First of all, suggested edits are - on the whole - made to improve the post, so it should be treated as such primarily.
When to approve
From my own experience, the kind of useful edits breaks down like this:
The bulk of edits are purely code formatting; new users tend to paste their code without realizing that it looks horrendous without a fixed font.
Next up is correcting grammar and rephrasing the question for those whose first language is not English.
The third place goes to adding (mostly) or removing tags.
The title is changed to better describe the question.
In above cases, the edit usually gets approved.
When to improve
If you agree with most of the edit, but you feel that it could do with some more TLC, you can click "Improve" and suggest an additional edit.
When to reject
To reject an edit you need experience to spot the tell-tale signs of a bad edit, this takes time to fine tune. When I started out, I used this trick to get a feel of what other editors are doing:
Click the "Reject" option; a list of reasons appears;
See if any reasons has a blue number next to them;
Close the dialog and read the edit again to see if you feel the same way.
Rejection breakdown
In my own experience, I typically close questions for the following reasons (in order):
Invalid edit - Something that looks like a comment or something completely unrelated would get proposed.
Radical change - An "improvement" of an answer is proposed, but unless it's fixing an obvious typo, it usually changes the overall meaning.
Too minor - Some like to go wild with backticks to highlight certain parts of the sentence that speaks of code or capitalize a few i's, etc. Exceptions are the cases in which it really improves the readability of the contents.
When to skip
Although your vote is not the only one that determines the fate of the edit, if you're simply unsure about what to do, it would be better to just "Skip" the edit altogether and let someone more qualified (one can hope) take care of it.