4 ms·

The magic of climate science is they literally use correlation as evidence of causation. Climate change getting worse means anything else getting worse at same

The magic of climate science is they literally use correlation as evidence of causation. Climate change getting worse means anything else getting worse at same time can be bucketed into climate change.

nullc 2y ago
Your comment might sound trite to some but is not without a basis. I recently followed a rabbithole on media reports that the recent California wildfires are due to climate change. The underlying study driving the report worked like this: They took a time series of the number of acres burned each year in a part of northern California and used a regression to predict it based on a number of climate figures-- air dryness, wind speed, mean temp, max temp, etc. After finding coefficients for their regression they took an off the shelf counterfactual climate model that predicts the climate without human input. The difference in acres burned according to their regression were then attributed to climate change. An obvious issue, out of many, with this approach is that if you plug in any inputs with a trend over that time-- say hours spend viewing porn online-- the regression is going to end up with non-zero coefficients on it and you could walk away with conclusions like "internet porn viewing is causing wildfires". If you knew ahead of time that these climate stats were actually the main drivers of wildfire then this approach wouldn't be a crazy method to estimate what might happen under different conditions. But this approach cannot tells us if those inputs are drivers, much less the main ones. A generalization that threw in a bunch of additional predictors around population densities, forestry management data, etc. might get closer-- but the data being fit was already so sparse that it was probably already overfit. And at best the result would still only be as good as our ability to speculate and measure possible causes. The question of human climate change's impact on things like wildfires is an interesting, important, and very difficult one. I don't fault the researchers for trying a method that might say something useful, but to take that kind of result and treat it as an established fact and drive public policy from it seems foolish indeed. I'm sure that not all of these sorts of conclusion are equally dubious but unfortunately with media, politicians, and the general public so science illiterate that they're unable to engage with this research on its own terms we really can't trust that they're not without going and looking case by case.
- graemep 2y ago
  The problem with climate change and wildfires is a moral risk. If climate change can be blamed, it means bad management of forests is not the problem, so there is no need to spend money and effort on it. In fact, anything that raises the risk should mean that you put more effort into better management. The other problem is that the models are extremely complex and there are limits on the data available for back testing. Its similar to the problems with economic modelling, rather than those typical in say physics. We also do not have enough planets or time to test hypotheses!
- drawkward 2y ago
  >the regression is going to end up with non-zero coefficients on it This is just not true. Plenty of regressions have coefficients that arent statistically distinct from zero. >But this approach cannot tells us if those inputs are drivers, much less the main ones. Again, untrue. There are plenty of statistically appropriate ways to estimate causality. You might consider looking into the latter-day work of Judea Pearl, a well known computer scientist. "The Book of Why" seems like a decent place for you to start, because ut is for the layperson, and you have a ton of fundamental errors in your statements of "fact." >unfortunately with media, politicians, and the general public so science illiterate that they're unable to engage with this research on its own terms You should also add "the confidently incorrect" to your list!
  - nullc 2y ago
    > This is just not true. Plenty of regressions have coefficients that arent statistically distinct from zero. Sure. But you are responding to something I didn't claim, or at least didn't intend to claim. If you throw in a spurious piece of data that happens to exhibit the same trend, it's going to end up a non-zero coefficient. But that doesn't mean there is a causal relationship. [ https://www.tylervigen.com/spurious-correlations https://www.tylervigen.com/spurious-correlations ] > There are plenty of statistically appropriate ways to estimate causality. I didn't claim otherwise, but that doesn't help for an extremely underpowered analysis which not only didn't even consider causality but didn't consider alternative hypothesis. (In particular, it didn't even consider the null hypothesis except perhaps in some p-hack sense that they may not have published at all if the none of their coefficients had significance according to R's GLM or whatever package they used). That wasn't it's goal, I'm not even accusing the authors of bad science (at a minimum it passed the bar to get published)-- but the conclusions the media were drawing from it couldn't be supported by the work. It's an easy error to make because there is a gap between what we want to know and what we have the data to tell us.
drawkward 2y ago
>The magic of climate science is they literally use correlation as evidence of causation. Ironic that you claim these climate scientists are making assertions without appropriate evidence...
aithrowawaycomm 2y ago
This reflexive cynicism is anti-science, because what you claim is not what's happening here. The authors have a sensible causal hypothesis, which is supported by an empirical correlation. A lot of the value seems to be the hard work they did collecting years of urban rat data from various different countries. Individually, sensible causal hypotheses and correlations aren't worth very much. But when they align and the hypotheses are supported by additional evidence, that's usually a good indication of real causality.

