Should Democrats look to the southeast and southwest?

In a previous post, I implied that Hillary Clinton’s 2016 losses in five states won by Barack Obama in 2012–Pennsylvania, Ohio, Michigan, Wisconsin and Iowa—resulted from white voters without a college degree (14 percentage points less Democratic in 2016 than in 2012) averaging 44.4% of these states’ electorates, while white voters with a college degree (10 percentage points more Democratic than in 2012) averaged 38.0%.[1]

This steep decline for Democrats in 2016 among midwestern/Rust Belt white voters without a college degree is the bad news.

Today, I will present some good news.

Four states totaling 80 electoral votes (EV)—North Carolina, Georgia, Texas and Arizona—may be trending slowly toward the Democrats. Clinton lost all four of these states, as did Democratic nominees in 2004, 2008 (excepting a squeaker win in North Carolina) and 2012. However, the average margin by which she lost to Republican Donald Trump in these four states was 5.3 percentage points, about half the average 2004-12 margin (-10.2).[2]

This slow pro-Democratic trend is even clearer when you adjust for the national Democratic margins over the last four presidential elections, as shown in Table 1.

Table 1: National and State-Level Margins (Democratic % – Republican %), 2004-16

Election National Margin Average Margin in   NC, GA, TX, AZ State Margin-National Margin
2004 -2.5 % points -15.6 % points -13.1 % points
2008 7.3 % points -6.3 % points -13.5 % points
2012 3.9 % points -8.7 % points -12.5 % points
2016 2.1 % points -5.3 % points -7.4 % points

Yes, on average, these four states were 7.4 percentage points less Democratic than the nation in 2016. But that is still a shift toward the Democrats of 5.6 percentage points compared to the 2004-12 average of -13.0 percentage points.

Put simply, as the country as a whole voted slightly more Republican in 2016, these states voted relatively more Democratic.

The maps in Figure 1, taken from Dave Leip’s indispensable Atlas of U.S. Presidential Elections, show the counties where Democrats performed better in 2016 than in 2012 (shades of pink and red[3]), and where they performed worse (shades of blue).

Figure 1. County-level change in voting margin (Democratic % – Republican %) from 2012 to 2016







North Carolina


In the five key states Clinton lost, she essentially matched Obama’s 2012 performance in the core urban Democratic areas of the state, while doing far worse than Obama elsewhere. In these four southeastern/southwestern states, however, Clinton actually improved on Obama’s 2012 performance in the urban (and surrounding suburbs) areas of the state. Specifically, Clinton improved on Obama’s 2012 performance in and around Atlanta and Savannah, Georgia; Austin, Dallas and Houston, Texas; Flagstaff, Phoenix and Tucson, Arizona (as well the border counties with Mexico); and Charlotte and Raleigh/Durham/Greensboro/Winston-Salem, North Carolina.

These four states also have populations that are far less white (average 55.0%, 65.5% of the 2016 presidential electorate) than the five key states Clinton lost (81.0%, 82.4% of the electorate). Their residents are also slightly more likely to have a college degree (27.9% to 26.2% of the population, 50.8% to 44.4% of the electorate).

And as Table 2 shows (data from here), while these electorates were less white overall[4], they averaged more white voters with a college degree (34.5%) than white voters without a college degree (31.0%). That said, while white voters with a college degree were far less Republican on average (R+24.2) than white voters without a college degree (R+46.8) in these states, they were still far more Republican than all such voters nationwide. In fact, the smaller the white proportion of these electorates, the more Republican white voters (with or without a college degree) were[5].

Table 2: Presidential Voting Margins Among White and Non-White Voters With and Without College Degrees, Nationally and in Four Democratic-Trending States

State White College White Non-College Non-White College Non-White Non-College
U.S. +4 R 37% +39 R 34% +50 D 13% +56 D 16%
NC +19 R 37% +44 R 33% +63 D 13% +59 D 17%
GA +41 R 30% +66 R 30% +66 D 20% +72 D 20%
TX +31 R 31% +55 R 26% +28 D 21% +47 D 22%
AZ +6 R 40% +22 R 35% +22 D 11% +31 D 14%

Now, just bear with me while I briefly describe multiple linear regression.

Multiple linear regression is a statistical method that reveals how much a given (dependent) variable (say, state 2016 Democratic presidential margin) would change, on average, if other (independent) variables (say, state % white and % adults over 25 with a college degree[6]), changed by a specified amount. It also tells you how much of the variance in the dependent variable is “accounted for” (phrasing I prefer to “explained by”) the group of independent variables. Finally, it gives you an equation that can be used to calculate the “expected” value of the dependent variable given specified values of the independent variables.

Thus, using 2016 presidential election data from the 50 states and the District of Columbia, I estimated the following equation:

2016 Dem Margin = -0.53 – 0.51*% White + 2.94 *% College Degree

On average, then, a one percentage point increase in the white population of a state (all else being equal) lowered the 2016 Democratic margin in that state by an average 0.51 percentage points, while a corresponding increase in the percentage of adults over 25 with a college degree increased the Democratic margin by 2.94 percentage points. These two variables alone accounted for a whopping 75.4% of the variation in the 2016 state Democratic margins[7].

In 2013, the population of Massachusetts was 75.1% white and 28.0% of its residents over the age of 25 had a college degree. Plugging those values into the equation yields a “predicted” 2016 Democratic margin of 25.2, barely below the 27.2 percentage points by which Clinton actually won Massachusetts.

Table 3: Predicted and Actual 2016 Democratic Presidential Margins, Based Upon % White and % Adults with a College Degree, in Four States

State % White % College Degree (Age >25) Predicted 2016 Democratic Margin Actual 2016 Democratic Margin
GA 54.8% 28.0% D+2.0 R+5.1
TX 44.0% 26.7% D+3.6 R+9.0
AZ 56.7% 29.6% D+5.7 R+3.5
NC 64.4% 27.3% R+5.0 R+3.7

As Table 3 shows, this simple model “predicted” Clinton would win three states she lost: Georgia, Texas and Arizona. The average 9.6 percentage point gap between predicted and actual margins likely resulted from white voters with a college degree in these states voting more Republican than their counterparts elsewhere. Had this group voted as their national counterparts did in 2016 (R+4), all else being equal, Clinton would have won North Carolina’s 15 EV by 2.2 percentage points, Georgia’s 16 EV by 6.6 percentage points and Texas’ 38 EV by 0.7 percentage points, though she would still have lost Arizona’s 11 EV by 2.5 percentage points. Those extra 69 EV would have raised her total to 296, 26 more than necessary[8].

It is quite possible that the future of the Democratic Party lies with a three-part coalition of non-white voters, white voters with a college degree and younger voters (voters aged 18-39—37% of the electorate—voted for Clinton by 16 percentage points).

