A concerned reader sent me this recent paper Water-vapor climate feedback inferred from climate fluctuations, 2003-2008, writing:

The following (ala Hansen) IMO should never have been accepted in a "peer reviewed" journal. "The existence of a strong and positive water-vapor feedback means that projected business-as-usual greenhouse gas emissions over the next century are virtually guaranteed to produce warming of several degrees Celsius. The only way that will not happen is if a strong, negative, and currently unknown feedback is discovered somewhere in our climate system."

After my previous post on detecting research bias by checking if the data justify the conclusions, this looked like a good candidate — no significance test, very few data points, idiosyncratic methodology, and bold claims. So I ran some significance tests (R script here).

Would someone like to read the paper and check the logic behind my choice of tests? Its very short.

Using data on water vapor and temperature fluctuations between 2003-2008 collected by the NASA satellite AIRS, Dessler et al. 2008 claim independent confirmation of a strongly positive feedback in the specific humidity parameter λ_{q} of 2.04W/m/K. However, they did not calculate an uncertainty for this value, raising the question of whether the results are distinguishable from the null hypothesis of no water-vapor feedback.

The values of the specific humidity feedback parameter λ_{q} for each of years 2003-2007 relative to year 2008 were 2.10, 2.69, 1.77, 2.69, and 0.94 W/m/K respectively. Direct calculation of the standard deviation of the five data poinst suggests an sd=0.73 with a standard error of 0.37, giving a lower limit to the 95% confidence interval of 1.32 — much greater than zero. However, this approach does not consider the way λ_{q} was calculated relative to a single 2008 value. The uncertainty in the 2008 value must be taken into account in estimating the range of the difference between the anomalous year of 2008 and the other years.

To incorporate the true uncertainty of the difference of two uncertain values, we use a t-test of the difference of means of unequal samples with equal variance. This will incorporate both the variance of the sample of five feedback values, and a sample of one value, zero, assumed to have the same variance as the sample of five. This test yields a t value of 2.54, which at 4 degrees of freedom gives a *p* value of 0.033, slightly larger than the *p*=0.025 critical value required for a conventional one-tailed test to be significant at the 95% confidence level. Below are the *p* values for other non-parametric tests of the difference of means.

Test | P | Significant |
---|---|---|

t.test | 0.03 | NS |

Kruskal | 0.14 | NS |

Wilcox | 0.33 | NS |

All *p* values listed in Table 1 indicate the alternative hypothesis, that water-vapor feedback is significantly greater than zero, falls short of the conventional 95% confidence level. The probability the results arose by chance is high, largely due to a methodology relying on one anomalous year as a reference point. Therefore Dessler et al. 2008 provides no credible justification for claims of the “existence of a strong and positive feedback ” which is “virtually guaranteed to produce warming of several degrees Celsius” (in this century).