Hey, remember back in July when we I asked and answered the question: does a cooler-than-normal July correspond to a colder-than-normal winter? (The answer was: no. No, it does not.) Well, now that we’re heading to a mid-November cold snap, I thought it might be useful to look at whether colder-than-normal November temperatures are followed by colder-than-normal winter temperatures (“winter” here meaning December, January and February). And since I’m lazy, I’m using exactly the same “Sherlock” memes and gifs to help explain things along the way.
We’re heading into a prolonged stretch of cold weather. Temperatures for the next 10-14 days look like they’re going to be below normal…but our on-air forecast only goes out to the next 7 days. Those 7 days are definitely going to be below-average — at least, in terms of temperatures. Whether or not they’re below-average in terms of life experience is kind of up to you.
If we take the data from the first 11 days of the month (about -3.5 degrees compared to “normal” so far), AND if we assume that our forecast for the next 7 days is accurate (don’t laugh), AND then if we optimistically hope that the remaining 12 days of November bring us a return to just-average temperatures, this would still add up to the 9th-coldest November on record. Exciting.
Because no one can ever be satisfied just enduring a cold snap while it’s here, I’ve been getting a LOT of questions about what our current cold weather is a harbinger of what’s to come this winter. (Yes, I used the word “harbinger.” Don’t judge.)
…but any halfway-decent scientist wants proof. So, I dove into the record books to look for answers. First, here are the numbers I’ll be referring back to throughout the rest of this post. These are the 30 coldest Novembers on record, along with the average December-January-February temperatures for the following winter (and their overall rank in the 143 years of Nashville records):
A few things of interest within those numbers:
1) The “normal” winter temperature over the last 143 years is 40.4 degrees. The 20 coolest Novembers on record were followed by 12 colder-than-average winters…leaving 8 as above-average.
2) Of the 20 coldest Novembers in Nashville, 4 were followed by Top-20-coldest winters.
3) Of the 30 coldest Novembers, 9 were followed by Top-30-coldest winters.
3) Looking at the flip side, of the 20 coldest Novembers, only 2 subsequent winters ranked in the Top 20 warmest.
5) The Top 30 coldest Novembers were followed by a Top-30-warmest winter only 5 times.
Those five facts above are fun bits of trivia, but they’re picked from a small sample size…they’re not the entire picture. If you want to definitively state that there is a direct correlation between November temperatures and following-winter temperatures, you have to look at all the data, and plot every November temperature against every winter temperature. If there’s a direct relationship, the chart would look like this:
Quickly glance at that, and you probably notice the line through the dots — that’s the “best-fit” line that attempts to summarize all the data points in a straight line. “AHA!” you say, “that line goes UP! Therefore, a cooler July equals a colder winter, and a hotter July equals a warmer winter!” And Moriarty says…
There’s a way to measure the accuracy of that “best-fit” line…basically, measuring how close the data “dots” correspond to the line. It’s called the “Coefficient of Determination,” and it’s abbreviated as R-squared. (Why is it R-squared instead of CD? Why isn’t it called something simpler? These are questions I asked my statistics professor in graduate school, and the answers were…unenlightening.)
Anyway…that “Coefficient of Determination” tells us whether the best-fit line does a good job of summarizing the data, or whether it’s just the software doing it’s best to pound a square peg into a round hole. An R-squared value of exactly 1 means that it’s a perfect fit, while values closer to zero mean that there’s really no relationship between the two things you’re trying to compare.
For November temperatures vs. Winter temperatures, the R-squared value is 0.0447. That’s really low, and tells us that with almost 150 years of data, there is virtually no predictive relationship between temperatures in November and temperatures the subsequent winter.