News for hackers. Stuff that matters.
Re:July 2015 was warmest month ever recorded by mikeash on Thursday August 20, 2015 @04:28PM (#10094447)
The exact same principle applies, yes. But the quantities are way different.

Variation at a point is high. Variation for a period of a single day is high. Samples are taken frequently. Thus new records will be set regularly.

Variation for the entire globe is low. Variation for a period of an entire month is low. There are twelve samples per year. Thus new records would not be expected to be set frequently.

Again, what exactly is your point?

Re:July 2015 was warmest month ever recorded by mikeash on Thursday August 20, 2015 @09:07PM (#10095465)
Thinking about this more, samples are actually taken at the same rate: yearly. The difference is how many different things you're tracking in each year.

For regional records, you're tracking 365 events per year. Each one is compared with that day on prior years.

For global records, you're tracking one event per year (since July will always be the warmest month).

Thus, if you'd expect to set a new record every 50 years, then you'd see on average one global record every 50 years, but you'd see about seven daily regional records every year.

Also consider that you should be seeing both record highs and record lows. Record lows continue at a good pace for regional daily records, but there hasn't been a global record low for well over a century.

Re:July 2015 was warmest month ever recorded by fluidcruft on Thursday August 20, 2015 @11:28PM (#10095838)
Records don't occur linearly for a stationary variable. That is you don't expect "an average record every 50 years". The longer you observe the signal the more more extreme the expected outlier.
Re:July 2015 was warmest month ever recorded by mikeash on Friday August 21, 2015 @10:41AM (#10097764)
Sure, that's just a simplification to illustrate the principle. Whatever the expected recurrance for records is, you'll get a lot more of them when you're testing 365 (really 730, since you have both highs and lows) separate values each year than when you're just testing one (realy two).