| Basic Commands | Probability Puzzles | Hypothesis Test, Count Data | Hypothesis Test, Measured Data | Confidence Interval, Count Data | Confidence Interval, Measured Data | Association / Correlation | Regression | Other Examples |

SRU-486 vs. Hormone Treatment

The French "abortion pill" RU-486 is tested as a morning-after contraceptive. 398 women are given a standard treatment (high hormone doses), while 402 women are given RU-486. Four of the "standard treatment" women got pregnant, while none of the RU-486 women did. (Reported in Science News, 10/10/92.) How likely is such a difference to arise by chance? We test the "null hypothesis" of no difference between the two groups by repeatedly drawing pairs of groups (of the same size as our actual groups) from a universe that has the pregnancy rate we would assume if we knew there were no difference between the two groups = (4+0)/398+400 = 4/800.

REPEAT 1000

Do 1000 trials

   GENERATE 398 1,800 A

Generate 398 "patients" represented by random numbers drawn between 1-800, where 1-4 = pregnant, 5-80 = not pregnant

   COUNT A <=4 AA

Count how many pregnant

   GENERATE 402 1,800 B

Same for a group of 402 "patients"

     COUNT B <=4 BB
     SUBTRACT AA BB C

Find the difference in # of pregnancies

     SCORE C Z

Keep score of the difference

   END

Go back and repeat until all trials are complete

   HISTOGRAM Z

Produce a histogram of trial results

   COUNT Z >=4 K

How often was the difference between groups >=4

   DIVIDE K 1000 KK

Convert to a proportion -- this estimates the "p=value"

F    +
r    +
e 300+
q    +
u    +
e    +
n    +                     *
c 200+                     *
y    +                   * *
     +                   * * *
*    +                   * * *
     +                   * * * *
Z 100+                 * * * * *
     +                 * * * * *
     +               * * * * * * *
     +             * * * * * * * *
     +             * * * * * * * * * *
    0+-------------------------------------------
       |^^^^^^^^^|^^^^^^^^^|^^^^^^^^^|^^^^^^^^^|
      -10       -5         0         5        10
           DIFFERENCE IN NUMBER PREGNANT

KK = .043  ("p-value" estimate)