nullc 2y ago

Your comment might sound trite to some but is not without a basis. I recently followed a rabbithole on media reports that the recent California wildfires are due to climate change. The underlying study driving the report worked like this: They took a time series of the number of acres burned each year in a part of northern California and used a regression to predict it based on a number of climate figures-- air dryness, wind speed, mean temp, max temp, etc. After finding coefficients for their regression they took an off the shelf counterfactual climate model that predicts the climate without human input. The difference in acres burned according to their regression were then attributed to climate change. An obvious issue, out of many, with this approach is that if you plug in any inputs with a trend over that time-- say hours spend viewing porn online-- the regression is going to end up with non-zero coefficients on it and you could walk away with conclusions like "internet porn viewing is causing wildfires". If you knew ahead of time that these climate stats were actually the main drivers of wildfire then this approach wouldn't be a crazy method to estimate what might happen under different conditions. But this approach cannot tells us if those inputs are drivers, much less the main ones. A generalization that threw in a bunch of additional predictors around population densities, forestry management data, etc. might get closer-- but the data being fit was already so sparse that it was probably already overfit. And at best the result would still only be as good as our ability to speculate and measure possible causes. The question of human climate change's impact on things like wildfires is an interesting, important, and very difficult one. I don't fault the researchers for trying a method that might say something useful, but to take that kind of result and treat it as an established fact and drive public policy from it seems foolish indeed. I'm sure that not all of these sorts of conclusion are equally dubious but unfortunately with media, politicians, and the general public so science illiterate that they're unable to engage with this research on its own terms we really can't trust that they're not without going and looking case by case.

graemep 2y ago
The problem with climate change and wildfires is a moral risk. If climate change can be blamed, it means bad management of forests is not the problem, so there is no need to spend money and effort on it. In fact, anything that raises the risk should mean that you put more effort into better management. The other problem is that the models are extremely complex and there are limits on the data available for back testing. Its similar to the problems with economic modelling, rather than those typical in say physics. We also do not have enough planets or time to test hypotheses!
drawkward 2y ago
>the regression is going to end up with non-zero coefficients on it This is just not true. Plenty of regressions have coefficients that arent statistically distinct from zero. >But this approach cannot tells us if those inputs are drivers, much less the main ones. Again, untrue. There are plenty of statistically appropriate ways to estimate causality. You might consider looking into the latter-day work of Judea Pearl, a well known computer scientist. "The Book of Why" seems like a decent place for you to start, because ut is for the layperson, and you have a ton of fundamental errors in your statements of "fact." >unfortunately with media, politicians, and the general public so science illiterate that they're unable to engage with this research on its own terms You should also add "the confidently incorrect" to your list!
- nullc 2y ago
  > This is just not true. Plenty of regressions have coefficients that arent statistically distinct from zero. Sure. But you are responding to something I didn't claim, or at least didn't intend to claim. If you throw in a spurious piece of data that happens to exhibit the same trend, it's going to end up a non-zero coefficient. But that doesn't mean there is a causal relationship. [ https://www.tylervigen.com/spurious-correlations https://www.tylervigen.com/spurious-correlations ] > There are plenty of statistically appropriate ways to estimate causality. I didn't claim otherwise, but that doesn't help for an extremely underpowered analysis which not only didn't even consider causality but didn't consider alternative hypothesis. (In particular, it didn't even consider the null hypothesis except perhaps in some p-hack sense that they may not have published at all if the none of their coefficients had significance according to R's GLM or whatever package they used). That wasn't it's goal, I'm not even accusing the authors of bad science (at a minimum it passed the bar to get published)-- but the conclusions the media were drawing from it couldn't be supported by the work. It's an easy error to make because there is a gap between what we want to know and what we have the data to tell us.

drawkward 2y ago

>The magic of climate science is they literally use correlation as evidence of causation. Ironic that you claim these climate scientists are making assertions without appropriate evidence...

aithrowawaycomm 2y ago

This reflexive cynicism is anti-science, because what you claim is not what's happening here. The authors have a sensible causal hypothesis, which is supported by an empirical correlation. A lot of the value seems to be the hard work they did collecting years of urban rat data from various different countries. Individually, sensible causal hypotheses and correlations aren't worth very much. But when they align and the hypotheses are supported by additional evidence, that's usually a good indication of real causality.