Introductory Statistics with Randomization and Simulation (ISRS) is a free statistics textbook from the OpenIntro project, written by David M. Diez, Christopher D. Barr and Mine Çetinkaya-Rundel. It is a first-course text for BS Statistics and BSCS students, and it teaches inference in an unusual order: you first shuffle cards and run simulations to see what chance alone can produce, and only then meet the normal and t distributions as shortcuts for the same idea. Please note the age of this book. It is the First Edition, copyright 2014. OpenIntro’s own book page says that its newer Introduction to Modern Statistics represents the evolution of this book, recommends the newer book to anyone adopting a text for a new course, and now de-emphasises this one in its catalogue. The older book is still free and still readable, but the newer one is the better starting point.
If you are a BS Statistics student, ISRS suits semester-exam preparation well. Each method comes with one real data question, a short list of conditions to check and a worked answer, and the 231 end-of-chapter exercises have answers to the odd-numbered ones at the back. If you are a BSCS student taking a statistics course, the simulation-first order will feel familiar: you watch a null distribution being built from repeated random trials before you are asked to trust a formula. That makes revision easier, because you can explain why a p-value is small, not only read it off a table.
Book Overview
| Course | Introductory Statistics (randomization and simulation first) |
| Degree Programs | BS Statistics · BSCS (as a statistics requirement) |
| Level | Undergraduate, introductory (no prior statistics, no calculus) |
| Edition | First Edition, copyright 2014. An older edition: OpenIntro recommends its newer Introduction to Modern Statistics for new courses |
| Author(s) | David M. Diez, Christopher D. Barr, Mine Çetinkaya-Rundel |
| Publisher | OpenIntro |
| Structure | 6 chapters and 3 appendices (A Probability, B End of chapter exercise solutions, C Distribution tables) |
| Exercises | 231 numbered end-of-chapter exercises; solutions to the odd-numbered ones in Appendix B; guided-practice questions in the text have answers in footnotes |
| Length | 354 pages (PDF) |
| Language | English |
| License | Creative Commons BY-SA 3.0 according to OpenIntro’s licence page (the PDF’s own copyright page says only that the book is available under a Creative Commons license) |
| Format | Free PDF from the official page (optional contribution; set the price to $0 to skip it); paperback sold separately |
Chapter List
Chapter 1: Introduction to Data
Difficulty: Easy · Key topics: stent experiment, data matrix, variable types, sampling, experiments, histograms, box plots, contingency tables
The chapter opens with a real randomized experiment: 451 patients at risk of stroke were split into a stent group (224) and a control group (227), and the stent group did worse, with about 20% having a stroke within a year against 12% for the controls. That result raises the question the whole book answers: is a difference real or just chance? The chapter then names the basics (cases, variables, data matrices), separates anecdotes from samples and observational studies from experiments, and covers sampling methods and the principles of experimental design. The second half teaches summaries: scatterplots, dot plots, histograms, mean, standard deviation, box plots, robust statistics, and contingency tables for categorical data.
Key Points:
- A data matrix has one case per row and one variable per column; variables are numerical (discrete or continuous) or categorical, and the book treats ordinal variables as categorical.
- Anecdotal evidence and convenience samples mislead; a simple random sample gives every case in the population an equal chance of being picked.
- A confounding variable is related to both the explanatory and the response variable, so an observational study can show association but not cause.
- Good experiments use control, randomization, replication and blocking; blinding with a placebo reduces bias in human studies.
- The median and IQR are robust to outliers, while the mean and standard deviation can swing a lot when one extreme value is added.
- A contingency table counts two categorical variables together; row or column proportions show whether they look associated.
Memory Tip: Random sampling decides who gets into the study; random assignment decides who gets the treatment. The first lets you generalize to the population, the second lets you talk about cause.
Common Mistake: Concluding that sunscreen causes skin cancer because heavy sunscreen users get more of it; the missing variable is sun exposure, which drives both.
Important Questions:
- Why can an experiment support a causal claim when an observational study cannot? Random assignment spreads other variables evenly across the groups, so confounding variables cannot explain a difference in outcomes.
- Which summary statistics are robust to outliers? The median and the IQR. The mean and standard deviation change a lot when an extreme value is added, as the chapter shows with one very long email.
