This was just what I needed to brighten my Sunday. I told you I wasn’t paying any more attention to polls, especially national polls for the rest of the cycle. But that doesn’t mean that something stupid in the poll can’t reach out of the tv and slap me in the face.
I get that some people take me to the wood shed for my spelling, and I appreciate all of your support in getting those people preached up. But at least those errors have a solid medical reason. But whether professional or a college poli-sci program, these silly f*ckers are getting paid by the sponsors for these polls. And they can’t even count!
The object of my sarcastic glee this time is the NBC News national poll released today. Full Disclosure NBC News normally pairs with either Marist University or Morning Consult for the polls. So I won’t tar either one of them today, NBC News can eat the whole humble pie.
This is a national presidential poll, and here are the results;
Harris 49% Trump 47% Undecided/other 5%
I wanted to make absolutely sure I wasn’t about to make an ass of myself instead of NBC News, so I triple checked it. Did the math in my head, did the math on my fingers, and pulled up the calculator on my laptop. And it all came out the same.
49+47+5 = 101%. And as Mel Brooks pointed out in The Producers, How many percent is that. Leo? 1500%. How many percent of a play Dan you sell? Max, you can only ever sell 100% of anything!
I gotta admit, I honestly have no idea how they accomplished this, but Alex Witt had the exact same graphic leading off every hour. And I mean, even if the pollsters compiling the figures were brain dead, and didn’t catch a simple error, wouldn’t you think that somebody at either NBC News or MSNBC would catch that when they created the graphic? I mean, it kind of sticks out to you, at least it did for me. Today’s Sunday Funny for you.
I thank you for the privilege of your time.
You are 101% right.
These polls don’t add up.
rounding…
46.6 –> 47
48.6 –> 49
leaves 3.4 + 1.4 = 4.8 percent left,
which rounds up to 5
Screw that…no matter what, you CAN’T have more than 100% of anything…Get it right, or don’t bother at all…
Not sure what you’re proposing to make it fit your definition of “right”…
It’s just how rounding and math works. Unless you’re going to show full resolution of every number, you will have some situations where it doesn’t add up to 100.000. If they had shown the tenths place, it could still have a similar outcome; adding up to 100.1 or 99.9. At what point would you be happy that it’s “right” from your perspective?
If it happened to be “46.666666…” (i.e. 46.6 repeating; 46 and 2/3rds) to “48.666666…” (48 and 2/3rds), that will always cause it to not add up to exactly 100, no matter how many digits you round to. The last digit will always be overcounted when you add the rounded versions.
So, they could have shown it onscreen as 46.66666667 to 48.66666667 with 4.66666667 remaining — yielding a total of 100.00000001. That would not have added any better understanding of the relative difference between candidates in the poll, would have looked ridiculous, would have implied a level or accuracy that the poll simply can’t predict (the margin of error is typically around +/- 3 percentage points anyway; why on earth would you round it to any decimal places), and still wouldn’t have added up to exactly 100.00000000.
It’s a math thing. It’s not a mistake. There are simply combinations of number that are impossible to get it right by your definition. The sum of rounded numbers don’t always add up to the sum of the unrounded raw numbers. Unless the network simply didn’t show percentages and only listed the actual total number of respondents for each candidate and undecided/other. Again, what does that gain? It just makes it more difficult to understand what was found in the poll.
Or would you want them to actively mis-report the numbers — i.e. give incorrectly rounded numbers for the sake of making the rounded numbers add up more cleanly? I would see this as a much worse route, as they would have to decide which number to actively report incorrectly rounded.
And, for clarity, the 2/3rds remainder numbers I gave above is just an easy example. I’m certainly not saying that was the actual numbers in the poll. But, it’s certainly not the only data set where this type of behavior happens. Depending on the total number of respondents that answer, and how many options are available to answer, there are many combinations where this type of behavior happens at various rounding levels.
The goal of percentages, in almost all cases, is to give a meaningful and relatively easy feel for results. If you want to do your own data analysis, use the raw numbers. Otherwise, realize that the presenter has chosen a level of resolution on the percentages based on the situation. In this case, they felt whole integers, on a poll tha likely had a margin of error around +/- 3%, seemed reasonable (and I personally wouldn’t disagree).
And if you insist that they “get it right” anyway, I’m just not sure what that means. In the example numbers I gave above, what would you recommend in order to make the right from your viewpoint?
Thank you for ding the math so I didn’t have to. Of course you are correct.
I have, however, occasionally seen pollsters and others who use percentages add a disclaimer to the effect that, because of rounding, results do not appear to add to exactly 100%. Perhaps this as a general policy would help the innumerate understand a bit better.
Hey, NOTHING about this election adds up—-this is a perfect example!
Nah, there’s a perfetly logical explanation for this. Evidently there’s a small cohort that is both undecided AND for Harris, and another that is both undecided AND for Trump.
I”m glad more people are starting to see just how effed up our polling institutions are-they suck. This election season in particular has really shined a light, and a bright one, on how pollsters are making polls say whatever the people writing the checks want them to say.
the people being polled don’t necessarily vote, some aren’t registered to vote, some have never voted, but they all have an opinion. that’s why polls don’t mean anything .
Hate to tell you this but this happens all the damned time. This isn’t the flub you think it is.
It can also go the other way where it doesn’t sum to 100%. And don’t get me started on multiple select items. I can’t tell you how many executives (people whose title starts with a C) don’t understand when you have multiple items to select it will never sum to 100%.
I have been presenting numbers to audiences for over 45 years and I can guarantee that you just DON’T present three numbers that are supposed to sum to 100% and do not, without at least an asterix saying it is due to rounding. The CEO in the audience will suddenly look up from his phone to say “Gotchya!” Or worse just think it. You could also just add one more significant digit (or rather insignificant digit)—it doesn’t matter with polls since almost none are accurate to 1% anyhow. Just don’t look like a moron who didn’t realize something obvious. But with the news media and the word “moron,” too often the shoe fits.
Absolutely agree they did a poor job explaining a somewhat complicated situation that looks like a mistake on initial inspection.
I like your suggestion of a footnote — even if there’s not room on screen (or on slide) for a fulsome explanation, a short footnote (e.g. “* rounded percentages — won’t always add up to 100%”) could tip viewers off that it’s known & correct, and not an overlooked mistake.