Another nail in the ATEOTD=manager coffin

« previous post | next post »

Some people are hard to persuade. In response to my post "'At the end of the day' not management-speak", Peter Taylor commented:

I argue that the first question to ask is whether hearing someone use the phrase "At the end of the day" conveys information on whether they are likely to be a manager…

Well, a definitive determination of the information gain involved, aside from its limited general interest, would require more resources than I can bring to bear over my morning coffee. But we can make a plausible guess, and the answer turns out to be that the "information gain" is probably pretty small, and is just about as likely to point away from the conclusion that the speaker or writer is a manager as towards it.

The information gain associated with an observation y is

In the current case, this cashes out as the probability that someone is a manager given that we've added to our background knowledge the fact that they said or wrote "at the end of the day"  — call it p(manager | ATEOTD,I) — multiplied by the base2-log of the ratio between this same term p(manager | ATEOTD,I) and P(manager | I), which is the probability that they're a manager given only our background knowledge I. We should sum this quantity over the alternatives in the distribution under consideration, here x=manager and x=not-manager.

Based on the distribution of this phrase in Mark Davies' COCA corpus, p(manager | ATEOTD,I) seems to be about one in a hundred, 0.01. (In other words, out of a hundred speakers or writers of the phrase "at the end of the day" in its figurative meaning, one is a manager, on average. See below for details.)

What about the probability that someone is a manager given only the background distribution, symbolized p(manager,I)? We don't want to use Census Bureau statistics, since being the source of speech or text in the COCA sample is already going to be highly skewed relative to the occupational distribution of the general American population, regardless of what phrases are used.

Before even trying to estimate this quantity, we can look at what the information gain would turn out to be, for a variety of values for this background proportion of managers. If one in a hundred sources of speech or text in COCA is a "manager", then the information gain is 0.01*log(0.01/0.01) + 0.99*log(0.99/0.99) = 0 bits. If it's one in a thousand, the information gain is 0.01*log(0.01/0.001) + 0.99*log(0.99/0.999) = 0.02 bits.

But I'd be very surprised if the proportion of "managers" in the COCA sample was even as low as one in a thousand — and it might very well be more than one in a hundred, not less.

If we search COCA for the phrase "in the final analysis", for example, the first 100 hits include four clear examples of a managerial source, and two other marginal ones. If the actual value of p(manager) is actually .04, then the information gain associated with saying or writing "at the end of the day" will be 0.01*log2(0.01/0.04) + 0.99*log2(0.99/0.96) = 0.024 bits, but tending towards the conclusion that x=not-manager.  Thus if this estimate of p(manager | I) is accurate, use of "at the end of the day" is a small piece of evidence against the hypothesis that the speaker or writer is a "manager".

Unfortunately, I don't see a good  way to get a representative random sample of COCA sources, which we would need in order to estimate p(manager | I) properly. But I invite readers to try to sharpen up these estimates, if they're interested — I've satisfied myself that saying or writing "at the end of the day" provides no useful evidence that someone is a "manager", and may in fact count as a small piece of evidence in the opposite direction.

And for those social scientists still reading this (if any), let me point out that setting up to do this sort of analysis in a more systematic way would provide an interesting laboratory for investigating the quantitative relationship between stereotypes and reality.  The comments on earlier posts in this series make it clear that many Americans are absolutely convinced that "at the end of the day" is manager-speak.  My guess remains that most linguistic peeves associated with despised groups will turn out to be similarly unsupported by evidence.


