內容簡介
Principles and Techniques、Design: Basic Principles and Techniques、The Art of Experimentation、Replication、Blocking、Randomization、Analysis: Basic Principles and Techniques、Planning Experiments、A Checklist for Planning Experiments、Real Experiment——Cotton-Spinning Experiment等等。
內頁插圖
目錄
Preface
1. Principles and Techniques
1.1. Design: Basic Principles and Techniques
1.1.1. The Art of Experimentation
1.1.2. Replication
1.1.3. Blocking
1.1.4. Randomization
1.2. Analysis: Basic Principles and Techniques
2. Planning Experiments
2.1. Introduction
2.2. A Checklist for Planning Experiments
2.3. A Real Experiment——Cotton-Spinning Experiment
2.4. Some Standard Experimental Designs
2.4.1. Completely Randomized Designs
2.4.2. Block Designs
2.4.3. Designs with Two or More Blocking Factors
2.4.4. Split-Plot Designs
2.5. More Real Experiments
2.5.1. Soap Experiment
2.5.2. Battery Experiment
2.5.3. Cake-Baking Experiment
Exercises
3. Designs with One Source of Variation
3.1. Introduction
3.2. Randomization
3.3. Model for a Completely Randomized Design
3.4. Estimation of Parameters
3.4.1. Estimable Functions of Parameters
3.4.2. Notation
3.4.3. Obtaining Least Squares Estimates
3.4.4. Properties of Least Squares Estimators
3.4.5. Estimation ofo2
3.4.6. Confidence Bound for ~r2
3.5. One-Way Analysis of Variance
3.5.1. Testing Equality of Treatment Effects
3.5.2. Use of p-Values
3.6. Sample Sizes
3.6.1. Expected Mean Squares for Treatments
3.6.2. Sample Sizes Using Power of a Test
3.7. A Real Experiment——-Soap Experiment, Continued
3.7.1. Checklist, Continued
3.7.2. Data Collection and Analysis
3.7.3. Discussion by the Experimenter
3.7.4. Further Observations by the Experimenter
3.8. Using SAS Software
3.8.1. Randomization
3.8.2. Analysis of Variance
Exercises
4. Inferences for Contrasts and Treatment Means
4.1. Introduction
4.2. Contrasts
4.2.1. Pairwise Comparisons
4.2.2. Treatment Versus Control
4.2.3. Difference of Averages
4.2.4. Trends
4.3. Individual Contrasts and Treatment Means
4.3.1. Confidence Interval for a Single Contrast
4.3.2. Confidence Interval for a Single Treatment Mean
4.3.3. Hypothesis Test for a Single Contrast or Treatment Mean
4.4. Methods of Multiple Comparisons
4.4.1. Multiple Confidence Intervals
4.4.2. Bonferroni Method for Preplanned Comparisons
4.4.3. Scheff6 Method of Multiple Comparisons
4.4.4. Tukey Method for All Pairwise Comparisons
4.4.5. Dunnett Method for Treatment-Versus-Control Comparisons
4.4.6. Hsu Method for Multiple Comparisons with the Best
reatment
4.4.7. Combination of Methods
4.4.8. Methods Not Controlling Experimentwise Error Rate
4.5. Sample Sizes
4.6. Using SAS Software
4.6.1. Inferences on Individual Contrasts
4.6.2. Multiple Comparisons
Exercises
5. Checking Model Assumptions
5.1. Introduction
5.2. Strategy for Checking Model Assumptions
5.2.1. Residuals
5.2.2. Residual Plots
5.3. Checking the Fit of the Model
5.4. Checking for Outliers
5.5. Checking Independence of the Error Terms
5.6. Checking the Equal Variance Assumption
5.6.1. Detection of Unequal Variances
5.6.2. Data Transformations to Equalize Variances
5.6.3. Analysis with Unequal Error Variances
5.7. Checking the Normality Assumption
5.8. Using SAS Software
5.8.1. Using SAS to Generate Residual Plots
5.8.2. Transforming the Data
Exercises
6. Experiments with Two Crossed Treatment Factors
6.1. Introduction
6.2. Models and Factorial Effects
6.2.1. The Meaning of Interaction
6.2.2. Models for Two Treatment Factors
6.2.3. Checking the Assumptions on the Model
6.3. Contrasts
6.3.1. Contrasts for Main Effects and Interactions
6.3.2. Writing Contrasts as Coefficient Lists
6.4. Analysis of the Two-Way Complete Model
6.4.1. Least Squares Estimators for the Two-Way Complete Model
6.4.2. Estimation ofo~ for the Two-Way Complete Model
6.4.3. Multiple Comparisons for the Complete Model
6.4.4. Analysis of Variance for the Complete Model
6.5. Analysis of the Two-Way Main-Effects Model
6.5.1. Least Squares Estimators for the Main-Effects Model
6.5.2. Estimation ofa2 in the Main-Effects Model
6.5.3. Multiple Comparisons for the Main-Effects Model
6.5.4. Unequal Variances
6.5.5. Analysis of Variance for Equal Sample Sizes
6.5.6. Model Building
6.6. Calculating Sample Sizes
6.7. Small Experiments
6.7.1. One Observation per Cell
6.7.2. Analysis Based on Orthogonal Contrasts
6.7.3. Tukeys Test for Additivity
6.7.4. A Real Experiment——Air Velocity Experiment
6.8. Using SAS Software
6.8.1. Contrasts and Multiple Comparisons
