memory for the most complex tables is the most desirable for census processing. Census processing, like most business data processing, has a high ratio of input/output operations as compared to calculations (internal processing).

6.22

Personnel. --Well-trained, experienced and capable personnel are the greatest asset, because they can overcome machine shortcomings with their ability and ingenuity. Aptitude, logic, and adaptability, however, often are more desirable assets than training and experience. Systems analysts must have a thorough knowledge of the census and the machine capabilities. They must be able to foresee the most efficient method of accomplishing the desired results. They must also have the logic to develop the detailed specifications necessary to implement the plan. The final method of tabulation usually cannot be determined until after the programming staff has shown its ability on early assignments. The systems analysts may have to demonstrate their adaptability by modifying the processing plan when necessary.

6.3 Programming considerations

This

There are basically two approaches to programming the tables--the individual table approach and the multi-table approach. is an over simplification, as there are several methods which may be used in each approach. The selection of a method will depend mainly on the resources available for the census.

6.31 Individual-table approach. --One table is produced each time the data file is read.

This is probably the least efficient but most frequently used method. The ability and/or experience of the analysts and programmers and the computer size often cause this method to be selected. There will be at least

one program per table. It may be necessary to tabulate the data in one program and write tabulated data on tape or disk. The file is sorted and then a separate program prints the data.

6.32 Multi-table approach.-- The multitable approach results in producing more than one table each time the data file is read into the computer. There are four methods of implementing this approach:

(1) This first method uses one general

(2)

(3)

purpose program which will tabulate and print a number of tables based on specifications read from punched cards or parameter cards. The number of tables that may be prepared at one time will depend on the memory available and the complexity and size of the tables. This method would seem to be the most efficient because only one program would be needed to produce all of the desired tables. The parameter cards, if well designed and documented, could be coded by nonprogrammers with a minimum of training. Actually this method is probably the most impractical of the four, because the extreme complexity of such a program would require a large number of the best qualified personnel, a long period of time for development, and a very large computer.

The second method uses more than one general-purpose program to tabulate and print a number of tables. The table specifications are read from parameter cards by each program. The number of tables that may be prepared by one running of the programs depends on the memory available and the complexity and size of the tables. This method can be very efficient.

The third method uses one program to tabulate and print specific tables. This method requires well-qualified personnel, a considerable amount of time to develop, and a medium-to-large computer.

(4) The fourth method uses more than one program to tabulate and print specific tables. This method requires less than method three in terms of personnel qualifications, development time, and computer size.

6.4 Intended use of print-outs

Computer print-outs are provided for

several reasons.

They serve as the final

tables, as worksheets for analysis and review, and as documents for historical purposes. For example, card-to-tape diary print-outs provide information for review and later serve as records of the processing.

6.41 Final publication. --Print-out tables can be programmed so that they will be complete and ready for the printing reproduction process. The title, column heads, and stubs are properly spelled, punctuated, centered, and spaced. The data in each cell are properly aligned and punctuated with decimal points and commas as needed.

6.42 Work tables.--Print-outs can serve as work tables from which final tables can be prepared. In some cases, categories in the work tables will be combined because the separate cells would disclose information for an individual, or the figures in the individual cells are insignificant. The analysts often require work tables when they do not have enough information from previous censuses to establish the precise detail in advance.

7. ADVANCE SAMPLE TABULATIONS

The concept of advance sample tabulations appears useful but difficult to develop into a working model. In principle, the purpose of the advance tabulations is to publish selected sample tabulations as soon as possible after the enumeration. The figures for Providencia would be simple distributions of selected items for the total country.

7.1 Value of advance sample tabulations

In the course of the pre-enumeration publicity and nationwide interest at the time of the census, there will be a great demand

from official and unofficial sources for some

figures from the NSO. By planning an advance tabulation procedure and following through with it, the NSO can eliminate some of the pressure for census figures as the processing develops.

Advance tabulations can also be an insurance against an incomplete publication of census data. Occasionally, some countries may start out with a plan that is too ambitious and they may not be able to complete the tables for all parts of the country. If this occurs, some parts of the country will have no figures at all unless advance tabulations have been prepared. (On the other hand, the census planning staff should be aware that too much effort on the advance tabulations may make it impossible to complete the regular census tabulation program.) Advance sample

tabulations are also beneficial to the morale of the NSO employees as they will be able to see some fruits of their labors early in the processing cycle.

Careful time records should be kept of the various stages of the processing of the tables. These records will be valuable for making estimates of the time and money needed to process the entire census. Some changes in plans may be required in order to have sufficient resources for the analysis and evaluation of the census results. A final benefit from producing advance tabulations is the opportunity to test the computer programs and other processing operations.

7.2 Planning a procedure

As discussed earlier in the census plan, there are several factors to consider in recommending the advance tabulations. Some are: (1) How much extra work will the advance tabulations create?

(2) How does the NSO develop a sample for the advance tabulations--one that is simple for the staff to carry out?

