It's tough to make predictions, especially about the future. I made my fair share of them in 2014, and now that the year is drawing to a close, it's a good time to revisit them.

This election cycle, I left the big predictions—Senate, House, governor—to the experts. Instead, I tried my hand at handicapping some less celebrated but equally important races: those for constitutional office in the 50 states. By Election Day, I had issued Cook-style race ratings for the nation's lieutenant governor races, attorney general races, auditor races, comptroller/controller races, and state superintendent races. (Final ratings are, for a limited time, still up on my "2014 Ratings" tab but are archived forever at the bottom of this post.) I wanted to provide a guide to elections no one else was really bothering with, to help and encourage people to understand them at a glance—but that isn't much of a help if those race snapshots are totally off the mark.

Here in December, of course, we know who won each of those 72 races, so it's time to go back and see how my inaugural constitutional race ratings turned out. The main takeaway? Politics is easier to predict than baseball. Here's how I did by office:

Of the 17 lieutenant governors elected separately from governors, I predicted a post-election breakdown of 12 Republicans and five Democrats. That's exactly where it ended up.
I predicted a post-election split among the nation's 43 attorneys general of 22 Democrats and 19 Republicans, with two tossups. It was actually 20 Democrats to 23 Republicans. This was my worst category, but it also offered the most chances to be wrong, with 31 races to handicap.
I foresaw Republicans taking a 12–11 lead among auditors, with one tossup. The end result was a 14–10 Republican auditing majority.
For the nine comptrollers/controllers in the country, I predicted each party would win four seats, with one tossup. That rubber match ended up going to Republicans, who took a 5–4 lead among this group of officers.
Finally, I said Republicans would hold four superintendent jobs, Democrats would hold three, and two would be tossups. All the tossups went to the GOP, as they maintained their 6–3 superintendent advantage.

Not all of those picks were created equal, though; as with Senate or governor, many elections were foregone conclusions, while others were harder to forecast. You get a better picture of where I may have gone wrong when you look at the results of the races in each rating bracket:

Democrats won 17 of the 17 races I rated as Solid Democratic, including one uncontested race.
Democrats won three of the three races I rated as Likely Democratic.
Democrats won just two of five races I rated as Leans Democratic.
Republicans won all six races I rated as Tossup.
Republicans won seven of the seven races I rated as Leans Republican.
Republicans won 11 of the 11 races I rated as Likely Republican.
Republicans won 23 of the 23 races I rated as Solid Republican, including four uncontested races.

From this, it's clear that my ratings were pretty good, but not great. I had the right general sense for the spectrum on which races went from Democratic-leaning to Republican-leaning, but I underestimated Republicans across the board. However, I'm in some pretty good company; most election forecasters this year expected Republican gains but were taken aback by the Republican wave that ended up forming. The fact that general punditry, not to mention the polls, were overly friendly to Democrats explains why the GOP took all of my tossup races—and also won three of my Leans Democratic races, something that would normally be a red flag. In this election, though, I think it makes sense—even if I still would have liked to have known better.

This still doesn't tell the whole story, though. It's not just about who won these races, but how much they won by. (This is, after all, what separates the Solids from the Leanses.) There's an important caveat here: a candidate's final margin of victory doesn't necessarily reflect their pre-election likelihood of winning the race; some states or races are more elastic than others. (For instance, the 13-point race that was California lieutenant governor was never even remotely within Republicans' grasp, whereas the 16-point race for New Mexico AG was definitely winnable for the GOP; those last 13 points are much, much harder to scrape together in California than in New Mexico.) However, I do think final margins are important to look at as the most obvious indicator of a race's closeness. The following chart contains the Democratic margin of victory (positive numbers) or defeat (negative numbers) in all the races I handicapped (numbers are unofficial counts from the AP):

The average margins for each category are at least in the right order, from biggest at Solid Democratic (+21.3 points) to most negative at Solid Republican (–46.0 points). Likely Democratic (+9.7 points) is just where you'd expect a generic Likely Democratic race to be, as is Leans Democratic (+3.8 points). Tossup (–9.8 points) and Leans Republican (–14.0 points) are definitely miscalibrated, but again, that's a function of the GOP's overachieving night.

However, those averages mask some pretty wide deviation in some of the rating categories—most glaringly Leans Democratic. In retrospect, none of the five races given this rating were truly Leans Democratic. Rhode Island LG (+20.5 points) and New Mexico AG (+16.2 points) fit better as Likely Democratic, Nevada AG (–0.9 points) should have been a Tossup, and Arkansas AG as well as Delaware auditor (both –8.4 points) should have been Leans Republican.

Of my Tossup races, Arkansas auditor (–19.8 points) and Nevada controller (–14.9 points) look the worst. Most of my Leans Republican races should've been classified as Likely Republican, but especially Iowa auditor (–14.0 points), Arkansas lieutenant governor (–18.5 points), Ohio auditor (–19.1 points), and Nevada lieutenant governor (–25.9 points). And the 33-point Democratic loss for Nebraska attorney general was a real stretch to put as Likely Republican.

