2012-01-27 09:10:11Null hypothesis argument
John Cook


Got this email, hadn't encountered this exact argument before:

One argument I'm increasingly coming across from AGW deniers is that the null hypothesis is that climate change is natural - it's as though they see it as their "get out of jail free" card when all else fails! They also seem to believe that it absolves them of all obligation to provide evidence of their own!

My recollection from my research days is that the null hypothesis applies to individual experiments rather than the evaluation of multiple lines of evidence. Does this "null hypothesis" argument warrant a rebuttal of its own on SKS?

2012-01-27 09:13:16
Daniel Bailey
Daniel Bailey

Paging Dikran Marsupial:  There's a rebuttal post needing to be written with your autograph at the top...

2012-01-27 09:30:01
Dana Nuccitelli

That's the Judith Curry argument.  She loves talking about how the null hypothesis should be 'no AGW' (or at least AGW isn't responsible for the majority of observed warming), and that climate scientists are trying to change the null hypothesis to 'AGW is happening' (or responsible for most of the warming) (i.e. here).

Really the null hypothesis can be whatever you want it to be.  It just depends what you're trying to prove.  The fake skeptics are trying to argue that we should be trying to prove AGW is correct, whereas climate scientists already know AGW is correct.  But as the emailer says, each experiment can have whatever null hypothesis it likes.  This is really just a new way for the deniers to complain about the existence of the consensus.

But Dikran is certainly more of a null hypothesis expert than me.

2012-01-27 16:50:53


So really, the "null hypothesis" is actually "the hypothesis that is supported if the outcome of this experiment / test is negative"?

That requires that the null hypothesis actually *be* a hypothesis, i.e. a means of explaining experimental / observational outcomes, that is based on our understanding of how the world actually works, that would be supported by particular outcomes.

Trying to argue that "it's natural" is a null hypothesis is just hand-waving.  You might as well say "God did it" (as I'm sure we've all seen argued on various fora).

It seems that the deniers often confuse "null hypothesis" for "no hypothesis at all".

Instead of saying "Well, if CO2 isn't causing the warming, then [factor X] is because of blah blah blah", they're saying "CO2 isn't causing warming, something else is (that we conveniently don't know anything about).  The fact that climate scientists don't know of anything other than CO2 that could cause the warming we're seeing is obviously a sign that these climate scientists are incompetent!"

2012-01-27 18:12:04
Tom Curtis


Bern, it is a favourite tactic of Creationists to treat "goddidit" as a null hypothesis.  Indeed, I have yet to see a tactic used by climate change deniers that was not first used by evolution deniers.


With regard to the null hypothesis itself, it is merely a statistical construct needed to perform statistical tests on data.  In that use, <b>it is never supported by the data</b>.  The most that can ever be said about it is that the statistical test fails to exlude the null hypothesis.  A better way to state it is simply that, the test is statistically significant (= "the null hypothesis is 'falsified'* "), or the test is not statistically significant (="the null hypothesis is not 'falsified'* " )  (*  "falisfied" is here used in a very loose sense, and certainly not in the way falsification is understood in Popper's theory).  


The reason the null hypothesis is never supported is because a test against a different hypothesis using the same data may well exclude the null hypothesis.  That is why the "null hypothesis" only has meaning in the context of statistical tests.  It is only with regard to a defined test relative to a defined hypothesis that it is null.


That, at least, is how I understand the situation.  Of course, any halfway cluey statistician should be able to say that in more accurate language than I have used.

2012-01-28 04:39:45
Dana Nuccitelli

I think that was a good explanation, Tom.  That's my understanding of the null hypothesis too.

The deniers of course think every experiment should have the null hypothesis of "no AGW" or something similar, so that they have to prove the warming is anthropogenic.  But it depends on the study.  If doing a fingerprinting or other attribution study, then that's probably the appropriate null hypothesis.  It's certainly not appropriate for every study though.  For example, research into the climate impacts of AGW isn't going to use the null hypothesis that AGW doesn't exist.

It's just a way for the denialists to claim that climate scientists are assuming, rather than proving the consensus.  It's BS.

2012-02-08 19:59:39
Dikran Marsupial
Gavin Cawley

The null hypothesis ought to be the thing that you need to disprove in order to have confidence that your experimental hypothesis is true.  So for AGW, "natural causes" would be a perfectly reasonable null hypothesis.  The problem lies with defining what is plausible if the null hypothesis is true.  The skeptics will just point to paleoclimate and say that any temperature seen in the past was caused by natural variation (which is true), however the point they miss is that it needs to be what is plausible for natural variability with current natural forcings.  That is a big difference.

For me the null hypothesis has already been rejected, there is a nice diagram in the IPCC report showing attempts to model 20th century climate with natural forcings, anthropogenic forcings and both, and both are needed to explain the observations.  The fact that no skeptic has made a GCM that explains the observed climate without anthropogenic forcings is an admission that that null hypothsis is not true.

Of course the skeptics will then say that the models are nonsense, in which case they need to state what outcomes are plausible under natural variability for the observed natural forcings.  It isn't a hypothesis if it can't be unambiguously stated and it isn't objectively testable.

2012-02-08 21:02:05
Rob Painting

Sounds like a rebuttal brewing Dikran........