Chapter 2: Foundation for Inference
Difficulty: Medium · Key topics: randomization, hypothesis testing, p-value, Type 1 and Type 2 errors, simulation, Central Limit Theorem, normal distribution, confidence intervals
This is the heart of the book, and it builds the logic of inference from simulation before any formula. A bank promotion study (21 of 24 male files recommended for promotion against 14 of 24 female files) and an opportunity-cost experiment with 150 students are analyzed by shuffling cards and repeating the shuffle many times. The chapter then names the parts: null and alternative hypotheses, p-value, significance level, Type 1 and Type 2 errors, one-sided and two-sided tests. Two more simulations (3 complications in 62 surgeries, 3 correct guesses in 120 tapping games) lead to the Central Limit Theorem, the normal model, Z scores, standard error and 95% confidence intervals.
Key Points:
- The p-value is the probability of seeing data at least as favourable to the alternative as the observed data, if the null hypothesis were true.
- A result is statistically significant when the p-value is below the significance level, usually 0.05; the level can be set lower when a Type 1 Error is costly.
- A Type 1 Error rejects a true null hypothesis and a Type 2 Error misses a false one; lowering one rate usually raises the other.
- Failing to reject the null hypothesis is not proof that it is true, just as a not-guilty verdict is not proof of innocence.
- A Z score is (observation – mean) / SD, and about 68%, 95% and 99.7% of a normal distribution lies within 1, 2 and 3 SDs of the mean.
- A 95% confidence interval is the point estimate ± 1.96 × SE, where the standard error measures how much the estimate varies from sample to sample.
- The normal model for a sample proportion needs independent observations and a large enough sample.
Memory Tip: Think of a courtroom. The null hypothesis is the presumption of innocence, the p-value is how surprising the evidence is under that presumption, and a Type 1 Error is convicting an innocent person.
Common Mistake: Saying there is a 95% probability that the parameter is inside the interval you just computed. The correct wording is that you are 95% confident the interval captures it, and about 1 in 20 such intervals misses.
Important Questions:
- What is a p-value? The probability of getting data at least as extreme as ours, in the direction of the alternative, if the null hypothesis were true. A small p-value is evidence against the null.
- What is the difference between a Type 1 and a Type 2 Error? A Type 1 Error rejects the null hypothesis when it is true; a Type 2 Error fails to reject it when the alternative is true.
Chapter 3: Inference for Categorical Data
Difficulty: Medium · Key topics: single proportion, difference of two proportions, success-failure condition, sample size, chi-square goodness of fit, two-way tables
Chapter 3 applies the normal model to proportions. For a single proportion it checks independence and the success-failure condition, builds an interval from a poll of 976 adults on Supreme Court approval, runs a test against a null value, and works out the sample size needed for a chosen margin of error. It then compares two proportions, including a different standard error when the null says the two are equal, using a dog-cancer herbicide study. Two special topics follow: the chi-square goodness-of-fit test, applied to the racial makeup of 275 jurors, and the chi-square test for independence in a two-way table, applied to three search algorithms tested on 10,000 searches.
Key Points:
- The sample proportion is nearly normal when the observations are independent and at least 10 successes and 10 failures are expected.
- The standard error of a sample proportion is √(p(1 – p)/n); intervals plug in the sample proportion, while tests use the null value p0.
- For a difference p1 – p2 the two variances add; if the null says p1 = p2, a pooled proportion is used in the standard error.
- When no earlier estimate exists, p = 0.5 is the cautious choice for planning a sample size, since it gives the widest margin of error.
- The chi-square statistic adds up (observed – expected)²/expected over all bins, and goodness of fit has k – 1 degrees of freedom for k bins.
- In a two-way table the expected count is (row total × column total)/table total and df = (R – 1) × (C – 1).
Memory Tip: A chi-square test asks one thing: how far are the observed counts from the counts the null model expected, measured against the size of those expected counts? A large chi-square value means a small p-value.
Common Mistake: Checking the success-failure condition with the sample proportion in a hypothesis test. A test works under the null hypothesis, so the check uses the null proportion p0; only intervals use the sample proportion.
