First, non-voters are — on average — only a bit more Democratic than voters, and a bit more supportive of liberal policies. For evidence, see mywork with Jack Citrin and Eric Schickler, this paper by Jan Leighley and Jonathan Nagler (now part of their book), the chapter by Eitan Hersh and Steve Ansolabehere in this book, and this paper by Ben Highton and the late Ray Wolfinger (helpfully cited by Dylan Matthews).
Now, note the fine print. “On average” means that Citrin, Schickler and I found that in some states in some election years, non-voters were actually a bit more Republican. That wasn’t the norm, but we should not assume that higher turnout or compulsory voting would always and everywhere benefit Democrats. (See also this piece by James DeNardo.)
And “a bit” means that the observed differences are not always large — usually a single-digit number of percentage points, and quite often in the low single digits. Any larger differences — as in this 2012 Pew preelection survey of self-reported likely voters and likely non-voters — are the exception, not the rule.
Here’s the second important finding: simulations suggest that compulsory voting would change the outcome of very few elections. This is not only because non-voters often aren’t that different than voters, but also because lots of elections — such as at the House and Senate levels — aren’t that close.
For example, when Citrin, Schickler, and I studied 246 Senate elections from 1990-2006, only eight outcomes changed when we simulated the impact of universal turnout. (See our chapter in this book.) In a very close presidential election, as in 2000, universal turnout could have made a difference. But such elections aren’t common.