Over the course of this year, I haven’t posted much here, but I have been busy on a project in my free time and that is creating maps of the primary race as it stands between the Democratic presidential candidates.
I have been posting past versions on my diary at Daily Kos, including this latest update, but I figured that this site could be of use as well and so, in the future I will likely post in both locations. Note that I have created a Maps page, which will be updated as new version are created.
So without further adieu, here is are the results of the 2008 Democratic presidential nomination process by county thus far:
Of course, there are many caveats, the biggest of which is that not all counties have the same number of voters and those numbers are not proportional to the size of the counties in the map. That will be addressed after the break.
Note that all maps will be presented with a main color scheme (Clinton-red, Edwards-green, Obama-blue) map as the large map and square thumbnails of the other 5 color schemes below it. Clicking on any of the thumbnails, including the large map, will lead you to their Flickr pages, where you can view and download a larger version of the map. Feel free to redistribute and copy the map, but if you do please give me credit by linking here or naming me (you can use either Ben Main or Meng Bomin).
Beyond voter density issues, it should be noted that the states of Alaska, Kansas, and North Dakota did not organize their caucuses in such a way that county results were available, and thus I made due with the smallest level of organization possible. In North Dakota and Alaska, that was the state house districts. In Kansas, it was the state senate districts. That is why, especially in Kansas, the division looks strange.
Another issue with this map comes from the Michigan and Florida contests. Since both states were stripped of delegates by the DNC Rules and Bylaws committee last fall for violating both the spirit and letter of the DNC scheduling rules and since no candidate campaigned in either state beyond a few private fundraisers after that decision, with Barack Obama, John Edwards, Bill Richardson, and Joe Biden taking themselves off the Michigan ballot, they lack a certain legitimacy that is not easily reflected in a map.
For that reason, I have made several variants of the map, which unlike the map above, do not treat them as normal contests
Map with modified Michigan results
Michigan is of particular interest because Senator Clinton was the only candidate on the ballot, with the biggest challenger being “Uncommitted”. Since Uncommitted is not one of the three major candidates represented in the map, Uncommitted votes are only reflected in the darkening of counties proportionally to the vote share of Uncommitted, which doesn’t differentiate them from Kucinich or Dodd votes. Unfortunately, this gives the incorrect impression to the casual observer that Senator Clinton had universal support in Michigan.
So in the following map, Uncommitted votes in Michigan are treated as if they were votes for Obama. There are several reasons that I did not originally do this. Among them were the fact that Edwards was still in the race and one cannot truly tell the mind of voters. Furthermore, many potential voters probably decided to stay home knowing that their vote wouldn’t count towards any national convention delegates. This is evidenced by the fact that Florida and Michigan were some of the few states where Republican primary voters outnumbered Democratic primary voters.
As a result of these issues, Michigan was significantly more “pro-Clinton” than the surrounding states, with the border counties in Michigan being more “pro-Clinton” than their neighbors in both Wisconsin and Ohio.
Map without Michigan results
Of course, instead of dealing with a monotone Michigan or one where one voting option was treated as if it was a vote for someone who was not on the ballot, Michigan can simply be removed from the map, and that’s what this version is.
Map without Michigan or Florida results
The final modification is to remove both states from the map to show that no legitimate primaries have been held in either state.
Of course, none of these maps deal with what I called the biggest issue above, and that is the issue of population distribution. In the general election, there were almost as many voters on the island of Manhattan than in the entire state of Nebraska, and since Nebraska tends to vote Republican and Manhattan Democrat, there are significantly more Democrats in Manhattan than Nebraska. However, Nebraska occupies a large portion of the map whereas Manhattan is a small speck.
This land area bias has a very strong impact on the map. Take Missouri, where, if you have been following the race, you would know that Barack Obama narrowly beat Hillary Clinton in the popular vote there, but looking at the map above, the average observer might think that Senator Clinton won by a landslide because of her strength in rural areas. The same could be said of this USAToday map of the 2000 election, which makes it look like a Bush landslide despite the fact that Gore won the popular vote.
This was a problem encountered by Robert J. Vanderbrei, a computer science professor at Princeton, who created the Purple America map, which was the original inspiration for my map. Some people asked him if he could make a map that was darkened in less populated areas to correct for this problem. Here’s his response:
Can you make the areas with high population density brighter than those with low density? Yes, I tried that. The trouble is that the big cities are so much more densely populated than everywhere else that the map appears black with just a few small bright counties. Unfortunately, computer monitors have a dynamic intensity range of just 256 and this is not enough for an intensity differentiated map. I experimented with some nonlinear transformations (such as a logarithmic or gamma-power law) and so doing was able to make a map with darker, but not black, unpopulated areas. But, this seems even more misleading because the viewer is told that the intensity represents population density and then thinks more people live in the unpopulated areas than actually do—the correct luminosity is virtually black. So, instead, here is a map from the International Dark Sky Association. It shows quite dramatically where most people live.
Eventually, he made a 3-D version where counties were assigned heights proportional to the number of votes per area in a given county. However, I think that the reasoning he used is a bit faulty. While it is true that there is a vast difference in population density between major urban centers and rural areas such that a proportional representation by lighting would leave most of the map dark. Indeed, if Manhattan were set at an intensity of 255 (the maximum), Cook County, IL, where Chicago is located, would have an intesity of 20, which would make it barely visible and in fact, most of the US outside New York City would be virtually or totally black.
