掌握统计基础与SPSS实践：管理与科学应用详解

需积分: 10 68 浏览量更新于2024-07-25 收藏 973KB PDF 举报

本资源专注于统计初级课程，旨在引导学习者理解和掌握基础的统计概念以及如何运用SPSS进行数据分析。内容覆盖了实验设计在管理、工程和科学领域的实际应用，强调了统计在决策制定过程中的关键作用。首先，书籍《ExperimentalDesign with Applications in Management, Engineering, and the Sciences》探讨了实验设计的方法，这对于理解不同领域的研究设计和优化实践至关重要。通过这本书，读者可以了解到如何设计有效实验，以获取可靠的数据并推动业务改进。其次，Albright和Zappe的作品《Data Analysis & Decision Making with Microsoft Excel》和《Managerial Statistics》展示了如何结合统计知识与电子表格软件，实现数据处理和直观分析，提升日常决策的精确度。这有助于非专业背景的人员也能有效地利用技术工具进行统计工作。 Berk和Carey的《Data Analysis with Microsoft Excel》同样关注Excel在统计分析中的应用，强调了软件在实际商业环境中的实用价值。此外，还有《Data Analysis in Plain English》由Brightman撰写，特别注重以易懂的语言阐述复杂的统计概念，便于初学者快速入门。对于更深入的统计理论，如概率论和回归分析，Devore的《Probability & Statistics for Engineering and the Sciences》提供了第五版的内容，适合工程和科学专业人士研读。Dielman的《Applied Regression Analysis, 3rd Ed.》则专门讲解回归分析，是数据科学家和分析师的必备工具。现代质量管理方面，Farnum的《Modern Statistical Quality Control and Improvement》涵盖了质量控制和持续改进的统计方法，对制造业和服务业具有指导意义。Hildebrand和Ott的《Statistical Thinking for Managers, 4th Ed.》则将统计思维融入管理决策，帮助管理者从数据角度洞察业务运营。 Hoerl和Snee的《Statistical Thinking: Improving Business Performance》深入剖析统计如何提升企业的经营效率，对于提升企业竞争力有实际指导价值。Johnson的《Applied Multivariate Methods for Data Analysts》则着重于多元统计方法，适用于数据分析师解决复杂问题。最后，涉及其他教材如《Statistics for Management & Economics, 5th Ed.》和《Modern Industrial Statistics》等，它们针对管理经济学和工业领域的统计需求，提供了全面且针对性的学习材料。这个资源集合涵盖了从基础实验设计到高级统计分析的各个方面，适合希望系统学习统计知识，或寻求在实际工作中应用统计技能的专业人士和学生。通过这些教材，学习者不仅能掌握统计基础知识，还能掌握实用的统计分析工具，以提升工作效率和决策质量。

. P

AUL

D. B

ERGER

, Professor of Marketing and

Quantitative Methods at the School of

Management, Boston University, has been teach-

ing experimental design for more than 30 years.

He currently teaches it in the MBA and doctoral

programs at Boston University. He has also taught

at Massachusetts Institute of Technology (MIT)

every summer since the mid-1960s, working until

1985 with Professor Harold Freeman of the

Economics Department at MIT, one of the all-

time greats in the field of statistics and experimental design, who, unfortu-

nately, passed away at age 88 in early 1998. He has also conducted in-house

courses at a large number of companies. He has consulted extensively in the

field of experimental design at marketing research firms, manufacturing

companies, and services providers. He has also published a large number of

journal articles on statistics, experimental design, and mathematical model-

ing. He coauthored a textbook in the field of direct marketing (Direct

Marketing Management, with Mary Lou Roberts, first edition 1989, second

edition 1999) and a case book illustrating statistical applications (Cases in

Business Statistics, with Ronald Klimberg and Peter Arnold, 1994). Dr. Berger

also won the university-wide Metcalf prize for teaching excellence at Boston

University. He has his bachelor of science, master of science, and doctoral de-

grees from the Sloan School of Management, MIT.

