Applied Linear Statistical Models 5e is the long established leading authoritative text and reference on statistical modeling. The text includes brief introductory and review material, and then proceeds through regression and modeling for the first half, and through ANOVA and Experimental Design in the second half. All topics are presented in a precise and clear style supported with solved examples, numbered formulae, graphic illustrations, and Notes to provide depth and statistical accuracy and precision. The Fifth edition provides an increased use of computing and graphical analysis throughout, without sacrificing concepts or rigor. In general, the 5e uses larger data sets in examples and exercises, and where methods can be automated within software without loss of understanding, it is so done.

At the December 1984 meeting of the ASA Board of Directors, the board authorized a long-range strategic planning effort by the ASA and approved a budget line item to support the activity. Incoming President John Neter asked incoming President-Elect Donald Marquardt to lead the strategic planning effort. The strategic plan will influence the course of ASA and the statistics profession for some years to come. In view of the potential for a wide impact, the planning process was organized to involve many people from a wide variety of constituencies in ASA. The participation of many ASA constituent organizations and members enriched the resulting plans. Such a process necessarily took much effort and time, but many of the most important future benefits derive directly from the discussions and the insights developed by the ASA members involved during the planning process. In preparing this report we represent the consensus of many people and, in particular, of the Board of Directors. Consensus does not necessarily mean unanimity. One of the strengths of ASA is the diversity of its members and their interests, complementing their unity of purpose in the field of statistics. A Strategic Planning Group carried out many activities, beginning in March 1985, and set the stage for a meeting of the combined 1985 and 1986 Boards of Directors at Gaithersburg, Maryland, on November 1-3, 1985. At that meeting, strategic issues and strategic goals were defined. After that meeting, three task forces worked on strategic issues and presented their reports to the Board of Directors in August 1986. (A summary of strategic planning activities is included in App. A.) Using the results from the task forces and comments from the board, the chairs of the three task forces, with John Neter and Don Marquardt, prepared this integrated document. At the same time the former Future Goals Committee was restructured as the Planning Committee, with the presidentelect serving as chair, to deal with tactical year-to-year planning of ASA program priorities. The Planning Committee does not do long-term planning, though it does have responsibility for preparing updates to the ASA strategic plan and determining when a new plan is needed. Though most organizations, including ASA, do fairly well at shortterm planning, frequently there is little emphasis on longrange strategic planning. Section 2 describes the foundation of the strategic plan. This material was initially developed at the Gaithersburg conference and has subsequently been refined. Section 3 includes a statement of the future directions of ASA, in light of the 14 goals set by the Board of Directors at its April 1986 meeting. Section 4 describes needed modifications in the structure and functions of the organization, including the roles of sections, committees, chapters, publications, meetings, membership, the Board of Directors, and the central office. There is a section on the interactions of ASA with other organizations. Finally, there is a section on the transition to a new organization.

Auditors usually have information on both the book amount and the audit amount available for each sample audit unit. Problems in using difference and ratio estimators, as well as some other estimators, for constructing large-sample normal confidence intervals when the population error rate is low are explored empirically in this article. The findings indicate the need for great care in using large-sample normal confidence intervals for sample sizes of 100 or 200 (sizes frequently encountered in auditing practice).

The choice of statistical sampling procedures in auditing encompasses selecting (1) a method of sampling, (2) one or more statistics to be calculated from the sample, and (3) a methodology for assessing the sample results. It is the premise of this paper that the audit objectives, the environment in which the audit occurs, and the audit procedures utilized greatly influence the choice made. Since these factors vary widely from one audit situation to another, no one statistical sampling procedure is likely to be appropriate for all circumstances. In this paper, we consider the effects of audit objectives, the environment, and audit procedures on the selection of a sampling procedure and the specific design of the sample. The discussion will focus on some of the major considerations in the selection of a sampling procedure and the design of the sample and is not intended to include every possible consideration. We conclude the paper with a simple, tentative model of a decision process which an auditor might utilize in choosing a statistical sampling procedure.

Professors Neter and Williams' paper entitled, Acceptability of Three Normative Methods in Insurance Decision Making describes an interesting and well done experiment.' The purpose of this note is twofold: (1) to illustrate how dissonance theory provides an alternative explanation of the results of their experiment, and (2) to suggest some personality variables which have proven useful in the analysis of decision making behavior.