In Mark Davies' COCA corpus, there are 2,438 examples of "at the end of the day' in 400 million words, for an overall frequency of 6.1 per million words. The frequency is greater in "spoken" material (basically news interviews) than in other genres:

Size (MW) Freq Freq per MW
SPOKEN 81.7 1078 13.2
FICTION 78.8 281 3.6
MAGAZINE 83.3 518 6.2
NEWSPAPER 79.4 409 5.2
ACADEMIC 79.3 135 1.7

And the rate of use has nearly tripled over the past couple of decades:

Size (MW) Freq Freq per MW
1990-1994 103.3 355 3.4
1995-1999 102.9 484 4.7
2000-2004 102.6 689 6.7
2005-2009 93.6 910 9.7

Checking the first (most recent) 200 COCA hits for this phrase, I determined that 51 of them were literal references to the end of daylight, or the end of a working day, or the end of a 24-hour period. That left 149 figurative uses, meaning something like "in the final analysis". Of these, just one was spoken or written by someone I would call a "manager", namely the spokesperson for a manufacturer of sex toys (Jessica Rae Patton, "Make Love, Not Waste: Bringing Environmentalism into the Bedroom", E: the Environmental Magazine, Sep/Oct 2008):

Day says, " We are working on reduction by offering products in larger quantities–lubricant in a 16-ounce bottle, for instance. Dildos that are glass or wood… will eventually go back to the earth, and if used as they're meant to be used, will last a very long time. " The store no longer carries products containing phthalates. " We offer a huge selection of rechargeable vibrators, " she says, but acknowledges, " At the end of the day, it is still a manufactured product that will eventually end up in the dump. That's the grim reality. " Day notes that the adult product industry hasn't yet figured out how to address this waste. " It's only a matter of time before that person comes forward who figures out how to recycle sex toys. Trust me, every company in the adult industry will use that service! " she says.

The only other source that was close to being a "manager" was a fashion designer ("Hottest, Newest, Latest", Harpers Bazaar, June 2009:

His strong, architectural silhouettes come together to compose a collection of 21 looks, including everything from a modernized interpretation of a tuxedo to a feminine white blouse made of frothy embellished flowers (at right) to exquisite column evening gowns. All in a primary palette of red, white, and black (with some touches of fur detailing), the pieces will be sold at such stores as Bloomingdale's in New York and DNA in Saudi Arabia. " At the end of the day, I didn't know what the reaction would be, but I'm a firm believer that if you do something with pure integrity, you'll find an audience, " Gurung says. Well, this audience is still applauding.

So depending on how you count, we get an estimate of 1 or 2 in 149, or about 0.0067 to 0.013 — let's call it one in a hundred, 0.01.

The other sources for ATEOTD in this sample are pretty diverse — a TV journalist, a rescue hero, a basketball player, a rapper, a country singer, a primatologist, and so on:

Let me press down on that. At the end of the day, are you really talking about over the course of your presidency, some kind of a grand bargain?

Mr-COLLIER: At the end of the day it worked for us and we did what we had to do. Mr-ELLIS: Having got those people off were, particularly in this case, nobody else could have gotten them out. Mr-COLLIER: It' s very satisfying.

Yeah, I mean, Shaq, you know, Kobe does really recognize that Shaq helped him to get three titles and Shaq got another title on his own without Kobe, but at the end of the day, both of them realized that they missed out on opportunities to do something special and – you know, when you' re a little bit younger, you' re a little bit immature and then when you get old and wiser, you reflect on things that would' ve – could have happened.

The rest of Relapse is even more grim. Many of Eminem's new songs depict his drug years in terms that seem to alternate between raw honesty and wild hyperbole. And though rumors have spread that his estranged and reportedly ailing mother, Debbie Nelson, is eager for a reconciliation, a song titled " My Mom " takes aim at her as viciously as ever. (" Don't get me wrong, " he said during last week's Sirius XM interview. " At the end of the day, she is my mother and I do love her. ")

When all this role-playing is over, the wife of country legend Tim McGraw and mother of three girls (Gracie, 12, Maggie, 10, and Audrey, 7) has no problem snapping back to reality. She washes her face, pulls her hair into a ponytail, and slips back into those sweats. Has she discovered anything from stepping into such glamorous shoes? " At the end of the day, for me beauty has a lot to do with comfort and being around my family. And, " she says with a laugh, " most likely no makeup! " Then, with a big hug and a cheery " Thank you! " to everyone on set, she's off to meet Tim at the girls' school for a basketball game.

