So, a change from 49% to 48% is a 1% change in the lead. But I think— just as a matter of perspective— that it's better to consider the swing: just .5% of people changed their opinions. Yes, these are identical assertions numerically, but this is how I would prefer it.

And when I hear "a sweeping High Court decision of 3 to 6!!", I grind my teeth. Yes, that does mean twice as many people went one way as another— but it also means that if just two people's minds had changed, it would gave gone the other way. A matter of two people out of nine is important, but suddenly seems not quite so epic as three against six.

But this swing-vs-lead thing is just me happening to like one numerical representation more than another.

It's the thing about margins of error that I consider to be a Real Problem with the way the news is reported.

Tags:

Current Location: Edmonton, Alberta
Current Mood: erroneous
Current Music: Grace Jones- Demolition Man

respond...

-¤- Powered by LiveJournal.com