Important Questions:
- What is the success-failure condition? The sample must be expected to have at least 10 successes and at least 10 failures (n × p ≥ 10 and n × (1 – p) ≥ 10), so the sampling distribution of the proportion is close enough to normal.
- How many degrees of freedom does a chi-square test have for a two-way table? (R – 1) × (C – 1). For the book’s search-algorithm table with 2 outcomes and 3 algorithms, that is 1 × 2 = 2.
Chapter 4: Inference for Numerical Data
Difficulty: Medium · Key topics: t distribution, paired data, difference of two means, pooled standard deviation, ANOVA, F test, Bonferroni correction, bootstrap
Chapter 4 moves the same toolkit from proportions to means. With small samples the normal model gives way to the t distribution, with thicker tails and n – 1 degrees of freedom. It is applied to one mean (Cherry Blossom Run times), to paired data (the same 73 textbooks priced at the UCLA bookstore and on Amazon) and to the difference of two independent means (stem cells in sheep, smoking and birth weight, two versions of an exam). ANOVA then compares many means with a single F test, using baseball on-base percentages by position, followed by Bonferroni-corrected pairwise tests. The chapter ends with a bootstrap for the standard deviation.
Key Points:
- The t distribution is bell-shaped and centered at zero with thicker tails than the normal; it approaches the normal curve as the degrees of freedom grow.
- The standard error of a sample mean is s/√n, so more data or less variable data gives a more precise estimate.
- With paired observations you analyze the differences with a one-sample t procedure instead of treating the two columns as separate samples.
- For two independent means, SE = √(s1²/n1 + s2²/n2), after checking independence and near-normality in both groups.
- ANOVA tests H0 that all group means are equal using F, the variability between group means relative to the variability within groups.
- After a significant ANOVA, pairwise comparisons use the Bonferroni correction, a stricter significance level, to control the Type 1 Error rate.
- A bootstrap sample is drawn from the original data with replacement, and the SD of many bootstrap estimates approximates the standard error.
Memory Tip: Read ANOVA as signal over noise: the spread of the group means is the signal, the spread of the observations inside each group is the noise, and F is their ratio.
Common Mistake: Treating paired measurements, such as the UCLA and Amazon prices of the same book, as two independent samples. The pairing is thrown away and the wrong standard error is used.
Important Questions:
- When should you use a paired t procedure? When every observation in one data set has a natural partner in the other, such as the same book priced in two stores. Subtract within each pair and run a one-sample t test or interval on the differences.
- Why use ANOVA instead of many two-sample t tests? Many separate tests make a false positive likely somewhere by chance alone. ANOVA uses one test of the null hypothesis that all the means are equal.
Chapter 5: Introduction to Linear Regression
Difficulty: Medium · Key topics: scatterplots, residuals, correlation, least squares, R-squared, outliers, leverage, inference for the slope
Chapter 5 fits a straight line to two numerical variables. It starts with measurements of 104 brushtail possums, explains residuals and correlation, and shows when a straight line is the wrong model. Least squares then gives an objective best line, worked through with the family income and gift aid of 50 Elmhurst College freshmen: the slope is R × (sy/sx) and the line passes through the point of means. The chapter interprets slope and intercept, warns about extrapolation, explains R-squared, handles a two-level categorical predictor, and sorts outliers by leverage and influence. It closes with inference for the slope: four conditions and a t test read from software output.
Key Points:
- A residual is the observed value minus the predicted value, and least squares picks the line that minimizes the sum of the squared residuals.
- The least squares slope is b1 = R × (sy/sx), and the line always passes through (x-bar, y-bar).
- Correlation R lies between -1 and 1 and measures only the strength of a linear relationship.
- R-squared is the share of the variability in y explained by the linear model.
- Extrapolating far beyond the range of the data is an unreliable bet.
- Points far from the cloud in the x direction have high leverage and may be influential, pulling the line toward themselves.
- Inference needs linearity, nearly normal residuals, constant variability and independent observations; the slope is tested with H0: β1 = 0.
Memory Tip: A good fit leaves a shapeless cloud in the residual plot. If you can still see a curve, a fan or a trend in the residuals, one of the four conditions is broken.
