The 2013 general election is now long over, and the declared winners of each race have each taken office and started leaving their official marks on the Virginia political scene. The Off on a Tangent election coverage went on through the evening of November 5, and then continued [as time permitted] until the final race concluded more than a month later on December 18. Overall, I am satisfied with my performance—both in my predictions and in my live coverage—but there was, as always, room for improvement. This post serves as a [very belated] look at what went well, and what didn’t, in Off on a Tangent’s 2013 election coverage.
Making the Calls: Successes and a Debacle
I am never afraid to make a call when the data supports making one (even when the major media outlets haven’t yet), but I am also very careful and will not make a call if the data doesn’t support it (even when the major media outlets already have). In 2004, I called the race for President George W. Bush (R) long before anybody in the media had the guts to. They were still overly gun-shy after the botched Florida call in 2000. However, in the 2012 presidential election, I refused to call North Dakota for former Governor Mitt Romney (R-MA)—even though most major media outlets already had—because the polls there were not even closed.
On this election night, many major media outlets called the race for now-Governor Terry McAuliffe (D-VA) around 10:00 p.m. I was reluctant to do so at that time for a number of reasons. The reported numbers coming from the Virginia State Board of Elections showed a nearly-even race between McAuliffe and Attorney General Ken Cuccinelli (R-VA), and the yet-unreported precincts were split three-ways between those that tend strongly Republican, those that tend strongly Democratic, and those that are true ‘toss-ups.’ There were strong indications that McAuliffe was going to win, and I said as much in my live update at 10:05 p.m., but I didn’t have enough confidence to make a formal call.
It was also very clear to me by that time that the Virginia State Board of Elections (SBE) was posting erroneous and incomplete data, which made me even less confident than I might have normally been. In other years, I probably would have made the call for McAuliffe around the time that the major media outlets did . . . but I just wasn’t certain I could trust the return data. Early in the evening, the SBE had claimed that forty precincts had reported-in with zero votes . . . an obvious impossibility. Then, a half hour later, there were results showing Goochland County with a 100 percent vote for McAuliffe . . . even though Goochland is a rural county that usually tends Republican by a 60/40 margin (give or take). It is not unusual for live SBE results to include errors, especially early in the evening, but they were much more badly awry than I had ever seen them. I thought, at the time, that things would shake out as the evening progressed. I should not have made that assumption.
Around 9:00 p.m., the SBE site went down entirely. I was able to continue coverage based on media reports and the returns being reported by local governments, but for a time I had lost my most valuable (and normally most accurate) election night resource. I posted at the time that the SBE was having a ‘healthcare.gov moment.’ I was making a glib joke, but I had no idea how right I was. After all of that, a bit after 11:00 p.m., it finally began to look like the SBE had resolved their issues. Their results site was live and working again, the numbers looked solid and there were no obvious errors. Goochland was about 60/40 for Cuccinelli, there weren’t a ton of ‘reported’ precincts with zero votes anymore. With apparently-reliable information now available, I began a thorough analysis of where things stood. Finally, at 11:13 p.m., I was comfortable enough with the results to call the gubernatorial election for McAuliffe.
Up until this point, my coverage had been ‘perfect,’ at least in the sense that I was reporting accurate numbers and making accurate calls. By this time, I had called all but two of the races I was following—and each one of those calls was right. With the exception of the McAuliffe call, which lagged behind the national media for the reasons outlined above, each of my other calls were timely—ahead of, or at least in-line with, all of the major local and national media outlets.
The only outstanding races were the Virginia Attorney General’s race and our local race for the 87th District in the Virginia House of Delegates. At 12:03 a.m. on November 6, I called the attorney general’s race for state Senator Mark Obenshain (R-VA 26th). At 12:15 a.m., I called the 87th District race for incumbent Delegate David Ramadan (R-VA 87th). Both of these calls were well ahead of the local and national media, but I believed they were solidly supported by the data. I concluded my live coverage at 12:20 a.m., and went to sleep happy and pleased with Off on a Tangent’s election night coverage (although not necessarily pleased with the each of the results themselves).