. R

OBERT

E. M

AURER

has more than 35 years of

industrial experience at Bell Telephone Labora-

tories. In his last assignment, he was responsible for

process and product design and manufacture of a

several-hundred-million-dollar product line of hy-

brid integrated circuits. Through his initiative and

guidance, the disciplines of statistical process con-

trol (SPC) and experimental design were deployed

throughout his organization, leading to improved

quality and reduced cost. Dr. Maurer has more

than 20 years of experience teaching in the areas of statistical communication

theory at the graduate school of engineering at Northeastern University and sta-

tistical analysis at the school of management at Boston University. He has pub-

lished numerous papers and holds patents in the communications area.

Dr. Maurer earned his bachelor and master of science and doctoral degrees in

electrical engineering from Northeastern University, and an MBA from Boston

University.

About the Authors

xvi

Frequently we have the latitude to select the levels of the factors under

study in advance of data collection; in these instances, we realize advantages

that increase the efficiency of the experimental effort. Sometimes, however,

we cannot specify the levels of the independent variables—we have to take

what we’re given. For example, if we want to study the impact of corporate div-

idends on stock price, we generally cannot manipulate the companies’ divi-

dends; they would be what they would be. Such situations can also arise sim-

ply because the data are already collected, and an analysis can be done only

ex post facto (that is, after the fact, or after the data are collected). However,

in both situations, it might be possible to sort the data to find a subset of com-

panies with the exact levels of dividends one would wish to choose.

1.2

The Growth in Experimental Design

Experimental design is a growing area of interest in an increasing number of

applications. Initially, experimental design found application in agriculture,

biology, and other areas of hard science. It has since spread through the en-

gineering arenas to the social sciences of economics and behavioral analysis.

Experimental design appears to have been used in traditional business and

management applications only since the mid-1960s; more recently, experi-

mental-design methodology has been applied in the nonprofit and govern-

ment sectors. There are many reasons for this progression, but the principal

one is the increased training in statistics and quantitative methods among pro-

fessionals in management and the latter areas, along with the resultant wide-

spread use of quantitative decision-making techniques.

This trend was further encouraged by the Total Quality Management

(TQM) movement originating in the mid-1980s and continuing today.

Indicative of the widespread acceptance of the virtues of experimental design

was its being heralded by the esteemed “establishment” outlet Forbes magazine

in a March 11, 1996, article entitled “The New Mantra: MVT.” MVT stands for

multivariable testing, a term the article uses along with experimental design,

using the term MVT to distinguish factorial designs from the vilified (both in

the article and this text) one-factor-at-a-time experiments.

That the trend continues is attested to by a more recent article, “The

Numbers Game: Multivariate Testing Scores Big,” in the Executive Edge sec-

tion of the April 2000 edition of Continental, the in-flight magazine of

Continental Airlines. The article cites many companies that rely more and

more on experimental design (including DuPont, American Express, Boise

Cascade, Circuit City, and SBC Communications), and reports in detail on its

use by the Deluxe Corporation.

Behind all the praise of experimental-design methods is the simple fact

that it works. The two articles note many successes, and Forbes briefly described

the history of the discipline. Experimental design is not, strictly speaking, a

new field. It originated at the beginning of the 20th century with Sir Ronald

2 CHAPTER 1 INTRODUCTION TO EXPERIMENTAL DESIGN

Fisher’s work; indeed, Fisher is often referred to as “the father of experimen-

tal design.”

The field is, in many ways, a simple one. The terminology and notation

may appear strange to the uninitiated, and the mathematical connections may

seem formidable, but as for any worthwhile new skill, one can overcome the

barriers to entry with study and practice. With mastery one can admire the in-

herent beauty of the subject and appreciate how its success is enhanced by

combining the discipline of statistics with the knowledge of process experts.

1.3

The Six Steps of Experimental Design

One can frame the experimental-design process as a six-step process, as seen

in Figure 1.1.

