Tony Finch - Twitter echo chamber fail

dotatfanf wrote
on 3rd October 2010 at 20:14
Previous Entry Add to Memories Share Next Entry

Twitter echo chamber fail

"Lots of people are WRONG on the Internet!"

Twitter users have a propensity to retweet, echo and rebroadcast interesting facts and links without very much fact checking. For example,

@_468 This October has 5 Fridays, 5 Saturdays and 5 Sundays all in one month. It happens only once in 823 years. waw.
(700+ retweets according to search.twitter.com)
@bryanthatcher This month has 5 Fridays, 5 Saturdays and 5 Sundays. It happens only once in 823 years.
(300+ retweets according to search.twitter.com)

Those retweet counts do not include the vast numbers of copies that didn't use twitter's retweet mechanism.

If you think about it, this happens whenever the month has 31 days and the first day of the month is a Friday. Because the years cycle through the days of the weeks fairly evenly, you would expect this to happen for a particular month about 1/7th of the time - in fact if you average over the entire Gregorian 400 year cycle it happens 1 in 7.1429 Octobers. Each year has 7 months with 31 days, and if you average over all of them you expect about one month in each year to have this property, and in fact you get precisely this result if you average over the entire Gregorian cycle. There happen to have been two months this year with 5 Fridays, Saturdays, and Sundays because January also started with a Friday.

1 in 7 and 1 in 1 are both rather more frequent than 1 in 823.


(Leave a comment)
From:naath
Date:2010-10-03 22:13 (UTC)
(Link)
I wonder where 823 years comes from... surely the rarest possible Interesting Date Thing in the Gregorian calendar would happen once every 400 years or never since the Gregorian calendar has a 400 year cycle. Oh well, I suppose unless you think Easter is Interesting (which it probably is, to be fair).
(Reply) (Thread)
From:cartesiandaemon
Date:2010-10-04 12:36 (UTC)
(Link)
I tried googling, but of course, found nothing other than this meme...
(Reply) (Parent) (Thread)
From:fanf
Date:2010-10-04 13:16 (UTC)
(Link)
Good point about 400 < 823 :-)
(Reply) (Parent) (Thread)
From:kotturinn
Date:2010-10-04 07:48 (UTC)
(Link)
What happens if you just do month=October?
(Reply) (Thread)
From:fanf
Date:2010-10-04 08:58 (UTC)
(Link)
That's the 1/7th case :-)
(Reply) (Parent) (Thread)
From:kotturinn
Date:2010-10-04 11:22 (UTC)
(Link)
Sorry - that'll teach me to wait until I'm awake before trying to read!
(Reply) (Parent) (Thread)
From:dougs
Date:2010-10-04 11:57 (UTC)
(Link)
2010, 2021, 2027, 2032, 2038 ...
(Reply) (Parent) (Thread)
From:deliberateblank
Date:2010-11-01 19:23 (UTC)
(Link)
No need for the "...", that's when the world will end.
(Reply) (Parent) (Thread)
From:samholloway
Date:2010-11-02 17:02 (UTC)
(Link)
If the world ends, does the calendar end, too?
(Reply) (Parent) (Thread)
From:fanf
Date:2010-11-02 17:06 (UTC)
(Link)
It restarts after the world has rebooted.

http://www.h-online.com/security/news/item/Spontaneous-worldwide-reboot-of-Check-Point-appliances-1129484.html
(Reply) (Parent) (Thread)
From:captain_aj
Date:2010-10-04 09:12 (UTC)
(Link)
This is one of the reasons I'm wary of joining Twitter. The urge to exploit reflex retweeters and spread deliberate disinformation around the twittersphere would be way too high. Just take some fears/prejudices that you can wrap up into 140 characters and stand back ...
"OMG the eu is forcing buckingham palace to be demolished to build a mosque. I'm not racist but this is human rights gone mad" or "Amazon.com is rounding up gays and mentals into concentration camps"
(Reply) (Thread)
From:vatine
Date:2010-10-04 10:33 (UTC)
(Link)
Did you know that Friday is the most common "13th of the month"? With one more occurrence over a 400-year cycle than any other weekday being "the 13th" (58 times, as opposed to 57 times).
(Reply) (Thread)
From:fanf
Date:2010-10-04 11:45 (UTC)
(Link)
There are 400x12 months so there are 4800 13ths. Three days of the week fall on the 13th 685 times, three fall 686 times, and Friday falls on the 13th 687 times. (Looks like your error was to count one 13th per year.)

