﻿Inconstancy 
  of 
  Improved 
  Races 
  779 
  

  

  found 
  that 
  it 
  is 
  highly 
  variable 
  and 
  complies 
  in 
  

   the 
  main 
  with 
  Quetelet's 
  law. 
  Madia 
  elegans, 
  

   a 
  garden-species, 
  has 
  on 
  the 
  average 
  21 
  rays 
  

   on 
  each 
  head, 
  fluctuating 
  between 
  16 
  and 
  25 
  or 
  

   more. 
  I 
  saved 
  the 
  seeds 
  of 
  a 
  plant 
  with 
  only 
  

  

  17 
  rays 
  on 
  the 
  terminal 
  head, 
  and 
  got 
  from 
  

   them 
  a 
  culture 
  which 
  averaged 
  19 
  rays, 
  which 
  

   is 
  the 
  mean 
  between 
  21 
  and 
  17. 
  In 
  this 
  second 
  

   generation 
  I 
  observed 
  the 
  extremes 
  to 
  be 
  22 
  and 
  

   12, 
  and 
  selected 
  a 
  plant 
  with 
  13 
  rays 
  as 
  the 
  

   parent 
  for 
  a 
  continuation 
  of 
  the 
  experiment. 
  

   The 
  plants, 
  which 
  I 
  got 
  from 
  its 
  seeds, 
  averaged 
  

  

  18 
  and 
  showed 
  22 
  and 
  13 
  as 
  extremes. 
  The 
  total 
  

   progression 
  of 
  the 
  average 
  was 
  thus, 
  in 
  two 
  gen- 
  

   erations, 
  from 
  21 
  to 
  18, 
  and 
  the 
  total 
  regression 
  

   from 
  13 
  to 
  18, 
  and 
  the 
  proportion 
  is 
  thus 
  seen 
  

   to 
  diminish 
  by 
  the 
  repetition 
  rather 
  than 
  to 
  in- 
  

   crease. 
  

  

  This 
  experiment, 
  however, 
  is 
  of 
  course 
  too 
  

   imperfect 
  upon 
  which 
  to 
  found 
  general 
  conclu- 
  

   sions. 
  It 
  only 
  proves 
  the 
  important 
  fact 
  that 
  

   the 
  improved 
  average 
  of 
  the 
  second 
  generation 
  

   is 
  not 
  the 
  starting-point 
  for 
  the 
  further 
  im- 
  

   provement. 
  But 
  the 
  second 
  generation 
  allows 
  a 
  

   choice 
  of 
  an 
  extreme, 
  which 
  diverges 
  noticeably 
  

   more 
  from 
  the 
  mean 
  than 
  any 
  individual 
  of 
  the 
  

   first 
  culture, 
  and 
  thereby 
  gives 
  a 
  larger 
  amount 
  

   of 
  absolute 
  progression, 
  even 
  if 
  the 
  proportion 
  

   between 
  progression 
  and 
  regression 
  remains 
  

  

  