I am trying to learn about non-response/dropout bias—and how to correct for it. I do not know this area well. Is it on-topic to ask for authoritative/canonical/highly-cited/classic references on this issue?

I feel as if this is not necessarily a matter of opinion, which would make it off-topic. For example, I would recommend Elements of Statistical Learning to someone who wants an introduction to machine learning. I would recommend Cohen, Cohen, West, & Aiken for someone interested in learning multiple regression. I know popular review papers on power, p-hacking, median splits, etc., to recommend if people are interested in those topics. These aren't necessarily opinion, as many of them are the "classics" that people always talk about.

Can I ask for good references on a topic I am unfamiliar with?

  • 3
    $\begingroup$ See stats.stackexchange.com/… for exemplary answers in threads of this type. I expect most of them to support your contention that they are not necessarily matters of opinion: they provide objective reasons to justify their recommendations. $\endgroup$
    – whuber Mod
    Commented Feb 8, 2018 at 23:06

2 Answers 2


As @whuber hints, there is no problem with this. Note that we even have a tag. One thing to bear in mind is that such questions often elicit a 'big list' of possible answers, and although you are right that there can be real canonical treatments, it can still be the case that there isn't a single response that is the right response or more right than all others by definition. As a result, questions like that sometimes become CW. If you were only after reputation, that would be a disincentive, but the information will be just as good, so I encourage you to ask away, if you have a good question and it hasn't been covered on the site before.

  • $\begingroup$ Sometimes it isn't that I didn't grasp it in a class, it is that I have never seen the idea and I'm looking for written basics by a decent and credible author. $\endgroup$ Commented Feb 15, 2018 at 0:59

The problem with this is that there is no such thing as "canonical" or "authoritative" references. What constitutes "good" is also subjective. Some articles are certainly written as (if they were) canonical or with authority. Top-level research like this can be nice because it is accessible. But some articles venture into domains of "hand-wavy" to flat-out wrong. Claims made in either type of article are targeted at a familiar audience and may elude someone who's coming up to speed with a subject.

For instance, many users of our site still miss the fact that it is the residuals, not the outcome, that must be normally distributed in a linear regression to justify the F-test; no doubt they encountered the concept in basic texts but still lack an understanding. Didactic materials give more attention to developing concepts.

I think a more on-topic request would be for reviews, applied, or didactic materials like research articles or texts.

  • 1
    $\begingroup$ You're right in abstraction and I've often said or thought similarly. But in turn (a) many people still want to know which books are worthwhile for particular purposes (b) a good recommendation will say why a book is good and who or what it's best for (and also who or what it's not good for) (c) any specific recommendation can be challenged, as for example you might challenge any mention of a textbook that was unreliable. (Note that some might want to point out that assumptions are about errors, not residuals. There's always a criticism on some different level.) $\endgroup$
    – Nick Cox
    Commented Feb 13, 2018 at 19:07
  • $\begingroup$ @NickCox I think you're helping to underscore the point of my post: be specific with requests. Asking for "A good book about the linear regression" should not be closed as off-topic, but could be closed for being too broad. Also thanks for NB about errors/residuals :) $\endgroup$
    – AdamO
    Commented Feb 13, 2018 at 19:16
  • $\begingroup$ Now I am puzzled, as your first sentence seems in direct contradiction to that of @gung. His very short answer to the question is evidently Yes, and I thought your very short answer to be thus No, although the interest is in precisely why and what you are recommending. But I am pleased if you think we agree. $\endgroup$
    – Nick Cox
    Commented Feb 13, 2018 at 19:30
  • $\begingroup$ @NickCox OP's question lands in me a few different ways, to which I have different responses [1] It's okay to ask for references (but name your purpose) [2] It's silly to ask for good references [3] There is no such thing as canonical/authoritative references. $\endgroup$
    – AdamO
    Commented Feb 13, 2018 at 19:35
  • 2
    $\begingroup$ We agree on (1). Counter-examples to (3) seem easy e.g. McCullagh and Nelder is canonical and authoritative on GLMs; Efron and Tibshirani on the bootstrap; etc. (2) is like (1): it can be discussed with details on purpose and background. I wouldn't say that requests for linear regression references are necessarily off-topic, although there could be dozens of different answers. $\endgroup$
    – Nick Cox
    Commented Feb 13, 2018 at 20:18

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .