Ishfaaq Mohammed Imtiyas writes:
I am a starting lecturer in statistics trying to figure out a curriculum that works for an intro stats class at a liberal arts college.
In this video, you mention that there is but one intro stats book that you like. What is this book?
My reply: I recommend this book by Llaudet and Imai. See also here. And I recommend this book which has lots of stories and class-participation activities.
I also like Rohan Alexander’s book, Telling Stories with Data, but it’s not really an intro statistics book on its own.
There are also some free online books which are ok, they’re about as good as the popular commercially-published intro statistics books, but I find just about all of them to be uninspired and stuck in a mode of statistics that does not seem real to me. Llaudet and Imai’s is the only one I could really imagine using for an introductory course.
Feel free to add your own recommendations in the comments.
I like “Statistical Rethinking” by Richard McElreath. It’s also how I found out about this blog!
Luciano:
Statistical Rethinking is an excellent book, but I’d hardly consider it appropriate for an intro stats class at a liberal arts college!
For that matter, I also recommend our book Regression and Other Stories, but, realistically, it moves too fast for an intro undergraduate statistics course.
Andrew, would you recommend Statistical Rethinking for a second course in statistics at the undergraduate level? For reference, this class is not required for any major except for the data science major, so the students are motivated.
Sorry, I’m not the most familiar with your education system. Regression and Other Stories is already in line as my next stats book, unless the bayesian workflow comes out before I find some time!
Only a few years ago, in my first two statistics classes, we used a textbook written by the lecturer. As I am not a fan, I would not recommend it. At over 1,000 pages, it was an intimidating monster, which is the last thing you need if you are a new student. Later, I took an econometrics class which used the Stock and Watson textbook. I found it accessible — at least, accessible enough to be an enjoyable learning resource. My understanding of statistics has developed quite a bit since then, so I am not sure if I was just excited to finally understand how deduction (in the assumptions) and induction (in the data) work together.
Throughout most of my studies, however, I learned from the lecture slides. Most university instructors (and most people, for that matter!) misuse slides. Rather than enhancing their presentation, they contain all of the information. This strongly discourages students from picking up a textbook. They also discourage good teaching practice by focusing solely on teaching input.
The only times I picked up a textbook during my studies were (i) for my econometrics class, as I felt I did not understand the slides/presentation, and (ii) in a class that focused entirely on making the content accessible. There were no slides, and it was the best class I ever took!
TL;DR: Do not look for the perfect textbook. The design of the class, including the slides, is so much more important. I recommend starting with setting learning outcomes (what competencies are my students supposed to acquire or train?) and reframe the class around those goals.
Andrew didn’t mention that the fourth link (not including the title) is to his own book with Aki Vehtari, Active Statistics, or that it comes with a free online pdf and complete course materials including software. It also has a long description of Andrew’s teaching philosophy, which I’m guessing Aki agrees with because he put his name on the book. As Andrew knows, I’m too much of a control freak to be the n-th author of a book I don’t completely agree with.
I wouldn’t personally use all of Andrew’s teaching methods, but I will say they are all thought provoking and worth thinking about. They would also be super useful if you’re the kind of person willing to take risks and experiment with teaching methods on your students (just don’t tell the IRB that’s what you’re doing and don’t get too carried away with conclusions based on a small data set with huge variance due to students and teacher—measuring the benefit of educational interventions is very very hard).
Andrew—was that book really meant to be standalone textbook for a stats class? The book itself says, “You can use this book as a supplement to Regression and Other Stories or as part of a course on applied statistics.”
Bob:
Active Statistics was not intended as a standalone textbook for a stats class, and we never presented it as such! In the above post, when I wrote, “I recommend [Active Statistics] which has lots of stories and class-participation activities,” I meant that I recommend it as a resource.
I recommend that any student who’s taking a statistics course (at any level) should buy (or download) Active Statistics and read through all the stories, and I recommend that any instructor who’s teaching a statistics course (at any level) should buy (or download) Active Statistics, for the stories (which can be told in class), for the activities (which can be done in class to keep students involved), for the computer demonstrations (these work out great in class), and for the drills and discussion problems (also good in class).
I think that students can get a lot out of reading the stories in Active Statistics, above and beyond whatever is covered in their classes, and I think that the book gives instructors lots of tools that can help make the classroom experience much more interesting and productive for their students.
