This is a great question. I'm not a cognitive psychologist by training, but I can offer a usability perspective on this from a person who is a trained cognitive psychologist. Jakob Nielsen recommends using icons when users don't have to guess what they represent. If the icon is vague, there's no additional benefit to including an icon with your label. It might actually confuse people. People aren't going to understand the meaning of your icon as well as you probably assume. This is because you have an expert bias in choosing your icon.
It doesn't sound like you're creating an interface. You say you'd like to visualize data using graphics like smiley faces. Emoticons (smiley faces) are the one exception to this icon rule. People immediately recognize emotions and faces. Our brains are hardwired to turn anything into a face. Every car has a face and an emotion, for example. The first thing you recognize when you look at a picture subconsciously will be the expression on everybody's face. In eyetracking studies, the first thing people notice in pictures is faces.
This is getting really deep, but on the subject of smiley faces, the more simplified the face, the more we project ourselves into the character. The character can be drawn very realistically, and it looks like a photographic picture. In this case, you are an observer looking at the character. Or the character can be less detailed. In this case, you become the cartoon character. This is why cartoons are so universal and permeate our culture.
Your representation of a face in your mind is like a cartoon. You don't see yourself wearing a photographic picture of your face. When you're smiling to someone, you see yourself with an upturned mouth. You see yourself wearing a very simplified smiley face. That's the abstract idea of your appearance in your mind. It's like looking at a bike. You don't see two wheels, a frame, and so on. In your mind, you see the abstract concept of "bike."
The more simplified the cartoon, the more we identify with it. This is demonstrated well in japanese comics. When the artist wants a character to seem otherly or distant from the reader, he draws them very detailed.
So if you want people to feel a certain way about your data, they will automatically empathize with the emotion of the cartoon characters you draw. Hope all this explanation answered your question and was relevant. ;)
If you want to dive deep into data visualization, buy any book by Edward Tufte. Envisioning Information was awakening for me. The Quantitative Display of Information might be a better place for you to start. The graphics aren't as colorful and pretty, but it really gets into data visualization. Tufte's books are seminal.