Omitted Variables Bias in Estimating the Rate of Global Warming

There has been a lot of discussion in the media and blogosphere of a supposed cooling of the global climate since 1998 that is being pounced upon by climate change sceptics. This supposed trend is almost certainly not a statistically significant break in the warming trend. Real Climate points out that whether you see a cooling effect or not depends on which data series you use. Apparently the Hadley data series is missing data from the parts of the Arctic where the most warming has been occuring:



The image shows the difference between the average temperatures in 1999-2003 and 2004-2008.

But the atmosphere is only a small part of the total heat budget of the planet. An excellent blog post from Climate Progress explaining the importance of the increase in ocean heat content in understanding global warming. These charts show the increase in heat content of the oceans till 2003 and from 2003-2008:





Climate Progress dramatizes this data nicely by pointing out that the ocean is warming at the rate of 190,000 GW... If you try to estimate the trend in atmospheric temperature while ignoring this massive storage of heat in the ocean you may fall victim to the classic econometric problem "omitted variables bias". When you estimate a regression model omitting some important variables that are correlated with those that you include in the regression your estimates of the effects of the included variables will be biased.

I wrote two papers on this topic. In the papers, I showed that taking into account the build up of heat in the ocean resulted in much higher estimates of the sensitivity of global temperatures to increases in greenhouse gases than when you just use atmospheric temperature to produce an estimate. Also, that just looking at the atmosphere you will estimate that temperature responds very fast to increases in greenhouse gases and that after just a few years the adjustment to a new equilibrium temperature is complete. These are symptoms of omitted variables bias.