Sentence length in academic articles

While reviewing a paper recently it struck me that the content was (very) good, but the writing was stereotypical academic. My first impression was that this was caused by monotonously long sentences. See this advice from Gary Provost (via Francis Diebold). Part of the reason why long sentences are undesirable is not only for aesthetic reasons though — longer sentences are harder to parse, hold in memory, and subsequently understand. See Steven Pinker’s The Sense of Style writing guide for discussion.

So I did some text analysis of the sentences. To do the text analysis I used the nltk library in python, and here is the IPython notebook to replicate for yourself if you care to do so. In the notebook I have saved two text corpuses, one my finished draft of this article. I compared the sentence length to Mark Twain’s Huckleberry Finn (text via here).

For a simple example getting started with the library, here is an example of tokenizing a string into words and sentences:

#some tests for http://www.nltk.org/, nice book to follow along
import nltk
#nltk.download('punkt') #need to download this for the English sentence tokenizer files

#this splits up punctuation
test = """At eight o'clock on Thursday morning Arthur didn't feel very good. This is a second sentence."""
tokens = nltk.word_tokenize(test)
print tokens

ts = nltk.sent_tokenize(test)
print ts

The first prints out each individual word (plus punctuation in some circumstances) and the second marks individual sentences. I have the line #nltk.download('punkt') commented out, as I downloaded it once already. (Running once in Wakari I did not need to download it again – I presume it would work similarly on your local machine.)

So what I did was transfer the PDF document I was reviewing to a text file and then clean up things like the section headers (ditto for my academic articles I compare it to). In Huckleberry I took out the table of contents and the "CHAPTER ?" parts. I also started a list of variables that were parsed as words but that I did not want to count after the sentences and words were tokenized. For example, an inline cite such as (X, 1996) would be split into 4 words with the original tokenizer, (, X, 1996 and ). The "x96" is an en-dash. Below takes those instances out.

#Get the corpus
f = open('SmallSample_Corpus.txt')
raw = f.read()

#Count number of sentences
sent_tok = nltk.sent_tokenize(raw)
ns = len(sent_tok)

#Count number of words
word_tok = nltk.word_tokenize(raw) #need to take out commas plus other stuff
NoWord = [',','(',')',':',';','.','%','\x96','{','}','[',']','!','?',"''","``"]
word_tok2 = [i for i in word_tok if i not in NoWord]
nw = len(word_tok2)

#Average Sentence length are words divided by sentences
print float(nw)/ns

There are inevitably more instances of things that shouldn’t be counted as words, but that makes the sentences longer on average. For example, I spotted a few possessive 's that were listed as different words. (The nltk library is smart and lists contractions as seperate words.)

So someone may know a better way to count the words, but all the articles should have the same biases. In my tests, here are the average number of words per sentence:

  • article I was reviewing, 28
  • my small sample article, 27
  • my working article (that has not undergone review), 25
  • Huck Finn, 20

So the pot is calling the kettle black here – my writing is not much better. I looked at the difference between an in-print article and a working draft, as responses to reviewers I bet will make the sentences longer. Hedges in statements that academics love.

Looking at the academic article histograms they are fairly symmetric, confirming my impression about monotonous sentence length. To make the histograms I used the panda’s library, which has a nice simple method.

sent_len = []
for i in sent_tok:
    sent_w1 = nltk.word_tokenize(i)
    sent_w2 = [i for i in sent_w1 if i not in NoWord]
    sent_len.append(len(sent_w2))

import pandas as pd

dfh = pd.DataFrame(sent_len)
dfh.hist(bins = 50);

Here is the histogram for my small sample paper:

And here it is for Huck Finn

(I’m not much of an exemplar for making graphs in python – forgive the laziness in the figures.) Apparently analyzing sentence length has a long history, see a paper by G. Udny Yule in 1939! From a quick perusal the long right tail is more usual for analyzing texts. The symmetry I see for this sample of academic articles is not the norm.

There could be more innocuous reasons for this. Huck Finn has dialogue with shorter sentences, and the academic articles have numbers and citations. (Although I think it is reasonable to count those things towards sentence complexity, "1" or "one" should have the same complexity.)

I will have to keep this in mind in the future (maybe I should write my articles in poem form)!