" It's hard to say what exactly precipitated this behavior, " said Colleen McCann, a primatologist at the Bronx Zoo. " At the end of the day, they are not human and you can't always predict their behavior and how they or any other wild animal will respond when they feel threatened. "

As for "in the final analysis", the four manager-sources in the first 100 hits were:

("Constellation calls off deal with Buffett unit", Business News, Dec. 2008) " In the final analysis, we concluded that the EdF investment represents significant enhanced value for our shareholders and serves the best interest as well of our customers, our employees, our regulators and the communities we serve, " Mayo A. Shattuck III, chairman, president and chief executive of Constellation Energy, told analysts on a conference call.

("Are you paying yourself enough?", Inc magazine, Nov. 2004) And that's something to keep in mind. In the final analysis, says Driskill, whatever you don't take from your company today should eventually come back to you. " Really, when I'm ready to retire, the compensation issue becomes moot, " she says, " because theoretically I will sell the company back to my employees. That's the nice thing — it all becomes your money in the end. "

(" In U.S. Plants and Wallets, The Other Iraq Standoff ", WaPo, Feb. 2003) " The Gulf War triggered a relief trade in equity markets and a brief surge in consumer business confidence, but in the final analysis, we didn't have a normal, self-sustaining recovery until' 93, " said David Rosenberg, chief North American economist at Merrill Lynch &; Co. in New York.

(" An Experimental Examination of Information Technology and Compensation Structure Complementarities in an Expert System Context. ", Journal of Information Systems, Spring 2003) Prendergast (1999) suggests that about one-third of the increase in performance attributable to PC incentives arises from attracting more skilled workers, with the remainder attributable to increased effort. In the final analysis, it is the combination of skill and effort that leads to task performance.


  1. David Cantor said,

    September 28, 2009 @ 7:56 am

    In reading many of the posts from those peeved by this phrase, I get a clear sense that they make an important underlying assumption: that managers are more likely to commit the crime of content-free speech than others.

    I confess that I am myself a manager, but that also means that I have many opportunities to observe both managers and non-managers each day. I'd say that neither group has a monopoly on vacuous speech.

  2. Peter Taylor said,

    September 28, 2009 @ 9:28 am

    The first thing of which I'm unconvinced is that you're asking the right question. I thought that was fairly clear in my previous post, but maybe not.

    The second thing of which I am unconvinced is that you're analysing your data correctly. That's where I joined the discussion on the previous post. This post first raises eyebrows with the suggestion that information gain can be negative. I think the reason you think that is that you obtained a negative result with an estimated proportion of managers in COCA of >0.01, but your calculations are wrong because they neglect the capital sigma in the equation you posted.

    [(myl) Oops, that's right. KL divergence (the fancy name for "information gain" is a function of two probability distribution, and indeed in this case, the whole distribution is p(manager) and p(not-manager), so I need to add up both terms. I believe that I've fixed the post to reflect this.]

    Suppose we accept your figure of p(manager | ATEOTD,I) ~ 0.01. That gives p(not manager | ATEOTD, I) ~ 0.99. If p(manager | I) is one in a thousand then the information gain isn't 0.033 bits but 0.01*log(0.01/0.001) + 0.99*log(0.99/0.999) = 0.020 bits. If p(manager | I) is 0.3 then the information gain is 0.45 bits (someone using the phrase gives you a lot of information suggesting that they're not a manager).

    But that figure of 0.01 is very imprecise. If you're aiming for statistical significance and you get 1 or 2 elements in one category then you should throw the data away and take a fresh measurement with ten times as many samples. It also fails to give any assurance that the 149 samples are all from different people, which is a rather key element of the figure estimating what it is claimed to estimate.

    [(myl) It's true that all the figures here are very imprecise. But we're doing a sort of back-of-the-envelope calculation, to try to get an idea of how a proper experiment would come out. In this sample, we got one out of 150 for p(manager | "at the end of the day") — which is supposed to be a manager-associated locution — versus four out of 100 for p(manager | "in the final analysis"), which is not supposed to be a manager-associated locution. This doesn't look good for the ATEOTD-means-manager hypothesis.]