6.8.2. Plots
6.8.3. One Observation per Cell
Exercises
7. Several Crossed Treatment Factors
7.1. Introduction
7.2. Models and Factorial Effects
7.2.1. Models
7.2.2. The Meaning of Interaction
7.2.3. Separability of Factorial Effects
7.2.4. Estimation of Factorial Contrasts
7.3. Analysis——Equal Sample Sizes
7.4. A Real Experiment——Popcorn-Microwave Experiment
7.5. One Observation per Cell
7.5.1. Analysis Assuming That Certain Interaction Effects Are egligible
7.5.2. Analysis Using Normal Probability Plot of Effect Estimates
7.5.3. Analysis Using Confidence Intervals
7.6. Design for the Control of Noise Variability
7.6.1. Analysis of Design-by-Noise Interactions
7.6.2. Analyzing the Effects of Design Factors on Variability .
7.7. Using SAS Software
7.7.1. Normal Probability Plots of Contrast Estimates
7.7.2. Voss-Wang Confidence Interval Method
7.7.3. Identification of Robust Factor Settings
7.7.4. Experiments with Empty Cells
Exercises
8. Polynomial Regression
8.1. Introduction
8.2. Models
8.3. Least Squares Estimation (Optional)
8.3.1. Normal Equations
……
9. Analysis of Covariance
10. Complete Block Designs
11. Incomplete Block Designs
12. Designs with Two Blocking Factors
13. Confounded Two-Level Factorial Experiments
14. Confounding in General Factorial Experiments
15. Fractional Factorial Experiments
16. esponse Surface Methodology
17. andom Effects and Variance Components
18. estde Models
19. plit-Plot Designs
A. ables
Bibliography
Index of Authors
Index of Experiments
Index of Subjects
精彩書摘
In the analysis of data, it is desirable to provide both graphical and statistical analyses. Plotsthat illustrate the relative responses of the factor settings under study allow the experimenterto gain a feel for the practical implications of the statistical results and to communicateeffectively the results of the experiment to others. In addition, data plots allow the proposedmodel to be checked and aid in the identification of unusual observations, as discussed inChapter 5. Statistical analysis quantifies the relative responses of the factors, thus clarifyingconclusions that might be misleading or not at all apparent in plots of the data.
The purpose of an experiment can range from exploratory (discovering new importantsources of variability) to confirmatory (confirming that previously discovered sources ofvariability are sufficiently major to warrant further study), and the philosophy of the analysisdepends on the purpose of the experiment. In the early stages of experimentation the analysismay be exploratory, and one would plot and analyze the data in any way that assists in theidentification of important sources of variation. In later stages of experimentation, analysisis usually confirmatory in nature. A mathematical model of the response is postulated andhypotheses are tested and confidence intervals are calculated. In this book, we use linear models to model our response and the methodofleast squaresfor obtaining estimates of the parameters in the model. These are described in Chapter 3.Our models include random "error variables" that encompass all the sources of variabilitynot explicity present in the model. We operate under the assumption that the error termsare normally distributed. However, most of the procedures in this book are generally fairlyrobust to nonnormality, provided that there are no extreme observations among the data. It is rare nowadays for experimental data to be analyzed by hand. Most experimentersand statisticians have access to a computer package that is capable of producing, at the veryleast, a basic analysis of data for the simplest experiments. To the extent possible, for eachdesign discussed, we shall present useful plots and methods of analysis that can be obtainedfrom most statistical software packages. We will also develop many of the mathematicalformulas that lie behind the computer analysis. This will enable the reader more easilyto appreciate and interpret statistical computer package output and the associated manuals.Computer packages vary in sophistication, flexibility, and the statistical knowledge requiredof the user. The SAS software is one of the better packages for analyzing experimental data.It can handle every model discussed in this book, and although it requires some knowledgeof experimental design on the part of the user, it is easy to learn. We provide some basicSAS statements and output at the end of most chapters to illustrate data analysis.