Posting my peer reviews on Publons and a few notes about reviewing

Publons is a service that currates your peer review work. In a perfect world this would be done by publishers – they just forward your reviews with some standard meta-data. This would be useful when one paper is reviewed multiple times, as well as identifying good vs. poor reviewers. I forget where I saw the suggestion recently (maybe on the orgtheory or scatterplot blog), but someone mentioned it would be nice if you submit your paper to a different journal to forward your previous reviews at your discretion. I wouldn’t mind that at all, as oft the best journals will reject for lesser reasons because they can be more selective. (Also I’ve gotten copy-paste same reviews from different journals, even though I have updated the draft to address some of the comments. Forwarding would allow me to address those comments directly before the revise-resubmit decision.)

I’ve posted all of my reviews so far, but they are only public if the paper is eventually accepted. So here you can see my review for the recent JQC article Shooting on the Street: Measuring the Spatial Influence of Physical Features on Gun Violence in a Bounded Street Network by Jie Xu and Elizabeth Griffiths.

I’ve done my fair share of complaining about reviews before, but I don’t think the whole peer-review process is fatally flawed despite its low reliability. People take peer review a bit too seriously at times – but that is a problem for most academics in general. Even if you think your idea is not getting a fair shake, just publish it yourself on your website (or places like SSRN and ArXiv). This of course does not count towards things like tenure – but valuing quantity over quality is another separate problem currently in academia.


In the spirit of do unto others as you would have them do unto you, here are a two main points I try to abide by when I review papers.

  • be as specific as possible in your critique

There is nothing more frustrating than getting a vague critique (the paper has multiple mispellings and grammar issues). A frequent one I have come across (both in reviews of my papers and seeing comments others have made on papers I’ve reviewed) is in the framing of the paper – a.k.a. the literature review. (Which makes sending the paper to multiple journals so frustrating, you will always get more arbitrary framing debates each time with new reviewers.)

So for a few examples:

  • (bad) The literature review is insufficient
  • (good) The literature review skips some important literature, see specifically (X, 2004; Y, 2006; Z, 2007). The description of (A, 2000) is awkward/wrong.
  • (bad) The paper is too long, it can be written in half the length
  • (better) The paper could be shortened, section A.1 can be eliminated in my opinion, and section A.2 can be reduced to one paragraph on X.

Being specific provides a clear path for the author to correct what you think, or at least respond if they disagree. The “you can cut the paper in half” I don’t even know how to respond to effectively, nor the generic complaint about spelling. One I’ve gotten before is “x is an innapropriate measure” with no other detail. This is tricky because I have to guess why you think it is innapropriate, so I have to make your argument for you (mind read) and then respond why I disagree (which obviously I do, or I wouldn’t have included that measure to begin with). So to respond to a critique I at first have to critique my own paper – maybe this reviewer is more brilliant than I thought.

Being specific I also think helps cut down on arbitrary complaints that are arguable.

  • provide clear signals to the editor, both about main critiques and the extent to which they can be addressed

Peer review has two potential motivations, one is a gate-keeper and one is to improve the draft. Often times arbitrary advice by reviewers intended for the latter is not clearly delineated in the review, so it is easily confused for evidence pertinent to the gate-keeper function. I’ve gotten reviews of 20 bullet points or 2,000 words that make it seem like a poor paper due to sheer length of the comment, but the majority are minor points or arbitrary suggestions. Longer reviews actually suggest the paper is better – if there is something clearly wrong you can say it in a much shorter space.

Gabriel Rossman states these different parts of peer review a bit more succintly than me:

You need to adopt a mentality of “is it good how the author did it” rather than “how could this paper be made better”

I think this is a good quip to follow. I might add “don’t sweat the small stuff” to that as well. Some editors will read the paper and reviews and make judgement calls – but some editors just follow the reviewers blindly – so I worry with the 20 bullet point minor review that it unduly influenced a reject decision. I’m happy to respond to the bullets, and happy you took the time, but I’m not happy about you (the reviewer) not giving clear advice to the editor of the extent to which I can address those points.

I still give advice about improving the manuscript, but I try to provide clear signals to the editor about main critiques, and I also will explicitly state whether they can be addressed. The “can be addressed” is not for the person writing the paper – it is for the editor making the decision for whether to revise-and-resubmit! The main critiques in my experience will either entail 2-3 main points (or none at all for some papers). I also typically say when things are minor and put them in a separate section, which editors can pretty much ignore.

