CARD TRANSLATOU EXPERIMENTS 395 



■_M) per cent in the capacity of a tratislatoi' i)\('r tlic (l('si<i;ii object ives. 

 Tlie reciuireiiKMit that an uncoded caitl he placed next to each .separator 

 should remain until the life test which is currently in progress is con- 

 chided. 'These uncoded cai'ds are to he included in the 100 cards per bin 

 fi|2;ure. A preliminary analysis of the data indicates that it is ])rol)al)ly 

 necessary tliat this nMiuirement he retained for held use. 



With reu;ai(l to the le\-eling of the machine, it was found that if the 

 translator table is leveled consistent with the normal practice of the in- 

 stallation department no further levelin<>; is necessary when the trans- 

 intoi' is installed on the table ))ro\ided an uncoded card is next to each 

 separator. W'lien only cards that were at least one removed from the 

 separator were considered it was found that the Translator could he 

 tilted 1 inch in 3 feet without seriously affecting the card drop time. 



C'onsitlering- the ranges of the several variables considered and the 

 results of the analysis of the data, it appears that there is no major un- 

 known variable having an effect on the card dropping time. It is also 

 believed that the results of the work on this machine can be considered 

 representative of the results that will be obtained on another new pro- 

 duction model translator. 



Appendices 

 i. analysis of variance 



The general theory of the analysis of variance has been formulated and 

 discussed at length by several authors.'' *' ^ Basically it reduces to the 

 concept that in any set of data obtained from a statistically designed 

 experiment the total sum of sciuares of deviations from the mean can 

 be partitioned into orthogonal components, and that under certain 

 restrictions the distribution of each component falls into known pat- 

 terns. Hence data taken from designed experiments can be examined 

 for conformance to the known pattern, and a lack of conformity indicates 

 an assignable cause of variation. Further, the distribution of the ratio 

 of mean square deviations under specified conditions has been tabulated 

 as the table of the F ratio. It has also been shown that when a treatment 

 variable is not a parameter or assignable cause of variation in the ex- 

 periment, the partitioned component for that variable must contain 

 only residual variation. Thus, the analj'sis of variance tests the hypothe- 

 sis that the treatment means for a given variable are all eiiual (i.e., the 

 variable is not a parameter) by testing the ratio of the mean s(iuares of 

 mean deviations for the variable to the residual mean s(iuare, i.e., the 

 F ratio. When this F ratio is larger than the critical value at the a" 

 level, th(^ variable is said to be significant at the a ' level. That is, let 



