Atnaujinkite slapukų nuostatas

El. knyga: Trustworthy Online Controlled Experiments: A Practical Guide to A/B Testing

4.39/5 (1095 ratings by Goodreads)
  • Formatas: PDF+DRM
  • Išleidimo metai: 02-Apr-2020
  • Leidėjas: Cambridge University Press
  • Kalba: eng
  • ISBN-13: 9781108601375
  • Formatas: PDF+DRM
  • Išleidimo metai: 02-Apr-2020
  • Leidėjas: Cambridge University Press
  • Kalba: eng
  • ISBN-13: 9781108601375

DRM apribojimai

  • Kopijuoti:

    neleidžiama

  • Spausdinti:

    neleidžiama

  • El. knygos naudojimas:

    Skaitmeninių teisių valdymas (DRM)
    Leidykla pateikė šią knygą šifruota forma, o tai reiškia, kad norint ją atrakinti ir perskaityti reikia įdiegti nemokamą programinę įrangą. Norint skaityti šią el. knygą, turite susikurti Adobe ID . Daugiau informacijos  čia. El. knygą galima atsisiųsti į 6 įrenginius (vienas vartotojas su tuo pačiu Adobe ID).

    Reikalinga programinė įranga
    Norint skaityti šią el. knygą mobiliajame įrenginyje (telefone ar planšetiniame kompiuteryje), turite įdiegti šią nemokamą programėlę: PocketBook Reader (iOS / Android)

    Norint skaityti šią el. knygą asmeniniame arba „Mac“ kompiuteryje, Jums reikalinga  Adobe Digital Editions “ (tai nemokama programa, specialiai sukurta el. knygoms. Tai nėra tas pats, kas „Adobe Reader“, kurią tikriausiai jau turite savo kompiuteryje.)

    Negalite skaityti šios el. knygos naudodami „Amazon Kindle“.

Getting numbers is easy; getting numbers you can trust is hard. This practical guide by experimentation leaders at Google, LinkedIn, and Microsoft will teach you how to accelerate innovation using trustworthy online controlled experiments, or A/B tests. Based on practical experiences at companies that each run more than 20,000 controlled experiments a year, the authors share examples, pitfalls, and advice for students and industry professionals getting started with experiments, plus deeper dives into advanced topics for practitioners who want to improve the way they make data-driven decisions. Learn how to • Use the scientific method to evaluate hypotheses using controlled experiments • Define key metrics and ideally an Overall Evaluation Criterion • Test for trustworthiness of the results and alert experimenters to violated assumptions • Build a scalable platform that lowers the marginal cost of experiments close to zero • Avoid pitfalls like carryover effects and Twyman's law • Understand how statistical issues play out in practice.

Getting numbers is easy; getting trustworthy numbers is hard. From experimentation leaders at Amazon, Google, LinkedIn, and Microsoft, this guide to accelerating innovation using A/B tests includes practical examples, pitfalls, and advice for students and industry professionals, plus deeper dives into advanced topics for experienced practitioners.

