1908.] NATURAL SCIENCES OF PHILADELPHIA. 419 



form in which they may be more easih' considered, a subsidiary table 

 has been compiled from the primary ones. This table consists of four 

 columns (p. 420a). 



Column 1 contains the number of the primary talkie for reference. 



Column 2 gives the number of experiments in the primary tables that 

 are favorable to the presence of a factor. 



Column 3 gives the number of experiments in the primary tables that 

 are favorable to the absence of a factor. 



Column 4 shows the number of experiments that are indeterminate. 

 To determine whether an experiment is indeterminate or not certain 

 rules are followed : 



1. If there has been a large mortality among the snails which were 

 the larger at the end of the experiment the difference was considered 

 indeterminate. The fact that they were the larger could be explained 

 by the fact that they were the fewer. If, however, the opposite 

 was true, i.e., the mortality was among the smaller snails, then the 

 probability is that they are fewer because the conditions have been the 

 more severe. 



2. An experiment has been considered indeterminate if there was a 

 large mortality on both sides of the experiment, notwithstanding the 

 fact that the remaining numbers are nearly equal. The reason for 

 this is the probability that an uncontrolled factor has been acting. 



3. When a known factor has acted on one portion of the experiment 

 and not on the other, the difference has been considered indeterminate. 



4. Those experiments where the difference is under 10 per cent, of 

 the greatest average has been believed to be indeterminate. This 

 purely arbitrary criterion has been devised to allow for two uncon- 

 trollable errors — individual variation and errors in measurement. The 

 obvious way to correct these errors would be to make use of large num- 

 bers of individuals in single experiments. As the number of eggs in a 

 case is small, and when the snails are crowded the mortality is large, 

 it has been found impossible to deal with large numbers. A limi of 

 error must be made that will be large enough to cover most unknown 

 errors (see next page). 



