952 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 



With this in mind, the product specification prescribes for any given 

 key characteristic a nominal value, N, maximum and minimum limits 

 for indi\adual units, \/V dr A, and adds the clause ^'subject to the distri- 

 bution requirements of (F)." This supplementary specification (F) gives 

 a statrstical description of the intent, together with control procedures 

 to be followed in the inspection of the product. 



3.0 Control Procedures Associated with Distribution Require- 

 ments 



Distribution requirements can be given operationally definite meaning 

 by providing procedures (of the nature of inspection procedures) that 

 define (a) the character and quantity of evidence needed regarding the 

 collective quality of the product as it is made day by day, (b) the criteria 

 for judging when such product may be considered conforming to the 

 intent of the specification, and (c) the treatment or disposition of product 

 units when these criteria are not met. 



Three such procedures have been prepared to meet the several con- 

 ditions that may be encountered in the production of L3 component 

 elements: 



1. control chart method, 



2. batch method, and 



3. three-cell method. 



The first two methods permit the use of sampling. For both of these 

 methods the criteria have been so selected that product should be found 

 to be conforming to the distribution requirements practically all of the 

 time if the process is so controlled as to maintain a distribution of indi- 

 vidual units with (a) an average within the band, A^ ± 0.1^, and (b) 

 a standard deviation not greater than O.SA. The third method requires 

 100 per cent inspection, and while this method may be used at any time 

 at the option of the manufacturer, its use is mandatory whenever a failure 

 to meet the criteria of the other methods is encountered. 



To insure that the product shipped continually meets the intent of the 

 distribution requirements, a provision is made for packaging the output 

 in groups of 5 units. This in effect furnishes the user either with random 

 sample groups of 5 from a process which has been shown to be in satis- 

 factory control (control chart and batch methods), or with specially 

 selected groups of 5, the units in each of which have been chosen to meet 

 a particular distribution pattern (three-cell method). 



The following sections give the general character of the statistical 

 models that have been set up for the three methods. 