If this is indeed the case, then newly-elected Democratic National Committee chairman Tom Perez may want to think about putting less emphasis on midwestern and Rust Belt states with a large proportion of whites without a college degree[9] to focus instead on south Atlantic states like Georgia and North Carolina, and southwestern states like Texas and Arizona, whose electorates will become even more dominated by this three-part coalition in the future.

Until next time…

[1] At a practical level, this means that she lost the vast majority of white, low-population, rural counties in these states by an average 363,000 votes—and 14.0 percentage points—than Obama had four years earlier, while holding her own in the Democratic core counties.

[2] For comparison, Clinton lost the five key Midwestern/Rust Belt states by an average of 3.8 percentage points.

[3] Unlike every other news outlet or data analyst, Dave Leip colors Democratic areas red and Republican areas blue.

[4] The mix of non-white voters varies across by state, with Arizona and Texas having a higher Latino population (30.3 and 38.4%, respectively) and Georgia and North Carolina having a higher Black population (31.4 and 22.0%). Black voters supported Clinton by a margin of 81% percentage points, while Latinos did so by a margin of 38 percentage points.

[5] A thoroughly irrelevant sidebar: this effect (white voters becoming more Republican as the non-white population increases) inspired the “demographic trait activation” hypothesis I was going to test in my political science doctoral program. The one I never finished (though I did get a Master’s Degree out of it).

[6] 2013 data

[7] The “predictive value” of a state’s percentage white or college educated was also higher in 2016 than in 2012. In 2012, a one percentage point increase in state percentage white decreased that state’s Democratic margin 0.37 percentage points and a one percentage point increase in state percentage of adults with a college degree increased the margin 2.65 percentage points. These two variables accounted for just over half (56.5%) of the variation in the 2012 state Democratic margins.

[8] The total would be 301, but for five “faithless” electors.

[9]Acknowledging, of course, the very close margins in Michigan, Pennsylvania and Wisconsin (as well as in Minnesota: Clinton+1.5), suggesting that simply activating more voters belonging to this three-part coalition could easily flip these states Democratic again.

A closer look at Hillary Clinton’s performance in five key states

In a previous post, I proposed a “three-election weighted relative Democratic margin” (3W-RDM) for each state and the District of Columbia (DC). The “RDM” is the arithmetic difference between each state’s voting margin (% Democratic – % Republican[1]) and the national margin in a given presidential election. I calculated every state’s average RDM over successive three-election cycles, starting with 1984-1992, using a 1-2-3 weighting scheme, to yield seven 3W-RDM for each state.

(Unless otherwise noted, all presidential election data are from Dave Leip’s indispensable Atlas of U.S. Presidential Elections).

In theory, if you add an actual or hypothesized national Democratic margin for a given presidential election to a state’s most recent 3W-RDM, you can “predict” that state’s Democratic margin in that election. I did this for the 2016 presidential election, summing each state’s 2004-12 3W-RDM and the 2.1 percentage points by which Democrat Hillary Clinton beat Republican Donald Trump.

Clinton lost five states—Pennsylvania, Ohio, Michigan, Wisconsin and Iowa—she was projected to win under this analysis. These states, plus Florida, were the ones she lost that Democrat Barack Obama had won four years earlier.

I want to look more closely at the results in these five states.

But first, just bear with me as I present some national vote totals.

Table 1: Voting Totals and Percentages in the 2012 and 2016 Presidential Elections

  2012 2016 2016-2012
# Votes % Vote # Votes % Vote # Votes % Change
Democrat 65,918,507 51.0% 65,853,625 48.0% -64,882 -0.1%
Republican 60,934,407 47.1% 62,985,106 45.9% 2,050,699 +3.4%
Libertarian 1,275,923 1.0% 4,489,233 3.3% 3,213,310 +251.8%
Green 469,015 0.4% 1,457,222 1.1% 988,207 +210.7%
Other 639,790 0.5% 2,315,043 1.7% 1,675,253 +261.8%
TOTAL 129,237,642 100% 137,100,229 100% 7,862,587 +6.1%

From 2012 to 2016, according to Table 1, the national Democratic margin dropped 1.8 percentage points, even as the total vote cast for president increased by nearly 7.9 million votes. For all the talk that Clinton did not turn out the Obama vote, she only won 64,882 fewer votes than Obama had in 2012; in fact, only Obama, in 2008 and 2012, ever won more votes for president than Clinton did in 2016.

Trump, by contrast, received just over 2 million more votes in 2016 than Republican Mitt Romney had in 2012, about one quarter of the vote increase from 2012 to 2016; just over half of the remaining increase was the result of Libertarian Gary Johnson receiving 3.2 million more votes in 2016 than he had in 2012.

If every state’s margin had dropped an identical 1.8 percentage points, the only state Obama won in 2012 that Clinton would have lost in 2016 would have been Florida. In reality, the Democratic margin dropped by more than that in 34 states (of 39 total where the Democratic margin dropped); it dropped more than five percentage points in 24 states and more than 10 percentage points in eight states.

Across the five states in question, the average drop in Democratic margin was 10.0 percentage points, ranging from 6.1 percentage points in Pennsylvania to 15.2 percentage points in Iowa.

Obama beat Romney by a little over 1.2 million votes across these five states, but Clinton lost these five states by just over 670,000 votes (all but about 79,000 from Ohio and Iowa). This marginal difference of just over 1.9 million votes is 90% of the net change in the Democratic margin in the presidential vote from 2012 to 2016.

So…what did happen in these five states?

Democrats generally win statewide elections by building up huge vote margins in a handful of urban and/or college-town counties while holding down the margin in the rest of the state.

Thus, when Obama won Pennsylvania in 2012, he won five southeastern counties (Philadelphia and its suburban ring: Bucks, Chester, Delaware, Montgomery) by about 615,000 votes and Allegheny County (Pittsburgh) by about 90,000 votes—for a total margin of about 706,000 votes. While Obama LOST the other 61 Pennsylvania counties by about 400,000 votes, he still won the state by over 300,000 votes overall.

The pattern is similar in Ohio (three northeastern counties—Cayuhoga [Cleveland], Summit, Lorain; Franklin [Columbus]; Hamilton [Cincinnati]), Michigan (Wayne [Detroit], Oakland; Washtenaw [Ann Arbor]; Ingham [Lansing]), Wisconsin (Milwaukee, Dane [Madison], Rock [Janesville[2]], and Iowa (Polk [Des Moines], Johnson [Iowa City], Linn [Cedar Rapids]).

Figure 1: Change in Absolute Democratic Vote Margin, 2012-16


Figure 2: Change in Percentage Democratic Margin, 2012-16


As you can see from Figures 1 and 2, Clinton won the core Democratic counties of each of these five states by about the same margin, in terms of both absolute vote and percentage of the vote, as Obama had four years earlier, She even improved on Obama’s margin in the Pennsylvania’s six core Democratic counties by 61,395 votes!

However, it was in the remaining 61-96 counties that Democratic support absolutely collapsed in these five states between 2012 and 2016, as shown in Table 2.

