Here’s an edited extract from an article from AirSafe.com, a website devoted to tracking fatal and other significant events involving commercial airlines.
The writer used this website’s accumulated data to answer the question “How frequent are sequences of 3 or more fatal/significant aviation safety and security events where the time between events is 10 days or less?”
With the exception of 2007 and 2009, every year since 1996 included at least1 sequence of 3 fatal or significant events that were separated by no more than 10 days. There was a sequence of 5 significant (nonfatal) events in January 2008.
There were 14 sequences of 3 or more events separated no more than 10 days. 3 were sequences of 5 events, 3 sequences had 4 events, and the other 10 consisted of 3 events each.
Most of the fatal and significant events since 1996 were not part of any sequence of 3 or more events.
There have been 5 fatal or significant events in a 70-day period between 20 December 2008 and 25 February 2009, but only 2 of them less than ten days apart.
For 12 of the past 13 calendar years (1996 to 2008), there has been at least 1 grouping of 3 or more fatal or significant events that occurred over a relatively short period. At the same time, no information indicate any connection among them, nor is there evidence that the earlier events in the sequence made a later event more likely.
He came to the conclusion that he can “no longer say that plane crashes don’t happen in 3’s. Since at least 1996, they have happened in 3’s, in 4’s, and in 5’s. Let us all hope that there are no 6’s and 7’s in our future.”
A poster immediately noted that the researcher only listed the events, without using a probability model to compare whether the events in 3’s are statistically significant sequences.
Personally, I see a problem with the fact that he jumbled “significant events” and fatal accidents together. We also don’t know whether his information is exhaustive, and his definition of “significant events” is elastic. It would be very interesting to obtain a viewpoint from forum members who are skilled in statistical analysis – perhaps you could post some ideas to the source?
I’m relieved that the article itself also states that this exercise was done “just for fun”, so I suppose anything goes. But if it was a serious study, I would have questioned the arbitrary time criterion of 10 days. One could just as easily pick 3 days study interval to disprove, or one month to support the idea of accidents happening in threes.
I agree with Sentinel, but I don’t think we can dismiss it out of hand without scrutinising his data and methods. For the record, he has a full definition of “significant event” here Definitions of key terms but:
Whether an event is included as a significant event is based on many factors. Every event that has one or more of these characteristics may not be included.
There’s the rub.
The reason why I am interested is that I have to wade through rumours, anecdotal remarks, newspaper clippings, the CAA, credit ratings of maintenance companies, EU regulations, IOSA compliance, word of mouth and financial reports to decide whether or not to ban airlines from use in my sphere of influence.
South African airlines have been “fortunate” (no doubt due to competency) in terms of fatal events but I have noticed that “insignificant events”, or least reporting of insignificant events come in multiples. And before I get a -1 for woo-woo nut, this can be substantiated from newspaper articles!
Let make this clear, I do not believe that there is some chappie with a checklist somewhere saying “oh, it’s time for another multiple of 3!”. I am asking if there is someone on the forum with sufficient scientific background to ask the right questions. The writer of the article is well known in airline risk analysis and his website provides information to the traveling public. If there are flaws in his research, it should be pointed out to him. Even if it was done just for fun, he has a responsibility to not use his position to promote superstition.
The site surely is a statistical joke: selection bias, loose definitions, ignoring large-number statistics and making assorted unwarranted inferences. Given the number of daily passenger flights, having three accidents in short succession (ten days) is hardly surprising. There are 365 different ten-day runs per year to choose from if we are allowed to include up to nine days of the previous or following year, and, worldwide, there are thousands of flights taking off daily. Note, however, that the weakness is not in the statistical analysis itself, but in the vagueness of the premises it is made from.
It panders to assorted fears of people. The “comes-in-threes” is a dead giveaway. “Three” is a magical number, isn’t it? Try “comes-in-twos” and people will yawn. Try “comes-in-sevens” or “comes-in-thirteens” and you have nothing convincing to work with and people will still yawn. “Threes” is the proverbial bull’s-eye for superstition-salesmen – just enough to exceed our tolerance for “coincidence” but hardly enough to make any kind of convincing empirical or inductive case.