Signs of the times? Not likely.

Following the tornado in Joplin, MO, I found myself in a discussion about whether this was another sign of the times. The argument presented to me was that these natural disasters are becoming more frequent. I countered that I did not believe this was the case, and before I knew it, evidence was being presented to demonstrate to me that these events were, indeed, more frequent now than in the past.

The evidence as it was presented was that since the 2007 adoption of the Enhanced Fujita scale--a scale the measures the strength of a tornado--there have been 5 EF5 tornadoes, four of which have occurred in 2011. Upon my stating that it was not possible to determine if that was any meaningful change without EF tornado data from before 2007, I was informed that there had been about 50 EF5 tornadoes since 1950.

The basic assumptions that immediately came to mind were that annual tornado frequencies follow a Poisson distribution. So 50 tornadoes over 57 years (1950 - 2006) suggests an annual tornado rate of 0.877. The annual tornado rate from 2007 - 2011 would be 5 tornadoes in 5 years, or 1.000. I didn't run a formal comparison, but my suspicion is that the difference in those rates wouldn't be noted as significant by a statistical test. And I said as much.

And then followed the counter argument: "You seem to have missed the fact that there were four tornadoes in 2011 alone."

Well, if you're going to give me silly arguments, you're going to get walloped with the evidence. So here we go.

I went to Wikipedia and pulled off the data for all of the EF5 tornadoes since 1950. I loaded them into a spreadsheet, and then imported the data into R 2.12.2 to do some analysis. First, let's take a look at the annual frequency of EF5 tornadoes.



As you can see, many of the years saw no tornadoes at all. Many saw only 1, and in 12 years, there were more than one EF5 tornadoes observed. I'll also point out that in four of the multiple EF5 tornado years, two or more tornadoes appeared in the same storm system. For instance, in 1974 we saw six EF5 tornadoes all on the same day. In 2011, three EF5 tornadoes were noted on one day in April.

Perhaps the most interesting observation, and this is strictly visual, is that there appears to be a more dense appearance of tornadoes on the left side of the graph than on the right. In other words, there were more EF5 tornadoes in the beginning of the study period than in the end.

Looking at the data in a slightly different way, we see how frequently multiple tornadoes are observed in a year. In the figure below, it's pretty clear that multiple EF5 tornadoes are pretty rare. Based on the data for 62 years, and having seen 2 years with 4 tornadoes, we can expect to see 4 EF5 tornadoes occur 3.2 times every 100 year. In other words, about once in every 33 years. That last time we saw 4 tornadoes in a single year was 43 years ago. Seems like we were due for a big year.


Now, just to solidify this even more, let's run a poisson regression model to see if there has been a change in the frequency of EF5 tornadoes between 1950 and 2011. Running the model (using the glm function in R), we find that there has been a 1.1% reduction in the frequency of tornadoes over time. However, this reduction is not statistically significant (p = 0.129). Based on the data, then, all we can say is that there is insufficient evidence to support the claim that the number of EF5 tornadoes has increased since 1950.

What does this have to do with signs of the times? Well, all I'm trying to say is that if you think that these natural disasters are God's way of telling us that the end is rapidly approaching, then you might want to rethink how that fits the data. If four EF5 tornadoes in 2011 is a warning from God, he must have been trying to send an even stronger warning in 1974. The sophist in me wants to ask if the end of the world was closer at hand in 1974 than it is now.

Ultimately, the point is that the Second Coming of Christ cannot be boiled down to a single event. It's a process that has been underway for a long time now. I think there's very little point in trying to interpret the "signs of the times," especially when we have so little data to be able to understand if anything we're experiencing is any different than what past centuries have experienced. It's a lot more productive to be mindful of what you as an individual can do to be worthy to meet your God, regardless of when that time comes.



Some limitations on my analysis. First, it's limited to EF5 tornadoes. In understanding the overall trend of destruction by tornado, we'd need to include all tornadoes. From what I've presented, we really can't determine if there has been an uptick in, say, EF4 tornadoes. Professionally, I'd be surprised if there was.

Second, the data don't actually follow a Poisson distribution, making the model I built somewhat biased. The figure below illustrates how the data are distributed...more of a zero-inflated Poisson distribution (such distributions have more 0 counts than we would expect and more counts in the right tail than we would expect if adhering to a Poisson distribution). Indeed, the data are much too overdispersed to be modeled particularly well by poisson regression, and zero-inflated adjustments would be most appropriate. At the same time, such methods aren't likely to change the conclusions.