Table 2: Changes in Voting Patterns in Non-“Core”-Democratic Counties, 2012-16

State Number non-D counties Change in absolute vote margin Change in percentage point vote margin
PA 61 -417,765 -11.2
OH 83 -565,766 -16.1
MI 79 -382,890 -12.5
WI 69 -223,606 -10.6
IA 96 -224,958 -19.7
Average 78 -362,997 -14.0

These are predominantly white rural counties. And while I have no county-level data on education level (yet), there is strong evidence that the fundamental change between 2012 and 2016 voting was among white voters with (10 percentage points more Democratic, according to Table 3) and without (14 percentage points less Democratic) college degrees.

Table 3: Changes from 2012 to 2016 in Presidential Voting Margins Among White Voters With and Without College Degrees

State 2016 2012
White College White Non-College White College White Non-College
Margin % voters Margin % voters Margin % voters Margin % voters
U.S. +4 R 37% +39 R 34% +14 R 36% +25 R 36%
PA +0 D 41% +32 R 40% n/a n/a n/a n/a
OH +25 R 37% +30 R 43% n/a n/a n/a n/a
MI +8 R 33% +31 R 42% n/a n/a n/a n/a
WI +12 D 39% +25 R 47% n/a n/a n/a n/a
IA +5 R 40% +20 R 50% n/a n/a n/a n/a

National voting margins were obtained here and state-level voting (not available for these subgroups in 2012) margins were obtained here.

All five of these states had electorates with a higher share of white voters overall—and white voters without college degrees—than the national electorate. It would not be a stretch to say that the percentage of the electorate who were white voters without college degrees was higher still in the non-core-Democratic counties in these five states.

And thus it did not matter that the margins among white voters without a college degree were actually more Democratic in these five states than nationally, because this solidly Republican group formed the largest voting bloc in four of these five states. Pennsylvania, basically evenly split between white voters with and without college degrees, also had the largest gap between these two groups: while Clinton broke even among whites with a college degree, she lost whites without a college degree by 32 percentage points, a 32 percentage point gap!

It is unfortunate that we do not have exit polling on the white college/non-college breakdown at the state level in 2012 for comparison, so we can see how these two groups changed both in terms of their relative proportion of the electorate and their propensity to vote Republican. The best we can do is assume they followed a similar pattern as the nation as a whole—roughly equal shares of the electorate but a strong pro-Democratic shift for white voters with a college degree and the opposite shift for white voters without a college degree.

Until next time…

[1] Of the total vote cast.

[2] Hometown of Republican Speaker of the House Paul Ryan


About those recent presidential approval polls…caveat emptor

During President Donald Trump’s recent combative press conference, he cited a new Rasmussen poll showing him at 55% approval.

What the…? I thought.

At first, I thought he had simply read the “disapprove” number as the “approve” number, because all of the presidential approval numbers I had been hearing (primarily from Gallup tracking polls) showed Trump’s approval percentage below 50%.

But then I checked the presidential approval polls on Huffington Post Pollster. Sure enough, the latest Rasmussen poll showed Trump at 55% approval and 45% disapproval, for a net approval of +10%. This poll of 1,500 likely voters was conducted from February 13 to 15, 2017.

My memory of the Gallup polls was also correct, however. The most recent Gallup survey (1,500 adults, February 11-13, 2017) showed Trump with 40% approval and 54% disapproval, for a net approval of -14%.

Umm, excuse me?

Since Trump’s inauguration on January 20, 2017, Gallup and Rasmussen have been surveying Trump’s approval rating using rolling three-day samples; Rasmussen presented four-day samples twice. That means that they began by surveying 500 people each day for three days to obtain an initial sample of 1,500 people. On the fourth day, they replaced the first day’s 500 respondents with a new sample of 500 respondents. This means that the sample of 1,500 respondents is completely updated every three days.

Besides the sampling universe (all adults for Gallup, likely voters for Rasmussen), another difference between the two polling organizations is that Gallup uses live telephone interviewers, while Rasmussen uses Interactive Voice Response (IVR—essentially automated telephone calls) supplemented by an online survey tool. Also, while Gallup allows respondents to provide an answer other than “approve” or “disapprove” (e.g., I don’t know), Rasmussen forces respondents to choose one of these two responses, and thus their percentages always sum to 100.

This was not the first time I had observed substantive differences between Gallup and Rasmussen presidential approval polls. In April 2016, I collected the 100 most recent polls on then-President Barack Obama’s approval (midpoints: January 27 to April 5, 2016). Overall, Obama’s averaged 48.1% approve and 48.4% disapprove, for a net approval of -0.3%. The 22 polls from Gallup, however, averaged 49.5% approve and 46.6% disapprove. The 16 Rasmussen polls averaged 49.2% approve and 49.9% disapprove; the average 0.9% neither approving nor disapproving likely results from rounding reported values to zero decimal places.

Thus, early in 2016, Gallup showed Obama’s net approval at +2.9% while Rasmussen pegged it at -0.7%, a not insubstantial difference of 3.6 percentage points.

What about in the first four weeks of the Trump Administration?

Overall, 40 non-overlapping (i.e., completely updated three-day samples, as appropriate) presidential approval polls have been conducted since January 20, 2017 by 13 different organizations[1]. President Trump has averaged 45.6% approval and 47.6% disapproval over this period, for a net approval of -2.0%. Examining values by consecutive five-day periods, however, shows a declining net approval, from +4.0% to -1.7% to -2.5% to -5.6% to -2.5%.

By contrast, the averages in the nine Gallup presidential approval polls were 43.0% approve/50.4% disapprove, for a net approval of -7.4%, while those in the seven Rasmussen presidential approval polls were 54.3% approve/45.7% disapprove, for a net approval of +8.6%!

In other words, two polling agencies, surveying American adults over the same four week period, found an astonishing 16.0 percentage point difference in President Trump’s net approval rating!

And the gap appears to be widening, especially over the last week, according to Figure 1 below:

Figure 1: Net Approval for President Donald Trump (% Approve – % Disapprove) in Post-Inauguration Polls


Dates shown in Figure 1 are the survey midpoint (or the day after, if the survey was conducted over an even number of days). If multiple surveys had the same midpoint, their values were averaged.

Besides the 16 total Gallup and Rasmussen presidential approval polls, Figure 1 also displays the other 13 presidential approval polls that sampled all adults (average net Trump approval -4.5%, average sample size 2,903), the seven that sampled registered voters (+0.9%, 1,600) and the other four that sample likely voters (-0.2%, 835).

There is admittedly a lot of noise in these data, especially for the 24 polls not from Gallup or Rasmussen. Still, the stark contrast between the Gallup and Rasmussen presidential approval polls is evident, as is the fact that the Gallup net approval numbers are clearly declining. It is less clear whether the same is true for the Rasmussen net approval numbers.