As we wrote on the book, and as you quote, you can use Active Statistics as a supplement to Regression and Other Stories or as part of a course on applied statistics.”
For the past few years, I’ve been using a modified version of OpenIntro’s “Introduction to Modern Statistics” (https://www.openintro.org/book/ims/). I like that the book is free and open source, has lots of good examples and exercises, and also has some online R tutorials. But really I think of the book as a resource for practice, not as a great way to learn the concepts which happens primarily in the lectures, activities, and discussions in class.
The main thing that I like about the OpenIntro book is that it has large sections on simulation. In my experience, the hardest thing for students to do is to understand that data could have come out differently. We have a sample, but to know what we can learn from that sample, we need to imagine other ways that our data could have looked. That requires using a model based on some hypothesis. Jumping straight into mathematical models like posteriors or sampling distributions is too abstract for most students. Instead, I’ve found that spending time getting acquainted with a few basic techniques like randomization and bootstrapping helps students get over the major conceptual hump that statistical inference is really an exercise in imagination.
Gec:
The Open Intro book is excellent for what it is, and I do like the simulation stuff, but I still am unhappy that it’s presenting the conventional take on statistics: inference for a proportion, inference for the mean, confidence intervals, hypothesis tests, etc etc. I prefer the Llaudet and Imai book (for reasons you can see if you follow the above link) because to me it seems to more directly address problems of interest. And I think the Llaudet and Imai book would be great for an intro course at a liberal arts college.
I really think that the question has to have context. Which kind of intro stats? Intro stats as the required general education course for all students, intro stats for students who have taken calculus, intro stats for those who have taken precalculus, intro stats for students majoring in disciplines where statistics are used (all the social/behavioral sciences, business, health sciences, biology, computer science), intro stats for students who are math majors, intro stats for statistics majors. And so on.
For ” intro stats class at a liberal arts” I would assume it is a course for non-STEM majors many of whom don’t like math but also will end up in majors where they need statistics. Also, depending how selective the college is, most students may or may not have had precalculus.
I have liked using Course Kata https://www.coursekata.org/ which is very interactive and modern, uses simulation-based approaches and incoporates real data projects throughout. I also like that I can flexibly decide which version to use based on the background.
Many of my colleagues use Passion Driven Statistics https://passiondrivenstatistics.wescreates.wesleyan.edu/ which is project based.
Both are products of NSF funded undergraduate education projects and have articles you can read about student outcomes and lots of supporting material.
The link to passion driven statistics does not seem to function.
I get to the page, but the video they link doesn’t work.
Yeah the video isn’t there, but probably it is this one https://www.youtube.com/watch?v=vjxUyqr2bYc&list=PL8nC8L7kjElekx3imIMvacweBs-OFRwWR
I don’t know if my post got lost in the spam filter or if I did something wrong. So I’ll say that I suggested Course Kata
https://www.coursekata.org/ and Passion Driven Statistics https://passiondrivenstatistics.wescreates.wesleyan.edu/.
(I’ll repost about why if my other post doesn’t show up.)
But the other thing I would strongly suggest if you are dealing with a gen ed course is that you do a good quality pretest (and post test) to understand where your students are. Unfortunately, you will often find they have much less understanding than you think they do. We use https://locus.thinkdataed.org/ but many others use CAOS (can’t find a not broken link at the moment).
ISLR was great for self study, personally.
This type of question is pretty popular on Reddit and I see the same meme recommendations here are mentioned. I’m happy to see at least some background of the reader is mentioned. I think Statistics by David Freedman and Statistical Methods work book would be a great place for the assumed background. The only issue with this book is there is no code for users to leverage. Outside of that, you get a lot of insight from big name Statisticans like Freedman which is absent in a lot of Intro to Statistics Books that don’t assume Calculus (even with those books it seems Freedman’s book still has a lot to offer over those books).
I’ll humble throw my effort with Matteo Lisi into the suggestion list (https://www.mheducation.co.uk/statistics-for-psychology-using-r-a-linear-models-perspective-9780335252626-emea-group).
It’s not Bayes, but hopefully it provides a solid enough foundation to make tomes such as Statistical Rehthinking more approachable for the typical psych student.
