Roy D G Allen - Collected Quotes

"A knowledge of statistical methods is not only essential for those who present statistical arguments it is also needed by those on the receiving end." (Roy D G Allen, "Statistics for Economists", 1951)

"All statistical data are subject to errors in collection." (Roy D G Allen, "Statistics for Economists", 1951)

"Any time series can now be plotted in two ways. Time is measured along the horizontal axis on a natural scale; the variable is measured along the vertical axis either on a natural or on a ratio scale. A graph of the second kind is the new construction; it is often called a semi-logarithmic graph since the ratio or logarithmic scale is used on one of the two axes of the graph."

"As with tabulation, however, skill in constructing diagrams is only acquired after long experience. The main point can be easily made; a graph or diagram should be clear and simple since it adds nothing to our understanding if it does not show up the trends and relations of our data more obviously than in the original tables. A chart is meant to 'help out' in drawing broad conclusions from a table which may be quite complicated. Inevitably the graph or diagram is less exact and shows less detail than the table; it is a step in the constant process of summarizing data. This must not be overdone. It is only too easy to simplify so drastically as to be misleading." (Roy D G Allen, "Statistics for Economists", 1951)

"Graphs and diagrams help to show up trends and relations but they do not define or measure them precisely. This can be achieved by calculations on the numerical data and, in particular, by the derivation of figures to summarize and relate the significant facts in a table. The main purpose of statistical analysis is to make comparisons. A single figure has no meaning by itself; it only becomes significant and "alive" when compared explicitly or implicitly with another figure. Our first task in analysis is to make the comparisons explicit, to express the relation between one figure and another." (Roy D G Allen, "Statistics for Economists", 1951)

"Graphs [for time series] can be misleading, however, and we need to subject our first impression to a closer scrutiny. We must develop more precise methods of analysis of time series. The variations of a time series are of many kinds which can be grouped under three heads. There is, first, the general direction of movement or the trend of the variable over the long period. Then there are oscillations of various types, of greater or less regularity, superimposed on the trend. Finally, there are residual or irregular variations which may arise from isolated events such as a war or general strike, or which may be due to the operation of random influences." (Roy D G Allen, "Statistics for Economists", 1951)

"It is only by experience that skill is acquired in the framing of tables. It is partly a matter of design, to get a neat and concise layout which is both cheap to print and easy on the eye. It is partly a question of making sure that no essential information is omitted so as to leave the meaning of the table uncertain." (Roy D G Allen, "Statistics for Economists", 1951)

"Not even the most subtle and skilled analysis can overcome completely the unreliability of basic data." (Roy D G Allen, "Statistics for Economists", 1951)

"One very simple but effective form of statistical analysis is to represent the tabular data by drawing graphs or diagrams. If made with skill and care in avoiding bias, a diagram will show the data in a graphical form in which the salient features leap to the eye. The risk is that diagrams can be misleading when drawn by the unskilled and they can be very dangerous tools in unscrupulous hands." (Roy D G Allen, "Statistics for Economists", 1951)

"Summarization of statistical data into tabular form is an art rather than a routine following a set of formal rules. Tabulation inevitably implies a loss of detail. The original data are far too voluminous to be appreciated and understood; the significant details are mixed up with much that is irrelevant. The art of tabulation lies in the sacrifice of detail which is less significant for the purposes in hand so that what is really important can be emphasized. Tabulation implies classification, the grouping of items into classes according to various characteristics. And classification depends on clear and precise definitions." (Roy D G Allen, "Statistics for Economists", 1951)

"The error in a sum or difference of any number of rounded figures is the sum of the errors in the separate figures. [...] The relative error in a product or quotient. of two rounded figures is approximately the sum of the relative errors in the separate figures. [...] It is generally safe to write a product or quotient as correct to one less significant figure than the less accurate of the two values in the product or quotient." (Roy D G Allen, "Statistics for Economists", 1951)

"The function of the regression lines, as approximate representations of means of arrays, is to isolate the mean value of one variable corresponding to any given value of the other; the variation of the first variable about its mean is ignored. A regression line is an average relation, and with it there is a variation of values about the average. In the regression of y on x, the variation ignored is in the vertical direction, a variation of y up and down about the line." (Roy D G Allen, "Statistics for Economists", 1951)