always open to those who sought counsel concerning their work, he made their problems his own, his suggestions were generous and helpful, his good-nature was unfailing. To have enjoyed such an association and to have shared some of the hours not given to labor, when his genial nature, his breadth and warmth of sympathy, and his keen and kindly humor brought into view another phase of his manly and exalted character, I shall always regard as among the highest of privileges.

But we turn from the past, characterized by great men and brilliant achievements, to a present full of promise. The direction of great statistical undertakings is full of exhausting labor, and requires the strength and initiative of young men. The great men of whom I have spoken came to their own while very young as compared with men called to equal responsibilities and prominence in other fields. Other young men have been called to fill the places which they filled, and have taken up their labors with equal devotion to duty, with purposes equally noble, and, we believe, with equal promise of successful achievement. Their labors are so linked to a great past that they cannot choose but to strive greatly. It will be our pleasure to listen this evening to statements of the plans and purposes of some of the best qualified among them.

CENSUS METHODS.

BY HON. E. DANA DURAND, DIRECTOR OF THE CENSUS.

Accuracy is the fundamental requirement of all statistical work. This means, first, accuracy in collecting the original material; second, accuracy in compiling and tabulating it; and, third, accuracy in analyzing and interpreting it. The first and third stages in the process of statistical work are extremely difficult, and it is concerning these that I wish especially to speak. It is a matter of comparative simplicity to secure accuracy in compiling and tabulating statistics. The principal problem here is to secure economy of time and money. A few words only as to the methods of compilation and tabulation contemplated for the coming census will be sufficient.

METHODS OF TABULATION.

The population census of 1910, like that of 1900, will be tabulated by the use of punched cards. Every person in the United States is given a card on which the facts with regard to sex, race, age, birthplace, birthplace of parents, and the like are indicated by the punching of appropriate holes. The number of persons possessing each specific characteristic or combination of characteristics which it is desired to show in the final tables is then counted by means of electrical tabulating machinery, electrical contacts being made through the holes punched in the cards.

The punching machines to be used at the present census differ very radically from those used before, and will, it is believed, not only increase the rapidity of the work, but tend to reduce the number of errors on the part of the clerks doing the punching. With the old punching machine, if an error was made in a single item, even though it might be the last item to be punched on the card,—the entire card had to be destroyed.

This consumed time, and also resulted in a tempation to the operator to let the error go uncorrected. With the machines now to be used, no hole is punched in the card until the keys for all the facts to be punched have been set, and, if the operator makes a mistake by depressing the wrong key, he can correct it before the card is punched. It was not found possible at the last census to check all of the cards back to the schedules, and it will probably not be feasible to do so at this census. Part of the cards punched by each operator are selected at random and compared back, and, if any appreciable percentage of error is discovered in those thus compared, the other cards punched by the same operator are likewise compared.

The tabulating machines to be used at the present census will likewise result, it is believed, in a material increase of rapidity and reduction of errors. This is principally brought about by the fact that the results of the count for each unit of area are automatically printed, whereas formerly they were registered on dials from which readings had to be taken and recorded by hand. The reading of these dials took a large amount of time, during which the machine was idle, and inaccurate readings were not uncommon.

THE MARGIN OF ERROR.

Turning now to the more important subject of the means of securing accuracy in the original census returns collected in the field, it is self-evident that the entire value of the census depends on the securing of approximate accuracy in the original returns. Absolute accuracy is out of the question, and, in fact, a small margin of error does not seriously affect the value of the statistics; but any considerable margin of error practically destroys their value. No degree of accuracy in tabulating and no degree of skill and judgment in analyzing statistics can give value to data which were incorrect in the first place.

It has too often been a vice of statisticians to present to the public tables purporting to show all sorts of important facts without due consideration of their truthfulness. The general

public who use the statistics-in fact, even the trained statisticians who use them-have in most cases no means of discovering the inaccuracy of such statistics and go on using them in misguided confidence. The compiler of statistics has often been, unconsciously perhaps, careless as to their accuracy because of the lack of any possibility of detection. The man who has charge of the collection of statistical data ought himself to be the severe and uncompromising critic of those data. He alone has approximately adequate means of judging the degree of accuracy which has been secured; and it is his duty, having done all that is possible to eliminate error, to inform the public fully and frankly of the extent to which error presumably still persists. So far from taking advantage of the fact that others cannot discover the errors which are hidden away in imposing-looking totals, he should from that very fact recognize the more clearly his own duty to explain just how much or how little the statistics may be trusted.

This does not mean, of course, that the statistician who knows that there are certain errors in his figures should straightway declare them of no value. Statistics on certain subjects may be of very little value unless almost absolute accuracy is secured, but there are many subjects as to which close approximations to the truth are almost as valuable as the exact truth itself. It is the mark of the competent statistician to be able to decide approximately what the margin of error actually is, and also to what extent the error vitiates the results. As Josh Billings said, "It is better not to know so much than to know too many things what ain't so."

At the census of 1900 and during the so-called intercensal period since that time the Census Bureau has done a great deal in the way of criticising its own statistics in the manner I have suggested; but I doubt whether it has yet been sufficiently emphatic in cautioning the public with respect to the existence of unavoidable errors. As one means of expressing the approximate character of the data, I think it desirable to express all very large totals of manufacturing and agricultural statistics in millions, and, in the case of the smaller figures

which appear in the statistics for individual localities and industries, to express the results in thousands.

To criticise statistics, however, is easy: the difficult task is to improve them. The Census Bureau will have done part of its duty to the public if it gives warning regarding the margin of error in the statistics it publishes, but it is far more important that it should reduce that margin of error.

METHOD OF SELECTING FIELD FORCE.

There are two means to this end. The first and more difficult is to secure better men to collect the statistics: the second is to simplify and clarify the inquiries.

As you are aware, the statistics of population and of agriculture are collected by a different force from that employed in gathering the statistics of manufactures. The population and agricultural data are secured by enumerators, of whom there will be about sixty-five thousand at the present census, they in turn being appointed by the supervisors, of whom there are about three hundred and thirty. The difficulty of securing competent and faithful enumerators is very great. The length of service is short, fifteen days in the cities and thirty days in the country districts. The period is thus too short to justify a man who has a good job in quitting it, while, on the other hand, it is too long in most cases to enable such men to get leave from their regular work to take the census. Moreover, the pay is small, averaging perhaps $3 per day in the country districts and a trifle more in the cities,-practically the pay of ordinary mechanics. Most of those who seek, from financial motives, to be enumerators are men who are able to command only moderate pay in their occupations, while many of them are men who cannot command regular employment and who are looking for odd jobs.

Consideration has been given by the census authorities from time to time to the plan pursued in Germany and some other European countries, by which the census is taken chiefly or wholly by men serving without pay, who either volunteer