Edited at 2010-10-04 12:37 (UTC)
(Reply) (Parent) (Thread)
From:vatine
Date:2010-10-04 12:45 (UTC)
(Link)
Er, yes.
(Reply) (Parent) (Thread)
From:cartesiandaemon
Date:2010-10-04 12:36 (UTC)
(Link)
I remember I was surprised to learn that "leap days" in the Jewish calendar were carefully tuned to keep certain festivals on the sabbath, or not on the sabbath, or not ADJACENT to the sabbath.

Imagine my surprise when I realised, that I had never realised before, that the gregorian calendar also synchronised with days of the week, because the 400 year cycle was a whole number of weeks.
(Reply) (Parent) (Thread)
From:simont
Date:2010-10-04 12:56 (UTC)
(Link)
(Well, even if that hadn't been the case, it'd have synchronised every 2800 years at worst.)
(Reply) (Parent) (Thread)
From:cartesiandaemon
Date:2010-10-04 13:13 (UTC)
(Link)
Oh yes, it'd obviously be quite difficult to design a natural-seeming calendar which DIDN'T eventually repeat. But somehow the two cycles lining up seemed suspiciously neat. And yet, was presumably a complete coincidence.

I wonder if anyone DID propose any alternatives to the gregorian calendar: obviously you could do something other than the 100-400 exceptions, especially if you're willing to alter the 4-year cycle[1], but I don't know if any would have made equal sense, or had ever been suggested.

[1] Although, given the hassle they had getting that going, they were probably wise not to.
(Reply) (Parent) (Thread)
From:simont
Date:2010-10-04 15:30 (UTC)
(Link)
And yet, was presumably a complete coincidence.

The only way it could not be a coincidence would be if they considered lots of possible periods (in years) for the Gregorian calendar and picked the one which made the number of days come out to a multiple of 7. (Equivalently, I suppose, look for a good rational approximation to 1/7 of the number of days in a year, and use its denominator as the calendar period.)

I wonder if anyone DID propose any alternatives to the gregorian calendar: obviously you could do something other than the 100-400 exceptions, especially if you're willing to alter the 4-year cycle[1], but I don't know if any would have made equal sense, or had ever been suggested.

At Eastercon 2003 someone gave a talk suggesting a 33-year cycle containing 8 leap years, which he claimed gets an even better approximation to the true length of a year than 97/400. (It appears to be a continued-fraction convergent of the true length.) He also had a very silly conspiracy theory about how some people actually planned to implement that version in the 16th century in order to make the date of Easter work out more sensibly, but I can't remember any of the details.
(Reply) (Parent) (Thread)
From:fanf
Date:2010-10-04 15:38 (UTC)
(Link)
"More sensibly"?! The Gregorian rules for Easter are deliberately awkward to minimise the likelihood of coinciding with Passover. Anti-semitism isn't very sensible even in the 16th C...
(Reply) (Parent) (Thread)
From:simont
Date:2010-10-04 15:50 (UTC)
(Link)
Oh god, I'd forgotten Passover Sense Multiple Festival / Collision Detect. I don't recall the Eastercon 33-year guy saying how he intended to get round that one...
(Reply) (Parent) (Thread)
From:cartesiandaemon
Date:2010-10-05 10:23 (UTC)
(Link)
What happened? I learned a lot about the date of passover and easter, but not this.
(Reply) (Parent) (Thread)
From:fanf
Date:2010-10-05 14:42 (UTC)
(Link)
I think the details are described in "Calendrical Calculations", and I can't remember them because they are somewhat arcane.

