5 ms·

They show 28 overlapping lines, that many lines simply can't be individually coloured. To accuse them of deliberately choosing a poor colour scheme is plain dis

They show 28 overlapping lines, that many lines simply can't be individually coloured. To accuse them of deliberately choosing a poor colour scheme is plain dishonest. It is a graph comparing the current year to the previous minimum area years, with the entire dataset for context. The choices they have made are completely legitimate for this purpose.

atemerev 10y ago
I suppose points 2) and 3) can (but unlikely to -- see below) be unintentional, but point 1) is clearly intentional and sufficient to consider this chart manipulative. As for the color scheme choice -- almost any other option would have been better. If the colors were random, it would have removed the ability to recover the trend in any way except of tedious manual comparison, but at least it would be certainly bias-free. If mapped to some color gradient -- well, fitting 28 points is hard, but at least close colors would have matched close years. But this scheme is different. It _looks_ like the gradient map, and invites the reader to mentally build the year-to-year trend, but some colors are nearly repeated in distant areas! This is a cardinal sin of using color mapping for dimensional representation, and it's hard for me to imagine this could be done unintentionally. Not to mention that this is not a good choice for representing YTY trend; 2D color map, like in https://xkcd.com/1410/ https://xkcd.com/1410/, would be so much better.
candiodari 10y ago
Fact of the matter is, the GP is right. Can we at least agree the color scheme could have been chosen better ? You are also obviously showing a bias towards one conclusion, and accusing opposition to this conclusion of being "plain dishonest" is not constructive. This is data interpretation. Even attributing 100% to your prior (ie. ignoring the data) is not considered dishonest. Furthermore, when push comes to shove, let me just say that as someone holding a masters in statistics, there is zero useful data on that chart. I think the the point is to show that we are more than 2 standard deviations from the norm ... And then we have a long list of issues with that conslusion. In increasing importance : 1) there is an obvious reason to pick the start of the time series, and that reason is NOT independent of the conclusion that is reached here. Needless to say, that is a huge no-no. 2) more generally the data supporting the conclusion is not randomly drawn from the distribution. Again, huge no-no. 3) is the relation between time of year and sea ice extent well established ? For instance, does the minimum drift ? How much ? (Why I pick that one: because if you ignore the time of year, flatten the data, we are suddenly very far removed from a 2 sigma deviation, and arrive at a much more sane conclusion : lowest extent ever, but not hugely different from before. And of course, seeing 2 extremes of a slowly evolving variable close together is not exactly strange. When you're climbing the hill, every step sets a "new height record". Focusing on that is misleading, to say the least). How often do we hit "lowest ever" because the minimum extent falls at a slightly different point every year ? (looking at the data: quite often, for instance, we hit lowest ever and highest ever extents last year. Lines that hit "lowest ever" at some point: dark red, dirty green, thick bright green, thin dark green, thin light green, purple, orange and another medium-dark green. I can't be bothered to look up the years, but I think the point stands: lowest ever is not a rare occurrence, and therefore "more than 2 sigma below the norm" is not an accurate way to report that). You are making assumptions just by choosing which 2 variables to graph against eachother. Are these assumptions valid ? Why ? If you don't have a reason, then please, shut up until you've done your homework. It may make a lot more sense to plot the minima at the same points every year, to prevent normal noise from generating these extremes, or just only compare minima and maxima, and even then, smoothing out over 4-5 years seems necessary. Or perhaps just don't report things like this. 3) more fundamentally, why would the data be normally distributed at all ? This seems to me unlikely in the extreme. Technically speaking, the data is contradictory: it leads to a conclusion that disproves itself. The