What lessons can I draw from my biggest forecasting misses? First, there's a pattern in the races I mischaracterized most badly. The same states keep popping up: Nevada. Arkansas. Ohio. Iowa. These are states where 2014 saw voters turning particularly hard, and particularly unexpectedly, toward Republicans. In Nevada and Ohio, non-serious top-of-the-ticket Democratic campaigns allowed GOP GOTV machines to operate completely uninhibited, turning these usual swing states into what Dave Wasserman called "orphan states." In Arkansas and Iowa, nominally competitive Senate races turned into laughers when polls failed to predict how utterly and completely the bottom would drop out for Democrats among certain voters there. Those Republican currents were strong enough to sweep away even downballot Democrats running separate, often quite capable campaigns, leading to blowouts of candidates who may have deserved better than they got.

To a certain extent, you can't guard against this. Even a constitutional race that is consistently tied or close in polls can fall victim to it and become a landslide. This is because even the best poll of a race like these includes far more undecided voters than your average Senate or gubernatorial survey—people just pay less attention to their state treasurer or insurance commissioner. The race may indeed be as close as it seems if each candidate wins over undecided voters equally—but often swing voters will all break the same way on Election Day. A race that was 38% to 38% in a poll (as Oklahoma superintendent was) thus can easily wind up as 56% to 44% (as Oklahoma superintendent did) without much imagination. Still, these currents are perceptible if you look carefully enough. In Oklahoma, undecideds were likely to break for the Republican given the overall conservativeness of the median voter there; in states like Nevada, there were warning signs that Democrats might roll over. I failed to pick up on the warning signs that the current was developing, and I underestimated how strong these currents would be.

I may have also been too idealistic in thinking that people would cast their votes based on the merits of each individual race. In many cases, I talked myself into seeing idiosyncratic strengths of, say, the Democratic candidate for Ohio auditor, or I banked on the scandal-tarred unpopularity of the South Carolina comptroller, when in fact I'd have been better off looking at the state fundamentals. Perhaps the biggest takeaway from this project is that it is essential to hone that sense for when to look at the state or partisanship and when the race is truly important enough to break through and stand on its own.

We had examples of both kinds of races this year. In certain states, it was clear from how closely downballot Democratic performances tracked with one another that bigger forces were at play. In states like Ohio, the race was decided by turnout; more of a certain shade of voter (in this case, red) was simply showing up to the polls than another shade. In states like Arkansas, voter anger at the amorphous scourge of "Obama" or "Reid" or "Pelosi" drove voters to make a statement and vote blindly against Democrats en masse. And in states like Texas and South Carolina, the normal cross-section of normally conservative-leaning voters just showed up like normal and didn't see anything in these lower-level races to cause them to break their Republican-voting habits.

Then there were other states where it was clear that voters were exercising independent judgment on each race. In Maryland, voters made a statement by electing Republican Larry Hogan governor—but my Solid Democratic ratings for Maryland's two constitutional officers proved right on the money. In Idaho, voters comfortably returned their Republican governor, lieutenant governor, and attorney general to office, but they almost elected a Democrat as superintendent of public instruction as well. That race gained a lot of independent attention because of how controversial education has been in Idaho in recent years. The key is knowing when a race qualifies as "special" enough that people will vote purposefully for it and not just follow their partisan instincts or the national mood. It's a subjective call to make, and it's why forecasting downballot races specifically can be so tricky.

Overall, I'm pleased with how my ratings turned out. I still pegged most races at their correct level of competitiveness, and I'm not too concerned about my less prescient calls. Constitutional race ratings will always involve a lot more guesswork than the traditional ratings on Cook or Daily Kos; downballot races have few polls or hard data to work off, and the polls that do exist tend to be less accurate than those of better-covered races. That inherent uncertainty will also always cause me to rate more races as "Tossups" or "Leans" than probably will deserve it in the final analysis. This year, only 13 races were within single digits, but considerably more than that were plausibly up for grabs, simply because no one knew enough to say otherwise. Some races I rate as close will inevitably be those 15-point routs, but that's a feature, not a bug. I stand by a couple of my "bad" ratings from this year for this reason: with as little as we knew about it, Nevada controller (–14.9 points) was indeed anyone's game.

Having tried my hand at race ratings for both Senate/governor/top-of-the-ticket races (in 2012) and downballot races (this year), I've definitely found these to be more of a challenge—and that's the way I like it. I plan on continuing to handicap and provide ratings for constitutional races in 2015 and 2016. I ran out of time in 2014, but with fewer seats on the ballot in the next two years, I hope and expect to branch out to secretaries of state, treasurers, and more, in addition to the offices I test-drove this year. My New Year's resolution to Baseballot readers: to preview every non-gubernatorial statewide office for you on these very pages. Stay tuned.