My only guess (and that's all I can do) is that most of the structures already established for hg18 would have to be updated to hg19, and sometimes that takes a while to do.

One answer is that there is a lot more annotation for the HG18 build, because it has been around for 4 years versus just over 1 year. A lot of people use the UCSC genome website where, for example, the HG19 build has just one regulation track (CpG islands), whereas HG18 has ~ 34 tracks for regulation: histone modifications, transcriptions factors and so on.

I think HG18 is seen as the "stable reference", more by virtue of its age than anything else. I'm sure HG19 will gradually catch up as more annotations are added. We might ask why this process has not become quicker in the intervening 3-4 years.

Here is my thoughts:

Always a reference dataset is required for the analysis. It is a general practice to freeze the dataset for analysis with stable version of available dataset at the time of first round of analysis. As suggested by Paulo & Chris changing the version requires considerable efforts. But the downside is we may end up doing analysis on a gene structure that may no longer exist or a transcript merged with another one. Even if it is takes considerable effort to switch, it is best to do the analysis on updated version of genome.

We provide alignment (mapping) software and services, and we just now are getting more requests for hg19 than hg18, so I can now say hg19 is finally "catching on" (after a year or more ?)

www.imagenix.com