Not familiar with several of the resources referenced here, but I’ll throw in another plug for Statistical Rethinking by Richard McElreath. I think he has a third edition nearly done, plus there’s a set of generously provided Youtube lectures. I’m not sure about the details of building a class around it, but it seems very accessible for that.
I honestly think that an undergraduate liberal arts college context could be fun and productive for teaching statistics, because the students will be one step removed from immediate pressure for application, presumably. The big challenge with teaching graduate students is that they are on a tight timeline to turn around their projects into publications, for which they will need their supervisor’s buy-in. Teaching anything other than the ‘business as usual’ practices in a field – which is still very often NHST +/- various commonly deployed fallacies and ad hoc ‘lexicographic rituals’ – turns into a minefield of inevitable frustrations for everyone involved.
Chris:
Statistical Rethinking is an excellent book, but I’d hardly consider it appropriate for an intro stats class at a liberal arts college!
For that matter, I also recommend our book Regression and Other Stories, but, realistically, it moves too fast for an intro undergraduate statistics course.
Maybe! I find McElreath’s pedagogical approach very conceptually lucid and ‘liberal artsy’ for lack of a better term. I will never get to replay how I learned statistics, but I’ve always felt disappointed that his book didn’t come out until after I’d gotten through the hard part of the learning curve…
Yes I believe Statistical Rethinking is actually ideal in this setting. It will free students to solve real problems in a direct manner. It would be a brave and somewhat risky choice because it doesn’t cover the traditional indirect methods, but it would do more good than any other choice I can think of.
It would be a risky choice because many undergraduates in a general education intro stats course do not know probability, how to calculate a mean from grouped data or what a variable is. He defines the audience for the book as (possibly) beginning Ph.D. students in the sciences or experienced researchers who are interested in Bayesian approaches.
I’ll offer a different take completely. I second the notion that we need to be careful of what we mean by the intro stats course. For an undergraduate course for liberal arts students, the mentioned texts are fine – at least I have nothing better to offer. For my students (MBA students, most likely to only take this one stats course, some with an undergrad course, but not all, and most not fond of, or proficient with math), I don’t use any traditional text. I’ve used Speigelhalter’s The Art of Statistics which is an excellent book but not at all a traditional text. But that is my point: my approach is to have students work with rich multidimensional data and explore a variety of predictive modeling approaches. They do not need, nor will they master, the technical details. But my hope is that they appreciate how data can improve decision making, the pitfalls they are likely to encounter with data, and gain an appreciation for the inherent uncertainty and potential benefits and costs of decisions based on data. For MBA students who are not going to do much analysis themselves, but who may hire others to do so and/or use such analysis to inform their decisions, those seem more important than what is covered in traditional texts.
I also find it interesting that almost all of the books people recommend (most of which are excellent), the course starts out with programming (usually R, but not all). We’ve had this discussion before. I respect the role of programming as well as the value it has in itself. But I respectfully disagree with its necessity in the intro course. In fact, I think the reason so many instructors like that approach is that they feel like students walk away with ‘something concrete.’ However, many liberal arts students will not continue to code, and are likely to forget it rather quickly. Then there is the growing ability of LLMs to do much of what they learn in the intro course. Rather than argue (as we’ve done before on this blog) endlessly about whether or not coding is necessary in the intro course, I would point out that the time spent on coding might be better spent on issues of measurement, sampling, visualization, multidimensional thinking, validation, and decision analysis. It is possible to do these things without coding (I personally use JMP for that reason – all of this can be covered in the course with no need for coding and with a rapid and intuitive learning curve). At least in my courses, the time is much more limited than I’d like, so I need to have a sense of priority over what I cover and what I don’t.
The project-based books people mention are much closer to my approach. I find the fact that they start with coding unnecessary and somewhat distracting, but at least the emphasis is on data and variability as enriching our understanding of the world rather than a disjointed set of mechanical techniques to analyze subsets of problems (e.g. inferences for 2 continuous variables, ANOVA, NHST, properties of regression models, etc.). So, I don’t think I am disagreeing with the recommendations people are making – they are good texts – but I am suggesting that the intro course is often improperly focused. It tries to do too many things and tries to work for too many different audiences. The one commonality I see for almost everyone that takes the intro course, is the need to understand what data is, and what it can and can’t be used for.