Recenzijos

'At the core of the Lean Methodology is the scientific method: Creating hypotheses, running experiments, gathering data, extracting insight and validation or modification of the hypothesis. A/B testing is the gold standard of creating verifiable and repeatable experiments, and this book is its definitive text.' Steve Blank, Adjunct professor at Stanford University, father of modern entrepreneurship, author of The Startup Owner's Manual and The Four Steps to the Epiphany 'This book is a great resource for executives, leaders, researchers or engineers looking to use online controlled experiments to optimize product features, project efficiency or revenue. I know firsthand the impact that Kohavi's work had on Bing and Microsoft, and I'm excited that these learnings can now reach a wider audience.' Harry Shum, EVP, Microsoft Artificial Intelligence and Research Group 'A great book that is both rigorous and accessible. Readers will learn how to bring trustworthy controlled experiments, which have revolutionized internet product development, to their organizations.' Adam D'Angelo, Co-founder and CEO of Quora and former CTO of Facebook 'This book is a great overview of how several companies use online experimentation and A/B testing to improve their products. Kohavi, Tang and Xu have a wealth of experience and excellent advice to convey, so the book has lots of practical real world examples and lessons learned over many years of the application of these techniques at scale.' Jeff Dean, Google Senior Fellow and SVP Google Research 'Do you want your organization to make consistently better decisions? This is the new bible of how to get from data to decisions in the digital age. Reading this book is like sitting in meetings inside Amazon, Google, LinkedIn, Microsoft. The authors expose for the first time the way the world's most successful companies make decisions. Beyond the admonitions and anecdotes of normal business books, this book shows what to do and how to do it well. It's the how-to manual for decision-making in the digital world, with dedicated sections for business leaders, engineers, and data analysts.' Scott Cook, Intuit Co-founder & Chairman of the Executive Committee 'Online controlled experiments are powerful tools. Understanding how they work, what their strengths are, and how they can be optimized can illuminate both specialists and a wider audience. This book is the rare combination of technically authoritative, enjoyable to read, and dealing with highly important matters.' John P. A. Ioannidis, Stanford University 'Kohavi, Tang, and Xu are pioneers of online experimentation. The platforms they've built and the experiments they've enabled have transformed some of the largest internet brands. Their research and talks have inspired teams across the industry to adopt experimentation. This book is the authoritative yet practical text that the industry has been waiting for.' Adil Aijaz, Co-founder and CEO, Split Software 'Which online option will be better? We frequently need to make such choices, and frequently err. To determine what will actually work better, we need rigorous controlled experiments, aka A/B testing. This excellent and lively book by experts from Microsoft, Google, and LinkedIn presents the theory and best practices of A/B testing. A must read for anyone who does anything online!' Gregory Piatetsky-Shapiro, Ph.D., president of KDnuggets, co-founder of SIGKDD, and LinkedIn Top Voice on Data Science & Analytics 'Ron Kohavi, Diane Tang and Ya Xu are the world's top experts on online experiments. I've been using their work for years and I'm delighted they have now teamed up to write the definitive guide. I recommend this book to all my students and everyone involved in online products and services.' Erik Brynjolfsson, Massachusetts Institute of Technology, co-author of The Second Machine Age 'A modern software-supported business cannot compete successfully without online controlled experimentation. Written by three of the most experienced leaders in the field, this book presents the fundamental principles, illustrates them with compelling examples, and digs deeper to present a wealth of practical advice. It's a 'must read'! Foster Provost, New York University and co-author of the best-selling Data Science for Business 'In the past two decades the technology industry has learned what scientists have known for centuries: that controlled experiments are among the best tools to understand complex phenomena and to solve very challenging problems. The ability to design controlled experiments, run them at scale, and interpret their results is the foundation of how modern high tech businesses operate. Between them the authors have designed and implemented several of the world's most powerful experimentation platforms. This book is a great opportunity to learn from their experiences about how to use these tools and techniques.' Kevin Scott, EVP and CTO of Microsoft 'Online experiments have fueled the success of Amazon, Microsoft, LinkedIn and other leading digital companies. This practical book gives the reader rare access to decades of experimentation experience at these companies and should be on the bookshelf of every data scientist, software engineer and product manager.' Stefan Thomke, William Barclay Harding Professor, Harvard Business School, author of Experimentation Works: The Surprising Power of Business Experiments 'The secret sauce for a successful online business is experimentation. But it is a secret no longer. Here three masters of the art describe the ABCs of A/B testing so that you too can continuously improve your online services.' Hal Varian, Chief Economist, Google, and author of Intermediate Microeconomics: A Modern Approach 'Experiments are the best tool for online products and services. This book is full of practical knowledge derived from years of successful testing at Microsoft Google and LinkedIn. Insights and best practices are explained with real examples and pitfalls, their markers and solutions identified. I strongly recommend this book!' Preston McAfee, former Chief Economist and VP of Microsoft 'Experimentation is the future of digital strategy and 'Trustworthy Experiments' will be its Bible. Kohavi, Tang and Xu are three of the most noteworthy experts on experimentation working today and their book delivers a truly practical roadmap for digital experimentation that is useful right out of the box. The revealing case studies they conducted over many decades at Microsoft, Amazon, Google and LinkedIn are organized into easy to understand practical lessens with tremendous depth and clarity. It should be required reading for any manager of a digital business.' Sinan Aral, David Austin Professor of Management, Massachusetts Institute of Technology, and author of The Hype Machine

