Library Collections: Document: Full Text

THE HOLLERITH METHOD OF STATISTICAL TABULATION.

THE returns of the enumerators at the next census, as we have elsewhere shown, will show for each of the 65,000,000 persons residing in the United States on the 1st of June, 1890, the color or race. the sex, age, relationship which each person bears to the head of the family, the conjugal condition, profession, occupation or trade, number of months unemployed during the census year, whether sick or disabled, whether deaf and dumb, blind, idiotic, insane, maimed: crippled, or bedridden, whether the person attended school during the census year, whether the person could read and could write, the place of birth of the person, and, finally, the place of birth of the father and the place of birth of the mother.

These records may be considered the raw material from which the Census Office is to manufacture the finished product -- the census reports. Although it will probably cost over $2,500,000 to collect this information, still, in the shape in which the returns are made to the Census Office, they will be of very little if any benefit to the country at large. From these returns, however, are compiled the various census reports, and to do this in the last census for a population of 50,000,000 cost nearly $2,000,000. By the various methods heretofore employed the returns have either been gone over and over, tallying out first one set of facts and then another set of facts by making little tally marks in squares on sheets of paper, and then counting and aggregating these tally fall marks; or else the returns have been transcribed to cards writing, and then these cards first sorted by hand, according to one scheme and counted, and then sorted according to another scheme and counted, and so on until finally all the desired data were obtained. Although at the last census the enumerators' returns contained answers to the above questions, still it was found impossible, within the limits of the appropriation, to compile even such valuable and essential information as the data regarding the civil conditions of our people, so that to-day we have not the (indecipherable)

A casual consideration of' the problems involved in the compilation of a census will show that unless some improved methods of compilation are adopted it will be found impossible to do much more at the Eleventh Census in 1899, with the $6,400,000 appropriated for that purpose, than was done at the Tenth Census in 1880.

Another item of very considerable importance in all matters relating to the tabulation of the census is the question of time. The returns of the enumerators are all made to the Census Office within a few weeks after the 1st of June, 1890. Much of the information contained in these returns loses in value each day that its publication is delayed, and it is, therefore, essential that the compilation be made as rapidly as possible. Again, in all statistical work, of course, accuracy of compilation is of prime importance.

During the compilation of the Tenth Census the attention of Mr. Herman Hollerith, a special agent of the Census Office, was called to the immense labor and enormous expense involved in the compilation of the population statistics, and he addressed himself to the work of devising a machine for the purpose of facilitating this portion of the census work. After several years' consideration of the problems involved he has invented the following method, which is now under consideration by the Superintendent of the Census for adoption in the compilation of the Eleventh Census. This method consists in transcribing the returns to cards by punching holes in such cards, the various combinations of such holes denoting the different characteristics of the given persons. These cards are then automatically counted and sorted in electrical devices, giving the number of persons falling under each characteristic or combination of characteristics, as may be required for publication.

Such punched transcripts are made by using small manila cards about three by six inches, which are first numbered in a numbering machine, these numbers corresponding with numbers given to the individual returns of the corresponding enumeration district. Across one end of the cards a combination of four or five holes is punched, which combination designates the enumerator's district, the supervisor's district, and the State to which the record relates. This combination is punched in the machine shown on this page. The cards punched with this combination, and having these numbers, are then given to a clerk, who, with a machine (as also shown on this page) transcribes the record for each individual person by punching holes in accordance with the given record. For example, the keyboard of this punch will have two spaces marked "male" and "female" respectively, and accordingly, as the pointer is pressed in one or the other of these holes a corresponding hole is punched in the card which will, in all future operation, either sort or count the card according to the given sex. The color, whether white, black, mulatto, quadroon, octoroon, etc., is in similar manner recorded by punching a hole in the proper position by depressing the pointer in one of the corresponding holes, as marked on the key-board. In this way all the simpler records are made, and where a larger scope of record is required, as, for example, in case of occupations, as many as 500 or 1,000 of which may be required, combinations of two holes are used to record each designation.

Page 2:

When the given cards have thus been punched they are passed to another division of the office, in which they are sorted and counted in the mechanism shown in our illustration. As the cards come from the punching clerks they are, of course. arranged by enumeration districts and minor civil divisions. The cards corresponding to a given minor civil division are placed, one by one, in a press, counted, and deposited in the open sorting-box. This operation is repeated until all the cards corresponding to the given district have been thus treated. The counters to the left will now show, for the given minor civil division. such information as may be desired. At the same time the cards will be distributed into the various divisions of the sorting-box, the cards for each of the corresponding divisions of the various districts being put together until the total cards for each characteristic or combination of characteristics, for the county are brought together. Each division of cards for the county is then treated in similar manner, being further subdivided according to statistical characteristics, while any desired data for the county are obtained on the counters. The cards are then thrown together by States, thus securing a finer statistical distribution, which will facilitate the compilation of further detailed data required by State totals.

When it is remembered, for example, that at the first handling of the cards they are divided into, say, twenty statistical divisions, and at the second handling each of these twenty divisions is further subdivided into twenty subdivisions, or 400 divisions in all, and at the third handling each one of these 400 divisions is counted or tallied according to, say, forty characteristics, on forty counters, it will be seen that the third handling of the cards will give a result equivalent to a table having 16,000 columns. Very little consideration of this question will show that with the comparatively few handlings of the cards a most complete and detailed tabulation is obtained.

Punched record cards, as compared with written cards, possess, for purposes of statistical tabulation, many manifest advantages. In the first place a misplaced card can readily be detected among a thousand or more in a few seconds, by the simple fact that the punched holes of the misplaced card will not correspond with the balance. This is readily detected by means of a long needle. With written cards, on the contrary, a misplaced card can only be detected by an actual inspection of each card.

Again, in the counting and sorting mechanism errors of tabulation are practically impossible, as the machine is so provided with mechanical checks that if a wrong card is placed in the machine, or if it is carelessly or improperly placed in position, the entire mechanism will fail to operate, and only when a proper card is correctly placed in position will there be any effect on the counters or sorting mechanism. In other words, this mechanism is so constructed that the counting and sorting must either be done correctly or not at all. If the mechanism fails to operate it is evidence to the clerk that such card does not properly belong among those being handled.