Agreed. If the student is never going to be implementing the stats themselves, they just need to know to make sure the results also apply to future data rather than whatever it was fit on.
You don’t need to know physics jargon to tell astronomers are doing something useful when they predict exactly when an eclipse will happen over and over again.
Heuristics to quickly sanity check for BS is the important thing, especially when they will be using llms to do it.
Yes hard agree on the programming issue. I do use R but I work hard to make sure that we do not spend the first weeks installing software etc. I think you would find that a lot of intro stats faculty working with the kinds of students you are end up with a fill in the blank model of programming and that’s okay. I also keep coming back to the idea that it’s really usefull to learn to do some basics of spreadsheets, which are the reality of most jobs.
Students all over have many challenges with understanding important foundational topics. How to figure out the correct denominator to use when calculating a proportion (and how the answer may be different for different purposes) is a struggle for many. So if you can get them through that and learning how to make some meaningful graphs, that is an accomplishment.
If you want to teach the students to think, especially, to think critically, then whether they know math or not, the book to use is Statistics by Freedman, Pisani, and Purves. It is the only book that takes students seriously enough to engage their minds without math and that has exercises that ask them whether real studies were justified in making their conclusions. It does not ask students to analyze data with the aim of making fake conclusions. However, most teachers don’t want their students to think like that. Other than those whose livelihood depends on their own thinking not like that, I don’t know why more of the others don’t teach that way.
Russ:
Unfortunately the Freedman et al. book has its own problems, for example a dogmatic treatment of statistics and sampling which doesn’t align with statistical sampling in the real world. That can be fine for a textbook–you have to decide where to be realistic and where to idealize–but I think the dogmatism can create problems for students. I recommend the Llaudet and Imai book, which isn’t perfect either but I think also does a good job at engaging with students. Also as I said in comments above, I really like our own book, Regression and Other Stories, but that’s at a more advanced level, appropriate for a second-semester undergraduate statistics class, I think.
Hi, Andrew.
Could you please give me an example of where FPP has “a dogmatic treatment of statistics and sampling which doesn’t align with statistical sampling in the real world”? Also, I don’t know what you mean by “That can be fine for a textbook–you have to decide where to be realistic and where to idealize–but I think the dogmatism can create problems for students.” After all, we are discussing textbooks and textbooks are usually for students.
Finally, the Llaudet and Imai book has no exercises and tells them that linear models “control” for confounders. This looked to me like the opposite of engagement. Rather, it is dogma.
It’s been many years since I’ve looked at the Freedman et al. book so I don’t remember the details. I’ll find a copy, take a look, and get back to you on this one.
Regarding my statement about textbooks being unrealistic: What I meant here is that no book can cover everything, and an intro book will necessarily involve some simplifications. I was just uncomfortable with some of the simplifications in the Freedman et al. book. In Regression and Other Stories we tried to be very careful and avoid blanket statements that sound good but don’t quite fit reality. Then again, we had the luxury of writing an intermediate-level, not an introductory, book.
The Llaudet and Imai book does has some exercises; see here: https://ellaudet.github.io/dss_instructor_resources/
But maybe not a complete set of exercises for the whole course; I’m not sure.
Regarding the statistics jargon of “controlling” for confounders: I agree, I don’t like that term either! I much prefer the term “adjust” in that setting. As I said, I don’t think the Llaudet and Imai book is perfect. But I don’t think their use of the unfortunately-standard term “control” is a form of dogma; I think it’s just a bad term, like “statistical significance,” “unbiased,” and other misleading expressions that have taken root in statistics. Maybe we can persuade them to change “control for” to “adjust for” in future editions of their book, but unfortunately this is the phrasing that’s used pretty much universally in social science.
I hope this reply gets in the right place; I can’t reply to your latest reply.
I look forward to seeing what you come up with in FPP.
As for L&I, see what they say on p. 149 in the box about adding all potential confounders and causality. I was not saying that the term “control” by itself is dogma; rather, their statement that a linear model is appropriate without even examining this is dogma. Even “adjust”, a far better term, would be misleading, because students (and others) assume it is an improvement without question.
FPP aims to prepare students to be discerning consumers of statistics. It does not pretend to prepare them to produce their own statistics. This is what is appropriate for the vast majority of students.
