Most scientific studies regarding the hiatus have considered time periods of a decade or longer. The science has looked for potential explanations for the "pause" and considered whether or not the discrepancy between expected versus observed warming might signal important shortcomings in the current understanding.
How long is the pause?
In order to claim there has been warming within a time period, we can't use just subjective judgement. Instead, the null hypothesis of only natural variability occurring has to be statistically rejected. This kind of hypothesis testing has been a routine part of climate science studies for decades.
In the inversion we employ an a priori null hypothesis for the GST history; that is, an initial estimate that there has been no climate change. This is a conservative hypothesis that is also fully independent of any extant models of climate change.
Huang et al. 2000
In statistics, a null hypothesis is what you expect to happen before you run an experiment. The idea is that if the results don't reject the null hypothesis, then you aren't finding anything new or surprising. The most common null hypothesis is the "no-change" or "no-difference" hypothesis.
Wikipedia
This website relies on the methodological choices of Foster and Rahmstorf (2011) and datasets (with some additions) selected by Dr Kevin Cowtan for the Skeptical Science temperature trend calculator to determine periods which do not reject the "no-change" null hypothesis.
The Skeptical Science temperature trend uncertainty calculator is a tool to allow temperature trends to be calculated with uncertainties, following the method in the methods section of Foster and Rahmstorf (2011) (Note: this is incidental to the main focus of that paper).
If the trend differs from some ‘null hypothesis’ by more than 2σ, then we say that the trend is statistically significant.
Skeptical Science
Despite the 2015-2016 El Niño (a natural warming spike, one of the strongest on record) most datasets do not indicate warming exceeding the 2σ error margin in the last 10 to 30 years.
Updates
9 February 2022: The "pause" time periods we consider now always end in the previous full month, instead of the end of each dataset. This way, we more accurately answer the question "does this data rule out non-warming in the last n months?"
Skeptical Science has stopped updating some datasets so that they are years out of date, e.g. NOAA. If we only consider data truncated to end on a warm phase, we may get false positives.
14 March 2017: HadSST from MetOffice was added.
12 March 2017: Dr. Cowtan made some changes to the trend calculator:
- The longest non-satellite dataset, Berkeley land-only, was removed.
- The NOAA land-only dataset was removed.
- HadCRUT4 was labeled "non-global", contradicting the data provider.
- NOAA land/ocean was labeled "non-global", contradicting the data provider.
- UAH 6.0 was added.
In order to provide as complete and objective analysis as possible, we have now re-added the removed datasets on this site. We will provide updates directly from the original data sources (Berkeley Earth and NOAA). The sources are linked on the graph page.
Dr. Cowtan's interpretations regarding "non-globality" were replaced with the original data providers' (the CRU, NOAA) view.
Finally, the "HadCRUT4 krig v2" dataset was marked "unofficial". The provenance of this data is undocumented. It is apparently Dr. Cowtan's own analysis and not the official HadCRUT 4 data which is separately listed.
Validating the calculation
To reproduce these uncertainty calculations on Skeptical Science, hover the mouse pointer over the "pause length" value given on this site (e.g. "16 years"). You will see the starting point of the calculation as a decimal number, e.g. "start = 2001.083". Next, go to the Skeptical Science calculator, select the dataset, and enter that number as "Start date". Leave the "End date" box empty and click "Calculate". You can then check that the trend is within the error margin.