Plan the Experiment

The planning process is vital to the success of the experiment. It is at this stage

that close collaboration between the people knowledgeable about the process

under study (the process experts) and those experienced in the design-of-

experiments methodology (the designers) is required. Diligence at this stage can

greatly increase the chance that appropriate, meaningful assumptions are made;

giving short shrift to this stage of the process can make everything that follows a

waste of time and resources. The planning stage itself consists of five steps:

1. Identify the dependent, or output, variable(s).

2. Translate output variables to measurable quantities.

3. Determine the factors (input, or independent, variables) that poten-

tially affect the output variables that are to be studied.

4. Determine the number of levels or values for each factor and what

those levels are.

5. Identify potential synergy between different factors.

The dependent, or output, variable to be studied should be chosen carefully.

Sometimes the choice is obvious, but other times it is not. The general guide-

line is to ensure that the right performance measure is chosen—typically, the

quantity that is really the important one; perhaps the one that tells how good

a product is, or how well it does its job, or how it’s received in the market-

place. For example, in an experiment to help develop a better cover for a spe-

cialty catalog, it’s unlikely that a wise choice of dependent variable is the

brightness of the cover; a wiser choice is the sales volume that results, or even

a measure of the potential customer’s attitude toward the cover.

One way to choose a dependent variable is to make sure that all vested in-

terests are represented—those of the user, producer, marketing department,

and relevant others—perhaps in a formal brainstorming session. If the goal is

CHAPTER 1 INTRODUCTION TO EXPERIMENTAL DESIGN 3

to produce popcorn that sells more, is it obvious to the consumer what the key

qualities should be? Fluffiness? Color? Taste? Texture? Percent popped?

Once the dependent variable is selected, it must usually be transformed

into a quantitative measure to make it useful. Many variables are subjective.

How do you measure taste or appearance? What about the units of measure-

ment? The choice of units is almost never important if those under consider-

ation are linearly related to one another (such as inches versus feet, or dollars

versus yen). However, how about the choice between circumference, area,

and volume to measure the size of a spherical product? A known value for one

measure determines the values of the other two measures, but if the value of

one measure changes, the values of the other measures do not change pro-

portionally. So although one measure may vary linearly with the level of a fac-

tor, another measure may vary nonlinearly. As we note later, varying linearly

versus nonlinearly may be an important issue.

Choosing the factors to study is sometimes quite straightforward; however,

other times it is not as easy as it might seem. In our consulting experience, the

process experts sometimes propose an unworkably large number of possible fac-

tors for study. An effective way to raise for consideration all candidate input vari-

ables is to begin with a formal brainstorming session, perhaps using Ishikawa (of-

ten called Fishbone or cause-and-effect) diagrams, a technique developed in the

context of quality control to help identify factors affecting product quality, but

adaptable to any situation where one desires to identify factors that affect a de-

pendent variable. This approach generally yields a nearly exhaustive list of choices

in categories such as people factors, equipment factors, environmental factors,

methods factors, and materials factors. This initial list may be pared down to its

essentials through a Pareto analysis, which in this context classically states the now

well-known concept that 80% of the identifiable impact will be associated with

20% of the factors, though the exact values of 20 and 80 are unlikely to be pre-

cisely realized. Of course, one should reduce the list in any obvious ways first. We

recall one case in which (in simplistic terms) the factor “temperature greater than

212°F or not” was identified along with another factor, “presence or absence of

steam” (with the experiment run at sea level in an unpressurized container).

One is usually motivated to minimize, to the degree possible, the number

of factors under study. In general, everything else being equal, a higher num-

4 CHAPTER 1 INTRODUCTION TO EXPERIMENTAL DESIGN

1. Plan the experiment.

2. Design the experiment.

3. Perform the experiment.

4. Analyze the data from the experiment.

5. Confirm the results of the experiment.

6. Evaluate the conclusions of the experiment.

Figure 1.1 The

process of experimental

design

剩余239页未读，继续阅读

FightingSW

粉丝: 0
资源: 1

掌握统计基础与SPSS实践：管理与科学应用详解

Statistics

computational statistics

Nonparametric Statistics

spssstatistics

dso statistics

statistics 模块

statistics python

statistics toolbox

@statistics

statistics下载

最新资源