This large difference between the presidential approval polls conducted in the last four weeks by Gallup and Rasmussen may simply result from differences in sample populations. Excluding Gallup and Rasmussen’s presidential approval polls, the difference in average net approval between polls conducted on all adults and those conducted only on likely voters is 4.3 percentage points. And while this gap is substantive, it pales next to the difference in average net approval of 10.9 percentage points when the Gallup and Rasmussen polls ARE included, and it really pales next to the 16.0 percentage point difference discussed earlier.

The remaining difference could be explained by interviewing strategies. Gallup uses live interviewers, who dial telephones and speak directly to respondents, while Rasmussen uses automated (forced-choice) and online techniques. There is a compelling notion that respondents will give more socially “acceptable” answers to a live interviewer than they would to an anonymous machine, but now that Trump is actually president, I suspect the level of “social unacceptability” has declined, if it ever existed.

Basically, differences in sample populations and interviewing strategies account for much of the disparity between the presidential approval polls conducted by Gallup and Rasmussen, but not all of it.

For now, the best course, as always, is to examine an aggregate of presidential approval polls, which suggest a net approval of about -4.0% over the last week (-5.9% on the latest Pollster chart)—which, while not in double-digit Gallup territory, is still lower than any recent president only one month into his first term.

Until next time…

[1] Besides Gallup and Rasmussen, there were four presidential approval polls each from Ipsos/Reuters, Politico/Morning Consult and YouGov/Economist; three from PPP; two each from Quinnipiac and Survey Monkey; and one each from CBS, CNN, Fox, Pew and Zogby. The averages across these 24 surveys were 44.3% approve, 46.9% disapprove (8.8% neither!), for a net approval of -2.6%.

The Democrats’ 2016 “blue wall” thesis

There was a great deal of talk during the 2016 presidential campaign about Democratic nominee Hillary Clinton’s “blue wall” in the Electoral College, with “blue” the color news organizations use to denote states won by Democrats.

The basis of this talk was simple. In general, states tend to vote similarly for president over time; the average correlation between state-level margins (Democratic % – Republican %) in successive elections since 1984 is 0.95. Eighteen states[1] and the District of Columbia (DC), with a total of 242 electoral votes (EV), had voted for the Democratic presidential nominee in six consecutive elections (1992-2012). Three other states—Iowa (6 EV), New Hampshire (4 EV) and New Mexico (5 EV)—had voted for the Democratic nominee in five of these six elections.

Thus, the thinking went, the 2016 Democratic nominee for president would start the general election with an all-but-guaranteed 242-257 EV (of the 270 needed to win the presidency). Add three Democratic-trending states won by Barack Obama twice—Colorado (9 EV), Nevada (6 EV) and Virginia (13 EV)—and there seemed to be at least 270 EV seemingly a lock for any Democratic nominee. 

And that was without two “swing” states that had voted for the Democratic presidential nominee in three (Florida; 29 EV) and four (Ohio; 20 EV) of the previous six elections, including for Obama twice.

Hillary Clinton was a shoo-in to beat Republican Donald Trump in 2016, according to this logic.

Despite winning the popular vote by 2.1 percentage points, she lost the Electoral College 306-232, 10 EV less than the bare minimum she supposedly had banked.



There were always flaws in the “blue wall” argument, but two are most relevant here.

One, state voting patterns change over time. From 1968 through 1988 it was the Republicans who had an even-more-impregnable “red wall,” with 22 states voting for the Republican presidential nominee in six consecutive presidential elections and 13 other states doing so in five of them. The Republicans won the White House in five of these six elections, averaging 417 EV.

Ask President George H. W. Bush how that “red wall” helped him in 1992, when he only won 168 EV (and 23 fewer states than in 1988) in losing to Democrat Bill Clinton.

Two, the Democratic nominee won the presidency four times between 1992 and 2012, averaging 327 EV.  The Democrats also won the popular vote in 2000, while losing the Electoral College and, thus, the White House. The one time Democrats lost the popular vote AND the Electoral College in these years, in 2004, John Kerry still won 251 EV (and 48.3% of the popular vote).

By definition, if you win at least two thirds of the presidential elections in a 20 year span, you are likely winning a significant number of states regularly. And you are likely winning some states by smaller margins than your national margins.

Consider Obama’s reelection bid in 2012, when he beat Republican Mitt Romney in the popular vote by 3.8 percentage points, capturing 332 EV.

Obama won two states—Ohio (D+3.0%) and Florida (D+0.9%)—by less than his national margin. Remove those two states, and his EV total drops to 283. Remove Virginia, which he won by 3.9%, just barely above his national margin, and his EV total drops to 270—exactly the number needed to win. If Obama had lost these three states plus either Colorado or Pennsylvania (both D+5.4%), Romney would have won.

Simply put, when one political party is generally winning the national popular vote, often by large margins, as the Republicans did from 1968 to 1988 (average R+9.6%), and the Democrats did from 1992 to 2012 (average D+3.9%), that party sweeps more “marginal” states into its column. Electoral college “locks” are illusory at worst, and transient at best.


As a long-time student of American electoral geography[2], I maintain an Excel workbook containing every state’s Democratic and Republican presidential vote for every election since 1984, using data from Dave Leip’s indispensable Atlas of U.S. Elections. This dataset allows me to calculate every state’s “Relative Democratic Margin” (RDM) since 1984.

RDM = (state D% – state R%) – (national D% – national R%)

Consider North Carolina, which Obama won by 0.3 percentage points in 2008, then lost by 2.0 percentage points in 2012. Obama beat Republican John McCain nationally in 2008 by 7.3 percentage points. Subtracting 7.3 from 0.3 yields -7.0; North Carolina was 7.0 percentage points less Democratic than the nation in 2008. In 2012, RDM was -5.8 percentage points (-2.0 – 3.8). Thus, North Carolina shifted 1.2 percentage points toward the Democrats, relative to the national vote, between 2008 and 2012.

Multi-election average RDM’s can also be calculated for each state: averages smooth out idiosyncratic voting behavior from a specific election. For example, presidential and vice presidential nominees tend to do slightly better in their home state than its partisan voting history would predict. Consider also Michigan. Democrats won the state in each presidential election from 1992 to 2004 by an average of 7.3 percentage points. Obama won Michigan by 16.4 percentage points in 2008 and 9.5 percentage points in 2012, margins that were likely inflated by strong approval of the auto industry bailout spearheaded by the Obama Administration.

We now know that this history was not especially predictive of 2016, when Trump won Michigan by 0.2 percentage points, dropping its RDM to D-2.3 from D+5.6 (2012) and D+9.2 (2012).

Like I said, oops.


I elected to “weight” average RDM by election recency. The formula I use to calculate a three-election weighted RDM (3W-RDM) is:

(Election 1 RDM + 2*Election 2 RDM + 3*Election 3 RDM)/6

The most recent election thus contributes 50% of the value of the 3W-RDM. Weights and time-frame are admittedly arbitrary. However, there is no discernible difference in the relative ordering of states using alternate weights and time-frames[3].

