Algebraic spaces are non-stacky algebraic (Artin) stacks, and DM-stacks, although stacky, have only finite stabilizer groups (or étale stab. groups; I'm sloppy here) for all points on the "underlying space".
To get to a scheme from an algebraic space, one can either pass to an étale cover or just restrict oneself to an open dense subspace. To get to a scheme from a DM-stack, one can still pass to an étale cover, or, if one prefers finite group actions, pass to a Zariski open, which supports a $G$-bundle with total space an affine scheme, and $G$ is a finite group. The DM-stack is covered by such Zariski opens (though with different groups $G$). But for a general Artin stack there is really a long way to get to a scheme (or alg. space): if schemes are floating on the surface of the ocean, then Artin stacks rest deep in the ocean --- they are "covered" by schemes but the relative dimension are usually positive.
This somehow explains (at least to me) why for algebraic spaces (or more generally, DM-stacks) it suffices to use the étale topology to define and compute cohomology of sheaves, whereas for general Artin stacks, lisse-étale topology is necessary.
Edit: This is actually a comment for 4) in unknowngoogle's answer. It is possible to have a continuously varying family of algebraic groups parameterized by a space (i.e. not iso-trivial, even if one passes to an algebraic stratification of the base space). Therefore, I don't think all algebraic stacks (say over $k$) are locally quotients of $k$-schemes by $k$-algebraic groups: one needs group schemes over a larger base. There exists a stratification of any algebraic stack such that the inertia is flat over each stratum, but one cannot make this flat family a constant family.