Prof. Geoffrey Pullum (Language Log) has a post that strikes me as so apt that it’s worth quoting at length, though with some extra emphasis on my part (see the original for the links, reader comments, and some more text):
See Plethoric Pundigrions for screen shots showing a version of Microsoft Word (I don’t know which one) that for levelheaded suggests correcting it to level-headed and for level-headed suggests correcting it to levelheaded. That should give rise to a frustrating morning of trying to finalize the draft, shouldn’t it?
You will probably want to know what The Cambridge Grammar of the English Language says about what the right answer is; and those who yearn not just for authority but for actual authoritarianism will be disappointed that it reports, “It is an area where we find a great deal of variation” (p. 1760, in the section on lexical hyphens).
If you think that nonetheless an answer should be stipulated, then go ahead and make up a stipulation. What The Cambridge Grammar is telling you is that you won’t have any basis for it. You might just as well have stipulated the opposite. Educated usage will not always match your stipulation (thus showing it to be a good one), and it won’t always fail to match.
There are two general tendencies, though. (1) The longer a compound has been in use, the more likely it is to have started being written without a hyphen. (2) American English is a bit less likely to favor hyphenating than British English is. Apart from those two rules of thumb, you are out there working with no net, trying to follow the shifting tendencies in the usage of other people. I think I would recommend simply finding a recent use of the term in the writing of an author whose work you really like to read, and following that. If Stephen King describes someone as levelheaded, and you like reading Stephen King, then write levelheaded. Nothing much will hang on it. Not everyone will agree with you (and Word may even disagree with itself), but hey, it’s a free country.
Does that make me sound like an anarchist? I hope not. I believe there are thousands of quite strict constraints on Standard English, constraints such that if you would be ill-advised to violate because you will look like a gormless illiterate. All I’m saying is that whether or not to hyphenate a compound like level(-)headed is not one of the areas of English in which a strict and widely respected constraint holds.
This strikes me as sound advice on this issue, but I particularly like two broader points:
1. Descriptivism stems in large part — it certainly does for me — not from some ideology about freedom or populism, but in the insistence that proposed rules have some actual foundation in something. I’m happy to be strict about the rules in math, or in physics, or for that matter in law, because there’s some basis for the rules. In law, that’s the judgment of some authoritative body (whether a court, a legislature, a constitutional convention, or the people voting for a law or a constitutional provision). In physics, that basis lies in the real world. In math, that basis comes in a combination of the real world and some conventions that are demonstrably more convenient for important mathematical principles, and that are therefore widely adopted by the great bulk of mathematicians.
But in the English language, the only such basis that I can see is the consistent pattern of usage (perhaps focusing on edited usage, just to set aside mistyping and similar things that even the user would on a moment’s reflection recognize as an error). If the great bulk of English speakers and writers say or write something, I simply see no basis for saying that it’s “against the rules.” You can say that it’s inelegant (an aesthetic judgment), or you can say that it’s potentially confusing, or you can say that it will alienate some readers, and you might well be right. But I just can’t see a basis for saying that it’s “incorrect.” (For a response to the argument that a common usage can be soundly labeled incorrect because it’s “illogical,” see here.)
2. Yet this hardly means that descriptivists don’t believe that there are any rules. There are lots of rules. Here’s one: Following “I am,” one says words like “eating” and not “eat.” No native adult speaker violates this rule; I’ve never heard even nonnative speakers violate it, nor have I heard uneducated speakers violate it.
Even my 4-year-old and my 6-year-old, who have trouble with other rules, have not to my recollection ever said “I am eat,” at least since they started speaking in complete sentences. This might make the rule sound trivial, because we don’t need to apply it while editing. But it actually means that the rule is extraordinarily strong, precisely because people almost never violate it.
Likewise, there are rules about the meanings of words — for every controversy about what a word “means,” which usually reflects the reality that the word has multiple meanings in common English usage, there are hundreds of entirely uncontroversial definitions. Each such definition is a rule. The same is true for rules about spelling, pronunciation, and so on.
So we descriptivists care a great deal about rules, and think it’s important that people follow them. We just insist that those who assert the validity of a rule have some evidence for that assertion. And the only evidence that makes sense to us (or at least to me) is Horace’s “will of custom, in whose power is the decision and right and standard of language.”