Regarding articles without abstracts that have abstract counts: all articles do have a landing page, which will include the abstract if one is available (and in which case will be linked to from the article's title in the ToC). This landing page is still available if there isn't an abstract to be read -- it's used for example as a landing page by Google Scholar and other indexing services, as it's the only page that uniquely identifies the article itself, but also provides links to all galley filetypes.
Take http://pkp.sfu.ca/ojs/demo/present/inde ... emojournal
for example. You'll see that the article at the bottom of the ToC doesn't include a link to an abstract -- there is no 'abstract' to link to. However, there is still a landing page: http://pkp.sfu.ca/ojs/demo/present/inde ... cle/view/5
. This page lists relevant article information (including information in the sidebar Reading Tools), links to all article galleys, etc., and is crawlable and visitable from search and indexing engines.
So that's one reason why you might be seeing some discrepancies, and abstract counts where abstracts shouldn't technically be counted. If there is a consensus as to whether this should be considered an abstract view count or not, I'd be happy to file a bug report.
Additionally, WRT comparing with COUNTER stats: COUNTER counts things a little differently -- most important, the protocol specifies certain situations where view counts shouldn't be recorded (such as more than x views per second). I don't recall the rules offhand, but if you'd like more info either check out the COUNTER website
or let me know.