# If it looks like a slowdown, and walks like a slowdown, and quacks like a slowdown, is it a slowdown?

If it looks like a slowdown, and walks like a slowdown, and quacks like a slowdown, is it a slowdown?

http://ift.tt/2ErQIJv

Guest essay by Sheldon Walker

At the moment, I am working on a statistical proof which will clearly show that the recent slowdown was real, and that it is statistically significant. Dark forces have been trying to prevent me from finishing this proof. “He who must not be named” (othewise known as Tamino, aka Grant Foster, oh no, I just named him) has been casting Harry Potter type spells at me, like “autocorrelation”, “heteroscedasticity”, and “multicollinearity”. Luckily I have a good supply of magic jellybeans, and I have managed to stay safe so far.

Sometimes I like to take a break from working on the serious stuff, and one of the things that I enjoy doing is experimenting with data visualisation. Sometimes it is possible to create striking images, which convey an idea or message in a way that a statistical test just can’t do.

I need to make everybody aware that the following images are not statistical tests. When you look at them, you may say “that really looks like there was a slowdown”, but the image does not provide an objective result which can be measured. You are getting a subjective impression.

You might now understand why I chose the title for this article. If it looks like a slowdown, and walks like a slowdown, and quacks like a slowdown, is it a slowdown? Subjective impressions are not always wrong. In fact, they are often right. We could not function efficiently as humans, if we could not rely on our subjective impressions. That pile of papers “looks” like it might fall over. That car “looks” to be out of control. It “looks” like it might rain.

Enjoy looking at the images, and see if they give you a subjective impression. Whatever you do, don’t tell a warmist that the images prove that there was a slowdown. You are likely to get a 30 minute lecture on how there is no statistical proof that a slowdown ever existed. You may even have to endure Tamino’s sermon on “how things are not always what they look like”. In other words, who are you going to believe, Tamino, or your own lying eyes?

Sorry, I need to give you some “objective” data about the images. For the first image I used 741 trends. These trends came from the date range 1970 to 2017. I only used trends which were 10 years long or greater. The maximum trend length was 47 years Every possible starting year, was paired with every possible ending year. So some example trends were 1970 to 1980, 1970 to 1981, 1970 to 1982, …, 1971 to 1981, 1971 to 1982, …, 2001 to 2011, 2001 to 2012, …, 2006 to 2016, 2006 to 2017, 2007 to 2017.

Hopefully I have explained that clearly. Every possible trend which is 10 years or greater in length, from the data range 1970 to 2017.

Then I picked the trends that I thought were “in” the slowdown. I made this decision based on other testing that I have done. I used the date range 2001 to 2015 for the slowdown. The trends in this group started in 2001 or later, AND ended in 2015 or before. There were not many of these trends, only 15 of them. Since there are not many of them, I will list them all.

 Start End 2001 2011 2001 2012 2001 2013 2001 2014 2001 2015 2002 2012 2002 2013 2002 2014 2002 2015 2003 2013 2003 2014 2003 2015 2004 2014 2004 2015 2005 2015

All that you need to know now, is that the slowdown trends were plotted in red, and the non-slowdown trends were plotted in blue.

There is a bonus point for anybody who can name the famous French monument that looks like Graph 1.

Graph 1

This graph surprised me when I plotted it. So I thought that other people might enjoy it. I didn’t expect the slowdown trends to be clustered so much to the left. I would call that a striking image.

A problem that people like me have, is that we are always anticipating the negative comments that other people might make. I thought that people might complain that Graph 1 was “unfair”, because the slowdown trends only go up to 14 years in length, but the non-slowdown trends go up to 47 years. The huge mass of non-slowdown points above 14 years in the middle, makes the red points look like they are squashed into the left hand side, making it look more like a slowdown.

I think that this would be a fair comment, and the last thing that I want to be is unfair. So I removed all of the non-slowdown points above trend length 14 years, and plotted a second graph. See graph 2.

Graph 2

This graph is probably not as striking as Graph 1, but it still manages to create the impression that the slowdown trends are “unusual” when compared to the non-slowdown trends.