Russ:
I think that the term “adjust” is appropriate and does not imply that the adjustment always makes things better. We’re doing statistics; at best we can just make things better on average!
Regarding your other point: I don’t have a great sense of whether it’s better to teach an intro course so as to teach people to do something or so as to teach people to be discerning consumers. In the English department, there are literature courses and there are writing courses. In music, there are performance courses and there are music appreciation courses. It’s my impression that in mathematical fields such as math, physics, and statistics, the courses are all focused on doing things, not on being a consumer. In an intro physics class, you learn how to solve intro physics problems, you don’t learn so much about being a consumer of physics. I’ve seen intro statistics classes that use the Freedman et al. book, and these courses too are focused on students learning how to solve intro statistics problems, roughly similar to what’s in the AP statistics curriculum. I agree that the reason for most students to take such courses is to be consumers of statistics, not practitioners, but it doesn’t seem that the courses do so well at that. I guess that’s one reason why it’s a luxury to write more advanced books. Currently I’m using Regression and Other Stories to teach a class of graduate students in political science. Some will grow up to be developers and advanced users of statistical methods, others will end up using statistical methods, and others will just be consumers, and I try to reach all of them in the course.
P.S. The blog service recommends limiting the number of indentations in the comments, I think to make the comments section readable on a phone.
Andrew,
Typo: I meant p. 146. I ran OCR, and here’s what that box says: “If, in the multiple linear regression model where X_1 is the treatment variable, we control for {\emph all} potential confounders by including them in the model as additional X variables, then we can interpret \hat\beta_1 as a valid estimate of the average causal effect of X on Y.”
The problem I was alluding to with “adjust” is what people think it means, not what it actually means. This must be addressed explicitly in the course.
An intro course in stats should definitely teach critical thinking. Obviously more advanced courses can teach technique, though they are not immune to the need for critical thinking. FPP should not be used to get students to do lots of arithmetic.
P.S. A reply button can still be included without the need to indent the reply; that’s a separate issue, which is perfectly reasonable. I presume the developers did not think of that.
I don’t believe in the hard distinction between learning to do vs learning to be better consumers of analysis. I think the best way to learn to be a good consumer is to learn how to do it yourself. This doesn’t mean you need to learn how to create an algorithm or how to implement it, but learning how to collect (broadly speaking), manipulate, visualize, and analyze data in the context of a meaningful problem is the best way to understand the value and limitations of somebody else’s analysis. I don’t think providing a list of yellow or red flags regarding analysis is nearly as effective as having to deal with the issues yourself. Experiencing the forking paths provides a concrete and visceral sense of what has gone into an analysis they might encounter as a consumer.
I think there are courses aimed at educating clinicians, lawyers, etc. about pitfalls in statistical studies – these are usually short courses providing lists and examples of things to look out for. While these can be useful, I don’t think they really provide the understanding I am describing. The experience of doing a real study where you have to decide what to do about an outlier or missing data, for examples, works better than a description of issues related to outliers and missing data.
Dale,
You are correct. However, FPP is designed for the vast majority of students who take statistics: they cannot do even algebra. Therefore, rather than rely on the students being able to understand what they are coding, it relies on examples and exercises that confront directly various pitfalls. I completely agree with you that a list of pitfalls is not very useful. The exercises are crucial, but I disagree with you that they need to involve coding, especially for the vast majority of students. (These days, AI will make such a focus even worse.)
For most students, being able to know how much to trust and how to interpret what they read or hear in the news when statistics are presented will be very useful to them as adults, whereas they will never have occasion to produce statistics themselves.
Russ
One misconception: I do not believe students need to do coding in order to experience doing the analysis themselves. I am perfectly willing to have them use software that makes the coding unnecessary (and I do so). I am not saying coding is bad for them – far from it – but I do question its necessity in the intro course. I also think it serves certain students well to have coding an integral part of the intro course – while doing a disservice to other types of students.
Dale,
I meant “coding” in the general sense of using some kind of software to calculate, not just a hand calculator.