But when I woke up the next morning, the vote totals in the attorney general’s race had inexplicably changed. Obenshain, who I had declared the winner the night before, was now showing behind state Senator Mark Herring (D-VA 33rd) by a hundred votes or so. I had called the 87th District race correctly, which landed me another ‘win’—an accurate election call made well ahead of the major media. But the attorney general race had turned into a complete and utter mess. The SBE numbers were jumping around, sometimes by thousands of votes. Sometimes Herring was ahead and sometimes Obenshain was. It didn’t make any sense. When I made the call just after midnight, I was pretty much certain that Obenshain was the winner. Some time overnight, everything had changed.
When I began live election coverage in 2004, I developed a particular, proprietary method for calling races. Aside from a few minor tweaks and adjustments over the years, I still use the same method today. Without giving away all of my secrets, I can say that it uses a mix of official returns, exit polling data [when available], media calls, and other sources to paint a very accurate view of where the race stands and assign a level of confidence. Once the level of confidence begins to approach 100 percent, I make the call. In this calculation, official returns are given the most weight . . . and in close races, I won’t make the call at all unless the official vote tallies back it up. Late on election night, I was poring through SBE result data—county by county, and sometimes precinct by precinct—to see if I could determine a winner in the attorney general’s race. I believed that the SBE’s errata and bad data had been fixed, and I was working on the assumption that the numbers that were in the system already were reasonably accurate. Shifts of a few votes here and there happen each year, but never in big enough numbers to change an outcome [or so I thought].
I went through and determined how many precincts across the state still hadn’t been counted, and then multiplied that number by the average number of votes-per-precinct to determine (very roughly) how many votes were yet to be counted. And then, I assumed that an absurd eighty-percent of those votes would go for Herring, who was then showing well behind Obenshain in the vote tallies. Even with an 80/20 bias for Herring—a bias far larger than would ever show in the real world—there just weren’t enough votes left for Herring to make up the difference and win. In other words, it was statistically impossible for Herring to win the race, which brought my confidence level up to very near 100 percent. Only then did I call the race for Obenshain.
In the morning though, with no explanation, the SBE numbers showed Herring ahead. I made the difficult decision to rescind my call for Obenshain at 9:10 a.m., the first time Off on a Tangent had ever done so. By noon, Obenshain had again pulled into a slim lead as the provisional and absentee ballots were added to the mix. There were rumors floating around that Fairfax County, Virginia, had yet to count ‘thousands’ of ballots. That sounded bizarre and far-fetched, but it was coming from sources that are normally reliable. Even though the numbers seemed to shifting back in line with my original call, albeit more narrowly, I decided not to re-call the race until I knew what was going on in Fairfax County. Later in the day, Fairfax County election officials categorically denied those rumors and stated that all votes (except provisional ballots) had been counted. With that information, at 10:40 p.m. on November 6, I again called the race for Obenshain. I explained my rationale later that evening:
Let’s assume that each county in Virginia has a number of provisional ballots equal to 0.19 percent of the vote (which is an over-estimate; the real number is likely more like 0.16 or 0.17 percent). And let’s assume that they are all accepted by county election officials (which won’t happen, as there are specific conditions for accepting and rejecting provisional ballots). And let’s assume that the provisional ballots in each county go for Herring by a whopping 10 percent greater margin than the rest of the ballots in that particular county did (also a huge over-estimate in Herring’s favor). Even with all of that, Obenshain still leads by about 275 votes. Excluding the slim possibilities of major counting errors, or election fraud during the recount(s), there’s no way for Herring to pull this out.
I thought that would be the end of it. I thought, as in so many other close races over the years where my system made it possible for me to make an ‘early’ call, the major media outlets would eventually crunch the numbers and build enough confidence to follow suit. I would be able to chalk up another ‘win,’ another accurate call ahead of the big names in election coverage.