    There is also the question of what would be a significant gain in information. Coming back to my first point about asking the right questions: 'whether hearing someone use the phrase "At the end of the day" conveys information on whether they are likely to be a manager' corresponds to the null hypothesis that the information gain is zero. It's pointless estimating the information gain if you don't also estimate a threshold for rejecting the null hypothesis at some reasonable confidence interval.

    [(myl) In my opinion, it's not very interesting to know whether the effect is statistically significant — the real question is whether it's what a biomedical researcher would call "clinically significant", or what we might call here "socially significant". That is, is there a big enough effect in the right direction to justify the stereotype? On the basis of what we've seen so far, I'd be willing to place a fairly large bet that (for instance) a complete analysis of the relevant aspects of COCA would not show this.]

    To wrap up, you have certainly proven that "at the end of the day" isn't used exclusively by managers. I don't think you've proven anything stronger, and I continue to believe that there isn't suitable data available to perform a breakfast experiment which could give stronger results.

    [(myl) With respect, I think these observations show something quite a bit stronger about this phrase than that it "isn't used exclusively by managers" — they shown that in a large and diverse sample of contemporary American text and speech, instances of this phrase are overwhelmingly more likely to be produced by non-managers than by managers.]

  3. Dan T. said,

    September 28, 2009 @ 10:58 am

    In addition to managers, marketers are known for buzzword-laden, content-free speech, as satired here in the marketing manager's statement. (He's a marketer and a manager, so you'll really get some buzzwords there!)

  4. Acilius said,

    September 28, 2009 @ 11:01 am

    Mark, I read all of your posts on the question "Is expression X an example of the jargon of sphere Y?" because I keep hoping that you will explain what you think the word "jargon" means.

    You keep proceeding as if an expression which is familiar in speech or writing outside sphere Y could not be jargon within that sphere. That would be a plausible way of handling the problem if a "jargon" were simply a list of words and phrases peculiar to a certain group of language users. It would then be reasonable to dismiss the question of how particular audiences (in the case of alleged management-speak, employees) respond to particular locutions (here, "at the end of the day") when particular speakers (the audiences' bosses) produce those locutions.

    If on the other hand we consider a jargon not as a list we might find in a book, but as a form of social interaction, we would have to take an entirely different approach to the question, "Is such-and-such an expression part of the jargon of such-and-such a group?" If jargon is a way we use language to signal our membership in certain groups, our right to express certain kinds of opinions, and our expectation that others will respond to us in certain ways, then we that question is one we can answer only by doing a qualitative field study. It is precisely how particular audiences react when particular speakers produce the locution that we must know if we are to know whether the locution is part of the jargon that sets the speaker apart from his or her listeners. We would actually have to go out and interview office workers and find out what it means to them when their bosses say "At the end of the day" and whether react to it differently than they would if they heard a soccer coach use the same phrase on television, or someone else in some other setting.

  5. JC Dill said,

    September 28, 2009 @ 1:58 pm

    I think you are being misled by attempting to use a written database (Google) to determine all uses of this phrase. It's a bit like trying to determine how often people say "um" by searching on Google – "um" is vocalized far more often than it is written.

    [(myl) The data in this post is from COCA, not from Google; about a third of the examples come from conversational transcripts; and many of the remaining examples are from interview quotations. And filled pauses are often edited out of transcripts and quotations, but whole phrases like "at the end of the day" are not.]

    Managers use the phrase "at the end of the day" speaking off-the-cuff in meetings far more often than they write it down in papers or presentations. This is what made the Dilbert strip so funny – it happens just as he wrote it, every day, in meeting rooms all over the US.

    [(myl) Right, just like women are always gabbing, and jews are always looking for discounts, and irishmen are always drunk, and so on through all the other evocative stereotypes. The jokes are there, and the laughs are there; the only thing that's missing is the evidence.]