前言/序言
Our initial motivation for writing this book was the observation from various students thatthe subject of design and analysis of experiments can seem like "a bunch of miscellaneoustopics." We believe that the identification of the objectives of the experiment and the practicalconsiderations governing the design form the heart of the subject matter and serve as thelink between the various analytical techniques. We also believe that learning about designand analysis of experiments is best achieved by the planning, running, and analyzing of asimple experiment.
With these considerations in mind, we have included throughout the book the detailsof the planning stage of several experiments that were run in the course of teaching ourclasses. The experiments were run by students in statistics and the applied sciences and aresufficiently simple that it is possible to discuss the planning of the entire experiment in afew pages, and the procedures can be reproduced by readers of the book. In each of theseexperiments, we had access to the investigators actual report, including the difficultiesthey came across and how they decided on the treatment factors, the needed number ofobservations, and the layout of the design. In the later chapters, we have included detailsof a number of published experiments. The outlines of many other student and publishedexperiments appear as exercises at the ends of the chapters. omplementing the practical aspects of the design are the statistical aspects of the anal-ysis. We have developed the theory of estimable functions and analysis of variance withsome care, but at a low mathematical level. Formulae are provided for almost all analyses sothat the statistical methods can be well understood, related design issues can be discussed,and computations can be done by hand in order to check computer output.
We recommend the use of a sophisticated statistical package in conjunction with thebook. Use of software helps to focus attention on the statistical issues rather than on thecalculation. Our particular preference is for the SAS sof~vare, and we have included theelementary use of this package at the end of most chapters. Many of the SAS program filesand data sets used in the book can be found at www.springer-ny.com. However, the book canequally well be used with any other statistical package. Availability of statistical soRwarehas also helped shape the book in that we can discuss more complicated analyses——theanalysis of unbalanced designs, for example.