But in essence, Pope Gregory did more than adjust the calendar to match the year more precisely. He also adjusted the rules for Easter, making them more complicated to satisfy various political goals. I can't remember now whether the adjustments also looked to the Eastern (Orthodox) church as well as the Jews- probably not since they continued to use the Julian calendar and the ecclesiastical equinox of 21st March, so their Easter would be unlikely to match the others. Dunno.
(Reply) (Parent) (Thread)
From:cartesiandaemon
Date:2010-10-05 15:13 (UTC)
(Link)
Thanks. I couldn't see why it would make any difference if Easter remains _normally_ just after passover, but it sounds like the sort of thing I'm not at all surprised happened.
(Reply) (Parent) (Thread)
From:cartesiandaemon
Date:2010-10-04 15:42 (UTC)
(Link)
The only way it could not be a coincidence would be if they considered lots of possible periods (in years) for the Gregorian calendar and picked the one which made the number of days come out to a multiple of 7.

Well, yes. It seemed unlikely that they WOULD have done, but also, they COULD have done. You could do the 33 year thing, or perhaps something similar but which decides only which normal 4-year leap years to skip. And making the days of the week repeat might be a mild advantage. (I can't see why it would be, but it would be tidy).

I was thinking of the 33-cycle thing. I saw it first invented by Mark Dominus (http://blog.plover.com/calendar/leapday.html) and I don't know if anyone suggested it before that or not. He said ancient persians (?) had something similar but more complicated.
(Reply) (Parent) (Thread)
From:cartesiandaemon
Date:2010-10-04 15:43 (UTC)
(Link)
I'm sure I heard some people discussing the 33 calendar on LJ. IIRC the advantages are that someone will hopefully remember the last 33-year anomoly, so there won't be any anomolies so rare people will not be ready for them. But the disadvantage is that you can't calculate leap years approximately by testing the year for division by 4, you need to do the division by 33.
(Reply) (Parent) (Thread)
From:nonameyet
Date:2010-10-05 07:05 (UTC)
(Link)
http://en.wikipedia.org/wiki/Iranian_calendar
(Reply) (Parent) (Thread)
From:nonameyet
Date:2010-10-05 21:15 (UTC)
(Link)
Sorry.
The Iranian calendar seems to have 8 leap days in 33 years: http://en.wikipedia.org/wiki/Iranian_calendar
(Reply) (Parent) (Thread)
From:pne
Date:2010-11-02 08:03 (UTC)
(Link)
This page is all about alternative leap year calculations (and his entire site has quite a lot of information on calendars and variations thereof, including his proposal for two perpetual calendars called "Symmetry 454" and "Symmetry 010").

He mentions the 8/33 rule but prefers a 52/293 one.

Apparently, which rule you choose depends on the length of the year you want to optimise for: time between successive equinoxes or between successive solstices, for example, will produce slight different lengths. (I'm afraid I don't remember details since I've only skimmed his pages - they're long and full of maths.)

Edited at 2010-11-02 08:04 (UTC)
(Reply) (Parent) (Thread)
From:fanf
Date:2010-11-02 09:34 (UTC)
(Link)
Interesting link, thanks! Re lengths of the year, the Gregorian calendar is intended to match the period between vernal equinoxes, because of the importance of Easter.
(Reply) (Parent) (Thread)
From:deliberateblank
Date:2010-11-02 00:51 (UTC)
(Link)
Which suggests the idea of the Penrose Calendar.

Of course, leap years *ought* to be based on a continually evaluated DDA using real measured year lengths.
(Reply) (Parent) (Thread)
From:fanf
Date:2010-11-02 01:28 (UTC)
(Link)
Ugh, no, observational calendars are a serious operational pain.
(Reply) (Parent) (Thread)
From:hsenag
Date:2010-10-08 20:00 (UTC)
(Link)
http://www.snopes.com/inboxer/trivia/fivedays.asp
(Reply) (Thread)
From:fanf
Date:2010-10-11 09:20 (UTC)
(Link)
Ah good :-)
(Reply) (Parent) (Thread)

(Leave a comment)

Powered by LiveJournal.com