(3)

(4)

How can bias be reduced in the special handling and processing of the questionnaires selected for advance tabulations?

Will the value offset the disadvantages of having the final tabulations delayed by the extra work necessary to produce the advance tabulations?

The sampling technician or statistician will need to develop a sample to use for the advance tabulations. A sample of ED's would need to be selected prior to the editingcoding operation. However, they should not be marked in a way that will identify them as belonging to a special group. The field personnel and the office clerical personnel should not be aware of the existence of such a designated sample.

n

th

An alternative plan is to select every establishment as the forms are received

in the NSO. A sample of this design requires fewer cases (consequently less punching and editing) and may avoid some bias, but is much more difficult to administer and control.

During the data processing operations, the advance sample ED's should be controlled by one supervisor who will monitor the progress of the advance work through all the stages of processing. This must be done without interrupting the regular flow of work.

The Providencia NSO decided against advance sample tabulations. However, they gave priority to some full-count preliminary

tabulations.

The basic purpose of a census is to produce statistics; for example, statistics from an industrial census might reflect the industrial capacity of a city, the fuels consumed by industry, number of employed persons in various industries, etc. These statistics are used primarily in making decisions which result in some action. The prime requirement is that they should lead to the correct decision.

The uses to which statistics are put should determine to a large extent how accurate they must be. Thus, the greater the precision required, the higher the quality of the statistics upon which the decision is based. On the other hand, it is "dis-economy" to require of statistics a quality higher than is actually necessary for the use for which they are intended.

2.

CONCEPTS OF QUALITY CONTROL

Quality control in a census consists of the procedures used to measure the accuracy of performance and to limit the number of errors to a degree that is acceptable. Quality control applies to many operations: (a) preenumeration operations such as question wording, printing and assembling forms, sampling specifications, preparation of maps, distribution of materials, (b) enumeration activities including pre-listing, completeness of questionnaires, observation and review of enumerator's work, and (c) data processing operations such as editing, coding, key punching, and tabulating.

This chapter emphasizes control of the quality of data processing operations--largely the clerical operations. The term "clerical operations" is defined to include such tasks as assembling and labeling materials, sorting, counting, transcribing information from one form to another, editing-coding, key punching, and the verification operations such as verification of editing-coding, and verification of key-punching.

Quality control applied to an operation attempts to limit the amount of error occurring in a batch of work; the limits are determined in advance. Quality is accomplished by reviewing the work and removing as many human errors as possible.

Quality control applied to individual employees requires that their work be measured by verification and that records be accumulated to provide data by which their performance can be evaluated. Employees who have the best productivity with acceptable quality are rewarded by advancement to more responsible assignments or given longer employment. Those with poor performances are re-trained, encouraged to improve, or removed from the work.

2.1 Objectives of quality control

The output of a clerical operation is the product of human rather than machine effort-a type of output which is highly variable in both productivity and accuracy. Clerical personnel should be aware that their performance is being observed and that poor work will be identified. The prime objective of quality control is to obtain the best possible

performance from each employee without the expenditure of excessive amounts of time and money. In cases where errors beyond tolerance are discovered, the work must be redone. In most cases, however, a few errors are acceptable provided they have no important effect on the final data that are to published.

The managers of the census are responsible for not only applying professional techniques to the processing of the data but also for accurately measuring the quality of the output. It is not sufficient to "believe" that the quality was controlled; there must be statistical evidence that a plan for doing so was developed and applied properly. Records of the performance of clerks must be maintained; an error rate, specifically defined, must be calculated both for the clerks and for each

of the operations. The measurement of an error rate for an operation is itself subject to some possible error; nevertheless, all information relating to its use and application should be part of the history of the census and should be available for analysis if questions arise as to the validity of the published data.

2.21

Verification of output.--Verification of the process itself as a system of controlling quality of output of clerical operations is usually the first to come to mind. Verification is the use of a second person (or system) to search for defects or errors, or to check or test the accuracy or correctness of an item or operation. Verification

of all items and verification of a sample of items are discussed in section 3.

2.22 Consistency checks.--Under a system of consistency checks, a questionnaire or other document, a tabulation, or a set of tabulations is subjected to one or more of the following tests:

(1) "Are quantities which should be equal to each other actually equal?" These are equality checks. They are most frequently employed in comparing stub and footing totals in statistical tables. A variation of these tests asks the question: "Are quantities within pre-determined limits?" In this form, the applications of these checks are limitless.

(3) "Are quantities, conditions, or ratios reasonable?" This is the check for reasonable (or unreasonable) magnitudes or conditions.

Consistency checks are primarily devices for isolating serious errors which can be corrected at the point of discovery. They are simple enough to be carried out, particularly when tabulating equipment or a computer is employed.

2.23 Error detection at later operational stages.--This technique calls for postponement of a decision whether the quality of a component is good until an assembly of components is verified. As a simple example, two different types of coding may be verified in one operation after both have been completed. Such a system is best when corrections made at the