The California legislature's attempt to reverse a 5-year-old 61%-39% referendum on same-sex marriage reminded me of something that political science professor Bruce Cain pointed out a while back — states have, in a sense, two electorates: The electorate as filtered through legislative election, and the direct electorate; and the two will inherently (in many states) yield different outcomes on many issues.

The most interesting reason Cain pointed to (and I realize there are many others, but I want to focus on this one) is that legislative districts have equal population, but not equal numbers of voters. Some districts consist nearly entirely of citizens, include relatively few children, and have high voting rates among eligible voters. Those districts cast many votes in statewide elections for governor, for U.S. senator, and on ballot measures. Other districts with the same population have many more noncitizens (whether legal immigrants or illegal aliens), many more children (for instance, Mexican Hispanics, Hispanics generally, and blacks have higher birth rates than whites), and lower voting rates among the eligible voters (for instance, nonwhites are more likely not to vote even when they are eligible). Those districts cast fewer votes in statewide elections. But each district elects one state assemblyman, state senator, or U.S. representative.

Consider, for instance, a simple and oversimplified example, with three districts, each with population 100,000, with the legislators faithfully tracking their residents' views, and with the breakdowns of views being the same among voters and nonvoters in the same district:

The legislature votes 2-1 against issue A, and that it is indeed the sentiment among the public as a whole (including nonvoters); but as a ballot measure, issue A wins by 1%. (In other models, the legislative vote is close in one direction, and the popular vote is a landslide in another direction.) So even without gerrymandering, political horse-trading, account being taken of the intensity of preferences (which, many assume, are reflected more in legislative votes than in popular votes), and the like, the different percentages of voters in each district are enough to make the legislative-filtered results be quite different from the direct results.

Now you can decide for yourself whether this is good or bad. Is the drawing of district lines by total population bad, because it gives each of the 20,000 voters in district 3 more influence over the selection of a legislator than each of the 80,000 voters in district 1? Is it good, because it leads each legislator to represent the same number of actual people, even if not the same number of voters? Should this lead us to like initiatives and referenda (or the decisions of statewide elected officials) more than the decisions of the legislature, or vice versa? Should it lead us to like a mixed legislative-popular system, or oppose it? I express no opinion on these subjects.

But I do think that this should remind us not to be surprised when Californians consistently elect Democratic House delegations and legislatures, but often elect Republican governors and U.S. senators, and enact more conservative ballot measures. And we should expect the same in other states that have a lot of immigrants (who are more likely to be noncitizens) and a lot of nonwhites (who are more likely to be underage, and more likely not to vote when eligible).

UPDATE: One of the comments complains that the 80%-20% disparity in the table above is unrealistic (though others point out that very substantial disparities, though not fourfold, do exist). But this is a stylized example, with only 3 districts. If you have 80 districts, you can have the same effects with much less stark differences.

Just to keep the arithmetic simpler, say you have 100 districts of 100,000 voters each. 55 are 55%-45% against proposal A, but they have on average only 45% turnout. 45 are 55%-45% in favor of proposal A, but they have on average 55% turnout. In the legislature, A loses 55-45. In a ballot measure vote, we have an exact draw -- 55 districts with 45% turnout are the same for statewide election purposes as 45 districts with 55% turnout. Adjust the numbers by a hair in A's favor, and A would win statewide, though it would still lose in the legislature.