Russ
I don’t understand what you are saying – why would students using software not serve the needs of liberal arts students to have a good intro course? The usual argument against using something like JMP is that it is a commercial product that they are unlikely to have in their future workplace. But we are talking about their initial exposure to exploring meaningful data and deriving meaning from it. I’m all for having them use powerful and easy to use software – and it permits them to explore large data sets of interest to almost everyone. Why limit them in that course? If the concern is about the availability of computers, there is hardly a college where students don’t already own computers or have them available through their school.
Dale,
Have you read FPP? Here’s a quote from their preface: “Why does the book include so many exercises that cannot be solved by plugging into a formula? The reason is that few real-life statistical problems can be solved that way. Blindly plugging into statistical formulas has caused a lot of confusion. So this book teaches a different approach: thinking.”
I am skeptical that students at the level I am discussing can get much from using software, and I highly doubt it would benefit them more than would using the time for more topics from FPP. If you haven’t read FPP, you cannot evaluate this for yourself. However, I admit that the last time I taught from FPP at this level was in 1998, and I never tried using software. On the other hand, others in my department did use software; if I recall properly, their students did not learn to think critically. Of course, this does not refute your point.
I think it would be hard to argue that our current system of teaching stats is successful. In fact, I recall Andrew agreeing with this in some earlier post and seeming to take some blame himself.
For students at a much higher level, but still undergrads, I did use software (Matlab). For them, I had them read FPP for the first few weeks; that was enough time for them to get the ideas before we got mathematical, which was needed because the course had no stats prereq.
Nevertheless, very high functioning math people have loved this book. Some of my (math) grad students felt it was a pedagogical masterpiece. A Fields medalist I recommended it to loved it. I admit that I did not quiz them and cannot vouch for what they learned.
Russ
I think we are talking past each other. I did read FPP long ago and can’t claim any clear memory of it, but I doubt that it is in conflict with using real data and software to help with critical thinking. I can think of many exercises involving statistical thinking that can be effective for students without having any data to accompany it – but I think having real data to explore alongside such exercises only adds to the effectiveness. For example (just off the top of my head), I can imagine a numerical example involving Simpson’s Paradox (sorry, Andrew, I know you don’t like that portrayal) that involves the students’ critical thinking about how aggregation can distort an understanding about relationships. Having data that produces such a paradox would enhance that critical thinking (in my view), provided that it doesn’t require a distracting focus on techniques or calculations (which is where good software should come in). I am certainly not advocating having students plug numbers into formulas – but I view real data (real and simulated) as essential to critical thinking. Without that, it is hard to see how students gain an appreciation for variability.
Dale,
Just to make sure you are aware that FPP uses lots of real data: Simpson’s paradox is explained in Chap. 2, Sec. 4, with a table of numbers. Each of the three sections of Chap. 1 has a table of numbers. I give these as examples to illustrate the importance FPP attaches to looking at numbers. But students at this level need guidance; they’ve never had to think about such things before.
Here are the questions in an exercise in Chap. 2 (without the table and other info that precedes it and without a continuation in the following exercise):
—
(a) Does screening save lives? Which numbers in the table prove your point?
(b) Why is the death rate from all other causes in the whole treatment group (“examined” and “refused” combined) about the same as the rate in the control group?
(c) Breast cancer (like polio, but unlike most other diseases) affects the rich more than the poor. Which numbers in the table confirm this association between breast cancer and income?
(d) The death rate (from all causes) among women who accepted screening is about half the death rate among women who refused. Did screening cut the death rate in half? If not, what explains the difference in death rates?
Russ
Just one more (I don’t think we are at “garbage time” but we are risking getting repetitious). The examples you cite sound good to me – the kind of things I agree should be in the intro course. I still think having the data and producing the tables referred to in the text are an important addition. The critical thinking is there with the questions being asked, but there is something important about producing the table yourself, if it is an easy thing to do (i.e., without any coding). There are almost always several ways to produce tables from given data, and I’d prefer the data to be multidimensional to begin with – in that way the question being asked requires the step of figuring out which variables could be used in the table and that there may be several alternatives. I’d be fine with the examples you cite as being asked first, with a followup exercise where a data set with several variables is provided and the students need to produce a relevant table and use that to answer the same sort of question. The link that I think is missing from the example you cite – and which I think is important – is going from the messy data-rich world, to the reduced table that illustrates the relevance of Simpson’s Paradox. I believe that even if students understand that the aggregated table can be completely misleading, there is something missing if they don’t see how that table might arise from the raw data. We might disagree on the importance of that – but if the software they use does not impose any unnecessary burden on the students, then I think that extra step is worth it.
