214 EXPERIMENTAL DESIGN 



a coin decide whether any individual rat becomes a member of the ex- 

 perimental group or a member of the control group. Tables of random 

 numbers prepared by computers, available in several of the statistics 

 books, can also be used in randomization. Each individual in the large 

 group must have an equal chance of being used in the experiment, and, 

 of those used, each must have an equal chance of being a "control" or an 

 "experimental." 



In many experiments the selection of the organisms is completely 

 random. Other cases allow the use of special statistical techniques in 

 which the deliberate selection of individuals and matching in pairs 

 produces more useful information than the complete randomization. For 

 example, suppose we needed five rats to serve as controls and five to re- 

 ceive an experimental treatment. We wish to test the effect of a new 

 drug upon gain in weight by the rats. The amount of gain, however, 

 might depend on the original weight of the rat. If we weigh the animals 

 initially and then arrange the weights in a table in descending order, 

 it is no difficult task to assign the rats in pairs that are more like each 

 other than like all the rest. One member of each pair receives the drug, 

 and we decide which one by flipping a coin. Each member of the pair 

 must have an equal chance to receive the experimental treatment. If 

 you find yourself wishing the coin would fall a certain way, you had 

 better beheve the coin rather than your feelings and wishes. 



One of the beauties of the use of microorganisms or subcellular partic- 

 ulates is that the experiment is conducted with a very large sample. There 

 is considerable variability among the individual cells in a yeast suspen- 

 sion, but the average behavior is surprisingly uniform. Only a few 

 minor precautions are necessary in sampling from such favorable popu- 

 lations. For reasons that are difficult to understand, the first pipetteful is 

 likely to be slighdy different in composition from all succeeding pipette- 

 fuls. Perhaps the dry inner walls of the clean pipette have some in- 

 fluence. Because cells in suspension setde quite rapidly, swirling or stir- 

 ring between samples is necessary. Obviously the number of cells per 5 

 ml is not uniform if the cell suspension in a 500-ml flask is more concen- 

 trated at the bottom than at the top. Finally, pipetting errors are hkely 

 either to increase or to decrease in taking a long series of samples. If all 

 the experimental samples are taken first and all the controls later, bias 

 may be introduced. The safe procedure, again, would be to withdraw a 

 pipetteful of cell suspension, and then let a coin decide whether it is to 

 be a control or an experimental sample. 



Many people have been successful research biologists without ever 