Common Mistake: Predicting far outside the range of the data. A line fitted to incomes in one range says little about much richer or much poorer cases.
Important Questions:
- What does the least squares line minimize? The sum of the squared residuals, e1² + e2² + … + en², which is easier to compute and punishes large misses more than small ones.
- How is the slope read in the Elmhurst gift-aid model? The estimated slope is about -0.0431 with both variables in thousands of dollars, so each extra $1,000 of family income goes with about $43 less gift aid on average.
Chapter 6: Multiple and Logistic Regression
Difficulty: Hard · Key topics: multiple regression, indicator variables, adjusted R-squared, model selection, residual diagnostics, logistic regression, spam filter
Chapter 6 extends regression to many predictors. The running example is 141 eBay auctions of Mario Kart for the Wii, with total price modelled from game condition, stock photo, auction length and the number of Wii wheels, using 0/1 indicator variables for the yes/no predictors. It explains why adjusted R-squared replaces R-squared, how p-values flag variables that add nothing, and how backward elimination and forward selection build a smaller model. Residual plots check the assumptions. The last section introduces logistic regression, a generalized linear model for a two-level outcome, by building a spam filter from email characteristics, then discusses probability cut-offs and ways to improve the predictors.
Key Points:
- Multiple regression has the form y-hat = b0 + b1x1 + … + bkxk, and each coefficient is read with the other predictors held constant.
- An indicator variable codes a two-level category as 1 or 0, so it can enter a regression directly.
- Adjusted R-squared uses df = n – k – 1 and can fall when a useless variable is added, which plain R-squared never does.
- Backward elimination starts from the full model and drops the variable with the largest non-significant p-value, refits, and repeats; forward selection adds variables one at a time.
- Residuals are checked with a normal probability plot, a plot of absolute residuals against fitted values, and a plot of residuals in order of collection.
- Logistic regression models the probability of a 0/1 outcome through the logit transformation, loge(p/(1 – p)), of a linear combination of predictors.
Memory Tip: Say ‘all else held constant’ every time you read a coefficient out loud. If you cannot finish the sentence that way, you have read it as a simple regression slope by mistake.
Common Mistake: Dropping every non-significant variable at once. The p-values change when the model is refit, so the book removes one variable at a time, largest p-value first.
Important Questions:
- Why compare models with adjusted R-squared? Plain R-squared never decreases when a variable is added, even a useless one. Adjusted R-squared penalizes extra predictors, so it can fall and help you choose a smaller model.
- When is logistic regression used? When the response has two possible outcomes, such as spam or not spam. It models the probability of the outcome, which stays between 0 and 1, instead of a straight-line prediction that could fall outside that range.
Appendices A-C: Probability, Selected Solutions and Distribution Tables
Appendix · Difficulty: Medium · Key topics: probability rules, conditional probability, tree diagrams, random variables, expected value, odd-numbered solutions, normal, t and chi-square tables
Appendix A is an optional, self-contained introduction to probability. It starts from the long-run proportion of die rolls and the Law of Large Numbers, then covers disjoint outcomes, the General Addition Rule, probability distributions, complements and independence. Next come marginal, joint and conditional probability, a 1721 Boston smallpox example, the General Multiplication Rule and tree diagrams. The last part treats random variables: expected value (a bookstore revenue example), variability and linear combinations. Appendix B gives short answers to the odd-numbered end-of-chapter exercises, and Appendix C holds the normal, t and chi-square probability tables that Chapters 2 to 4 rely on.
Key Points:
- Probability is the proportion of times an outcome would occur over many repetitions, and the Law of Large Numbers says the observed proportion settles toward it.
- For disjoint events P(A or B) = P(A) + P(B); otherwise the General Addition Rule subtracts P(A and B).
- For independent events P(A and B) = P(A) × P(B); in general P(A and B) = P(A | B) × P(B).
- Tree diagrams organize conditional probabilities along each branch.
- The expected value of a random variable is its probability-weighted average, and expected values of sums add.
- Appendix B answers only the odd-numbered exercises; Appendix C has tables for the normal, t and chi-square distributions.
Memory Tip: AND for independent events multiplies, OR for disjoint events adds. If the events overlap, subtract the overlap once.
