So, this is going to be bad news for Haskell programmers and good news for people who want to hire them. Based on the Stack Overflow Developer Survey 2015 I was able to get rough salary figures for programmers in the U.S. I assumed a log-normal distribution for the upper half of the distribution, and that distribution probably applies to the lower half once you take out part-time workers, students, and low-end freelancers. Because of reporting bias and the discretization, it was hard to get a sense of any distribution over the low end: my methodology was to linear-regress log(salary) to z-score at each of the bucket points, throwing out non-responses and "I'd rather not say". I also don't have a strong sense of the high end because the top bucket was "over $160,000", which was 8% of U.S. programmers.
The median programmer makes about $87,800 and the multiplicative standard deviation is 52.2%. So we'd expect a programmer who's 1 standard deviation above the mean (85th percentile) to make about $134,000. At 2 standard deviations (98th percentile) we'd expect to see $203,000. This was remarkably consistent across languages (except Scala). Here are some salary estimates for a few languages:
Language 25th 50th 75th 95th 99th
All 66 88 117 175 233
Java 65 88 118 182 246
Haskell 61 86 120 193 269
Scala 90 114 143 200 253
I'd be wary of taking the 99th-percentile estimates too seriously. Those were extrapolated (the regression I described above) because the survey maxed out at $160,000 per year.
I also analyzed Python, Ruby and a few others but they didn't differ substantially from Java's numbers. In fact, the only one that had a significant departure from the general programming population was Scala. The rest all had the same salary curve on the high end and low.
Here's the thing, though. I've spent enough time working with, interviewing, and hiring programmers to know that those curves shouldn't be equal. A 50th percentile Haskell engineer would be 80th percentile in the Scala community and 95th percentile in the Java community. Assuming these data are accurate, and assuming that the salary distribution maps to the competence distribution (which we know to be false in the real world, but gives us a first-order starting sense of what salaries "should be") then the median Haskell engineer is a $182,000-per-year engineer at $86,000: a 53% discount. Of course, we know that salary reflects more than just engineering competence (for one, negotiation skill is a huge factor) and I tend to have some distrust of self-reported data, so the actual discrepancy is probably closer to 20%. Among people with a year or more of production Haskell experience, I'd guess that the median salary's closer to $125,000 per year (which is still low, considering that equivalent Java engineers make much more).
As for what Haskell engineers "should" make, that gets subjective and complicated. I'd guess that you can probably do as much with a team of 5 Haskellers as with 100 average Java engineers or with 15 good (95th-percentile) Java engineers (because even good engineers will be held back by the language). This would theoretically justify a salary of either $546,000 (3x good Java engineer) or $1.76 million (20x median Java) but I wouldn't hold my breath for that to happen. Realistically, if you're hiring programmers, you shouldn't have trouble hiring a capable Haskell engineer, even in the Bay Area, at standard salary levels.