I would be so surprised if we get another record hot year.

Gavin typically uses a 67% confidence bar on his plots, so if that is the case for his image in Reply# 1471, this would mean that his is projection something over a 10% chance that 2017 may be warmer than 206

I replicated Gavin's calculation, and he appears to be using a 95% CI for the 2017 prediction. I superimposed my version on top of Gavin's, using black circles (and black error bars for 2017):

That gives the following:

99.5% chance that 2017 will exceed 2014 (third place or higher)

59.2% chance that 2017 will exceed 2015 (second place or higher)

1.8% chance that 2017 will exceed 2016 (first place)

or:

40.3% chance of third place

57.4% chance of second place

1.8% chance of first place

But there is a (slightly) better way. Gavin's calculations predict the annual mean based solely on the Jan-June mean. But, logically, one might expect that along with the Jan-June

**mean**, the Jan-June

**trend** matters (if temps are rising in the first half of the year the annual mean will likely be higher; if they are falling, it will likely be lower).

Sure enough, both the Jan-June mean and trend are highly significant as predictors, so why not use them both? That gives a model with a slightly lower predicted value for 2017 (1.09, instead of 1.11, relative to Gavin's 1880-1899 baseline) and a smaller standard error (0.044 instead of 0.051). That, in turn, gives the following outcomes for probability:

57.9% chance of third place

41.4% chance of second place

0.2% chance of first place

Clearly this is broadly similar to the first version, but it's a somewhat better fit to the observations.