Daugiau informacijos

This practical guide for students, researchers and practitioners offers real world guidance for data-driven decision making and innovation.
Preface - How to Read This Book xv
Acknowledgments xvii
PART I INTRODUCTORY TOPICS FOR EVERYONE
1(78)
1 Introduction and Motivation
3(23)
Online Controlled Experiments Terminology
5(3)
Why Experiment? Correlations, Causality, and Trustworthiness
8(2)
Necessary Ingredients for Running Useful Controlled Experiments
10(1)
Tenets
11(3)
Improvements over Time
14(2)
Examples of Interesting Online Controlled Experiments
16(4)
Strategy, Tactics, and Their Relationship to Experiments
20(4)
Additional Reading
24(2)
2 Running and Analyzing Experiments: An End-to-End Example
26(13)
Setting up the Example
26(3)
Hypothesis Testing: Establishing Statistical Significance
29(3)
Designing the Experiment
32(2)
Running the Experiment and Getting Data
34(1)
Interpreting the Results
34(2)
From Results to Decisions
36(3)
3 Twyman's Law and Experimentation Trustworthiness
39(19)
Misinterpretation of the Statistical Results
40(3)
Confidence Intervals
43(1)
Threats to Internal Validity
43(5)
Threats to External Validity
48(4)
Segment Differences
52(3)
Simpson's Paradox
55(2)
Encourage Healthy Skepticism
57(1)
4 Experimentation Platform and Culture
58(21)
Experimentation Maturity Models
58(8)
Infrastructure and Tools
66(13)
PART II SELECTED TOPICS FOR EVERYONE
79(46)
5 Speed Matters: An End-to-End Case Study
81(9)
Key Assumption: Local Linear Approximation
83(1)
How to Measure Website Performance
84(2)
The Slowdown Experiment Design
86(1)
Impact of Different Page Elements Differs
87(2)
Extreme Results
89(1)
6 Organizational Metrics
90(12)
Metrics Taxonomy
90(4)
Formulating Metrics: Principles and Techniques
94(2)
Evaluating Metrics
96(1)
Evolving Metrics
97(1)
Additional Resources
98(1)
SIDEBAR: Guardrail Metrics
98(2)
SIDEBAR: Gameability
100(2)
7 Metrics for Experimentation and the Overall Evaluation Criterion
102(9)
From Business Metrics to Metrics Appropriate for Experimentation
102(2)
Combining Key Metrics into an OEC
104(2)
Example: OEC for E-mail at Amazon
106(2)
Example: OEC for Bing's Search Engine
108(1)
Goodhart's Law, Campbell's Law, and the Lucas Critique
109(2)
8 Institutional Memory and Meta-Analysis
111(5)
What Is Institutional Memory?
111(1)
Why Is Institutional Memory Useful?
112(4)
9 Ethics in Controlled Experiments
116(9)
Background
116(5)
Data Collection
121(1)
Culture and Processes
122(1)
Sidebar: User Identifiers
123(2)
PART III COMPLEMENTARY AND ALTERNATIVE TECHNIQUES TO CONTROLLED EXPERIMENTS
125(26)
10 Complementary Techniques
127(10)
The Space of Complementary Techniques
127(1)
Logs-based Analysis
128(2)
Human Evaluation
130(1)
User Experience Research (UER)
131(1)
Focus Groups
132(1)
Surveys
132(1)
External Data
133(2)
Putting It All Together
135(2)
11 Observational Causal Studies
137(14)
When Controlled Experiments Are Not Possible
137(2)
Designs for Observational Causal Studies
139(5)
Pitfalls
144(3)
SIDEBAR: Refuted Observational Causal Studies
147(4)
PART IV ADVANCED TOPICS FOR BUILDING AN EXPERIMENTATION PLATFORM
151(32)