But then, on Monday, November 11, almost a week after the election, Fairfax County officials announced that they had ‘found’ over three thousand ballots that had not been counted properly . . . and, unsurprisingly, those ballots were from one of the most strongly Democratic-leaning precincts in the state. With those new numbers added to the SBE tallies, I re-ran my analyses . . . and yet, even then, Obenshain was still coming up as the winner. But the margin had narrowed enough that it was now possible for provisional ballots to change the result, so again the race’s outcome had come into question. I decided not to rescind the call again just then, but on November 13—when Herring declared himself the victor by 164 votes, and with strongly-Democratic Fairfax County and Richmond City continuing to ‘find’ new votes—I was forced to move the race back to ‘too close to call’ for the second time.
The state ran a recount, more votes were . . . ahem . . . ‘found,’ and ultimately the SBE certified Herring as the winner by about eight-hundred votes. Obenshain conceded defeat on December 18, but, as I noted in my live coverage, I had by then completely lost confidence in the integrity of the official results. Thousands of ballots in Fairfax County appear to have been handled improperly and processed incorrectly. There were other counting irregularities out of Richmond City. I am still not able to accurately project who won the race or by what margin, although there is some evidence that my original call was correct and Obenshain actually received the most real votes. This is a moot point. The SBE has certified Herring as the winner, and Obenshain conceded defeat and has not challenged the outcome. Herring is now the Attorney General of the Commonwealth of Virginia. That is that.
So I am left in the strange position of having made a ‘wrong’ call that, quite possibly, wasn’t actually wrong. The back-and-forth on this race was embarrassing, and it marks the only time that Off on a Tangent has ever had to rescind an election call, and the only time that Off on a Tangent has ever called a race for a losing candidate . . . and yet I am hard-pressed to identify any actual error on my part.
The only criticism I can levy at myself about my 2013 election coverage is that I probably should not have made my original call for Obenshain at 12:03 a.m. on election night . . . but only because I should have realized that the errors and oddities in the SBE results might still be happening. I should have played it a bit safer there, even though I had a high level of confidence in the result. But the second call I made for Obenshain—10:40 p.m. on November 6—is hard to criticize. The data was solid, the errors had shaken-out, and it was still statistically impossible for Herring to win the race except through fraud—’found’ ballots and counting shenanigans in the most staunchly left-wing jurisdictions in the Commonwealth.
Mistakes get made in elections, and the numbers will always move a bit as the results get reviewed or recounted. But the counting errors that get resolved during this process will not show a strong bias for one candidate or the other. Statistically speaking, counting errors will add votes to each tally roughly in proportion to the overall percentages. In other words, in a 60/40 race, about sixty percent of the errors will benefit the leading candidate, and about forty percent will benefit the trailing candidate. Machine counting errors don’t have a partisan bias.
That is why actual counting errors don’t change the outcome of a race, except when the race is so narrow that a minuscule handful of votes could make the difference. We’re talking about races where five or ten or fifty votes separate the leaders, not races where thousands do. When the ‘correction’ of ‘errors’ exhibits a strong bias toward one candidate or another—giving one candidate a net gain of thousands of votes over his opponent in an otherwise evenly-split race—that is a sure sign that the numbers are being manipulated, not corrected.
The Predictions: Almost Right-On
For the second year, I made a post shortly before the election explaining in detail what I thought was likely happen at the polls. I am well aware that predicting the future is a dangerous business; it is much easier to be wrong than to be right. In 2012, I posted three likely presidential scenarios, and I got fairly close with the second (which I said had a forty percent chance) . . . although I did underestimate President Barack Obama’s (D) performance.
If you take a look at my post-election analysis of that race and how I did, you’ll see that I had the right idea but mis-judged the magnitudes. The Ohio polls were skewed, just not as much as I suspected. But I did mis-judge both Florida and my own home state of Virginia, both of which I thought were pretty safe for former Governor Mitt Romney (R-MA) but ended up going—narrowly—for Obama, leading to the president outperforming all of my scenarios. In the end, I gave myself a mixed grade . . . which really wasn’t bad for my first attempt.
I did consider getting out of the prediction game, but I decided to give it another shot in 2013. I posted a new set of election scenarios on November 4, including three likely outcomes in the gubernatorial race and a brief review of the down-ticket races. In short, I thought it was most likely that Terry McAuliffe (D) would win comfortably (fifty percent chance), very possible that McAuliffe would win narrowly (forty-five percent chance), and that there was a small but real chance for an upset by then-Attorney General Ken Cuccinelli (R-VA) (five percent chance).