好的,這是一份關於《實驗設計與分析》一書的圖書簡介,內容詳盡,力求展現其學術深度與實用價值,同時避免提及AI生成。 --- 《實驗設計與分析》(Design and Analysis of Experiments) 一部全麵而深入的統計學與方法論專著 本書《實驗設計與分析》是一部旨在為讀者提供堅實理論基礎和精湛實踐技能的權威著作。它係統地梳理瞭從經典到前沿的實驗設計原理、方法論和統計分析技術。全書內容豐富,結構嚴謹,不僅適用於統計學、工程學、生命科學、醫學、心理學及社會科學等領域的學生和研究人員,也是緻力於提升數據驅動決策能力的專業人士不可或缺的參考指南。 核心理念:從問題到洞察的係統路徑 本書的核心目標是指導讀者如何科學、高效地組織研究,確保實驗結果的有效性、可靠性和可推廣性。我們深知,一個精心設計的實驗是得齣可信結論的前提,而恰當的分析則是解鎖數據深層含義的關鍵。因此,本書的編排緊密圍繞“設計—實施—分析—解釋”這一完整的科學研究循環展開。 第一部分:實驗設計的基石與原理 本部分深入探討瞭實驗設計的基本概念和核心原則。我們將從統計推斷的本質齣發,闡述隨機化、重復和局部控製(或區組化)這三大支柱在構建有效實驗中的作用。 統計基礎迴顧: 針對需要快速迴顧或建立基礎的讀者,本章提供瞭必要的概率論、隨機變量分布(如正態分布、泊鬆分布、二項分布)以及參數估計、假設檢驗等核心統計學知識的綜述。重點強調統計功效(Power)和顯著性水平的選擇如何影響實驗的決策質量。 實驗的要素與術語: 詳細界定瞭因子(Factors)、水平(Levels)、響應變量(Responses)、處理組閤(Treatments)以及實驗單元(Experimental Units)等關鍵術語,確保讀者對實驗構建的元素有清晰的認知。 完全隨機化設計(CRD): 作為最基礎的模型,CRD的適用場景、模型構建、方差分析(ANOVA)的原理及其實施步驟進行瞭詳盡的闡述,並探討瞭其局限性。 第二部分:經典實驗模型與高級布局 本部分是本書的重點,係統介紹瞭在不同研究情境下最常用且功能強大的實驗設計模型。 隨機化區組設計(RBD): 詳細解釋瞭如何通過區組化來處理已知的或預期的異質性來源,提高實驗的精確度。書中包含瞭如何選擇區組大小、如何進行模型擬閤和殘差分析的實例。 拉丁方設計(LSD): 探討瞭在需要同時控製兩個異質性源(行與列)的場景中LSD的應用。同時,本書深入分析瞭LSD在存在缺失數據或非正交模型時的處理方法,強調瞭其在農業試驗和工業質量控製中的價值。 析因設計(Factorial Designs): 這是理解多因子交互作用的基石。本書詳盡地展示瞭$2^k$、$3^k$以及混閤因子設計的構建、分析和解釋。特彆強調瞭交互作用的含義及其對主效應解釋的影響,並介紹瞭部分析因設計(Fractional Factorial Designs)在篩選大量因子時的效率優勢。 嵌套設計與重復測量設計(Nested and Repeated Measures Designs): 針對復雜的分層結構數據(如多級抽樣或同一受試者在不同時間點的測量),本書介紹瞭如何應用混閤效應模型來正確處理組內相關性,避免得齣誤導性的標準誤估計。 第三部分:應對復雜性和非標準情況 現代研究往往需要處理更復雜的實驗結構和數據特徵。本部分聚焦於應對這些挑戰的專業技術。 不完全區組設計(Incomplete Block Designs, IBD): 針對處理數過多而無法在一個區組內包含所有處理的場景,本書詳細介紹瞭平衡不完全區組設計(BIBD)和部分平衡不完全區組設計(PBIBD)的構造、效率評估和數據分析,特彆是如何利用損失函數最小化原則。 響應麯麵法(Response Surface Methodology, RSM): 專為工藝優化和係統探索而設計。本書從中心復閤設計(CCD)和Box-Behnken設計入手,詳細闡述瞭如何通過二次多項式模型擬閤,找到最優的操作條件,實現精細化控製。 交叉設計(Crossover Designs): 在醫學和藥理學領域極為重要。本書全麵覆蓋瞭基本的兩周期交叉設計,並擴展到多周期和多序列交叉設計,重點講解瞭如何處理序列效應(Carryover Effects)和如何進行必要的正交性檢查。 第四部分:統計分析與模型診斷 實驗設計方法的有效性最終依賴於穩健的統計分析。本書將統計軟件的應用與理論推導緊密結閤。 方差分析(ANOVA)的深入探討: 不僅停留在單因素和多因素ANOVA的錶層,還深入講解瞭效應模型的建立(固定效應與隨機效應)、模型選擇的標準(如AIC/BIC)以及如何進行事後檢驗(Post-hoc Tests,如Tukey, Dunnett, Bonferroni)。 模型假設的檢驗與修正: 強調瞭ANOVA模型的四大基本假設(正態性、方差齊性、獨立性)。本書詳細指導讀者如何通過殘差圖進行診斷,以及在假設不滿足時如何選擇替代方法,如數據變換(Transformation)或非參數檢驗。 迴歸分析在實驗中的應用: 展示瞭如何將方差分析框架轉化為綫性模型迴歸的視角,特彆是對於協方差分析(ANCOVA),解釋瞭協變量(Covariates)如何幫助提高實驗精度。 總結與展望 《實驗設計與分析》力求成為一本既具有深厚理論底蘊,又極具操作指南價值的教材。它不僅僅是一本統計公式的匯編,更是一部指導研究人員如何“像科學傢一樣思考”的指南。通過對大量真實世界案例的剖析,讀者將學會如何根據研究目標和資源限製,量身定製最優的實驗方案,並以最可靠的統計方法解析數據,從而確保研究發現的科學嚴謹性和影響力。 本書的結構設計,旨在引導讀者逐步建立起一個全麵的實驗科學思維體係,最終目標是賦能讀者在任何領域內,都能設計齣經得起推敲、足以支持決策的有效實驗。