Tufte defines Sparklines as “intense, simple, word-sized graphics”, but I think the “word-sized” part of that definition is probably overly restrictive. More important is the idea that graphics have a very high representational value that’s sustained even if the reader doesn’t absorb 100% of the data being presented. I don’t have to know the meaning of every data point if the overall graphic communicates the point the author is trying to make.
In short, they’re data-dense but somewhat deliberately opaque about the data sources which informed their creation. The liberating constraint placed on the graphics is that it’d be impossible to provide a key detailing each item in the space provided, so the reader is freed from the burden of having to know what each point means: All forest, no trees.
Sparklines would be good for use on mobiles given the limited display area that is available.