steps work like this : if the sea ice extent is indeed dropping over time (the conclusion of the graph), then obviously it's distribution over time cannot be normally distributed. Unfortunately that is a necessary precondition for interpreting the data in the first place. If the data is not normally distributed, then there is nothing strange about a 2 sigma deviation, which at that point is nothing but an arbitrary numerical value that has nothing to do with the distribution. If you want to prove that the data is indeed less than the data of the previous years, there's tests for that. This is not it. 4) that grey band obviously covers 98% of the data in the way years are reported here. Years are reported as special, when the 2-sigma deviation is based on days. The time series regularly dive outside of that band, and it's easy to see why: over (2016-1978)*365 days, we should expect ~2% 2-sigma events (days). That's a LOT of days. 5) we should keep our heads screwed on and ask ourselves if the data is really showing what we think it's showing. I would argue no. It looks to me like something prevented sea ice from reforming between August and December of last year. It seems to me unlikely in the extreme that such a sudden and big change would result from the minute temperature change that global warming contributed this year. Maybe not entirely out of the question, but that's an extraordinary claim, requiring extraordinary evidence. And frankly, this article is not off to a good start. Or to put it more technically : it looks 2016 is drawn from a different distribution than the other years. That means you should find out what happened using other means, and stop using statistics. It cannot help you in this case. Conclusion: there is something weird happening, true, but based on intuition alone. And let's be honest : chances are very good it's not global warming. Something changed, and it was not a 0.2 degree temperature change. This is way too big for that to be a valid explanation.
- DennisP 10y ago
  0.2 degrees is the global average temperature change over the last year. What matters to the arctic is the temperature in the arctic, which has gone up a lot more than that. More importantly, the graph shows area. It doesn't show thickness, which has been steadily decreasing for decades. Now the ice is quite thin, so it doesn't have to thin much more for the area to drop rapidly. If you want to read about the physical processes in detail, and find out why a large sudden drop in ice area due to global warming isn't surprising at all, read A Farewell to Ice by arctic researcher Peter Wadham.
- Oletros 10y ago
  Yes, the one obvious reason to pick the start of the time series is the launch of the instruments to measure it
  - candiodari 10y ago
    Exactly. And that launch was done because of the problem to be analysed. Feedback loop closed. Disastrously wrong results to follow.
    - Oletros 10y ago
      > And that launch was done because of the problem to be analysed. Feedback loop closed. Oh, yes, the Planck satellite was launched because of the problem to be analysed. Feedback loop closed. Meteo satellites were launched because of the problem to be analysed. Feedback loop closed. GPS satellites were launched because of the problem to be analysed. Feedback loop closed. Are you aware that you have arrived to a cnoclusion and you're trying very hard to bent reality to adapt to your conclusion even when your arguments are absurd? >? Disastrously wrong results to follow. And exaclty why the results are disatrously?
      - candiodari 10y ago
        Look you're attacking the weakest of the points. I think you're being unfair in your comparison. But pedantically the point stands. The planck satellite studies a phenomenon that exists independently of the satellite or the humans studying it. This satellite was launched to study climate change by people who claim they're causing it. There's a clear difference. > And exaclty why the results are disatrously? Because using data about something you're doing yourself to study that phenomenon invalidates the basic axioms of statistics. If you do that, all your conclusions about the data become wrong. Why ? Because you yourself are now a factor in the phenomenon under study. At that point, and I've seen this many times, there is no longer any real hope of being correct, as there either a big gaping hole in the model, or circular reasoning. Both obviously invalidate the model.