    A completely unrelated question – is there a way to contact Language Log contributors (as a group) when we see something and think "This is one for Language Log!"?

  6. JC Dill said,

    September 28, 2009 @ 4:23 pm

    Mark, you said:

    [red][(myl) The data in this post is from COCA, not from Google; about a third of the examples come from conversational transcripts; and many of the remaining examples are from interview quotations. And filled pauses are often edited out of transcripts and quotations, but whole phrases like "at the end of the day" are not.][/red]

    None of these are from *meetings*. Meetings have their own special language – that's what prompted the creation of buzzword bingo. When was the last time you heard someone say "low hanging fruit" or "how do we monetize this?" or "feature creep" or "incentivize" in a meeting? Have you ever actually played buzzword bingo at a meeting?

    [(myl) Point me to some meeting transcripts of the kind you have in mind, and I'll analyze them.

    You could be right, that managers speak in a special way in private meetings behind closed doors, and none of this way of talking leaks out into their writings or their interviews or their public presentations. This was certainly not the sense that I got from 15 years in an industrial research lab, nor is it the sense that I get from 20 years of meetings with academic staff at various levels. Still, you might be right — but wouldn't it be nice to have even a tiny piece of evidence to back up your opinion?]

  7. Stephen Jones said,

    September 28, 2009 @ 5:54 pm

    One of the reasons for the apparently skewed connections is, I believe, the fact that non-managers experience of management speaking disproportionately contains them speaking at meetings. And at meetings, particularly 'inspirational' ones, content-free discourse is much more likely than in other situations.

  8. Mark F said,

    September 28, 2009 @ 6:13 pm

    I realize that this doesn't speak directly to the question at hand, but I still don't see what is vacuous about the phrase itself. I assume that people don't think it's more vacuous than "when all is said and done", or "in the end" (when used in the same context), or other similar phrases. Right? Does anybody think it is more vacuous than those phrases?

    Perhaps there is something to be said against relying too heavily on newly-fashionable figures of speech, as a matter of style. But style is always hard to get right.

  9. Rubrick said,

    September 28, 2009 @ 6:27 pm

    [(myl) Point me to some meeting transcripts of the kind you have in mind, and I'll analyze them.]

    I think this is the whole problem, and the reason this debate will rage on: corporate secrecy being what it is, there simply isn't a corpus of such transcripts to analyze. So it's the (untrustworthy) gut instincts of one set of people regarding one set of data versus actual statistical analysis of an entirely different set of data.

    I think saying "My conclusions carry the day because I have data to analyze and you don't" is a bit unfair.

  10. J. W. Brewer said,

    September 28, 2009 @ 6:45 pm

    I, for one, am now fully convinced that ATEOTD is, in isolation, not distinctively characteristic of the managerial subset of Anglophones — although stepping back I'm not sure if Dilbert was asserting that it was (as opposed to asserting that a sentence composed of a string of three cliches is distinctively managerial even if none of the individual component cliches are distinctively managerial on a freestanding basis).

    But I'm not sure about the broader claim that peeves about the language use of "despised" groups are inherently likely to be empirically bogus in their descriptive assumptions (as constrasted with the negative inferences drawn therefrom). Giving managers a break, three groups of English-speaking Americans who have been from time to time despised are Jews, blacks, and Southerners, all of whom are generally said in the linguistics literature to exhibit various distinctive sorts of language usage, whether of pronunciation, lexicon, or syntax. It is presumably not the case that all of those perceived distinctive features are illusory and caused by social prejudice combined with confirmation bias. Is it only the perceived features of the language use of those groups that are peeved about that are unlikely to be empirically true? Which would those be? Surely if a group is sufficiently despised, peeves will attach themselves even to objectively neutral (but empirically accurate) distinctive features of its language use. Probably pretty much every distinctive feature of Southern speech has at one time or another been used as evidence of the general claim that Southerners are dumb hicks. Similarly, probably pretty much every distinctive feature of BVE/AAVE has been stigmatized as incorrect or ungrammatical. But I have never, for example, seen a defender of BVE's basic legitimacy as a variety of English claim that there was no empirical basis for a black/white difference in the prevalence of "ax" as a (peevologically-targeted) variant pronunciation of "ask", as opposed to denying the validity of the inference that employing that variant pronunciation was evidence of ignorance or illiteracy or what have you.

