Originally, I responded to ChemBark's post on the Dorta paper on Facebook. CEN's Carmen Drahl subsequently requested permission to use parts of this comment in a news article on the situation. To clarify my position, I provided Carmen with a (long-winded) comment on elemental analysis. Here, is more or less that commentary edited to be bloggable.
When was the last time you looked at a paper and went to the experimental or SI desperate to see the elemental analysis? When was the last time you bothered to look at it period? I suspect the answer is seldom or never, unless you were trained during a different era of chemistry. EA was the gold standard up until 30-35 years ago, but is it today? I argue no. Many journals still list EA as a requirement (or option), but it's really inconsistent on whether that standard is enforced. I suspect that EA was deemed a more reliable metric of purity than NMR in the past because impurities <5% wouldn't show. I'm not so sure that is as true with modern spectrometers. While it has its own weaknesses, analytical HPLC is a much better technique for organic compounds. Coordination compounds/organometallics may not be as amenable to HPLC, but a combination of things X-ray/NMR/powder diffraction/mp/etc. individually, or in combination can give an adequate or even better indication of purity.
I do believe EA has value if someone is reporting an unexpected or unusual result. An example that immediately comes to mind is a putative HBr complex, a porphyrin ligand from Lippard in 1998. In my recent experience, we had a chelating ligand with 5 donor groups. We got exactly what one would expect in the X-ray. The EA was off by a large margin. It was nonsense. It may not have been 100% pure, but it certainly wasn't half garbage. Hence, I suspected incomplete combustion because the company did not routinely use combustion aid, which we always used for perchlorate/metal complexes when I was a student. Fortunately, the HRMS detected the molecular ion, and that deemed a sufficient form of additional characterization. There was nothing unusual or unexpected about this complex, but we needed to meet the journal requirements.
Back to the original point, of "who looks at EA?" (and perhaps why was it deemed "acceptable" to make up numbers?). I think everyone looks at EA as a major hassle because they have had issues getting data. Of the 5 universities I've been at (CWRU, MIT, Berkeley, UConn and WPI), only Berkeley had on-site EA, which may indirectly indicate how valuable the community finds EA. This means getting analysis required sending out a sample. This becomes even more non-trivial with sensitive compounds. Even if the compounds aren't expected to be sensitive, an off analysis always makes one suspicious of decomp in transit, or a problem with the contracted lab. Then there's sending multiple samples trying to get a good analysis, it can become very tedious. Furthermore, I suspect a large percentage of people who report an EA have used the trick of including solvent in the calculation to make the analysis hit. I had to do this with my first sample at MIT. We must have analyzed dozens of crystals from many batches of one complex by X-ray (it made beautiful uniform, individual xtals). It was always the same thing. We thoroughly dried the complex, but the analysis wouldn't hit after many attempts. That's when we modeled in the water (fortunately water was in the crystal lattice so this was a reasonable assumption). I suspect others have done the same, and perhaps even added in solvents that weren't seen in the X-ray structure (but were in the synthesis) to make their data fit. If you're essentially massaging the data because it's hard to get what you expect, or you're jumping through so many hoops to get a hit (with what you are 99% sure is a pure complex/compound), it doesn't impart much faith in the technique.
In the end, I prefer alternatives to EA whenever possible. If I were to project a on Dorta, I doubt he's really trying to pull one over on the journal/community with bad data. He's looking at the inconvenience of getting the data, and perhaps has little/no faith in the technique but does have confidence that the complexes/chemistry is correct (i.e. does not believe it warrants the hassle of adding more confirmation). It's not the right way to do science, but somewhat understandable. I think there is an opportunity here to talk about reasonable standards, the review process, as well as scientific integrity. You really can't prevent falsifying data for purity, almost every technique has a workaround that would not be detectable without close scrutiny or even having to reproduce the author's work.