EXECUTIVE SUMMARY:   
   - We examine an interesting category of FBI missing people --   
     unidentified but living". Most unidentified missing people are just   
     bodies or body parts discovered various places.  But the FBI also   
     has a category of people found alive but unable to say who they   
     are. I'll use the mental model "amnesia" here.   
   - While the category shows strong statistical links with the "usual   
     suspects" using previously-described s/w -- variables like   
     Antarctic temperatures, phytoplankton blooming in the Southern   
     Ocean, positions of Saturn, mid-Atlantic earthquakes, etc. -- we ramp   
     up the pressure here to make sure the links are validated.  I.e. 1/3   
     of the data is used to estimate the relevant link and then the other   
     2/3 of the data is checked to ensure it has the same link.   
   - On top of that we removing all seasonal and annual trends and cycles   
     from all the data to leave what can be thought of as "noise" or the   
     data's "fingerprint".  If the fingerprint of X is also found in Y it   
     suggests the link between the 2 is rather stronger than just an   
     "association" and possibly of the "causes" type.   
   - We find certain UFO types seem to be causally connected with amnesia   
     cases found by the FBI. It also seems some links are -ve meaning   
     some UFO types may be "friendly".   
   - But overall we must remember just as UFO's are a complex combination   
     of phenomena -- "not just the one thing" -- even individual   
     categories are likely to have a lot of complexity.  In a human   
     community some people do some things and some do the opposite.  We   
     can't judge individuals just from the average for the whole community.   
      
      
   I was interested to see some comments from Luis Elizondo about cases   
   reported to the secret UFO unit he led concerning military people that   
   had been kidnapped by UFOs.   
      
   We've taken a look at the civilian data from the FBI that suggests   
   there is some connection between UFO activity and certain types of FBI   
   missing persons cases -- of which there are 10s of 1000s every year.   
      
   But here I'd like to look at one very special category the FBI is   
   interested in. Persons found wandering around and unable to remember   
   who they are.   
      
   While we can assume there are sometimes reasons people WANT to forget   
   or pretend to forget who they are and what they have done, we'll just   
   let the stats sort out whether there is some connection between these   
   cases -- usually 20-30 per year since at least 2015 -- and UFO   
   activity across the US.   
      
   And we'll tighten the elastic bands on our stats programs to try to   
   make it absolutely sure there is a connection far beyond the certainty   
   usually provided by simple statistical tests.   
      
   But how do we ensure some "X" really is the cause of some "Y"?   
      
   This is a growing hot topic in data science and there are a growing   
   set of methods to determine causation.   
      
   I will use a grab bag of them here, starting with the ideas first   
   introduced in the Surgeon General's report that linked cancer and   
   smoking. In that report it was suggests that 3 things must be true if   
   X causes Y. First, if X changes at time T then sometime later Y must   
   change. Second, if Y changes at time T then X must have changed before   
   that. And third, in states or age-groups or other ways of categorizing   
   the data with different levels of X there must be commensurate values   
   for Y.   
      
   Part of this is handled by the idea of "cross validation" in   
   statistical model building. The idea is to take part of the data to   
   determine relevant trends, and then see if that same trend explains   
   equally well parts of the data that were not involved in estimating   
   the trend.   
      
   My s/w does an even more extreme thing. It divides time series into 3   
   equal parts. It uses the middle part to estimate trends and then   
   checks that the same trend is seen in the first and third parts --   
   i.e. it makes sure the model "back casts" and "fore casts" results not   
   seen when estimating the model.   
      
   To make it even more solid we wont just use the original data. We will   
   manipulate it to extract JUST THE NOISE.   
      
   The idea here is that the noise in a data series is like a   
   "fingerprint".  In humans and animals most of the cell's DNA is   
   "noise" and even identical twins have different so-called "junk DNA".   
      
   So if X causes Y then we would expect the noise part of X should also   
   end up being part of the noise inside the Y data.   
   (I came across a similar idea when working on a production line as a   
   teenager. Working a machine with unfamiliar controls and very little   
   instruction from the foreman I found I could get a handle on what knob   
   caused what action of the machine by flicking the knobs and levels to   
   their max and min extent. Throwing a level to the extreme top quickly   
   tells you what is connected with it; then throwing the level to the   
   extreme bottom confirms the connection if the same thing moves to its   
   opposite extreme. Don't worry. Companies are insured for this kind of   
   thing :).   
      
   If some X and Y passes all these tests we might argue it is highly   
   suggestive of a causal link.   
      
   The data for the FBI's "unidentified and living" category that   
   features people that generally can't remember who they are (the rest   
   of the unidentified are dead bodies or body parts -- and we will look   
   at this part of the data sometime when I'm not eating) from 2015-2021   
   which are available online from the FBI looks like:   
      
   year.mo         Num of new unident living/amnesia cases   
   2015.04         15   
   2015.12         25   
   2015.21         15   
   2015.29         25   
   2015.38         15   
   2015.46         35   
   2015.54         15   
   2015.62         15   
   2015.71         25   
   2015.79         15   
   2015.88         15   
   2015.96         15   
   2016.04         11   
   2016.12         21   
   2016.21         14   
   2016.29         11   
   2016.38         12   
   2016.46         15   
   2016.54         20   
   2016.62         9   
   2016.71         11   
   2016.79         17   
   2016.88         16   
   2016.96         9   
   2017.04         10   
   2017.12         19   
   2017.21         10   
   2017.29         17   
   2017.38         12   
   2017.46         19   
   2017.54         26   
   2017.62         22   
   2017.71         19   
   2017.79         22   
   2017.88         15   
   2017.96         16   
   2018.04         9   
   2018.12         16   
   2018.21         18   
   2018.29         13   
   2018.38         20   
   2018.46         21   
   2018.54         15   
   2018.62         14   
   2018.71         11   
   2018.79         25   
   2018.88         23   
   2018.96         14   
   2019.04         12   
   2019.12         18   
   2019.21         17   
   2019.29         29   
   2019.38         28   
   2019.46         17   
   2019.54         14   
   2019.62         23   
   2019.71         20   
   2019.79         10   
   2019.88         21   
   2019.96         17   
   2020.04         25   
   2020.12         12   
   2020.21         27   
   2020.29         11   
   2020.38         23   
   2020.46         15   
   2020.54         10   
   2020.62         26   
   2020.71         31   
   2020.79         40   
   2020.88         24   
   2020.96         14   
   2021.04         27   
   2021.12         19   
   2021.21         29   
   2021.29         17   
   2021.38         22   
   2021.46         27   
   2021.54         28   
   2021.62         38   
   2021.71         36   
   2021.79         33   
   2021.88         13   
   2021.96         22   
      
      
   [continued in next message]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)