Being a quantitative guy, the ones that frustrate me the most are complaints about model specifications. Some are legitimately major complaints, but often times it will be things that are highly unlikely to greatly influence the reported results. Examples are adding/dropping/changing a particular control variable and changes in the caliper used for propensity score matching. Note I’m not saying you shouldn’t ask to see differences, I’m asking that you clearly articulate why your suggestion is preferable and make an appropriate judgement as to whether it is a big problem or a little problem. A similar complaint is what information to include in tables or in the main manuscript or appendix. The author already has the information, so it is minor editing, not a major problem.


While I am here I will end with three additional complaints that don’t fit into anywhere previously in my post. One, multiple rounds of review are totally a waste. So the life cycle of the paper-review should be

paper -> review -> editor decision -> reject/accept
                                          or
                                      revise-resumbit -> new paper + responses to reviews -> editor decision

The way the current system works, I have to submit another review after the new paper has been submitted. I rather the editor take the time to see if the authors sufficiently addressed the original complaints, because as a reviewer I am not an unbiased judge of that. So if I say something is important and you retort it is not, what else do you want me to say in my second review! It is the editors job at that point to arbiter disagreements. This then stops the cycle of multiple rounds of review, which have a very large amount of diminishing returns in improving the draft.

This then leads into my second complaint, generally about keeping a civil tone for reviews. In general I don’t care if a reviewer is a bit gruff in the review – it is not personal. But since reviewers have a second go, when I respond I need to keep an uber deferential tone. I don’t think that is really necessary, and I’d rather original authors have similar latitude to be gruff in responses. Reviewers say stupid things all the time (myself included) and you should be allowed to retort that my critique is stupid! (Of course with reasoning as to why it is stupid…)

Finally, I had one reviewer start recently:

This paper is interesting and very well written…I will not focus on the paper’s merits, but instead restrict myself to areas where it can be improved.

The good is necessary to signal to the editor whether a paper should be published. I’ve started to try in my own reviews to include more of the good (which is definately not the norm) and argue why a paper should be published. You can see in my linked review of the Xu and Griffiths paper by the third round I simply gave arguments why the paper should be published, despite a disagreement about the change-point model they reported on in the paper.

Learning to write badly in the social sciences

I recently finished Michael Billig’s book, Learn to Write Badly: How to Succeed in the Social Sciences, and I was largely convinced of Billig’s thesis so I shall reiterate it briefly here. Billig argues that much of social scientific writing is difficult to understand because of the excessive use of nouns instead of verbs. If we (ironically) use the word nominalization to describe the process of turning verbs into nouns, it would sound pretty similar to old hat of avoiding the use of jargon in scientific writing.

Billig goes a bit further though then the usual avoid jargon advice (which is uncontroversial), but gives many examples of where this change from verbs to nouns has negative consequences on how writing is interpreted. Two of these are:

  • Verbs are often much more clear about what actor is performing what actions (and in turning the verb to a noun both the actor and the action can become ambigious)
  • Replacing verbs with nouns gives a false sense of authority that the noun actually exists

The first is important for social scientists because we are pretty much always describing the actions of humans. The second Billig likens to marketing strategies to promote ones work (similar to how advertisements promote products), which I imagine the analogy turns a few academics stomachs.

As an example from my own work, I will use the title to one of my papers, The Moving Home Effect: A Quasi Experiment Assessing Effect of Home Location on the Offence Location. The title is really awful, and in a bit of self-deprecation the few times I presented the work I would make fun of my title making skills at the opening of my talk. The first part of the title before the semi-colon, "The Moving Home Effect", is an example of an ambiguous use of nouns. First, describing my findings as the moving home effect is rather ambiguous, it could mean effecting anything and everything. My particular study is much more restricted, I examined the distance between crimes before and after offenders move. Second, the effect of the added distance between crimes when moving I found to be rather small, which is probably one of the more interesting points of the paper. So saying "The Moving Home Effect" places unwanted emphasis that it exists and is real.

The same exercise can be used for the second part of the title. The use of quasi-experiment is simply econometrics jargon and is unneeded. It is an appeal to associate with a particular camp of analysis, and really does nothing to describe the nature of the work. So, I propose possible rewrites of my title to be:

  • Moving one’s home slightly changes the average distance between offences
  • When an offender moves, it slightly changes the location of where they offend

These are much more descriptive (and shorter) than the original title. Reviewing my own work such examples are rampant, so I have a bit of work to do to live up to Billig’s ideal, but I am convinced it will lead to improved writing. Also it seems to me it is a good exercise to make ones writing more concise, which is always welcome.