For example, the 3W-RDM for Pennsylvania, 2004-2012 is thus:

(5.1 + 2*3.1 +3*1.5)/6 = (5.1+6.2+4.5)/6 = 15.8/6 = 2.6

Using this method, Hillary Clinton’s projected 2016 margin in the Keystone state would have been 2.6 + 2.1 = 4.7 percentage points.

Instead, Ms. Clinton lost Pennsylvania by 0.7 percentage points, 5.4 percentage points lower than expected.

Again, oops.


The “projected” 2016 margin for each state can be compared to its actual 2016 margin to see how well the “blue wall” thesis actually held up. Table 1 presents theses values.

Table 1: Projected and Actual 2016 State-level Margins (Democratic % – Republican %)

State EV Projected 2016 Margin (D% – R%) Actual 2016 Margin (D% – R%) Actual Margin – Projected Margin
West Virginia 5 -21.7% -41.7% -20.0%
North Dakota 3 -19.1% -35.7% -16.6%
Iowa 6 4.1% -9.4% -13.6%
South Dakota 3 -17.2% -29.8% -12.6%
Missouri 10 -7.8% -18.5% -10.7%
Maine 4 13.1% 3.0% -10.1%
Michigan 16 8.9% -0.2% -9.2%
Rhode Island 4 24.6% 15.5% -9.1%
Indiana 11 -10.0% -19.0% -9.0%
Ohio 18 0.8% -8.1% -8.9%
Kentucky 8 -21.9% -29.8% -7.9%
Montana 3 -12.9% -20.2% -7.4%
Wisconsin 10 6.3% -0.8% -7.1%
Wyoming 3 -39.6% -46.3% -6.7%
Tennessee 11 -19.4% -26.0% -6.6%
Delaware 3 17.1% 11.3% -5.8%
Pennsylvania 20 4.7% -0.7% -5.4%
Vermont 3 31.6% 26.4% -5.2%
Arkansas 6 -21.9% -26.9% -5.0%
Minnesota 10 6.1% 1.5% -4.6%
New Hampshire 4 4.4% 0.4% -4.0%
Hawaii 4 36.1% 32.2% -3.9%
Alabama 9 -24.4% -27.7% -3.3%
Nevada 6 5.3% 2.4% -2.9%
Mississippi 6 -15.2% -17.8% -2.6%
Connecticut 7 16.0% 13.6% -2.4%
Oklahoma 7 -34.2% -36.4% -2.2%
Nebraska 5 -23.2% -25.1% -1.8%
New York 29 24.3% 22.5% -1.8%
South Carolina 9 -12.9% -14.3% -1.4%
Louisiana 8 -19.1% -19.6% -0.6%
New Mexico 5 8.1% 8.2% 0.1%
Florida 29 -1.3% -1.2% 0.1%
Illinois 20 16.7% 16.9% 0.2%
Oregon 7 10.4% 11.0% 0.6%
New Jersey 14 13.3% 14.0% 0.6%
Idaho 4 -32.5% -31.8% 0.7%
North Carolina 15 -4.8% -3.7% 1.2%
Kansas 6 -21.8% -20.4% 1.4%
Colorado 9 3.0% 4.9% 1.9%
Washington 12 12.4% 15.7% 3.3%
Virginia 13 0.9% 5.3% 4.5%
Maryland 10 21.9% 26.4% 4.6%
Massachusetts 11 22.5% 27.2% 4.7%
DC 3 81.9% 86.8% 4.8%
Georgia 16 -10.2% -5.1% 5.1%
Alaska 3 -20.3% -14.7% 5.5%
Arizona 11 -10.9% -3.5% 7.4%
Texas 38 -17.5% -9.0% 8.5%
California 55 19.4% 30.0% 10.6%
Utah 6 -42.6% -17.9% 24.7

On average, Hillary Clinton’s state-level margin was 2.3 percentage points lower than projected; she underperformed in 31 states (average=-6.7 percentage points), while over performing in 19 states plus DC (+4.5 percentage points). While 2.3 is not an especially large deviation given that each projection is based on only three data points, she still underperformed 61% of the time.

In fact, she lost five states (70 EV) she was projected to win by an average of 5.0 percentage points: Michigan, Wisconsin, Pennsylvania, Iowa and Ohio, although the latter was projected to be very close. Hillary Clinton lost these contiguous Rust Belt/Midwestern states by an average of 3.8 percentage points; Obama had won them by an average of 6.1 percentage points four years earlier. Overall, these five states shifted 8.2 percentage points more Republican relative to the nation from 2012 to 2016.

By contrast, Trump did not lose a single state he was projected to win.

There were 19 states in which Hillary Clinton underperformed her projected margin by five or more percentage points, six of which she underperformed by 10 or more percentage points: Maine (where she lost 1 EV), West Virginia and four contiguous Midwestern states (Missouri, Iowa, South Dakota, North Dakota). In fact, she underperformed in the 12 Midwestern states by 7.7 percentage points, by far the worst of any region.

Hillary Clinton over performed her projected margin by five or more percentage points in only six states: Georgia, Alaska, Arizona, Texas, California and Utah. Third-party candidates, including Utah native Evan McMullin and Libertarian Gary Johnson, did better than expected in Utah (27.8% combined) and Alaska (12.2%), holding down Trump’s percentage in those states. After DC (D+86.8%) and Hawaii (D+32.2%), California was Hillary Clinton’s best state: she won the state by 30.0 percentage points.

Georgia, Arizona and Texas are intriguing future targets for the Democrats, however. Hillary Clinton lost their combined 65 EV by an average 5.9 percentage points, a five percentage point improvement from 2012. They also shifted an average 7.6 percentage points more Democratic relative to the nation from 2012 to 2016.


The final question, then, is what other state- and regional-level shifts can be discerned from these data. One way to assess this is to examine changes in consecutive 3W-RDM. For example, following the 1992 presidential election, Nevada had a 3W-RDM of -8.5%. Over the next six elections, Nevada’s had 3W-RDM scores of -6.9, -5.0, -2.5, 2.0, 3.2 and 2.0, a weighted-average shift of 1.4 percentage points more Democratic every election[4].

