162 EMPIRICAL TOOLS AND EMPIRICISM 



havior of balls rolling on an inclined plane relevant to understanding 

 of free fall? Can I hope to gain understanding of the souring of milk 

 by studying a sugared yeast water that contains no milk? How shall 

 I design an artificial system that is relevant, what variables shall I 

 control, what observables shall I note? Before ever I make an experi- 

 ment on my problem I must bring fully to bear my speculative ideas 

 about its solution ( s ) . 



The choice of variables. In an experiment we study a phenomenon 

 in circumstances that are narrowly determined, but never totally de- 

 fined or definable. In practice we seek to measure and/or control 

 only the relevant variables. To do more is a pointless waste of time; 

 to do less is to aggravate unnecessarily the conceptual (interpretive) 

 problems we will later face. Which are the relevant variables? Pre- 

 sumably just those found relevant in previous studies of closely re- 

 lated phenomena. However, unless we are concerned only with 

 phenomena already thoroughly explored, such relation is only some- 

 thing we hypothesize. Even striking resemblance may here be a 

 wholly insufficient clue: the similar array of colors displayed by rain- 

 bow and by peacock tail we find, to our surprise, represent very dif- 

 ferent phenomena of light. No matter how we are guided by careful 

 observation and long experience, a judgment of "close relation"— and 

 the identification of relevant variables it implies— rest alike on the 

 insecure but essential foundation of those of our preconceptions that 

 dare forecast the results of experiments we have yet to begin. 



Can we not entirely bypass so unnervingly uncertain an identifica- 

 tion of relevant variables? Rather than dealing with them explicitly, 

 in a controlled experiment, might we not resort instead to use of an 

 experimental control? That is, let us contrive two experimental sys- 

 tems identical in all respects save one. Could we not then conclude 

 that any difiPerences in the results obtained are certainly attributable 

 to the action of the one variable in which the systems diflFer? Let us 

 then go on to set up mantj systems, each pair differing only in some 

 one of the many variables conceivably relevant. Could we not then 

 easily establish, by experiment, the identity (and also the specific 

 eflFect) of each relevant variable? 



Indisputably, use of an experimental control gives us an enor- 

 mously powerful technique. Probably that power is most evident in 

 biological work where— built into the experimental subject itself— a 

 multitude of ill-defined variables elude our control. We cope with 



