Regression to the Mean

I'm often reminded of the simplistic view that a lot of people have about climate and weather. For example, the belief that "it's snowing therefore global warming must be a hoax" is probably the most egregious example; but it exists nonetheless.

This post over at Capital Weather Gang seeks to answer the question, "since it was unseasonably warm in DC this winter, does it follow that it will be unbearably hot in DC this summer?" The author's answer to the question is: not necessarily. If you look at the scatterplot Ian Livingston created, you'll see that there is a positively sloped relationship between average winter temperature and average summer temperature.

(from afagen on Flickr)

However, the relationship appears statistically weak (they didn't post the r-squared value, so I can only estimate). What I see when I look at the scatterplot is a classic regression toward the mean. In other words, unseasonably warm winters are a data point in a much larger dataset, not necessarily evidence of a trend. Next winter the average temperature could fall anywhere on the scatterplot, but the most likely out come is that it will fall near the regression line.

Now, I'm not a scientist and there may be a legitimate meteorological reason to predict one thing or another for this summer. But it seems like a lot of weather perceptions are based on superstitions and fallacious logic, or driven by agendas.

I still think it's a shame that the term "global warming" is what stuck, rather than "climate change". Very small changes in temperatures are all that it takes to trigger changes in climate; but very small changes in temperature don't pop out on scatterplots and bar charts and other statistical graphics. Small changes in climate look like almost no change.

If DC does have a painfully hot summer, surely there will be people who believe it's karma - the universe's punishment for a mild winter.


    I can't speak to popular perceptions, but weather forecasts are very sophisticated and have very good scientific rationale behind them. Massive models, supercomputers, etc, and a very good track record of accuracy within established parameters.

    There are also the long term forecasts based on seasonal trends. For example, knowing this is a La Nina year, the forecasts did call for a warmer than average winter from the start.