So what is the best option? It’s one that Vanderbrei tried and discarded and that is the logarithmic scale. While it is true that if one were to take an average of the color levels of the pixels in a map that was weighted logarithmically, the result wouldn’t accurately reflect the popular vote, but that doesn’t neglect the usefulness of a logarithmic scale, which has many applications including population dynamics, and chemistry (pH is simply the negative base ten logarithm of the concentration of hydrogen ions in a solution given in moles per liter). And there’s even evidence that children naturally start out with a logarithmic understanding of numbers.
However, the primary vote adds another wrinkle to the equation: unlike a general election, different states have very different ways of running their contests, some opting toward caucuses, others opting toward closed primaries, and still others using open primaries. In the end, there is a great difference in voter turnout between states that are otherwise quite similar, such as Minnesota, which opted for a caucus, and Wisconsin, which opted for an open primary. So in the end, I chose this:
This is a map on of Kerry voter density by county weighted on a logarithmic scale. The most intense county is New York County (Manhattan). Since I set up the scale before trying to find the emptiest county in Alaska, Nebraska, or Montana, I took a short-cut in the formula by taking the logarithm of the density of Kerry voters in voters per square mile <em>plus one</em> as a logarithm of a number less than 1 is negative and there are several counties in the above mentioned states among others with Kerry voter density significantly below 1 voter per square mile.
Now, weighting the map based upon Kerry voter density is far from a perfect solution, but I thought that it was the best at the time because it (1) represents a contest where voting procedures were more similar between states than the current primaries, making them giving a more accurate picture of the distribution of Democratic voters, (2) gives a more accurate reflection of the weighting of delegates to the national convention, as that system is in part based upon the 2004 election (as well as 2000 and 1996) and since (3) allows a look toward the future of the race, as states that have not yet voted are included on the map, something that is not possible with a primary election turnout map (4) doesn’t rely upon projections for caucus state turnout (5) allows greater clarity of votes in rural areas; New York City had a higher proportion of voter turnout in the primaries than in the general election in comparison to the rest of the country, and would thus effectively darken the rest of the nation further, obscuring the patterns that the map is meant to show.
That is not to say there are drawbacks. Obviously, this doesn’t precisely reflect actual voter turnout and the density of voters from county to county within states may actually be somewhat different in the primaries than in the general election among other problems relating to the fact that this is not weighted by the actual voters, but I personally feel that this is outweighed by the underrepresentation of caucus results in such a map (a state earns a set number of delegates whether 1,000 voters show up or 1,000,000) and the inability to look to the future of the contest.
For the following maps, I strongly recommend that you click through to look at the bigger version, since they are significantly darker than the unweighted maps and detail is lost in the smaller versions of the map that are used in this post.
Map weighted by Kerry voter density
This is quite simply a weighted version of the Modified Michigan map above. Because of the nature of the logarithmic weighting, I felt that using the generic version of the map where Clinton support is shown in a linear intensity scale would give the false impression both that the Detroit area is less populated than it actually is and that it has stronger Clinton support than it actually does (counties with greater voter density tended to vote more heavily for Uncommitted and other non-Clinton candidates, as can be seen in the unweighted maps above).
One of the trends that this map makes abundantly clear is that in most cases throughout the country, urban areas vote more heavily for Obama than the surrounding suburban and rural areas. This is perhaps most clear in states like Tennessee, Ohio, and Missouri, where the rural areas vote strongly for Clinton and the urban areas lean Obama, leading to much closer contests and in the case of Missouri, a different winner than one would suspect by looking at the unweighted map.
It also drowns out the statistical noise that comes from low voter turnout like in one counties in rural Nebraska where the solitary voter voted for Clinton, thus making the county seem like a strange blip on the unweighted map as well as showing where a candidate’s true support base in a state is. For instance, it is clear that many of the Clinton votes in Texas came from the El Paso area and from Latino communities in the very south of Texas rather than the Texas panhandle, which though strongly leaning toward Clinton is rather sparsely inhabited.
Map weighted by Kerry voter density without Michigan and Florida results
Since one of the strengths of the Kerry voter density weighting that I cited was the better reflection it gives of delegate power (though still far from perfect), I had to make a map removing the two states that have been stripped of delegates.
Map weighted by Kerry voter density with future contests in gray
Another strength I cited was the ability to look forward using the Kerry voter density weighting. In past diaries, commenters have noted that if broad trends continue as expected, Kentucky, West Virginia, and Pennsylvania should go to Clinton, with Oregon, Montana, South Dakota, and North Carolina going to Obama and Indiana being a tossup.
In this version, I’ve added a gray coloring to states that have yet to vote and in fact, they are colored as they would be if voters in each of the remaining counties voted in equal parts for Clinton, Obama, and Edwards, which is unlikely. What this shows, though, is that Pennsylvania may not be such a lost cause and that Indiana may have a more pro-Obama disposition than would otherwise be predicted.
Map weighted by Kerry voter density with Iowa county convention restuls and future contests in gray
The one point of difference between this final map and the previous map is that instead of using the results of the Iowa caucuses where Edwards did quite well, I have used the results of the Iowa county conventions, which shifted toward Obama, as I described here. I added this map, not because I think that it better reflects the will of Iowa voters (it doesn’t), but rather because it better reflects the regional Clinton-Obama leanings and because, with the better performance by Edwards here, Iowa looks rather similar to states that have not yet voted, and thus clouds our view toward South Dakota, which shares a border with Iowa, a border along which the highest concentration of voters is present.
That’s all for now. Between now and the Pennsylvania primary on April 22, I will release some maps of state contests not shown in the above maps, including the Texas caucuses/county conventions, the Iowa county conventions, and the Washington primary.