Poli-Sci Perspective is a weekly Wonkblog feature in which Georgetown University's Dan Hopkins and George Washington University's Danny Hayes and John Sides offer an empirical perspective on the issues dominating Washington. In this edition, Sides looks at why gerrymandering doesn't explain the polarization of American politics. For past posts in the series, head here.

In a recent interview in The New Republic, President Obama said this about the politics surrounding gun control:

That does not mean that you don't have some real big differences. The House Republican majority is made up mostly of members who are in sharply gerrymandered districts that are very safely Republican and may not feel compelled to pay attention to broad-based public opinion, because what they're really concerned about is the opinions of their specific Republican constituencies.

Obama expressed a common view: that gerrymandering has created a bunch of safe seats for each party, making representatives responsive only to their partisan base and unwilling to forge bipartisan compromises.

It would be nice if this view were true, because it would suggest a clear solution to our polarized politics: draw more competitive districts.  But unfortunately it is not true.  The most important influence on how members of Congress vote is not their constituents, but their party.  This makes them out-of-step not only with the average American -- the "broad-based public opinion" that Obama mentioned -- but also, and ironically, with even their base.  Members are more partisan than even voters in their party.

The easiest way to see how little constituency matters is to compare how representatives vote to the partisanship of their constituents.  Here is what the 113th House looks like so far, based on calculations (pdf) by Stanford political scientist Simon Jackman

The vertical axis is a measure of candidate ideology based on roll call voting. Higher numbers indicate more conservative views, and lower numbers indicate more liberal views.  The horizontal axis captures how well Obama did in that district in 2008.  The red dots are Republican House members and the blue dots are Democrats.  All of the red dots are higher than all of the blue dots.  Polarization in the 113th Congress is already evident.

The important thing in this graph is the black lines that capture the relationship between, essentially, how liberal or conservative the member's constituents are and how liberal or conservative the member is.  Those lines should slope downward: the more liberal the district, the more liberal the member.  But the lines are mostly flat, with only a slight downward slope among Republicans.  No matter whether Obama won 20 percent or 50 percent of their district, Republican representatives have voted similarly -- that is, they have taken conservative positions on average.  No matter whether Obama won 50 percent or 80 percent of their district, Democratic representatives have taken liberal positions, on average.  Constituency hasn't affected anyone's overall voting behavior that much.

And the 113th Congress is no exception.  Here, via political scientist Nolan McCarty, is a picture of ideology in the 108th House compared to the 2004 vote for George W. Bush in their district:

The lines, which now should slope upward, do so a bit.  But party is still the more important influence.

What about the Senate?  Same thing.  Just think of states with split delegations.  How ideologically similar are, say, Chuck Grassley and Tom Harkin?  Or David Vitter and Mary Landrieu?  Not very, even though they ostensibly represent the same voters.  And it's not only Congress where this happens.  Here's the same graph but for state legislatures, drawing on research (pdf) by McCarty and Boris Shor:

In sum, Democrats and Republicans are just polarized, no matter whether their district is red, blue or purple.  It's hard to imagine that creating more competitive districts will mitigate polarization.  Members in purple districts are pretty polarized, too.

But surely there is some price to pay for being ideologically out-of-step with your constituents?  The answer is yes.  Being too ideologically extreme (pdf) or too loyal to party is associated with a greater chance of losing (not that many incumbent actually lose, of course).

There is a catch, however.  If an incumbent of one party gets booted by a challenger from the other party, this challenger won't be similar to the average voter.  Instead, they'll be more extreme than this voter, just in the other direction.  Political scientists Joseph Bafumi and Michael Herron call this "leapfrog representation."  Party power changes hands, but the average voter gets leapfrogged: their representative merely jumps from one side of the ideological spectrum to the other.  In fact, Bafumi and Herron find, representatives are often more ideologically extreme than even the average member of their party, as this graph (modified by me from their original) illustrates using data from California:

All of this begs the question: where did this polarization come from, if not from gerrymandered districts or even from voters generally?  That deserves another post, but here are two possibilities.  One is that polarization has deep roots in fundamental structural transformations of American politics -- the realignment of Republican and Democrats on civil rights and even the rise of economic inequality.  Another possibility has to do with who controls local and state party organizations, who play a large role in selecting new candidates to run for office.  In research on one of the most polarized state legislatures, California's, Seth Masket finds that local party organizations have been captured by activists for whom ideological fealty is paramount.

None of this suggests that we shouldn't reform the process of redistricting.  But when it comes to polarization, the people drawing the district boundaries -- that "league of dangerous mapmakers" -- aren't so dangerous after all.