A  FIRST  EXPERIMENT

33

est  in  the  second  test  was  eighth  in  production,  and  the
girl  who  was  second  was  fifth  in  production,  etc.  There
may  be  a  difference  all  along  the  line,  showing  that  there
was  a  lack  of  agreement  or  a  low  correlation  between  the
performance  of  the  girls  in  this  test  and  their  performance
at  their  regular  work.
In  every  case,  the  degree  of  agreement  or  correlation
between  tests  and  production  is  determined,  not  by  guesswork, ­
  but  with  mathematical  exactness,  by  means  of  the
method  already  described  and  certain  simple  formulae  discussed ­
  in  the  Appendix  under  the  heading  Correlations  .
It  would  be  too  cumbersome  to  go  into  further  details  on
this  subject  here.  Suffice  it  that  the  degree  of  correspondence ­
  between  two  sets  of  rankings,  the  ranks  of  a  group
°f  subjects  in  production  and  their  ranks  in  a  given  test,
can  vary  between  plus  i.oo  and  minus  i.oo.  Plus  i.oo
is  a  perfect  correlation.  Minus  i.oo  is  an  absolutely
negative  correlation.  Plus  .70  or  .60  is  considered  good
for  shop  and  factory  conditions.  A  test  which,  under
factory  conditions,  shows  a  correlation  with  production
°f  .40  or  more  is  considered  a  valuable  or  a  significant
test.
The  method  of  computing  the  value  of  tests  just  described ­
  was  applied  carefully  to  each  of  the  tests  given  in
this  experiment  with  the  following  results:
Table  of  Correlations
Card  sorting  5®
Tapping  x 4
Cancellation  ^3
Easy  directions  x 4
Number  group  checking  7 2
Accuracy  3  8
Steadiness  2 4