atemerev 10y ago

I suppose points 2) and 3) can (but unlikely to -- see below) be unintentional, but point 1) is clearly intentional and sufficient to consider this chart manipulative. As for the color scheme choice -- almost any other option would have been better. If the colors were random, it would have removed the ability to recover the trend in any way except of tedious manual comparison, but at least it would be certainly bias-free. If mapped to some color gradient -- well, fitting 28 points is hard, but at least close colors would have matched close years. But this scheme is different. It _looks_ like the gradient map, and invites the reader to mentally build the year-to-year trend, but some colors are nearly repeated in distant areas! This is a cardinal sin of using color mapping for dimensional representation, and it's hard for me to imagine this could be done unintentionally. Not to mention that this is not a good choice for representing YTY trend; 2D color map, like in https://xkcd.com/1410/ https://xkcd.com/1410/, would be so much better.

candiodari 10y ago

Fact of the matter is, the GP is right. Can we at least agree the color scheme could have been chosen better ? You are also obviously showing a bias towards one conclusion, and accusing opposition to this conclusion of being "plain dishonest" is not constructive. This is data interpretation. Even attributing 100% to your prior (ie. ignoring the data) is not considered dishonest. Furthermore, when push comes to shove, let me just say that as someone holding a masters in statistics, there is zero useful data on that chart. I think the the point is to show that we are more than 2 standard deviations from the norm ... And then we have a long list of issues with that conslusion. In increasing importance : 1) there is an obvious reason to pick the start of the time series, and that reason is NOT independent of the conclusion that is reached here. Needless to say, that is a huge no-no. 2) more generally the data supporting the conclusion is not randomly drawn from the distribution. Again, huge no-no. 3) is the relation between time of year and sea ice extent well established ? For instance, does the minimum drift ? How much ? (Why I pick that one: because if you ignore the time of year, flatten the data, we are suddenly very far removed from a 2 sigma deviation, and arrive at a much more sane conclusion : lowest extent ever, but not hugely different from before. And of course, seeing 2 extremes of a slowly evolving variable close together is not exactly strange. When you're climbing the hill, every step sets a "new height record". Focusing on that is misleading, to say the least). How often do we hit "lowest ever" because the minimum extent falls at a slightly different point every year ? (looking at the data: quite often, for instance, we hit lowest ever and highest ever extents last year. Lines that hit "lowest ever" at some point: dark red, dirty green, thick bright green, thin dark green, thin light green, purple, orange and another medium-dark green. I can't be bothered to look up the years, but I think the point stands: lowest ever is not a rare occurrence, and therefore "more than 2 sigma below the norm" is not an accurate way to report that). You are making assumptions just by choosing which 2 variables to graph against eachother. Are these assumptions valid ? Why ? If you don't have a reason, then please, shut up until you've done your homework. It may make a lot more sense to plot the minima at the same points every year, to prevent normal noise from generating these extremes, or just only compare minima and maxima, and even then, smoothing out over 4-5 years seems necessary. Or perhaps just don't report things like this. 3) more fundamentally, why would the data be normally distributed at all ? This seems to me unlikely in the extreme. Technically speaking, the data is contradictory: it leads to a conclusion that disproves itself. The steps work like this : if the sea ice extent is indeed dropping over time (the conclusion of the graph), then obviously it's distribution over time cannot be normally distributed. Unfortunately that is a necessary precondition for interpreting the data in the first place. If the data is not normally distributed, then there is nothing strange about a 2 sigma deviation, which at that point is nothing but an arbitrary numerical value that has nothing to do with the distribution. If you want to prove that the data is indeed less than the data of the previous years, there's tests for that. This is not it. 4) that grey band obviously covers 98% of the data in the way years are reported here. Years are reported as special, when the 2-sigma deviation is based on days. The time series regularly dive outside of that band, and it's easy to see why: over (2016-1978)*365 days, we should expect ~2% 2-sigma events (days). That's a LOT of days. 5) we should keep our heads screwed on and ask ourselves if the data is really showing what we think it's showing. I would argue no. It looks to me like something prevented sea ice from reforming between August and December of last year. It seems to me unlikely in the extreme that such a sudden and big change would result from the minute temperature change that global warming contributed this year. Maybe not entirely out of the question, but that's an extraordinary claim, requiring extraordinary evidence. And frankly, this article is not off to a good start. Or to put it more technically : it looks 2016 is drawn from a different distribution than the other years. That means you should find out what happened using other means, and stop using statistics. It cannot help you in this case. Conclusion: there is something weird happening, true, but based on intuition alone. And let's be honest : chances are very good it's not global warming. Something changed, and it was not a 0.2 degree temperature change. This is way too big for that to be a valid explanation.

DennisP 10y ago
0.2 degrees is the global average temperature change over the last year. What matters to the arctic is the temperature in the arctic, which has gone up a lot more than that. More importantly, the graph shows area. It doesn't show thickness, which has been steadily decreasing for decades. Now the ice is quite thin, so it doesn't have to thin much more for the area to drop rapidly. If you want to read about the physical processes in detail, and find out why a large sudden drop in ice area due to global warming isn't surprising at all, read A Farewell to Ice by arctic researcher Peter Wadham.
Oletros 10y ago
Yes, the one obvious reason to pick the start of the time series is the launch of the instruments to measure it
- candiodari 10y ago
  Exactly. And that launch was done because of the problem to be analysed. Feedback loop closed. Disastrously wrong results to follow.
  - Oletros 10y ago
    > And that launch was done because of the problem to be analysed. Feedback loop closed. Oh, yes, the Planck satellite was launched because of the problem to be analysed. Feedback loop closed. Meteo satellites were launched because of the problem to be analysed. Feedback loop closed. GPS satellites were launched because of the problem to be analysed. Feedback loop closed. Are you aware that you have arrived to a cnoclusion and you're trying very hard to bent reality to adapt to your conclusion even when your arguments are absurd? >? Disastrously wrong results to follow. And exaclty why the results are disatrously?
    - candiodari 10y ago
      Look you're attacking the weakest of the points. I think you're being unfair in your comparison. But pedantically the point stands. The planck satellite studies a phenomenon that exists independently of the satellite or the humans studying it. This satellite was launched to study climate change by people who claim they're causing it. There's a clear difference. > And exaclty why the results are disatrously? Because using data about something you're doing yourself to study that phenomenon invalidates the basic axioms of statistics. If you do that, all your conclusions about the data become wrong. Why ? Because you yourself are now a factor in the phenomenon under study. At that point, and I've seen this many times, there is no longer any real hope of being correct, as there either a big gaping hole in the model, or circular reasoning. Both obviously invalidate the model.