The use of computers in teaching statistics can make an impact upon students in two ways. First, it can affect the amount and rate of learning of statistics. Second, it can affect student attitudes toward computers, statistics and quantitative methods. This note reports a study on the effects of the computer in teaching statistics on student attitudes. The data are based on pre- and post-course questionnaires administered in a first-year statistics sequence. The results partly confirm expectations, and partly raise questions for further investigation.

task of writing a self-correcting problem book in statistics that can be used with many different texts is most difficult. Whitmore, Neter and Wasserman make such an attempt in Self-Correcting Problems in Statistics. scope of the material is good, including all of the topics covered in most introductory business and economics texts such as descriptive statistics, classical inference and hypothesis testing, Bayesian analysis, regression, time series analysis and index numbers. There are also sections on multiple regression, tests of goodness of fit, contingency tables and analysis of variance. In an attempt to minimize the problem of incompatibility, the authors provide an extensive glossary of symbols and a table crossreferencing sections of their book with chapters in thirty standard texts. authors opt for a minimum of exposition in tying together the exercises and problems. A consequence of this decision is a tendency to appear cook-bookish and formula oriented, especially where the subject matter is relatively difficult. This tendency is compensated for to some extent by introducing new material in the problems and trying to get the student to generalize about the nature of the statistical process in question by observing what happens in particular problem situations. This process has disadvantages as well as obvious advantages. While the ability of the student to observe what is happening and to generalize from particular situations is promoted, the situations are contrived so that the student is led to make the right generalization on the basis of very little evidence. Perhaps he should also be warned that when he tries this in non-structural, real world situations, he might be so lucky. New material, unfortunately, is frequently introduced in review problems. first time the student is confronted specifically with the notion that s can be used as an estimator of a is in the answer to review question 2b (p. 84). fact that op is a maximum when p .5 is found in the answer to review question 2d (p. 92). sign test is introduced in review problem 1 (p. 166). fact that exponential smoothing can be more than single order is noted only in the answer to review question 2c (p. 316). Within the conditions just stated, and remembering that this is a supplement to a text rather than a text itself, most of the topics are well covered. exposition is usually clear and there are exceptionally few typographical or computational errors. worst sections by far are the Bayesian ones, where it is clear that the authors' hearts really are elsewhere. Decision making under uncertainty is left entirely in the air: the basis for a choice among maximax, maximin or minimax of regret criteria is completely ignored. argument for decision making under uncertainty is weak. Further, no attempt is made to relate the Bayesian sections to those on classical inference. Caught in the no man's land between a problem book and a full text, the authors sometimes fall into other types of strategic errors. They do define terms carefully. For example, they do define probability (pp. 36-37), discrete (p. 49), simple random (p. 83), or unit normal loss function (p. 267). Important concepts are sometimes explained. For example, on p. 21, in developing the formula X = 2;fX/n, no mention is made of the fact that X in this context represents the mid-values of the classes, or that the sample mean so obtained is an approximation contingent upon a number of assumptions about the frequency distribution. On p. 28, s= <2/(X C) 2/(n 1) is defined as the sample standard deviation rather than an estimator of the population standard deviation obtained from a sample. No mention of the use of s as an estimator for ar is made until the answer to a review problem on p. 84. No explanation is given for the use of n 1 in the denominator. hypergeometric distribution (not defined as such) is calculated only by enumeration, and is related to the earlier discussion of the general multiplication theorem. Though the formula for the binominal distribution is given, most students will miss its relationship to the special multiplication theorem. Though power and OC curves are derived and are called for in the solution to many problems, no mention is made of how they might be used. Nowhere do the authors distinguish between statistical and practical significance. In the section on inferences concerning variances they develop one problem in which the interval estimate for the population variance of the adhesive quality of a tape is between .23 and 6.32 with no mention of the units in which strength is measured or the practical limits which are feasible for such a product. To keep things simple, problems are often trivial or unrealistic. In many problems, sample sizes and levels of significance are assumed with no justification. In Example 1 of Section 20, the breakeven point is assumed to be .05. authors then ask If it is important to . . . if p = .04, is the suggested decision rule an effective one? student might well wonder when it would ever not be important to proceed when p was less than the break-even point. In a problem involving the number of man-hours lost due to accidents in each of the 20 production units of a company, the problem states The company wishes to test whether or the safety program has changed the number of man-hours lost due to accidents within the production units. change from the previous year is then tested using the sign test, with the conclusion that the program in fact had reduced accidents. It was assumed without question that the data presented were sample data and that a test of significance was appropriate. Nowhere was it mentioned that accidents might have declined for reasons other than the institution of the safety program. In spite of the weaknesses discussed here, one must conclude that the strengths of the book outweigh its weaknesses, and that with judicious interpretation by the instructor, the use of Self-Correcting Problems in Statistics could be of considerable benefit to the typical student of business or economics. PAUL JEDAMUS UIniversity of Colorado

In an experiment, thirty-six professional insurance men employed the utility method, the worry method, and the comparison method in two insurance problems to determine the appropriate insurance coverage. It was found that the best act with the utility method almost always was no insurance, whereas the best act with the other two methods tended to be complete insurance. The utility method best act rarely agreed with the subject's actual preferred act, while the comparison method best act usually was the same or almost the same as the actual preferred act. While the worry method suggests that the subjects typically are risk averse, the utility functions obtained tend to show that the subjects typically are not risk averse.

