com-
valueless. For example if we were to compare cost of living figures which included rents for the years 1914 and 1939 we would be at least making a parison, but if we attempted to set a figure for 1914 which included rents beside à figure for 1939 which excluded rents we should not be making any sort of comparison at all-our alteration for 1939, even though it might appear to he justified on the score that rents for that year were quite abnormal, would simply result in the whole of the statistical records of the last quarter of a century becom- ing valueless.
This aspect of statistics has been impressed on me again and again both by government statistical officers and theoretical statisticians. I have emphasized the point because I think that it means that it would be a grave mistake to attempt too rapid or too wide a reorganization of our statistics. A slow and calculated develop- ment of the new department which will enable it to function on a rigidly set course for several decades and to shew continuity with our statistics for the last twenty or thirty years is likely I think to produce in the long run something of value, while a rapid reorganization without reference to past statistics, which may require modification year after year, will produce nothing at all.
Considering statistics on the theoretical side we have the following methods :—
Index Numbers. Rates (e.g. Birth and Death rates), etc.
The general nature of these devices is of course already familiar. They can be used in a variety of ways although some technical care is necessary to ensure that they give accurate information.
Methods of Measuring Correlation and Independence.
These are much less familiar devices although in my opinion they offer very considerable
in administrative work. They are scope
methods by which the relationship between two series of statistics can, in certain circumstances, be measured. If for example it is desired to ascertain whether any relationship exists between overcrowding and crime or between railway returns and New Terri- tories road expenditure or between overtime worked and sick leave, or in general between any two parallel series of statistics these methods, if they can be applied offer in many cases an incontrovertible answer that such a relationship exists or does not exist. The merit of this method is that instead of a mere vague expression of opinion, however expert "The committee is strongly of the opinion that
we can often get a mathematically certain statement "the two phenomena are related to to the extent of
-%- so that a change of so much in the one
has in the long run invariably been accompanied by a change of so much in the other". It is true that a definite statement of this character might be distrusted by the layman who would continue to prefer the "opinion" right or wrong of the experts. This is however merely a matter of prejudice and if it once became familiar its use should be as acceptable as index numbers are to-day. It would appear that one of its principal uses in government work would be the analysis of data for committees appointed from time to time to report on special problems.
Sampling.
This is in effect the taking of a sample survey of some aspect of government work without the expense of time and trouble involved in taking a complete "census" of every individual. Suppose for example that it was desired to ascertain the number of persons in the whole of Hong Kong suffering from tuberculosis or the number of persons totally unemployed, or the number of persons living in conditions that complied with a certain definition of overcrowding, all that would be necessary would be to make a very thorough examination of perhaps a thousand cases picked more or less at random from appropriate districts and from appro- priate strata of the population. On the basis of this small sample it would be possible to say for example that the chances were 10,000 to 1 that the total number of persons in the Colony shewing the investigated quality lay between 52%
In other words that if 10,000 such samples were taken 9999 would shew the given result
correctly, one would be incorrect.
3
and 58% of the total population. An actual census of the total population would of course give a more exact result but while it would be obviously impossible for the medical department, for example, to examine the total population for tuberculosis a sample examination of a thousand persons might be feasible. For all practical purposes of course the statement "between 52% and 58%" is as valuable as an exact percentage. This method is adopted in England in connexion with cost of living
surveys.
Interpolation and Graduation.
These are the processes by which it is sometimes possible to ascertain with considerable accuracy the statistics applying to a period for which no recorded statistics are available, or of ascertaining the most probable course that insufficient or inaccurate statistics would have taken had they been exact. In rare cases fore- casts of future trends can also be made.
By these methods it might for example be possible to ascertain with some degree of certainty the export and import figures for Hong Kong for the period during which those returns were suspended. To take a second example it is possible to ascertain with considerable accuracy the number of women of child- bearing age in any census even when there is a certainty that 20% of the ages given are substantially incorrect.
Intercensal estimates of various kinds fall under this heading.
The methods of theoretical statistics are in general so gumbrous and com- plicated that their practical value is seriously impaired. Really valuable results can be obtained but if two or three days work is required to reach those results it may be better to forego them. For this reason I employed a considerable part of the period devoted to theoretical study in considering and adapting various simplifying methods that have been proposed.
The above are some of the principal methods by which statistical technique might be of assistance in administrative work. But this aspect of statistics, which we may call the aspect of the statistician is almost completely separate from the collection and presentation of statistics which we may call the aspect of the statistical officer.
Take for example the case of the Registrar of Births and Deaths; his work is primarily to ensure that the machinery of collecting data, issuing certificates and compiling totals is properly done. This is the work of a statistical officer. The statistician on the other hand is concerned with an analysis of the data. He will distinguish between crude and corrected rates, he will seek to correlate causes of deaths in certain localities with other phenomena, perhaps drainage or water supply, in those localities. He will try to use his birth registration data to give an accurate birth rate figure and from this to get an accurate figure for the female population, and thence an accurate figure of the male population in intercensal years. In Singa- pore intercensal estimates have increased in accuracy from a 30% error to a 1%. error from the use of such methods.
Similarly with imports and exports-the present statistical department concerns itself with the collection of statistics a statistician will be chiefly interested in shewing what tendencies these statistics indicate, in correlating statistics of trade with other statistics of population or prices, or revenue.
Thus there is a distinct line of cleavage between the aspect of the statistician and that of the statistical officer. In effect the former is simply an administrative officer using refined scientific methods to indicate what lines government policy might profitably take while the statistical officer is simply a collector of facts.
It does not
For this reason it is perhaps a matter of little moment whether the work of collection and the work of analysis should be done in one department. matter very much. from the statistical aspect whether, for example, the Registrar
6