Table 2: Average Weighted Trends in State-Level 3W-RDM, 1984-1992 to 2004-2012 and to 2008-2016, and Projected Democratic Margins in 2020

State EV Weighted Average Election Year Change in 3W-RDM,


Weighted Average Election Year Change in 3W-RDM,


Projected 2020 Margin (D% – R%) Adjusted for Trend
Hawaii 4 6.1% 4.5% 38.8%
California 55 2.8% 3.7% 26.8%
Vermont 3 5.2% 3.2% 30.9%
Maryland 10 2.7% 2.7% 25.3%
Virginia 13 2.4% 2.5% 3.9%
DC 3 1.7% 1.8% 83.9%
Colorado 9 1.8% 1.7% 3.9%
New Jersey 14 1.8% 1.5% 13.4%
Nevada 6 2.5% 1.4% 3.4%
Washington 12 1.2% 1.4% 13.4%
Connecticut 7 2.4% 1.4% 14.1%
New Mexico 5 1.5% 1.2% 7.7%
Alaska 3 0.3% 1.2% -18.0%
Delaware 3 2.6% 1.1% 13.6%
Oregon 7 1.3% 1.1% 9.8%
Illinois 20 1.4% 1.0% 15.7%
North Carolina 15 1.0% 1.0% -5.0%
New York 29 1.4% 0.8% 22.4%
Utah 6 -3.8% 0.6% -32.5%
New Hampshire 4 1.6% 0.5% 0.6%
Florida 29 0.5% 0.4% -3.0%
Georgia 16 -0.7% 0.3% -9.2%
Massachusetts 11 -0.3% 0.2% 22.3%
Texas 38 -1.5% 0.1% -15.1%
Arizona 11 -1.4% 0.0% -9.8%
Nebraska 5 0.0% -0.1% -25.9%
South Carolina 9 0.0% -0.2% -15.9%
Maine 4 1.5% -0.3% 5.6%
Wisconsin 10 0.5% -0.6% 0.1%
Michigan 16 0.9% -0.7% 1.6%
Mississippi 6 -0.6% -0.8% -19.2%
Indiana 11 0.4% -0.9% -17.1%
Ohio 18 0.4% -1.0% -6.8%
Rhode Island 4 0.4% -1.0% 17.0%
Kansas 6 -1.7% -1.1% -24.5%
Pennsylvania 20 -0.3% -1.1% -1.5%
Idaho 4 -1.8% -1.2% -35.4%
Minnesota 10 -0.9% -1.3% 0.2%
Montana 3 -1.0% -1.8% -20.4%
Iowa 6 -0.4% -2.2% -6.9%
Alabama 9 -3.3% -2.9% -31.3%
North Dakota 3 -0.8% -2.9% -32.3%
South Dakota 3 -1.9% -3.2% -29.0%
Missouri 10 -2.8% -3.7% -19.7%
Louisiana 8 -5.0% -3.9% -26.1%
Oklahoma 7 -5.2% -4.2% -42.4%
Kentucky 8 -4.4% -4.5% -33.2%
Wyoming 3 -4.9% -4.6% -50.3%
Tennessee 11 -4.8% -4.6% -30.5%
Arkansas 6 -7.4% -6.5% -34.7%
West Virginia 5 -7.6% -8.8% -44.2%

What jumps out first from Table 2 is that there was no clear a priori evidence that Hillary Clinton was in a precarious position in the five key states she lost (adding some support to the “blue wall” thesis). In fact, there was evidence that Democrats’ relative position Michigan, Ohio and Wisconsin was IMPROVING by a weighted-average 0.6 percentage points each election cycle. There was slight evidence of movement away from the Democrats (-0.3%) in Iowa and Pennsylvania, but even then Democrats were still projected to win those states by about four percentage points in 2016.

“Trends” often become apparent only after the fact, unfortunately. Once the 2016 election results are added, the weighted-average trend shifted an average of 1.3 percentage points more Republican in these five states.

In other states, however, 2016 continued a clear strong pro-Republican trend. Nine states had a weighted average trend of at least -2.8 percentage points after both the 2012 and 2016 presidential elections; excepting Wyoming, these states form a contiguous band running from West Virginia (shifting away from the Democrats’ national margins at a rate of 8.8 percentage points per election!) south and west through Kentucky, Tennessee and Alabama, through Missouri, Arkansas and Louisiana into Oklahoma. Add in the Dakotas (and the five Rust Belt/Midwestern states discussed above), and there is clear evidence that a large swath of “flyover” states are trending strongly and steadily away from the Democrats.

At the same time, in 2016, Democrats halted trends away from them in four states: Utah, Georgia, Texas and Arizona. The reversal in Utah likely results from McMullin’s strong 2016 performance there; the Democratic presidential nominee is still projected to do 32.5 percentage points WORSE there than nationally in 2020. And while there are hopeful signs for the Democrats in Arizona, Georgia and Texas, these projections suggest they would still need to win the 2020 presidential by 10-15 percentage points to have a chance to WIN any of those states. Still, as 2016 showed, election-to-election shifts of eight or more percentage points can occur.

Five states continued to trend strongly pro-Democratic (minimum +2.4 percentage points after both 2012 and 2016): Virginia, Maryland, Vermont, California and Hawaii. The latter four, however, were already firmly in the Democratic column (Hillary Clinton won them by an average 28.8 percentage points).

Three other states with at least 1.0 percentage point pro-Democratic shifts in 2012 and 2016 are worth noting: Colorado, Nevada and North Carolina. Along with California, New Mexico and Arizona, Democrats are steadily improving their relative position in the southwestern United States. And Virginia, North Carolina and Georgia are all southeastern states bordering the Atlantic with large urban centers.


It is, of course, possible that the narrow (but Electoral-College-decisive) Trump victories in Michigan, Wisconsin and Pennsylvania that delivered him the White House were idiosyncratic results of the unusual 2016 elections. This would mean that projected presidential vote margins, such as those presented here, still have broad predictive value.

It is perhaps more likely, however, that the broad geographic center of the country is moving steadily away from the Democrats, while a U-shaped swath of states—New England and the Mid-Atlantic south along the Atlantic seaboard to Georgia (minus South Carolina), jumping west to Texas and the southwest, moving up the Pacific Coast to Washington, plus Hawaii—is moving steadily toward the Democrats. Add in the Democratic strongholds of Illinois and Minnesota (maybe), and this latter group of states (and DC) contains 304 EV, although that includes the combined 65 EV in Arizona, Georgia and Texas (and 25 less-certain EV in Minnesota and North Carolina).

I conclude, then, with this geography question for Democrats:

Does your future lie with the disaffected white rural blue collar voters of the Rust Belt (whose dramatic shift toward Trump in 2016 proved decisive), or does it lie with the African-American, Hispanic and college-educated white voters of the southeast and southwest?

Until next time.


[1] California, Connecticut, Delaware, Hawaii, Illinois, Maine, Maryland, Massachusetts, Michigan, Minnesota, New Jersey, New York, Oregon, Pennsylvania, Rhode Island, Vermont, Washington, Wisconsin

[2] Literally, as a political science major at Yale, and later as an ABD doctoral student at Harvard.

[3] I have calculated five-election weighted averages [weights=1,2,3,4,5], used a three-election weight scheme of 1,3,5) and used no weights at all.

[4] Weight scheme=1,2,3,4,5,6


NOIR CITY 15: How Noir is “NOIR?”

My affiliation with the Film Noir Foundation (FNF) began with a “Henchman”-level donation in March 2010.

