HADCRUT4: Statistics, Science and Spin
Continuing from the previous post about the newly developed HadCRUT4 global temperature dataset, it continues to come out in frustratingly small bits and pieces. The paper describing the HadCRUT4 revisions is here.The latest is a press release issued by the UK Met Office on 19th March that contains nothing new about the adjustments made to HadCRUT3 to obtain the new dataset. It does however list the top ten warmest years in both datasets. Click on image to enlarge.
This is an exercise in how not to present statistical data, and in drawing too much from it. It does what any good science teacher instructing A-Level science would admonish pupils for if they did the same.In HadCRUT3 the warmest year was 1998 with a temperature anomaly of 0.52 dec C, followed by 2010 with 0.50 and 2005 with 0.47. Although bear in mind this is the ‘other’ version of HadCRUT3 being used, link not available, that is quoted far less frequently than the ‘usual’ version of HadCRUT3 in which 1998 has a temperature anomaly of 0.53, 2010 has 0.47 and 2005 also 0.47. You will note that the major change between the two HadCRUT3 versions is that 2010 has been increased by 0.03 deg C.Fig 2 shows HadCRUT3 and HadCRUT4 together. Click on image to enlarge.
The media have almost unanimously said that HadCRUT4 now makes 2010 the hottest year. But what HadCRUT4 actually shows is that 2010 ties with 2005 for the hottest year and a trivial 0.01 deg C above 1998. Given the errors (0.1 deg C or ten times the difference being quoted) the only scientifically respectable way to describe the warmest years would be to say that 1998, 2005 and 2010 all tied, but that would have perhaps been a little to inconvenient. I note that NasaGiss, and here, faced a similar problem but chose to follow the respectable route in not calling a record if a year has a temperature difference of only 0.01 deg C.In fact a better way to summarise HadCRUT4 is that one cannot rank with confidence the top eight years (ranging from 0.53 to 0.48 deg C) so they should all be declared statistically indistinguishable.Bad PracticeIt is regrettable, and in my view poor form, to release the HadCRUT4 top ten years and not simultaneously release the complete dataset so that a more detailed look can be made at the time the media are compiling their stories based on the Met Office press release. Not releasing the full dataset could lead to accusations of ‘spinning’ the data. The HadCRUT4 complete data is said to be available in a few days. There is no reason why the Met Office’s press release of the 19th March could not have been held for a few days to coincide with the full data. In most other scientific disciplines this is standard practice.Some parts of the media are claiming that HadCRUT4 shows an increased warming since 1998 of 0.11 deg C (said to be 0.04 deg C more). However, until the full data is available it will not be possible to look at this figure in more detail though looking at the data that is available I don’t think that figure will stand up.Also mentioned in the Met Office press release is the global warming ‘signal’ seen since 1900, which they say is 0.75 deg C. Of course that ‘signal’ is not homogenous as we have discussed because it includes the periods not influenced by mankind and those said to be under his influence.Overall the new HadCRUT4 dataset does not seem to change anything. It even seems to emphasise the lack of warming in the past 10-15 years more than HadCRUT3 did. Although we do not yet have HadCRUT4 data for 2008 it is obvious that HadCRUT4 has less variance in this short period. HadCRUT3 had a range of 0.12 between the top ten warmest years with a fairly even spread. HadCRUT4 has the top three years separated by 0.01 and years 4 -8 also separated by 0.01 deg C, in fact years 4,5,6,7 are identical numerically.When the HadCRUT4 data is available the GWPF logo will be updated.HadCRUT4 is reported to be released shortly and goes up to 2010. Given that 2011 was a rather cool year and that we have previously discussed how its inclusion alters the statistics of trend lines, I hope any trends calculated ending in 2010 are qualified appropriately before they are related to the media.Finally it has been suggested that if HadCRUT4 was used instead of HadCRUT3 I would have lost my recent climate bet. It wouldn't have made any difference to the outcome.Feedback: david.whitehouse@netzerowatch.com