Dale,
You might be right. As I wrote earlier, I haven’t tried that. However, the text one chooses is crucial, and I think FPP is the only one that honestly and thoughtfully teaches critical thinking. The instructor is always free to add to it.
Thank you very much for the kind mention.
Gelman and Hill very much changed my academic direction – thank you for writing that.
Rohan:
I’m glad our book had a positive effect! And I like Regression and Other Stories even more.
Way back when I took undergrad statistics and we had the old edition of Statistics: A Guide to the Unknown. The chapter about the polio trials and the one about the Federalist papers have stuck with me. In fact, I just used a newer write up of the design controversies around the polio trials for students to see both how politics is always present and to think about why the randomized trials had different results than the observational ones. No matter what textbook you use, I’d encourage supplementing with engaging reading at an appropriate level.
Elin,
“The Salk vaccine field trial” is the name of Chap. 1, Sec. 1 of FPP (Freedman, Pisani, Purves). It is 4 pp long, and these polio trials are indeed a great introduction to statistics and various kinds of comparisons.
Do you have a recommend textbook for highschool students?
Any recommendations from the authors or the comments for a good intro textbook on probability theory specifically?
Dan:
I like the classic book, Feller Volume 1.
Depends on the background. You’re question will get different levels of text books because you didn’t specify the pre-reqs. Introduction to Probability by Blitzstein is a good book for example, but I think you aught to have some calculus and Discrete Math to read it first. Ian Hacking’s book on Probability is far more approachable for most people in comparison. One of the best books out there has to be Durrett but this is another example of a book that expects quite a bit from the reader.
Dan,
I highly recommend Pitman’s “Probability”. I once tried to use Feller vol I for the senior-level class, but it was too hard for the students.
Grimmett and Stirzaker.
One research-based intro statistics textbook, CourseKata Better Statistics, is very well-done. I participated in a summer PD workshop and really enjoyed it. It seems excellent for ensuring that students do the readings and exercises (interactive R exercises and active learning modules). The underlying idea is modeling from the beginning to the end.
https://www.coursekata.org/college/materials
Yes I really like that they will actually talk to you about how to use the book and materials. Also, if you do find an error or somehing your students were confused by, they fix it right away.
Jay:
I followed the link and took a look at Course Kata, and I have mixed feelings.
Positives:
– It’s free, it has lots of interactive exercises, etc., just a good learning package overall.
– I like all the coding and the way that the whole thing is simulation-focused.
– I like its discussion in the first chapter on measurement and sampling. This is done much better than in usual statistics books. The content is solid and it’s written clearly too.
Negatives:
– I don’t like how they deal with data description, filling up students’ heads with silly trad-stat things like means and medians, outliers, boxplots, etc. I just don’t think that the skill of learning about the one-dimensional distribution of a dataset is so important. I much prefer how we do things in chapter 2 of Regression and Other Stories, where we jump right into time series plots and scatterplots.
– I don’t like all the hypothesis testing. I just don’t think this is a good way to do science. This is a problem I have with just about every intro statistics textbook, other than the Llaudet and Imai book.
So, yeah, mixed feelings on the Course Kata statistics book. Conditional on it being a book that covers the standard material (which, yeah, I get it, that’s what most instructors want), I think it’s excellent. But I remain bothered by most of the material that it covers.
Andrew,
Oddly, L&I write, “rejecting the null hypothesis _is_ the same as accepting the alternative hypothesis” (p. 214). In Sec. 7.3, they apply this to an observational study: “We conclude, then, that receiving Russian TV likely had a non-zero average causal effect on the probability of voting for a pro-Russian party in the 2014 parliamentary election for all Ukrainians living close to the border with Russia, not just for those who participated in the survey.” They have no discussion of the validity of the mathematical assumptions behind their model.
Russ:
What can I say? I agree that the Llaudet and Imai book is imperfect. It seems that the slipped up on this one, in the same way that many researchers and authors of statistics books do. I wish they’d run their book by a careful reader such as you or me before publishing, and I hope they fix this in their next edition.
I still like the book a lot–but, yeah, if someone’s gonna teach from it, they should point out this and any other errors.