I had learned about the FNF through Eddie Muller’s film noir DVD commentaries (Muller, the “Czar of Noir,” is President and Founder of the FNF), but it was not until I received my free t-shirt and NOIR CITY 8 program and poster that I became aware of the FNF’s flagship film festival. NOIR CITY, a 10-day screening of 24-27 “noir” films, is held late every January at the Castro Theatre in San Francisco.

Donations and festival ticket sales (individual films or all-film Passport) are the primary funding sources for the FNF, dedicated in large part to the rescue and restoration of 35 mm prints of noir films at risk of being “lost or irreparably damaged.” At a time when the arts feel under siege, and our own history is in danger of being rewritten, this cause seems even more vital.

In the fall of 2013, I was fortunate enough to be able to make a Kingpin donation, which provided me with a Passport to NOIR CITY 12. Still, I wavered on actually flying from Boston to San Francisco for 11 days until I watched the NOIR CITY 12 preview.

I was hooked.

I was even more hooked by the festival itself. Beyond the 27 films themselves (noir-city-12-film-list; 20 I had never seen before), I was seduced by the 1940s-era outfits, the restored glory of the Castro Theatre (down to the old-fashioned snack bar, wide carpeted staircases and live organist), the nightly drinks on the mezzanine and, especially, my fellow attendees—welcoming, passionate fellow-enthusiasts.

Intended to be a one-off lark, I have now attended four consecutive NOIR CITY festivals as a Kingpin contributor, although I had to leave NOIR CITY 14 early due to a family medical emergency, and was trapped for two additional days in San Francisco after NOIR CITY 13 by snowmageddon.

Still, while I enjoyed NOIR CITY 15, something felt very different about this year’s festival. The energy level was lower, the mood was less celebratory, and my impression was that attendance was down from the previous three years I had attended. There was no featured “restoration” as in 2014 (Too Late for Tears [1949]), 2015 (Woman on the Run [1950], The Guilty [1947]) and 2016 (Los Tallos Amargos [The Bitter Stems; 1956]), and each day did not have a “sub-theme” in the program (e.g., “Humphrey Bogart: Artist” from NOIR CITY 14). The official author-signing event for the NOIR CITY ANNUAL 2016 (a collection of the best stories and essays from the year’s FNF quarterly e-magazines) was not as widely-advertised as in the past. And so on…

There are feasible explanations: the festival opened on the day President Donald Trump was inaugurated, sparking massive protests in San Francisco; politics may have trumped art for many potential festival attendees (pun intended); and rain soaked the city for the first four days of the festival.

But the likeliest explanation may be the simplest: NOIR CITY 15 was simply less “noir” than in the past.

Each NOIR CITY has a theme, an organizing principle for the selection of that year’s films. For the first three NOIR CITY’s I attended, the themes were international (films from Argentina, England, France, Germany, Japan, Mexico, Norway and Spain, as well as from the United States), marriage and the arts.

The theme for NOIR CITY 15 was “The Big Knockover: 24 Criminal Capers From Around the Globe.” Muller’s stated intent was to screen a chronologically-, geographically- and tonally-diverse range of “caper” films. Patrons expecting 10 days of (primarily) black-and-white films from the 1940s and 1950s may have been disappointed by the 13 films from 1964 (Kenju Zankoku Monogatari [Cruel Gun Story]) through 2015 (Victoria), all but two of which (Cruel Gun Story, Once a Thief [1965]) were in color. Two of the 11 “classic-era” films (Violent Saturday and The Ladykillers, both 1955) were also in color; I will stretch a point and include the black-and-white Classe Tous Risques (The Big Risk) and The League of Gentlemen (both 1960) as “classic-era.” Thus, only nine of the 24 films (38%) screened at NOIR CITY 15 were classic-era black-and-white films (and only four of them were American: Criss Cross [1949], The Asphalt Jungle [1950], Kansas City Confidential [1951] and The Killing [1956]).

This raises two questions:

1.      Just how “noir” has NOIR CITY been since its inception in 2003?

2.      Has NOIR CITY become less “noir” over time?

The lack of a universally-accepted list of film noir titles makes answering these questions difficult; examine 40 different published film noir “lists,” and you will get 40 different, albeit more-or-less overlapping, sets of titles.

One solution would be to aggregate the lists, not unlike the way political analysts and news organizations aggregate polling data, to see which films are most (and least) often cited as examples of film noir.

I will offer further details in a series of articles, but this is precisely what I have done: collect a wide array of published film noir lists (film-noir-database-sources), both explicit (e.g., encyclopedias, websites) and implicit (books containing a minimum 125 film noir titles within their text, often supplemented by a “filmography”), and entering these films[1] into an Excel workbook. Call it a “film noir database,” if you like. As of this writing, this “database” contains 4,316 titles, although only 4,065 have complete information entered.

Every film in the database has two “noir-consensus” scores:

1.   LISTS: a simple count of how many official lists include the film. As of this writing,  LISTS ranges from 1 to 27. All lists are weighted equally.

2.    POINTS: LISTS plus 0-2 for sub-listing (e.g., the Chronology in Bourde and Chaumeton’s A Panorama of American Film Noir: 1941-1953[2]) plus 0-2 for appearing in a published text discussing 25-124 titles (e.g., 77 films noted in Paul Schrader’s “Notes on Film Noir”[3]). As of this writing, POINTS ranges from 1 to 52.

Let me be very clear: I am NOT saying that films with higher LISTS/POINTS scores are intrinsically more “noir” than films with lower LISTS/POINTS scores.

What I am instead saying is that the higher the LISTS/POINTS score, the higher the level of consensus that a particular title is film noir, because more writers who have studied these films have denoted it as such. At the same time, because a higher POINTS score partly results from inclusion on more-exclusive lists, films with a higher POINTS score can be considered more exemplary of film noir.

I am agnostic as to when or where or in what colors a listed film was released. If it appears on an official list of film noir titles (even if designated “proto-” or “neo-”), it will be entered into the database. Film release years range from 1912 (The Musketeers of Pig Alley) to 2015 (Victoria), with 43% between 1940 and 1959. These films come from a total of 65 nations, with 62% of them produced at least in part in the United States. Finally, 55% of these films are entirely black-and-white, with an additional 2% partially black-and-white. Thus, only 30% of the films in this film noir database are American (in part) black-and-white films released between 1940 and 1959.

Table 1: Distribution of LISTS and POINTS for all titles and for titles screened at one or more NOIR CITY festivals, 2003-17


All Titles (n=4,316)

Noir City Screenings (n=294)






Average (SD*)

3.6 (4.9)

4.0 (6.1)

13.4 (7.9)

16.2 (11.3)




































* Standard deviation (square root of variance), a measure of how values are clustered spread around the mean: the higher the SD, the wider the spread

The midpoint if values are sorted from largest to smallest; half of values are above the median, half are below

Five films appear on all 27 lists: Double Indemnity (1944), Kiss Me Deadly (1955), Laura (1944), The Maltese Falcon (1941), and The Postman Always Rings Twice (1946), with 493 films (11%) appearing on 10 or more lists (see Table 1); more than half (56%) of these films appear on one list only. The average LISTS is 3.6, with a median of 1. Eleven films earned 40 or more POINTS, topped by Out of the Past (1947; 46 POINTS), The Maltese Falcon (47) and Double Indemnity (49), with 531 (12%) films having 10 or more POINTS; more than half (55%) of these films earned one POINT only. The average POINTS is 4.0, with a median of 1.