Common Mistake: Adding the probabilities of two events that can happen together without subtracting the overlap, which counts the shared outcomes twice.
Important Questions:
- What is the General Addition Rule? P(A or B) = P(A) + P(B) – P(A and B). It reduces to the simple sum when A and B are disjoint.
- What does it mean for two events to be independent? Knowing one happened does not change the probability of the other, so P(A and B) = P(A) × P(B).
How to Study This Book
Read this book in the order that fits your course, because the authors say the order is flexible. The preface notes that you can begin with Chapter 2 and pick up tools from Chapter 1 as you need them, that most of Chapter 5 could be covered right after Chapter 1, and that probability, if taught, works best at the very start or the very end of the course. Chapter 6 is described as an introduction to multiple and logistic regression for an accelerated course, so many one-semester courses stop after Chapter 4 or 5.
Chapter 1 is vocabulary and descriptive statistics. Read it quickly if you have already met histograms and box plots, but do not skip Sections 1.3 to 1.5 (data collection, observational studies and sampling, experiments), because the whole book keeps asking whether a result can be generalized and whether it can be called causal. Sections 1.4.2, 1.6.7 and 1.6.8 are marked as special topics (three sampling methods, transforming data, mapping data) and can wait.
Spend the most time on Chapter 2. Sections 2.1 to 2.4 are best read in order, and the card-shuffling and coin-flipping simulations are worth repeating by hand or on a computer, because every later test is the same story told with a formula. Sections 2.5 to 2.8 then explain why the formulas work: the Central Limit Theorem, the normal model, standard error and confidence intervals. If you rush this chapter, Chapters 3 and 4 turn into lists of unrelated recipes.
Chapters 3 and 4 follow one pattern: check the conditions, find the point estimate and its standard error, then build an interval or run a test. Sections 3.3 and 3.4 (chi-square tests), 4.3.5 (pooled standard deviation) and 4.4 (ANOVA) are labelled special topics, so check your syllabus before assuming you need them. The only bootstrap in the book is Section 4.5, and it is used for the standard deviation; the preface explains that the authors postponed the bootstrap on purpose.
Chapters 5 and 6 are the regression chapters. Chapter 5 (simple regression) is the one most courses cover, and it relies on Chapter 1 and the t test logic from Chapter 4. Chapter 6 needs all of that plus comfort with reading regression output tables. Treat Sections 6.1 to 6.3 (multiple regression, model selection, diagnostics) as the core and Section 6.4 (logistic regression) as an extra unless your course asks for classification.
Appendix A is a short probability primer, not a full probability course. Read it first if your course teaches probability before inference, or keep it as a reference. The preface points to Chapter 2 of OpenIntro Statistics for more probability content and exercises. Use Appendix B only after you have tried an odd-numbered exercise yourself, and Appendix C for the normal, t and chi-square tables while you practice.
Read the book with its age in mind. The examples mostly use data from the early 2010s, and OpenIntro keeps a list of known typos for the first edition on the book page. The statistics itself (hypothesis tests, intervals, regression) has not changed, but Introduction to Modern Statistics is the authors’ newer take on the same randomization-first idea, with bootstrap intervals added, and it is the better first choice for a new course. What this book never covers: Bayesian inference, time-series modelling, and experimental design beyond the principles in Chapter 1.
Used In These Programs
This book is likely to be used for the Introductory Statistics course in: BS Statistics · BS Computer Science. Browse all Statistics-tagged books, more on statistical inference and regression analysis, or the full Statistics category on Freebooks.pk.
Who Should Read This
ISRS is written for a first course in statistics. It assumes no earlier statistics and no calculus, only algebra, so it fits BS Statistics and BSCS students in an introductory or probability-and-statistics course, and anyone who wants to see why a p-value works by simulating one. It is a good match if your instructor has assigned it, or if you want a short, focused book that reaches inference in Chapter 2. If you are free to choose, start with Introduction to Modern Statistics, which OpenIntro recommends for new courses, or with the full-length OpenIntro Statistics if your course follows a conventional order with probability and distributions before inference; Introductory Statistics from OpenStax is another option for a standard first course. This book is not a calculus-based mathematical statistics text and it does not teach a programming language, so students who need proofs about distributions should pair it with a probability text such as Introduction to Probability by Grinstead and Snell.
