Non-fiction: Statistical Adjustment of Data

Overview
W. Edwards Deming's 1943 Statistical Adjustment of Data offers a concise, applied treatment of how to reconcile, correct, and combine numerical information from surveys and experiments. Written from the vantage of his early government and industrial consulting, the text foregrounds practical problems of bias, measurement error, and conflicting estimates, and sets out statistical principles for producing coherent, defensible summaries. The tone is technical yet pragmatic, aimed at practitioners charged with making data from different sources mutually consistent and suitable for policy or operational use.

Core Methods
Deming emphasizes the identification and separation of different error sources, distinguishing systematic bias from random variability and advocating explicit modeling of each component. He presents classical tools such as ratio and regression adjustments, weighted averaging, and least-squares reconciliation, explaining how these can reduce bias and improve precision when auxiliary information is available. Variance estimation and the propagation of uncertainty are treated as essential complements to point adjustment, with guidance on how to preserve honest measures of reliability after data are altered.

Modeling and Assumptions
A central theme is the careful specification of assumptions: which relationships among variables are being imposed, when it is appropriate to treat auxiliary values as fixed, and how model misspecification can induce misleading adjustments. Deming contrasts design-based ideas, where sampling design dictates inference, with analytic adjustments that rely on models linking estimates to true quantities, and he repeatedly cautions about the consequences of unjustified presumptions. Practical rules for assessing model fit and sensitivity to assumptions appear alongside mathematical derivations, encouraging disciplined judgment rather than blind mechanical application.

Handling Imperfect and Conflicting Data
Deming addresses common problems such as nonresponse, incomplete measurement, inconsistent tabulations, and overlapping surveys. Techniques like post-stratification, calibration to known totals, and reconciliation of independent counts are presented with an eye to preserving important margins while harmonizing internal inconsistencies. He advocates transparent documentation of the adjustments applied and recommends analysts report both original and adjusted figures so users can judge the impact of the procedures.

Applications and Examples
Examples draw from government statistical work and industrial measurement, reflecting Deming's experience with censuses, agricultural surveys, and quality-related measurement systems. Case discussions illustrate how modest analytical corrections can materially affect estimates used for planning, allocation, and regulatory decisions, and how improper adjustment can create greater harm than leaving raw but imperfect figures. The examples reinforce the principle that statistical adjustment is a tool to be wielded with technical competence and ethical restraint.

Legacy and Influence
Statistical Adjustment of Data helped codify a pragmatic approach to reconciling empirical information that influenced mid-20th-century survey practice and institutional data processing. Its combination of theory, practical rules, and professional caution fed into later developments in survey sampling, calibration estimation, and measurement error modeling, and it complements Deming's broader advocacy for statistical thinking in administration and industry. The work remains a succinct reflection of an era when statisticians were formalizing methods to make disparate measurements speak to one another coherently and reliably.