I also really like what Kosuki and Lluadet somewhat confusingly call “DSS”. But let’s go with it.
There are two problems with the text. First, it assumes that everybody has a machine that can run R and RStudio. They don’t; often the only device they have is their cellphone. Sure, you can try to get around this by having students share a computer (who’s?) and doing everything in class, but that defeats the flipped classroom idea. Second, coding is hard and the students will do everything to avoid it, especially if they don’t have independent computing capabilities. DSS is set up to make it easier; most of the code needed is right there in the book. But … that is sort of monkey see, monkey do, isn’t it?
What I will do the next time I teach a methods course is use DDS and do the exercises in Jamovi. That will take some work and it doesn’t address the “I don’t own a computer!” problem. But the windowing system for R in Jamovi makes generating the code necessary pretty easy and there’s a “module” in the app that allows the student to see the code. The curious ones will inevitably start fooling around with coding – you can do that right in Jamovi – to see what happens.
Well, maybe this will happen and the poor students that sit through the first interaction will, perhaps, not suffer too much. And I’m betting they’ll still use Jamovi when they leave the class.
PS: I have already used the R Commander in methods classes as well. Successfully too. But Jamovi is easier to install and use.
One wish I had for statistics education that I hope would lead to someone writing a book about is that one can generalize many of the standard methods to a single type of semiparametric model as detailed here: https://hbiostat.org/rmsc/ordsurv#special-cases-of-ordinal-cumulative-probability-models . As shown in the diagram there, cumulative probability ordinal semiparametric models encompass the spectrum from comparing proportions and means to virtually all of single-event survival analysis including the Cox proportional hazards model. CPMs can be equally well used in a Bayesian or in a frequentist framework. Just think of how many special cases don’t really need to be taught. I’ve done this on a small scale in https://hbiostat.org/bbr/nonpar .
I’d like to throw my book into the mix, which applies both frequentist and Bayesian statistics on the same problems. The frequentist stuff, I explain concepts first using resampling methods, and only then show analytical approximations. For Bayesian, I use the grid approximation to introduce basic ideas, then switch to MCMC in Bambi (which is based on PyMC). There is also a chapter on linear models, because we all know how useful they are, all that prefixed with prerequisites on data and probability…
You can see TOC and extended preview PDFs here:
https://minireference.com/static/excerpts/noBSstats/
I’d love to hear your feedback on the particular topics chosen. Anything missing? Anything that oculd be dropped to save space?
More background info on the announcement blog post:
https://minireference.com/blog/noBSstats-prerelease/
I took a brief look. I can’t comment on the virtues of the book (which may be many), but as an intro book for liberal arts students all I can say is that you must see different students than I do. Try using this book in a 7 week course that meets once/week or a semester course for a sociology major who is taking 15 credits, or a student who is afraid of equations but interested in evidence about the world. I could not use this book in any of the intro courses that I’ve taught. As for topics missing – there may well be, but at 1100 pages I suspect the reverse. Couldn’t something be dropped in the intro course? For example, probability theory is important, but do intro students really need to learn and apply the myriad rules governing probability problems?
Oops, I totally missed the “at a liberal arts college” part of the question, and I agree it would not be appropriate. I definitely had STEM majors in mind, and specifically ones that *want* more details than what is offered in a regular STATS101 class.
As for the 1100 pages, it’s really not as bad as it sounds because the paper format is 5.5″x8.5″ so it reads like a pocket book. Also, there is a lot of intentional foreshadowing and repetition of concepts (e.g. sampling distributions are introduced first as a probability concept in CH2 then again as outputs of estimators computed from random samples in CH3). That being said, your point still stands—probably impossible to go through all of it in a single semester.
I’ll look thought the PROB chapter to see if I can drop some things, but it’s going to be hard, because the way I chose the topics is only specific prerequisites I needed in the statistical inference chapters (e.g. the Student’s t-distribution for frequentist stuff and Beta distribution for Bayesian).
Instead, I’m thinking I could cut some of the model-fit diagnostics in the linear models chapter (CH4) and instead link out to Sec 16.9 of Ethan Weed’s adaptation of Danielle Navarro’s “Learning Statistics with R” which covers most essentials: https://ethanweed.github.io/pythonbook/05.04-regression.html#model-checking