Applicable Universities
This book is useful for students at Pakistani universities offering BS Statistics and BSCS degrees, including Punjab University, Virtual University, COMSATS, FAST, UET, NUST, and other HEC-recognized institutions.
Official Copy of Introductory Statistics with Randomization and Simulation
Click below to open the official OpenIntro page for this book, where the free PDF is offered with an optional contribution (set the price to $0 to skip it). A reminder from the same page: this is the older First Edition, and OpenIntro recommends its newer book, Introduction to Modern Statistics, for new courses.
→ Visit Official PageFAQs
Is Introductory Statistics with Randomization and Simulation free?
Yes. The official OpenIntro book page offers the PDF as a free download; it shows an optional contribution and says you can set the price to $0 to skip it. OpenIntro’s licence page lists its statistics textbooks under Creative Commons BY-SA 3.0, while the PDF’s own copyright page says only that the book is available under a Creative Commons license. A black-and-white paperback is sold separately.
Which edition is this, and is it still current?
This is the First Edition, copyright 2014, and the PDF has 354 pages. It is not the newest OpenIntro text: the official page says Introduction to Modern Statistics represents the evolution of this book, recommends that book for anyone adopting a text, and has moved this one to the end of its catalogue. The core ideas of hypothesis tests, intervals and regression have not changed, but for a new course the newer book is the better choice.
How many chapters and exercises does it have, and are there solutions?
Six chapters and three appendices (probability, selected solutions, distribution tables). There are 231 numbered end-of-chapter exercises, from 16 in Chapter 6 to 50 in Chapter 1. Appendix B gives short solutions to the odd-numbered exercises, and the guided-practice questions inside each section have answers in footnotes. The book does not print solutions to the even-numbered exercises.
Is it good for complete beginners?
Yes for the statistics itself: it assumes no earlier statistics course and no calculus, and it explains each idea with a real study. Chapters 1 to 4 are comfortable for beginners. Chapter 6 (multiple and logistic regression) is written for an accelerated course and is noticeably harder.
Does it cover randomization, simulation and the bootstrap?
Randomization and simulation drive Chapter 2: the bank promotion, opportunity-cost, medical-consultant and tapping case studies are all analyzed with repeated random trials before the normal model appears. The bootstrap appears only in Section 4.5, for the standard deviation, because the authors chose to postpone it in this edition. The newer Introduction to Modern Statistics gives bootstrap intervals a full chapter.
Do I need R or other software to use it?
No. The book teaches the statistics, not a programming language, and it shows software output as printed tables. The data sets used in the book are available on the OpenIntro website and in a companion R package if you want to repeat the analyses yourself.
Related Books
- Introduction to Modern Statistics PDF Download – Çetinkaya-Rundel & Hardin (2nd Edition)
- OpenIntro Statistics PDF Download – Diez, Barr & Cetinkaya-Rundel (4th Edition)
- Introductory Statistics PDF Download – Illowsky & Dean (OpenStax, 2nd Edition)
- Introduction to Probability PDF Download – Grinstead & Snell (2nd Edition)
Introductory Statistics with Randomization and Simulation is a compact, randomization-first route from data collection to regression, and it is still free from the authors’ own project. Treat it as the older edition of a story that OpenIntro now tells in Introduction to Modern Statistics, and browse more Statistics books on Freebooks.pk.
Introductory Statistics with Randomization and Simulation, First Edition (2014), by David M. Diez, Christopher D. Barr and Mine Çetinkaya-Rundel. Published by OpenIntro under a Creative Commons license; OpenIntro’s licence page gives it as Creative Commons Attribution-ShareAlike 3.0 (CC BY-SA 3.0). Official source: openintro.org/book/isrs.
© Copyright Policy
Freebooks.pk is an educational resource for students in Pakistan. Textbooks and other books belong to their respective boards, authors and publishers, and all rights remain with them. If you are the copyright holder of any material on this page and want it changed or removed, please contact us or file a request under our DMCA policy. We review every request and remove reported content promptly.
