3 ms·

Conflating six heads in a row with a periodic new record is unproductive. For any given area of the country we generally have several record highs per year, and

Conflating six heads in a row with a periodic new record is unproductive. For any given area of the country we generally have several record highs per year, and record lows per year, and have every single year since 1880. That's the nature of having a very small period of records. People acting outraged about a simple statement of fact like "since 1880", or the notional, obvious statement that a sporadic high or low is, by itself, not compelling, diminishes everyone.

mikeash 11y ago
Daily records at a point are completely irrelevant to this discussion. The fact that points set multiple daily records per year has no bearing whatsoever on the significance, or lack thereof, of a monthly, global record. You're right of course, that a single record high is not compelling. Even when it's a global, monthly record high. My point in my original reply is that pointing this out is irrelevant and annoying. Was your point merely that this data point, by itself, is not compelling? If so, my original comment, which you complained about missing the point, was exactly about that. Was it something else? If so, I'm having a seriously tough time figuring out what your point is supposed to be.
- inversionOf 11y ago
  The fact that points set multiple daily records per year has no bearing whatsoever on the significance, or lack thereof, of a monthly, global record. Exactly the same principle applies. Look at the title of this submission. Stop and before you knee-jerk a response, contemplate that this isn't "it's hot, yo, and getting hotter". It is specifically that July was the hottest month "ever recorded", which someone quite rightfully noted was since 1880.
  - dang 11y ago
    > before you knee-jerk a response > zealots such as yourself Stuff like this breaks the HN guidelines, and there is much of it in your comments. Please post civilly and substantively, or not at all, even when others are wrong. In fact, especially when others are wrong.
  - mikeash 11y ago
    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?
    - mikeash 11y ago
      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.
      - fluidcruft 11y ago
        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.
        mikeash 11y ago
        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).
fluidcruft 11y ago
You've choesn to insist on replacing the actual continuous data with a weaker derived parameter (period between record breaking events). Of course tests on weakend data have lower power. Should I be surprised that a test which relies on data extremes has less trend detection power? But I'll also note that you haven't even bothered to go back and compute the probabilities in the first place. It was one thing to make this argument in the 90's, but two decades of data later it's plainly ridiculous and it doesn't take nuanced data analysis. https://www.ncdc.noaa.gov/sotc/service/global/glob/201506.gif https://www.ncdc.noaa.gov/sotc/service/global/glob/201506.gi... https://www.ncdc.noaa.gov/sotc/service/global/lo-hem/201506.gif https://www.ncdc.noaa.gov/sotc/service/global/lo-hem/201506....

mikeash 11y ago

Daily records at a point are completely irrelevant to this discussion. The fact that points set multiple daily records per year has no bearing whatsoever on the significance, or lack thereof, of a monthly, global record. You're right of course, that a single record high is not compelling. Even when it's a global, monthly record high. My point in my original reply is that pointing this out is irrelevant and annoying. Was your point merely that this data point, by itself, is not compelling? If so, my original comment, which you complained about missing the point, was exactly about that. Was it something else? If so, I'm having a seriously tough time figuring out what your point is supposed to be.

inversionOf 11y ago
The fact that points set multiple daily records per year has no bearing whatsoever on the significance, or lack thereof, of a monthly, global record. Exactly the same principle applies. Look at the title of this submission. Stop and before you knee-jerk a response, contemplate that this isn't "it's hot, yo, and getting hotter". It is specifically that July was the hottest month "ever recorded", which someone quite rightfully noted was since 1880.
- dang 11y ago
  > before you knee-jerk a response > zealots such as yourself Stuff like this breaks the HN guidelines, and there is much of it in your comments. Please post civilly and substantively, or not at all, even when others are wrong. In fact, especially when others are wrong.
- mikeash 11y ago
  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?
  - mikeash 11y ago
    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.
    - fluidcruft 11y ago
      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.
      - mikeash 11y ago
        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).

fluidcruft 11y ago

You've choesn to insist on replacing the actual continuous data with a weaker derived parameter (period between record breaking events). Of course tests on weakend data have lower power. Should I be surprised that a test which relies on data extremes has less trend detection power? But I'll also note that you haven't even bothered to go back and compute the probabilities in the first place. It was one thing to make this argument in the 90's, but two decades of data later it's plainly ridiculous and it doesn't take nuanced data analysis. https://www.ncdc.noaa.gov/sotc/service/global/glob/201506.gif https://www.ncdc.noaa.gov/sotc/service/global/glob/201506.gi... https://www.ncdc.noaa.gov/sotc/service/global/lo-hem/201506.gif https://www.ncdc.noaa.gov/sotc/service/global/lo-hem/201506....