In other words, while well over four thousand films have been denoted (explicitly or implicitly) by at least one writer as “film noir,” only some 550 titles are cited by even one in three of these writers, and only some 150 by as many as two in three. Many films can be argued, however idiosyncratically, to be noir, but relatively few are widely considered to be so.

Did I mention that one of these 27 lists is the list of unique films screened at the first 15 NOIR CITY festivals? To generate this list, I first counted any film listed in the programs for NOIR CITY’s 8 and 11-15 (the programs currently in my possession). I then supplemented that list using the master NOIR CITY list. This method yielded 294 films screened at one or more NOIR CITY festivals between 2003 and 2017. This list is likely incomplete (26 films listed in the programs for NOIR CITY’s 8 and 11 were not on the master list), but it serves the purposes of this post.

As Table 1 reveals, films screened at NOIR CITY are, on average, far more widely considered noir than is true of the typical film in the database. The average NOIR CITY film appears on 13.4 LISTS, with 16.2 POINTS; median LISTS and POINTS are 14 and 15, respectively. That is, half of the films screened at NOIR CITY fall within the top 6% of LISTS or POINTS in the database. Fully two-thirds of the NOIR CITY films appear on 10 or more LISTS and/or have 10 or more POINTS. Only 21 films appear in the database solely due to their screening at NOIR CITY, and many of them are “pre-code” proto-noirs (e.g., A Kiss Before the Mirror, Laughter in Hell; both 1933), “lost” foreign films (e.g., La Citta Si Defende [Four Ways Out; 1951], El Vampiro Negro [The Black Vampire; 1953], Los Tallos Amargos) or still too recent for fuller noir consideration (Victoria).

Figure 1: NOIR CITY LISTS and POINTS over time


However, there is evidence that the “noir-ness” of NOIR CITY is slowly decreasing. Figure 1 shows that as recently as 2010 (NOIR CITY 8: “Lust and Larceny”), NOIR CITY films were widely considered noir, based upon the noir-consensus measures of LISTS (average=15.2, median=16) and POINTS (18.5, 17.5), anchored by 20+ POINTS films The Asphalt Jungle, Human Desire (1954), Niagara (1953), Odds Against Tomorrow (1959), Pickup On South Street (1953), Pitfall (1948) and The Postman Always Rings Twice.

Average LISTS and POINTS declined sharply just three years later, albeit remaining much higher than the typical film in the database. These values spiked upward in 2015, perhaps because the 20+ POINTS films Caught (1949), Clash by Night (1952), No Man of Her Own (1950), The Set-Up (1949) and Woman on the Run were screened, as well as The Suspect (1944; 19 POINTS). However, they declined in NOIR CITY 14 and even more sharply in NOIR CITY 15.

In fact, average LISTS and POINTS were 56% and 52% lower, respectively, in 2017 than they had been seven years earlier. This decline occurred despite the screening of four “classic era” American films (Criss Cross, The Asphalt Jungle, Kansas City Confidential, The Killing), plus the 1955 French film Du Rififi Chez Les Hommes (Rififi; 16 LISTS, 20 POINTS), with average LISTS and POINTS of 24.0 and 32.8, respectively.

It is the other 19 films that bring down NOIR CITY 15’s average LISTS and POINTS: their LISTS and POINTS averages are 2.7 and 2.8, respectively. Of the 21 films whose appearance in the film noir database is due solely to their screening at one or more NOIR CITY festivals, seven were first screened last month (Blue Collar [1978], Four Ways Out, Cruel Gun Story, The League of Gentlemen, I Soliti Ignoti [Big Deal on Madonna Street; 1958], Thunderbolt and Lightfoot [1974], Victoria). Put another way, 29% of the films screened at NOIR CITY 15 are not cited as film noir anywhere else (in my 36 other current sources, at any rate) as noir.

So here is my synopsis of NOIR CITY 15 which, while far less “noir” (using these noir-consensus measures) than previous NOIR CITY festivals, was still more noir than the typical database film. The first four days of the festival—Friday January 20 through Tuesday January 23—were rain-soaked and overshadowed by the nascent Trump Administration. The 10 films screened over those four days had LISTS and POINTS averages of 13.0 and 17.5, respectively, and were released between 1949 and 1958. This was, arguably, the “classic noir” segment of the festival—and who knows how many potential attendees stayed away because of the rain, the protests or the politics. Once the skies and roads cleared (if not the politics), the program quickly veered into the colorful 1970s and beyond, with a very low noir-consensus series of films. Potential attendees who had missed the first 10 films may then have been more inclined to stay away from the festival.

None of this is meant in any way as a critique of NOIR CITY 15, or of any film screened as part of that festival. I am not a film critic, nor do I have a particular set of a priori criteria to apply to films to say “this one is noir” and “that one is not.” Not yet, anyway.

Speaking solely as a fan, I loved the first 10 films screened[4]…but I also enjoyed the later 14 films, including three (The Ladykillers, Charley Varrick [1973], The Brink’s Job [1978]), I was looking forward to seeing again. Granted, Thunderbolt and Lightfoot, Straight Time (1978) and Sexy Beast (2000) did not thrill me as much as The Taking of Pelham One Two Three (1974), Blue Collar (1978), El Aura (The Aura; 2005) and Victoria.

Too often underappreciated is the difficulty of programming a festival like this year after year, while still keeping it fresh, and the commitment to an annual theme which results in selections that may irk the purists and the “fair-weather” fans (pun also intended). Muller will tell anyone who asks that he is not close to running out of classic-era black-and-white films (my language, not Muller’s) to screen. Still, the number of non-English-language films (26 in the past four years), the number of post-1959 films (21 in the last four years), and the number of color films (19 in the past four years) suggests a desire to advance a conception of film noir beyond that American, black-and-white, 1940-59 30%.

And thus Muller will continue to screen films that challenge the conventional definition of film noir (amorphous as it is), and I will put any film into my database that a published list includes as noir, regardless of era or nationality or color scheme.

Definitions advance. Consensus evolves.

Only, what, a little over 49 weeks until NOIR CITY 16?

Until next time…


[1] Plus alternate titles, year, BW/color, director[s], cinematographer[s], country/ies of production, primary studio, etc.

[2] Borde, Raymond and Chaumeton, Etienne. 2002. A Panorama of American Film Noir: 1941-1953. Translated from the French by Paul Hammond

[3] Schrader, Paul. 1972. “Notes on Film Noir.” Film Comment 8:1, pp. 8-13

[4] Well, OK, I didn’t love The Big Risk nearly as much.