Although the computer-generated charts are statistically meaningless, the persistence of number '17' in humans likely has its reason.
The first thing that comes to my mind is that this happens because 17 the number from 1 to 20 that we come into most rarely. When we learn multiplication at school, it doesn't have any factors so we tend to skip it, and for the same reason it's rare to encounter it in various problems. This happens also to '7', but '7' is small enough to occur in many occasions such as counting a number of items etc. So, maybe we are more inclined to pick '17' as 'random' because we rarely meet it. Maybe.
13 has pervasive cultural significance, but I agree that 19 seems as "rare" as 17. Perhaps at a glance 19 is less obviously prime than 17. Also note that 17 and 19 are both ages at which nothing really notable happens. 16 is driving age for lots of states and 18 is a big "legal" milestone.
I'd bet it's more the "7" effect. When asked to pick a number between 1 and 10, 7 is picked more often than other numbers. Someone below mentioned that magicians often exploit this for their tricks.
I'm guessing people think of 7 first, since people are used to choosing between one and ten. Then, on reflection, they choose 17 instead since they want to have used the whole range of options.
My economics teacher once used the number 17 as a random number twice in one class, without noticing.
A note on the number 7: magicians actually use the frequency of that number being picked for their tricks. If an amateur card magician asks you to pick a number, for example, "between 5 and 10" or the more daring "between 1 and 10" they want you to say 7. Now that you know this, though, I'd rather you humor us if you're ever asked..
This is a really bad analysis. I'd be willing to guess the author doesn't know much about statistics. (Or more charitably, he's just trying too hard to write for people who don't understand statistics.)
That said, it does look like the sample was large enough for the results to be valid, at least for "17".
Comment 10 from the original poster: "My stats knowledge is baby-simple. I first use an unpaired t-test, and if that doesn't give good results, I move on to a paired test (p<.05). That's pretty close to the limit for me."
What's so bad about it? I thought that generating 347 random numbers was a neat way to demonstrate the variation in a random distribution with the same sample size actually polled.
That doesn't really tell us anything about variance. After all, in that single trial of generating 347 numbers, his computer could just as easily have spat out a uniform distribution.*
To see variance, it would have been better to generate 347 numbers many times over, then plot the standard deviation of the count of each number using error bars.
*You can't actually evenly distribute 347 trials over the range [1,20], but you see my point.
his computer could just as easily have spat out a uniform distribution
But that's the point, isn't it? An exactly even distribution is unlikely. So unlikely, that the first distribution he found was probably fine for use in the article.
I suppose my only real point here is that error bars are boring, and it's perfectly possible to understand what the variation in a typical random set is by looking at the set, without having to explicitly put error bars around a straight line.
That wouldn’t be as intuitive. We have better statistical tools, sure, but his graph was immediately understandable, even if you don’t know what standard deviation is or what error bars are. I would use that graph as an illustration for the uninitiated even if I had exhausted the statistical toolset.
(Also, there are better tools but it’s not as though what he did was meaningless and provided no evidence for the point he was making. It’s not as strong but it is there.)
And of course, there is the small probability that his computer could have picked the number 10 as many as 347 times in a row. Something tells me he would have run the computer simulation again in that case.
You'll lose more than half the time. If you try a more complicated betting structure than "Bet you $5", you'll have to explain it to them, which will make them suspicious.
Let's assume the data in the article is perfectly representative, or that it is chosen "almost 18" percent of the time. I'll say an even 18 to make this easier:
"I'll bet that if you pick any random number from 1 to 20, I can guess it. If I miss, I give you a dollar. If I get it right, you give me ten."
Over 100 trials, you would lose $1 92 times, but gain $10 18 times: net gain of $88. It works almost all the way down to getting $5 per correct guess.
Intuition would tell the guesser that they've got a nineteen-to-one chance of gaining a dollar. You know (in theory) they've got only slightly better than a one-in-five chance of winning.
23 is. Though I am pretty rational person, I can't seem to explain why I see 23 all around. (Guessing it is some sort of psychological bias -- but the evidence from my everyday experience just baffles me).
The first thing that comes to my mind is that this happens because 17 the number from 1 to 20 that we come into most rarely. When we learn multiplication at school, it doesn't have any factors so we tend to skip it, and for the same reason it's rare to encounter it in various problems. This happens also to '7', but '7' is small enough to occur in many occasions such as counting a number of items etc. So, maybe we are more inclined to pick '17' as 'random' because we rarely meet it. Maybe.