The primary purpose of incorporating a set of internal controls in the financial information system is to enhance the system's reliability-i.e., to maintain a high probability of preventing, detecting, and eliminating errors, irregularities, and fraud in the financial information system. The demonstrated reliability of the system provides evidence as to the quality of the output of the system. It is well accepted that the effectiveness of internal controls must be taken into account in determining the extent and nature of the audit procedures appropriate in a given examination.' The more reliable the system, the less extensive the tests the auditor need conduct. Recognizing this inverse relationship between effectiveness of internal control and audit scope, the American Institute of CPAs requires all auditors to initially evaluate the reliability of internal controls as a matter of audit standards.2 Recently, the Committee on Auditing Procedures of the AICPA released several statements on the subject of internal controls which re-emphasize the importance of the study of their reliability.3 But, despite this emphasis, the auditor currently does not possess a means to objectively evaluate the reliability of the internal control system. Conventionally, the auditor uses questionnaires, flow charts, and tests of transactions for evaluation pur-

BECAUSE of empirical difficulties in obtaining necessary experimental measurements or survey data, there has been a persistent gap between index number research and practice. Frisch [6], in his survey of priceindex problem for Econometrica, distinguished between and ap-proaches to measuring price level changes. In atomistic approach, prices and quantities are considered in main two sets of independent variables. The research focus under this approach has been on analysis of comparative merits of competing index number formulas and on development of tests for identifying desirable formulas. The most influential work in atomistic tradition has been that by I. Fisher. In a 1920 paper, (later amplified into The Making of Index Numbers [5]), Fisher presented his conclusion that index number formula called was best for general purposes. As is well-known, this formula measures a binary price change between a given period and base period by taking of a geometric mean ofthe Laspeyres and Paasche indexes. Although this formula attracts continuing attention, it has never been employed widely in practice. For one -thing, index number practice has been concerned mainly with compilation of ongoing, internally consistent time series of index values rather than with individual binary measurements per se. While in principle chaining of ideal indexes could be used to construct a time series, formula requires both base-period and given-period for each binary comparison. The data for simply have not been available to permit ideal formula to be applied on any scale in compiling index numbers in real-world situations. Instead, Mitchell's [9] lead has been followed in that most index number series compiled today are based on fixed weights, where are typical for some period of time and are revised new editions of index are issued periodically. Mitchell cited, a practical reason in favor of this approach, the heavy expense in time and labor required for frequent revisions of weights [9, p. 66]. The use of held relatively fixed over an extended sequence of time periods has given rise to extensive and continuing controversy. A major criticism of fixed weight indexes is that they do not allow for routine substitutions among commodities relative prices shift. In functional approach, certain characteristic relations are presumed to exist between prices and quantities, and objective in constructing a functional price index is to measure money expenditure variation under constant [6, p. 22]. If a person who in situation 0 spends an amount pO is as well off when in situation 1 he spends an amount Pl, and assuming that expenditures in both cases are allocated optimally with respect to satisfaction obtained, two sets of quantity combinations are considered and functional price index for 1 on a 0 base is (pip/o) X 100. Frequently theory is extended to a group of consumers for whom individual consumer is typical or representative. Under ordinalist point of view which emerged in 1930's, quantities have usually been defined in terms of points on an indifference surface at some indifference level, I. All points on this surface represent equivalent combinations of commodities since each yields satisfaction level I to consumer. Thus, a central operating problem in compiling functional price indexes is to determine quantity combinations by which a given level of I can be reached most economically at prices of situations 0 and 1. The difficulties in making empirical determinations of

This article deals with the use and misuse of the correlation coefficient when the dependent variable is of a dichotomous 0,1 nature. It focuses particularly on problems relating to curvilinearity and the nature of the prediction being made. The prediction of consumer purchases from reported subjective probabilities provides a vehicle for illustrating problems discussed. It is noted that with a 0,1 dependent variable, the correlation ratio is likely to be a better measure of the degree of relationship than the coefficient of determination because it is free of restrictions on the functional form of the relationship. The article then considers the mean error probability and the average conditional entropy as alternative measures. Finally, the article emphasizes that the purpose for using the relation between the independent and dependent variable should govern the development of an appropriate model and the measure to be used for deciding which of several independent variables is best.

Empirical findings on measurement errors in reports by consumers of expenditures made in the past are reviewed, the implications of these findings for survey design are considered, and the needs for future research are discussed.

A linear programming food price index permits the possibility of routine substitutions in response to price changes, while a fixed weight index does not. This paper cites a variety of evidence indicating that consumers do trade off among foods in response to both long-term and seasonal price changes. Implications of this for food price indexes are discussed. A linear programming food price index is then developed, based on an empirical menu-planning model, and compared with corresponding fixed weight price indexes over a nine-month period. The linear programming index shows a significantly smaller increase in food prices over the nine-month period than the fixed weight index with initial period weights.