EXECUTIVE SUMMARY:   
   - We look again at UFO activity and missing persons, this time   
     long-term missing persons in Australia listed on a govt website.   
   - Adjusting for population changes in the US and AUS over the period   
     1930-2020 we find there is a strong and expected statistical link   
     between UFO activity and a chunk of the missing person data.   
   - We also show a strong likelihood of a causal connection by showing   
     the statistical model found on the middle 1/3 of the data also   
     predicts the relationship in the other 2/3 of the data, sight   
     unseen.   
   - Together with prev analysis of other sets of missing persons in the   
     US and UK we seem to have shown there is likely a global (or at the   
     very least the English-speaking part of it) and long-standing   
     pattern of UFO-related missing persons.   
   - In a future post we shall see almost no segment of the UFO data   
     (e.g. breaking sightings down by shape, color, time or location   
     seen) does NOT find a similar causal link.   
      
      
   We've previously looked at links between UFO activity and missing   
   persons in the US and UK.   
      
   While "links" were seen -- in the sense that years or regions of high   
   UFO activity (using sightings mostly in the US as a proxy for global   
   UFO activity) were also those of the higher numbers of missing persons   
   and contrariwise -- that doesn't constitute very good evidence of a   
   causal connection.   
      
   But we can start here to outline methods where causal connections can   
   be shown.   
      
   I'll again use the NUFORC database of sightings (mostly in the US, but   
   some in Canada and some in other parts of the world) as a proxy of   
   global UFO activity and we'll look at the Australian missing persons   
   data to show the "association" is not limited to just the N Hem.   
      
   The Australian govt has a website listing around 450 "long term   
   missing persons" going back to the 1930s. A missing person case ends   
   up on the list of there is otherwise no clue to the fate of the person   
   concerned and they have been missing for some time. Some, perhaps   
   many, of the cases likely involve mundane explanations.   
      
   But we can correlate our proxy for UFO activity against the   
   year-by-year data -- the year in which the victims were first noticed   
   to be missing -- to see if there is a strong statistical relationship.   
      
   We've done that before. But this time we'll add a twist.   
      
   We'll set up a "validating" procedure. We'll "hide" most of the data   
   from the statistical procedure, see if there is a high-probability   
   link between the proxy and missing person data in the middle 1/3 of   
   the dataset, and then check whether the years before that "training   
   set" and years after the training set both show the same relationship.   
   If they do then we have evidence of a causal relationship.   
      
   As we might recall the US Surgeon General's Report in the 1960s showed   
   the first scientific evidence for causality between smoking and lung   
   cancer by outlining types of statistical links.  It was shown   
   following years of higher levels of smoking there were higher levels   
   of lung cancer, preceding years of higher levels of lung cancer there   
   were higher levels of smoking, and in regions of the US with higher   
   levels of smoking there were higher levels of lung cancer.   
      
   We are going to do much the same with UFO data and long-term missing   
   persons in Australia.   
      
   Because UFO sightings and missing persons are also affected by the   
   size of the population we will adjust the data to rates per bn   
   population. As the population of the US and Australia grew over the   
   years since the 1930s we will "deflate" both sets of numbers by the   
   current population size.   
      
   And another adjustment that is required happens on the UFO side.  The   
   NUFORC started a web reporting form in Mar 2006 to capture a wider   
   and more immediate set of sightings data. By no coincidence that   
   followed the Nimitz incident in 2004. Unfortunately for statisticians   
   the character of reports after 2006 are much different from those   
   leading up to 2006. For one thing the annual totals of sightings after   
   they would be reported via the web form rose 10-fold.  In addition,   
   some classifications for UFO appearance were added -- before 2006   
   certain "shapes" were not seen as a bona fide category while after   
   that time it turned out they were relatively common.   
      
   So these considerations require the UFO data to be manipulated to   
   offset the change at 2006.   
      
   In the simple "variance adjustment" scheme used here we simply find a   
   scaling factor for the pre 2006 numbers that makes the sample variance   
   of the pre 2006 data match the post 2006 data.   
      
   After all these adjustments the procedure then find the following:   
      
   **** train ****   
   beta in 0.00154104 +- 0.00131404 90% CI   
   P(beta>0.000000) = 0.987977   
   |r| > rc (0.377000 2-sided) at 5%; reject H0:not_connected   
   r2 = 0.25219767   
   model=[y = 0.00154104*x + 7.02939]   
   **** validate ****   
   train: 1977 1997 19.7088   
   back: 1930 1977 21.9323   
   fore: 2000 2020 18.8736   
      
   The first parameter spat out is the "beta" for the relevant time   
   series regression between UFO sightings per capita for each year   
   between 1977 and 1997 (the middle 1/3 of the dataset) and long-term   
   missing people in Australia.  For each 1000 UFO's seen per mn   
   capita in the US an associated 1.5+-1.3 missing persons per mn   
   capita were noted in Australia.  The stats show the link in this   
   section of the dataset is highly robust. The "p-value" for the T-test   
   says there is less than 2% probability this link is just by luck.  The   
   Rank test shows there is only a 5% chance the order of years by UFO   
   sighting and the order by long-term missing persons could be so   
   similar, just by luck.  Together we might argue there is only a tiny   
   chance -- perhaps as small as .05% -- that pattern could come up just   
   by luck.   
      
   The R2 also is interesting. At least in this middle "training" part of   
   the dataset UFO sightings vary year by year and predict about 25% of   
   the year by year missing persons data. It seems UFO's "explain" up to   
   1/4 of Australian long term missing persons.   
      
   But now the most interesting part. We have only shown the statistics   
   1/3 of the data. Did that predict the other 67% of the data?  If we   
   have really captured a causal connection then it should.   
      
   The "validate" part of the output confirms there is a close match   
   between the patterns found in the middle third and the first and last   
   thirds of the dataset.   
      
   In the training portion -- data between 1977 and 1997 -- the model   
   predicts missing persons with an error of "19.8".  This turns out to   
   mean an error within 70% of the actual value across that part of the data.   
      
   The remaining numbers show within the other 67% of the data the error   
   is the same or even less.   
      
   It seems there is almost no chance there is not a causal connection of   
   the type and proportion found in the training dataset.   
      
   UFO activity is likely and in some manner causally connected to   
   long-term missing persons in Australia.   
      
   In a subsequent post we will look at splitting the UFO data into   
      
   [continued in next message]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)