    And some regional/dialect differences can cut both ways in the hands of stereotype-wielding peevesters: your group talks too slowly, evidencing that you are slow-witted — no, your group talks too fast, evidencing that you are rude, brusque, and impatient. The possible peevological use of such differences does not lead me to assume that the null hypothesis should be that there are no differences in average words-per-minute among different varieties of English.

  11. Acilius said,

    September 29, 2009 @ 11:34 am

    @Rubrick: "I think saying "My conclusions carry the day because I have data to analyze and you don't" is a bit unfair." Well, we can always wait for another day, when someone else will bring data. So in that sense I agree with you. Surely, however, in a contest between data and no data, data wins. Unless, of course, the data are irrelevant to the question at hand, as I believe Mark's is in this case.

  12. Acilius said,

    September 29, 2009 @ 3:17 pm

    Er, emend that last bit to "as I believe Mark's to be in this case."

  13. D.O. said,

    September 29, 2009 @ 6:32 pm

    If I understand Acilius, he is for recognizing popular right to be annoyed. Folks have a right to be annoyed if managers use particular turn of phrase frequently, or use it frequently specifically in conversations, or use it during meetings, or use it in meetings with non-managerial staff, or do not use it that frequently after all in either circumstances. People just have the right to be annoyed. Even I have a right to be annoyed though I do not go to meetings with managers and, when I did, I cannot recall a single use of ATEOTD. Come on, they are managers, what good can they do?

  14. Bloix said,

    September 29, 2009 @ 9:57 pm

    As long as we're reading the funnies, how about:

  15. Faldone said,

    September 29, 2009 @ 11:17 pm

    Sure. People have the right to be annoyed, but do they have to go whining about it where more tolerant people are inundated by their whining? I mean, I don't like celery. In fact, I abhor celery but I don't go running off to recipe websites crying that people who use celery in their recipes are all a bunch of ignorant louts. I'm with Mark F here. Cliches are easily understood and don't require some sort of new inspiration every time you want to say something that has been said before.

  16. Acilius said,

    September 30, 2009 @ 8:45 am

    @DO: What I'm saying is that when people label a locution as jargon, they aren't always complaining that the expressions are unfamiliar. Instead, they often mean that the speaker is using that locution to mark himself or herself as part of a privileged group which has some claim to make on them. So, you might imagine employees complaining that their boss uses "at the end of the day" to send the message "because I outrank you, I am closing this discussion." The speech act is what usually bothers people, not the relative frequency of the expression in various speech communities.

    That's why I say that the question "Is locution X an example of jargon?" is not one that can be settled by corpus analysis, but requires field work.

  17. Dan Lufkin said,

    September 30, 2009 @ 5:30 pm

    Just wanna report an impressive field sighting of ATEOTD. Sen. Saxby Chambliss (R, GA) appeared on the McNeil-Lehrer Report last evening and used ATEOTD five or six times in a couple of minutes of discourse (on the evils of a public option in the health-care bill). This may well represent the record for ATEOTD saturation. Of course, the senator is a professional, so the amateur record may still be in contention.

  18. mae said,

    October 1, 2009 @ 12:00 pm

    I just went to Q&A session by a low-level diplomat. He used "at the end of the day" in the answer to every (really, every) question. Sometimes more than once. I think he was being evasive most of the time — maybe it's a sign of evasiveness?

  19. Doug S. said,

    October 3, 2009 @ 4:49 am

    Sorry, I couldn't resist…

    At the end of the day, you're another day older…

RSS feed for comments on this post