07

Are planes crashing any less than they used to?

Recently, I've been spending more of my hiking time looking for old plane crashes in the mountains. And I've been looking for data that helps me do that, for instance the last post. A question that came up in conversation is: "are crashes getting more rare?" And since I now have several datasets at my disposal, I can very easily come up with a crude answer.

The last post describes how to map the available NTSB reports describing aviation incidents. I was only using the post-1982 reports in that project, but here let's also look at the older reports. Today I can download both from their site:

$ wget https://app.ntsb.gov/avdata/Access/avall.zip
$ unzip avall.zip    # <------- Post 1982

$ wget https://app.ntsb.gov/avdata/PRE1982.zip
$ unzip PRE1982.zip  # <------- Pre 1982

I import the relevant parts of each of these into sqlite:

$ ( mdb-schema avall.mdb sqlite -T events;
    echo "BEGIN;";
    mdb-export -I sqlite avall.mdb events;
    echo "COMMIT;";
  ) | sqlite3 post1982.sqlite

$ ( mdb-schema PRE1982.MDB sqlite -T tblFirstHalf;
    echo "BEGIN;";
    mdb-export -I sqlite PRE1982.MDB tblFirstHalf;
    echo "COMMIT;";
  ) | sqlite3 pre1982.sqlite

And then I pull out the incident dates, and make a histogram:

$ cat <(sqlite3 pre1982.sqlite 'select DATE_OCCURRENCE from tblFirstHalf') \
      <(sqlite3 post1982.sqlite 'select ev_date from events') |
  perl -pe 's{^../../(..) .*}{$1 + (($1<40)? 2000: 1900)}e'   |
  feedgnuplot --histo 0 --binwidth 1 --xmin 1960 --xlabel Year \
              --title 'NTSB-reported incident counts by year'

ntsb-histogram-by-year.svg

I guess by that metric everything is getting safer. This clearly just counts NTSB incidents, and I don't do any filtering by the severity of the incident (not all reports describe crashes), but close-enough. The NTSB only deals with civilian incidents in the USA, and only after the early 1960s, it looks like. Any info about the military?

At one point I went through "Historic Aircraft Wrecks of Los Angeles County" by G. Pat Macha, and listed all the described incidents in that book. This histogram of that dataset looks like this:

macha-la-histogram-by-year.svg

Aaand there're a few internet resources that list out significant incidents in Southern California. For instance:

I visualize that dataset:

$ < [abc].htm perl -nE '/^ \s* 19(\d\d) | \d\d \s*(?:\s|-|\/)\s* \d\d \s*(?:\s|-|\/)\s* (\d\d)[^\d]/x || next; $y = 1900+($1 or $2); say $y unless $y==1910' |
  feedgnuplot --histo 0 --binwidth 5

carcomm-by-year.svg

So what did we learn? I guess overall crashes are becoming more rare. And there was a glut of military incidents in the 1940s and 1950s in Southern California (not surprising given all the military bases and aircraft construction facilities here at that time). And by one metric there were lots of incidents in the late 1970s/early 1980s, but they were much more interesting to this "carcomm" person, than they were to Pat Macha.