Count Iblis II has been hot on my trail. Unfortunately he doesn’t really know a lot about statistics claiming that “P values that are not very low are meaningless”. This is unfortunate for him, however he has been pushing me to do some analysis to see if there is a significant difference in the current trend of temperatures and an increase of 0.6 degrees per century. So, I did and here are the results:
There is no evidence to suggest significant lower maximum temperatures than a 0.6 degree increase at Mawson (F = 0.72, p= 0.4; -0.006 +/- 0.013 per year – note that even though the difference is not significant, the trend is still amazingly at a negative 0.6 degree less trend per year).
However, there is evidence to suggest a significant lower minimum temperature trend than a 0.6 degree in crease at Mawson (F=5.26, p = 0.026, -0.017 +/- 0.015 per year).
Tests were done using Mawson’s (Antarctica) maximum and minimum temperatures. So we can conclude that despite what we would normally expect of a 0.6 decrease in maximum temperature (as compared to an increase in 0.6 degrees), the result is not significant. One must conclude that we need more data in this case, as results only go back to the 1950s for Mawson’s records.
However when looking at minimum temperatures, we can conclude that the difference is significant. So in summary:
There is no evidence to suggest tat Mawson’s maximum and minimum temperatures are significantly increasing or decreasing. There is no evidence to suggest that Mawson’s maximum temperatures are significantly lower than a 0.6 degree increase per century, however there is significant evidence to suggest that Mawson’s minimum temperatures are significantly lower than a 0.6 degree increase per century.
In other words, Mawson (Antarctica) just isn’t heating up, and not even close to a 0.6 increase per century.
3 comments:
"Unfortunately he doesn’t really know a lot about statistics claiming that “P values that are not very low are meaningless”."
Perhaps not completely meangless, but it isn't proof of the assumed hypothesis. Unless the P value for some alternative hypothesis becomes very low. But in that case you again have a low P value for that alternative hypothesis.
In science people usually draw exclusion plots based on their data. With more and more data, you can make the excluded regions bigger and bigger, thereby narowing down the region where the quantity you want to measure is hiding.
Global warming is a rise in global temperature. Not a temperature rise of every point on the planet. Temperature can decrease in some place and increase in others and still have an average increase. Right? You do stats.
I would just reiterate some concerns I raised in an earlier post about the use of significance testing generally and linear regression in time-series analysis.
If you have hundreds of observations, a trend that for all intents and purposes is flat can be highly statistically significant.
More important here however, one is also putting ones conclusions at risk by drawing inferences from linear regression when observations are not independent, which they rarely are in time-series. Your slope may be way off and the significance levels cannot be trusted. Even that old Cochrane-Orcutt technique is preferable to linear regression if you’re not comfortable with Box-Jenkins methods. Just yesterday I ran an example contrasting various methods for a colleague and the linear regression results proved highly misleading.
Post a Comment