12 Client-Side Experiments
153(9)
Differences between Server and Client Side
153(3)
Implications for Experiments
156(5)
Conclusions
161(1)
13 Instrumentation
162(4)
Client-Side vs. Server-Side Instrumentation
162(2)
Processing Logs from Multiple Sources
164(1)
Culture of Instrumentation
165(1)
14 Choosing a Randomization Unit
166(5)
Randomization Unit and Analysis Unit
168(1)
User-level Randomization
169(2)
15 Ramping Experiment Exposure: Trading Off Speed
Quality, and Risk
171(1)
What Is Ramping?
171(1)
SQR Ramping Framework
172(1)
Four Ramp Phases
173(3)
Post Final Ramp
176(1)
16 Scaling Experiment Analyses
177(6)
Data Processing
177(1)
Data Computation
178(2)
Results Summary and Visualization
180(3)
PART V ADVANCED TOPICS FOR ANALYZING EXPERIMENTS
183(63)
17 The Statistics behind Online Controlled Experiments
185(8)
Two-Sample t-Test
185(1)
p-Value and Confidence Interval
186(1)
Normality Assumption
187(2)
Type P/II Errors and Power
189(2)
Bias
191(1)
Multiple Testing
191(1)
Fisher's Meta-analysis
192(1)
18 Variance Estimation and Improved Sensitivity: Pitfalls and Solutions
193(7)
Common Pitfalls
193(3)
Improving Sensitivity
196(2)
Variance of Other Statistics
198(2)
19 The A/A Test
200(9)
Why A/A Tests?
200(5)
How to Run A/A Tests
205(2)
When the A/A Test Fails
207(2)
20 Triggering for Improved Sensitivity
209(10)
Examples of Triggering
209(3)
A Numerical Example (Kohavi, Longbotham et al. 2009)
212(1)
Optimal and Conservative Triggering
213(1)
Overall Treatment Effect
214(1)
Trustworthy Triggering
215(1)
Common Pitfalls
216(1)
Open Questions
217(2)
21 Sample Ratio Mismatch and Other Trust-Related Guardrail Metrics
219(7)
Sample Ratio Mismatch
219(3)
Debugging SRMs
222(4)
22 Leakage and Interference between Variants
226(9)
Examples
227(3)
Some Practical Solutions
230(4)
Detecting and Monitoring Interference
234(1)
23 Measuring Long-Term Treatment Effects
235(11)
What Are Long-Term Effects?
235(1)
Reasons the Treatment Effect May Differ between Short-Term and Long-Term
236(2)
Why Measure Long-Term Effects?
238(1)
Long-Running Experiments
239(2)
Alternative Methods for Long-Running Experiments
241(5)
References 246(20)
Index 266
Ron Kohavi is a Technical Fellow and corporate VP of Microsoft's Analysis and Experimentation, and was previously director of data mining and personalization at Amazon. He received his Ph.D. in Computer Science from Stanford University. His papers have over 40,000 citations and three of them are in the top 1,000 most-cited papers in Computer Science. Diane Tang is a Google Fellow, with expertise in large-scale data analysis and infrastructure, online controlled experiments, and ads systems. She has an A.B. from Harvard and an M.S./Ph.D. from Stanford University, with patents and publications in mobile networking, information visualization, experiment methodology, data infrastructure, data mining, and large data. Ya Xu heads Data Science and Experimentation at LinkedIn. She has published several papers on experimentation and is a frequent speaker at top-tier conferences and universities. She previously worked at Microsoft and received her Ph.D. in Statistics from Stanford University.