Much has been said by a certain Presidential candidate with a poor track record in the caucus states denigrating their value because they are "undemocratic" and not reflective of how a candidate will perform in the general election. The former is laughably false because it is simply democracy of a different form that what is practiced in the general election and the primaries. In fact I would say that any republic, meaning a representative democracy where politicians are elected by district, is also nowhere near as democratic as some alternatives, but I digress. The latter is a far more serious criticism because there is a kernel of truth to it, but it ignores two very important factors: first, the general election will not be a rerun of the primary for any number of reasons - higher turnout and a different opponent being chief among these; and second, the fact that the caucuses are different from the primaries and the general election doesn't mean that they're worthless, it just means that they give you different information about the candidates that ballot box races do. It is when one stops to consider what caucuses do tell you rather than bemoaning what they do not that they expose their true worth.
In the most abstract sense, an election is a process that candidates are forced to go through and the results of that election can tell us something about both the candidate and the people who support her. In short, elections serve are tests, plain and simple. Just like any other test what the results tell us will depend on the form the test takes. For instance, consider the difference between a 100 m dash and a marathon. Superficially both tests are the same thing, but deep down they measure very different things about the athletes who participate in them. The 100 m dash measures a person's explosive power whilst a marathon measures his stamina and ability to pace himself. So, too, primaries and cuacuses are tests that are superficially similar but that tell us very different things about the candidates who are able to win them, and a little consideration will go a long way in helping us figure out what these two processes are good for.
In a primary, candidates ask citizens to cast a ballot for them. The citizens are allowed to cast that ballot in a specific location for a finite amount of time - usually about 12 hours on a given day, but "early voting" and mail in voting are becoming more popular as time goes by. This does not require a large commitment on the part of the voter, and the commitment required is being reduced as time goes by. Because of that, ballot box elections tend to have relatively high voter turnout and so the essential factor in determining who will win is: how many people know about the candidate and like him enough, or dislike his opponent enough, to show up and vote. There are, then, two barriers that a candidate must overcome - the people have to know she exists, and know enough to form a presumedly positive opinion of her. So ballot box elections have become primarily a measure of one's ability to manipulate the media, raise money, and to a lesser extent increase turnout with get out the vote efforts. That is, assuming that the candidate doesn't already have a built in high level of notoriety (ahem). These are primarily skills that help one win the general election, so they are important, but winning elections is not the only job of a candidate. To all but the most blind of zealots, the job the candidate will do once they win the election is probably more important than whether or not she wins the election. In fact, it's safe to say that if the electorate could tell reliably the job a candidate would do beforehand then the election would largely hinge on the job the candidate would do. That's where caucuses come in.
In a caucus more is asked of a citizen in terms of time, timing, and know-how. The voter must show up by a specific time, be willing to commit upwards of a few hours, and know how to caucus. None of these things are trivial matters, and it has often been said that the bulk of a campaign's performance comes down to their ability to build a ground organization beforehand. Doing this requires planning, foresight, and the ability to execute on time. These skills may not be crucial in a general election, but they are among the essential skills of an effective leader. Thus some caucus wins are an essential part of a Presidential resume, in my humble opinion, precisely because it utilizes the skills a candidate will need to have when she is no longer a candidate but an elected official.
That's not all you learn from a caucus in comparison to a ballot box election, though. You also learn something about the candidate's supporters. See, supporters who are willing to make the commitment to show up to a caucus are not voters who will be easily swayed. In other words, a caucus allows you to get some feel for the depth of support a candidate inspires. Now, in a general election that would be pretty meaningless, but in a pre-general election contest that's valuable information. See, a candidate who inspires more commitment from his supporters gives him a base that is not going to be as easily swayed or dissuaded from voting by negative advertisements, lies, and dirty tricks - all hallmarks of a Republican campaign. That itself is valuable information because the general election campaign will be run against a Republican, so it's good to know whose supporters will stick with the party come what may, unlike the so-called "Reagan Democrats." This also means that they're the type of supporters are are not likely to take it lying down if they feel the race were wrongfully stolen from their candidate like Bush stole the election from Gore in 2000.
Now, is a caucus a perfect instrument for measuring these features of a candidate? Well, not really, I must confess. If one were to set out to design a test to measure these things I think a better one could be designed. In comparison to ballot box elections, however, this is what they tell you.
Naturally, there are other useful and objective metrics that are worth considering, even if not all of them are commonly considered: fundraising, number of donors, number of volunteer hours committed (though this might be easy to falsify), general election matchup polls, overall performance in all swing states, etc, but these are qualitatively different because they don't entail actual performance from the candidates in a formal process or are derived from the processes described above.
That said, I would actually like to see a few more different fundamentally democratic ways for the allotment of state delegates if only because the two we use now don't do a sufficient job of plumbing the depth of a candidate's abilities.