EXECUTIVE SUMMARY:   
      
   - The NUFORC have a network of detectors that are believed to detect   
     magnetic anomalies associated with UFO's.  Almost 2 years of data on   
     these sightings is now available.   
   - We analyze the monthly data with an AI s/w that "understands"   
     something about causal links as well as statistical correlations.   
   - Culling through 1000s of datasets it finds the most likely causes of   
     the MADAR sightings are associated with several common UFO "types"   
     most notably Triangles, as well as even more closely matching major   
     earthquakes in key regions in central and east Asia as well as ocean   
     ridges in the NE and SE Pacific.   
   - Some of these relationships seem to predict more MADAR activity in   
     subsequent months after certain events; some (in particular   
     earthquake events) predict less MADAR activity.   
      
      
   In getting the latest couple months of NUFORC's summary sighting data   
   I noticed their MADAR system now has 20 months of data -- enough to   
   start getting some interesting patterns.   
      
   The MADAR boxes are located across America and are designed to detect   
   unusual changes in the local magnetic field. While there is no   
   guarantee whatever caused the change is a UFO, it's believed unusual   
   flying objects sometimes have significant magnetic fields and   
   reputedly have affected (either "intentionally" or as part of their   
   "normal operation") car ignition system and other electrical and   
   electronic equipment.   
      
   The summary data currently looks like:   
      
      
   Year.Month      Number of MADAR detections   
   2020.04         89   
   2020.12         65   
   2020.21         48   
   2020.29         67   
   2020.38         90   
   2020.46         57   
   2020.54         55   
   2020.62         42   
   2020.71         50   
   2020.79         54   
   2020.88         44   
   2020.96         37   
   2021.04         36   
   2021.12         28   
   2021.21         33   
   2021.29         43   
   2021.38         26   
   2021.46         22   
   2021.54         12   
   2021.62         6	<- incomplete month   
      
   The same AI s/w I've used before can now grind through this data and   
   try to establish a list of "similar looking" datasets of the 10s of   
   1000s it has on file. The reasoning part of the AI is continually   
   updated -- well the whole thing is subject to tinkering 24/7 -- and   
   now tries to deduce which data series -- if found to be significantly   
   connected with the target it was given -- actually might be due to a   
   causal link.   
      
   The reasoning uses a set of in-built "known causal relationships", a   
   set of rules that enable other causal relationship to be deduced   
   (e.g. if A causes B and B causes C then A causes C; if A causes B and   
   C has similar meta-data to A then C might cause B too) plus other   
   relationship it learns over time. If it sees a strong statistical link   
   between A and B many times but A is not know to cause B and the link   
   can't be deduced then it will add it to the list of possible links. The   
   strength of the link is incremented a little each time; when it passes   
   a limit it's thereafter "believed" to be a causal link and can be used   
   in other deductions.   
      
   This part of the S/W neatly extracts bogus highly-correlated "suspect"   
   series from consideration as being causal. E.g. it turns out many   
   UFO-related things seem to highly correlate with the distance of Pluto   
   from the sun. But it turns out this is only due to most data being of   
   such short duration we don't have enough to correlate with a large   
   sample of Pluto's orbit; many high correlations are caused by   
   Pluto parameters that slowly increase or slowly decrease over the same   
   period as changes in a target dataset. Many things falsely match because   
   they are simply slowly changing -- have a modest trend -- like many of Pluto's   
   parameters. If more data were available to match against a good   
   chunk of Pluto's 248y orbit it might be found to be uncorrelated.   
      
   So with the new "causal filter" in place the s/w goes through its list   
   of data series and finds the most likely "causes" for the MADAR hits   
   listed above are:   
      
   Suspect         Lag   Log     R2              Beta        90% CI   
   gavqmongolia     4      y     0.92738793      -0.837399   0.120558   
   mqnepacrise      4      y     0.87520070      -1.5497     0.316871   
   gavqchina        4            0.81670900      19.4889     4.75022   
   gavqbandmag40    4            0.80431245      -69.8693    17.7315   
   mufo-Triangle    4            0.76624152      0.996401    0.249105   
   mqmagseq30       4      y     0.75440893      1.23887     0.36368   
   mqtajikistan     4            0.73884750      39.346      12.6664   
   mqseg110         4      y     0.66785991      0.073287    0.0279852   
   mufo-Light       2            0.62905482      0.241974    0.0783983   
   gavufo-Circle    4            0.62762349      0.486872    0.169748   
   mufo-Diamond     3            0.61939424      3.98103     1.36208   
   mqband30         4            0.61693743      2.6062      1.0566   
   mqsepacrise      3            0.61632600      -19.7465    8.92965   
      
   The "Suspect" column is the AI's code for each data series it found to   
   be causally linked with the MADAR hits. Some of the suspect series   
   were modified in some ways to improve the robustness of the match or   
   amplify it. E.g. "m" means the s/w added missing values to a data series   
   (usually the global average of available data in the series).  "gav"   
   is an operation that smooths the data and is analogous to calculating   
   soil moisture from rainfall -- some of the soil moisture carries over   
   from the last period and some is due to rainfall in the current period.   
      
   The "q" data series relate to major (mag 5+) quakes in various regions.   
   The "ufo" data series are NUFORC data for particular types of UFO's   
   (usually determined by the "shape" assigned by curators at NUFORC).   
      
   Many other series were also found to be possible suspects (e.g.  solar   
   events of certain types), but the table above is just the "most   
   likely" or "best explanations".   
      
   At the top of the list we find big quakes originating in Mongolia   
   track more than 90% of the month-to-month changes in MADAR hits.  The   
   beta is -ve. For each quake in Mongolia there are statistically fewer   
   MADAR hits in 4 months across the 20-m dataset.  Each big quake in   
   Mongolia approx 1/2s the MADAR hit rate in 4m time "for some reason".   
      
   Quakes in other regions have a similar effect. Quakes near the NE   
   Pacific Rise greatly decrease MADAR hits 4m later.  Quakes nr the SE   
   Pacific Rise almost stop MADAR hits 3m later.  OTOH quakes in the   
   Tajikistan region greatly increase MADAR readings 4m later.  Some   
   latitude bands are associated with big increases and some with big   
   decreases in MADAR readings in subsequent months.   
      
   In the past my mental model was that stimuli that increase UFO   
   activity might be analogous to "home" regions. Hit the hornet nest   
   with a stick and you get a lot of insects buzzing around. Whereas   
   stimuli that decrease activity may be "hunting grounds". If a farmer   
   mows down the wildflowers in a field you are likely to see fewer bees   
   in the area until they grow again.   
      
   Of course mental models can be backwards.  Or totally wrong. :)   
      
      
   [continued in next message]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)