CARD TRANSLATOR EXPERIMENTS 371 



S. Baktnccd versus ['Nbalanccd Loading — Since the earlier iii.stal- 

 latit)iis will have machines that are less than full, it was necessary to 

 learn whether or not any special loading considerations with regard to 

 the j)osition of (he cards shc^uld be specified. 



1). 7'/// - - (Machine) This has to do with the accuracy with which 

 the translators should l)c leveled on installation. 



10. Life — ^ This variable considers the effect of repeated operations 

 on cards and what effect, if any, these repeated operations have on the 

 dropping time. 



The tests on the first nine of the variables have been completed and 

 a discussion of their effect on card dropping time is possible at this time. 

 The tests on the tenth variable, life, are still in progress and have not as 

 3^et proceeded far enough to draw any conclusions. Therefore no con- 

 sideration will be given to the effects of usage in this article. The effect 

 of most of these variables was considered when the translator was oper- 

 ated both with and without the card support bars being used. The card 

 support bars operate at the same time the card is dropped. Their opera- 

 tion is recjuired for every card in the machine. Since their operation 

 is slower than the free fall time of the operating card the card may ride 

 down into the operated position on these bars and the observed drop 

 time will simply be the operate time of the card support bars. This masks 

 the effect of the various friction, gravity, and magnetic forces on the 

 cards. Operating without the card support bars gives information on 

 the free dropping time of the cards uninfluenced by these bars. Thus the 

 masking effect of the card support bars can be removed and the tests 

 made more sensitive to any physical effects which may be taking place 

 at the cards themselves. 



II. DESIGN OF EXPERIMENTS 



Theory of Designed Experiments 



In this problem it was recjuired to evaluate the effect of each of the 

 nine variables independently. It might have been convenient to take 

 one \'ariable at a time and vary it o\'er its range while holding the as- 

 sociated variables at sets of constant values. Then a second independent 

 \'ariable could be chosen, the first variable placed in the group of as- 

 sociated variables, and a new set of runs made. This could be repeated 

 until all the variables had been tested. This is obviously a logical, 

 straightforward but unwieldy procedure. 



A simplei' procedure would be to vary the chosen variable over its 

 range and at the same time let the associated variables vary over their 



