When I posted last week about plurals as the first elements of N + N compounds, Jon Lighter added the comment:
When I mentioned to my students in 2002 that “jobs market” had become the preferred cable-news form of “job market,” they seemed not to believe me.
I wonder what today’s reaction would be.
What I wondered about was the basis for Jon’s claim that jobs market was the preferred cable-news variant in 2002.
There’s some history on ADS-L of people (including Jon) making claims about frequencies of a variant (“X is rare”), relative frequences of variants (“X is preferred to Y”), and changes in these (“X is becoming common”) — claims based on impressions. Similar claims are made in lexicographic works, and until recently these were also based on impressions. Granted, impressions of people with a wide experience of texts and examples from texts. But these impressions can be inaccurate, and in some cases demonstrably are. The problem is that they are guided by various unconscious biases, which lead people to various versions of the Frequency Illusion.
As in the case at hand.
Some background: in my posting last week, I wrote:
My earlier posting noted a collection of cases where compounds with Npl first element alternated with compounds with a bare-stem N (interpreted semantically as plural) as first element; alongside the examples cited above [schools chief / superintendent / administrator, comments spam, jobs market] are: school chief / superintendent / administrator, comment spam, job market. And so it is with jobs creation, jobs creators, and jobs growth.
In fact, the jobs market case in my earlier posting was from Lighter, who took it to be “hypercorrect pluralization of attributives”. I wouldn’t automatically label such things as hypercorrections, but instead treat them as motivated by a desire for more semantic transparency in the form of compounds. In any case, alternations between Npl and bare-stem N are very common in certain precincts in the world of compounds.
The nub of the problem is in there: Lighter takes the bare-stem variants to be the correct ones, also the ones he himself would use. As a result, he wouldn’t notice them in what he reads and hears; they’re just what he expects. But he does notice the Npl variants, which annoy him. So it’s entirely possible that his assessment of the relative frequencies of the two variants is an illusion.
His assessment is testable, at least to some approximation; I can’t see any easy way to look at cable news only, but there are some stand-ins.
[What follows is a substantial revision of the material in the first version of this posting. Something was seriously messed up in the Google searches I did several days ago. Thanks to Ben Zimmer for pointing out the problem.]
As background, a Google search gets 26.2m job market raw ghits to 2.57m jobs market, for a ratio of 10.2:1.
Narrowing down to Google News search, I get 37,500 job to 8,790 jobs, for a ratio of 4.3:1. Ben Zimmer got 41,200 job to 9,190 jobs, for a ratio of 4.5:1.
In any case, all these searches substantially favor job market, though jobs market is certainly well attested.
Then: the Google Ngram Viewer tracks things by date, with tighter constraints on sources, and it shows job market sweeping the field: jobs market has an extremely low frequency, scarcely distinguishable from 0, while job market rises from 1940 through 1980, dips a bit and recovers, then begins a decline from 2000. The graph:
Ah, but this is all from books.
Finally, there’s the COCA database, which allows searching over time periods and also restriction to specific genres. I chose 1990-2010, and looked at occurrences in newspapers and occurrences in magazines. The results look much look the Ngram results for this period:
jobs market: newspaper 3 hits, magazine 0
job market: newspaper 357 hits, magazine 222
In all these more constrained searches, job market maintains its supremacy. On the web in general, it’s way ahead of jobs market, even more so than when I did a Google search in 2010 (as reported in my posting on Npl variants), when it was ahead 3:1. (The raw ghit figures are quite variable, but the difference between the two variants seems to be stable.)
In any case, I can’t find any evidence of a news preference for jobs growth in 2002. Jon Lighter was probably noticing all of the (few) instances of it in cable news reports ten years ago. Selective attention is powerful.