Take a look at this blog if you have time. There's a lot of content here about video game design/concepts (specifically horror) that I think you'll find interesting. I've pretty much read every single post and feature already
While I agree with some of what was said above, I think the suggestion that you begin studying general Psych is not the most helpful. Psychology is fascinating and (as was mentioned) you shouldn't worry so much about wasting time reading inapplicable material; however, for the specific purpose of effectively eliciting emotions, I would suggest you focus specifically on studies of emotion (often referred to as affect, likely because this sheds some of the overly broad connotations of emotion) in cognitive psychology, evolutionary psychology, and perhaps philosophy of the mind. Specifically, there are quite a few reasonably well-defined models of affect, derived from various social, evolutionary, neurological, or functional perspectives, and any of these could be applied to help you predict or deduce what design elements might elicit emotion. Human psychology can be unpredictable (which is why, IMHO, designers favor the artist role to the scientist role, after all, artists are really just acting on an internal, possibly informal, model of their audience) but it is also full of quirks and kinks that can be exploited, such as those mentioned by Jeffery above. Finally, although you understandably want to avoid emotional gimmicks, keep in mind that the most broadly effective tools may be necessarily generic. Hope this helps. P.S. I had planned to provide some links for more focused research articles, but my access to Google Scholar and EBSCO has apparently ended, now I'm sad.
Wikipedia has a good overview (as always) of affect in psychology, but unfortunately few web sources: http://en.wikipedia.org/wiki/Affect_(psychology) The linked subtopics Affect Infusion Model and Affect Heuristic both apply, but are perhaps not specific enough to be terribly helpful. Two areas that have been studied a great deal and formalized to the point that you could incorporate them into a game are facial expression and emotional contagion. It seems entirely plausible to intentionally manipulate a player's emotions via game character facial expressions, given high quality rendering. Of course, this wouldn't really be revolutionary, but it may offer a start. In particular, broad, easily interpreted, but highly realistic expressions could go a long way to establishing particular emotions in players.
Build personality templates. Sure, there are limits to everything, but flesh out some templates, even if just /randomed, and have them be set in stone. People don't change that much, especially not NPCs.
You can give them moods as well dependent on time/recent activity in the area/etc.