The race ended up closest to my second scenario, although not quite as narrow as I had estimated. In that scenario, I estimated 48 percent for McAuliffe, 47 percent for Cuccinelli, and 5 percent for Robert Sarvis (L). The actual outcome was 48 percent for McAuliffe (exact match), 45 percent for Cuccinelli (2 percent less than predicted), and 7 percent for Sarvis (2 percent more than predicted).
The main error in my second scenario was that I overestimated how many potential Sarvis voters would defect on election day. I was right in assuming that many who said they would vote for Sarvis when talking to pollsters would end up voting for one of the two major-party candidates anyway, but I under-weighted how deeply unlikable both McAuliffe and Cuccinelli were to many of those who lean in a libertarian direction. It would also appear that undecided voters split roughly evenly between McAuliffe and Cuccinelli, which was more in-line with my first scenario than with my second (and partially accounts for Cuccinelli’s numbers falling between those two predictions).
All-in-all, at the top of the ticket, I think I did a good job. It wasn’t perfect, but the real outcome matched reasonably closely with my second scenario, and my overall analysis of the issues in-play appears to have been mostly correct.
Down-ticket, I generally did even better.
In the Lieutenant Governor of Virginia race, I predicted that it would go 55 percent for state Senator Ralph Northam (D-VA 6th) and 45 percent for Reverend E.W. Jackson (R). That prediction was within half of a percentage point of the real results—55.1 percent for Northam and 44.5 percent for Jackson.
In my local 87th District House of Delegates race, I predicted 50.5 percent for Delegate David Ramadan (R-VA 87th) and 49.5 percent for Air Force Major John Bell (D). The actual result was 50.3 percent for Ramadan (off by 0.2 percent) and 49.4 percent for Bell (off by 0.1 percent). I’m especially proud of this one, since I had virtually no poll data to work with and considered this a ‘low confidence’ prediction.
On the Loudoun County bond referendums, I estimated that the transportation and fire & rescue bonds would pass with 75 or 80 percent of the vote, and that the schools and parks bonds would pass more narrowly with 65 or 70 percent. I considered each of these ‘low confidence’ predictions, although I was reasonably certain that they would all pass—and they did. I overestimated support for the transportation bond referendum, which passed with 68 percent. The fire & rescue bond referendum passed with 77 percent, which was right in the middle of my estimated range. The school bond referendum passed with 68 percent, which was also right in the middle of my estimated range. Lastly, the parks bond referendum passed with only 54 percent—well below my estimate. Although I got the outcomes right, I need to work on improving my margin estimates on bond referendums.
And what about the big mess of an election, the race to be the Attorney General of Virginia? I, quite presciently, called this race a ‘true toss-up’ between state Senator Mark Herring (D-VA 33rd) and state Senator Mark Obenshain (R-VA 26th). I believed that if my first gubernatorial scenario came to pass, the race would likely go narrowly to Herring. And if my second gubernatorial scenario came to pass, it would likely go to Obenshain. I ultimately put down a prediction of 50.5 percent for Herring and 49.5 percent for Obenshain, but said I had a very low level of confidence in those numbers.
Well, my second scenario came to pass . . . but with a bit of the first thrown in. So, quite logically, this race turned into a true tie. The official numbers show both candidates at 49.9 percent, with Herring edging out a win, but (as discussed earlier) I have little confidence in the accuracy of the results and consider it highly likely that Obenshain actually received the most votes. Of course, that doesn’t really matter since Obenshain conceded defeat and chose not to challenge the results. Regardless, although I was slightly off in the actual numbers, my synopsis of this race was essentially spot-on—I labeled it a tie and said it would likely go very narrowly one way or the other.
So, although there is still room for improvement, I think I did much better in my 2013 election scenarios than I did in 2012 . . . and I have learned a lot, so hopefully I’ll do even better in 2014. This is a difficult business and, as I mentioned at the beginning, it’s a lot easier to make a wrong prediction than a right one, but I have a lot of fun doing it (and doing my election coverage in general). Thanks for tuning in, and I hope you’ll come back in November!