Unemployment Map FAQs

What is this a map of?
This is a map of seasonally adjusted unemployment by county by month in the United States. You can view it either as a population-based cartogram or in an ordinary Mercator projection.
What does "seasonally adjusted" mean? Why do you need to adjust?
Many places have seasonal ups and downs. For example, there are not many agricultural jobs in Minnesota in the winter. Counties with large universities tend to have fewer jobs during the summer. If you do seasonal adjustments, it is easier to see when a change in unemployment is something to pay attention to and when it is normal and expected.
How did you do the seasonal adjustment? Why didn't you just get those numbers from the Bureau of Labor Statistics?
The Census Bureau does not provide seasonally adjusted numbers at the county level, only at the national level. They use a complex program called ARIMA to do seasonal adjustments. ARIMA has several parameters, and it looks to me like the parameters would be different (and require hand-tweaking) for each of the approximately 3100 counties. Instead, I used a function built in to the STL function of the R statistical package to smooth the unadjusted by-county data that I got from the Bureau of Labor Statistics.
Why is the map all distorted?
It is a population-based cartogram, meaning that the areas are distorted so that the area of each county is proportional to the county's population.
Why does the Continental US cartogram look so much bigger compared to the standard (mercator) projection?
Because it sucked up a lot of the pixels from Alaska. Alaska has a huge land area, which looks even bigger than it really is because of the distortions from a Mercator projection. Meanwhile, Alaska has a very low population compared to the Continental US, so the Continental US essentially "sucked up" all the area from Alaska.
Why didn't you just cut out Alaska then? Wouldn't that make it look better?
I did try it without Alaska, but that took a fair bit of extra work, didn't change the shape of the Continental US states (just the size), and Alaska is a legitimate member of the United States. Alaskans have unemployment, too!
Why do all the states look "fat"?
They don't. Populous states -- which tend to be on the oceans and so perhaps the ones you see first -- end up kind of fat-looking because they suck up area from other states. However, states with a low population end up looking scrawny. Not just Alaska, but the northern mountain states (e.g. Wyoming and Montana) look sucked-in.
Why do the city names appear and disappear as you move the map around?
People did not build cities in locations which were convenient for map makers. Dallas and Fort Worth, for example, are so close together that showing both on a map (especially a non-cartographic map) would be ugly and/or confusing. The Big Map Players probably make lists of which cities to show at every zoom level. I'm sorry, but I don't have the budget to do that.

Instead, I go through the cities in order of population. If I come to a city which is too close to a city which has already has its name on the map, I don't show that city's name. When I get to 25 cities on the viewable map, I stop.

This means that if you move a city so that it is no longer visible, that might "unblock" a smaller city. For example, if you move the map to the right until New York isn't visible, Newark will probably appear (at the default zoom level).

Why don't all of the county borders get drawn sometimes?
If the counties are too small, then the borders take up so much of the room that it's hard to see the unemployment colour.
Why did you just say "colour" there?
I am a US citizen, but I live in Canada now.
What controls aren't obvious?
If you click on the county graph, the map will change to show the country's unemployment as of the date you clicked on.
What is interesting about these maps?
The cartograms are interesting for showing just how many people live in cities compared to the country.

What I found more striking about the unemployment data was:

  1. The crash of 2008 was very bad. The difference between July of 2008 and November of 2009 is stark.
  2. Hurricane Katrina was very bad. If you look at the period between September 2005 and June 2006, you will see that the area around New Orleans is grey. That means they couldn't even collect data. This is the only time and place since January 1990 where they couldn't collect data. (Hurricane Sandy, by comparison, doesn't have much of an effect on the maps.) Counties in the vicinity of New Orleans which did have data had extremely high spikes in unemployment.
  3. It sucks to be Detroit. (There are other counties which also have chronic unemployment, like Imperial County CA and Yuma County AZ, but they aren't nearly as populous.)
  4. Populous counties tend to have smaller seasonal variations than smaller counties.
Why do the Katrina-ravaged Louisiana counties show some non-zero unemployment value if no data was collected?
Because of how R works, the seasonal adjustment was applied to all months, not just the months which had data. (If you look carefully at the pale blue line, you'll see that it actually is flat at zero during that period.
Can you see the effects of other hurricanes on unemployment?
Nothing as dramatic. There seem to be small decreases in unemployment in Florida right after Hurricane Andrew (Aug 1992) and in New York after Sandy (Oct 2012), but the decreases are minor. For your reference, here are the dates of some destructive hurricanes:
  1. Andrew Aug 1992
  2. Charley Aug 2004
  3. Ivan Sept 2004
  4. Wilma Oct 2005
  5. Ike Sept 2008
  6. Irene Aug 2011
  7. Sandy Oct 2012
How did you make the cartograms?
I used the program ScapeToad, which implements of the Gastner/Newman [2004] cartogram technique. (Newman has another implementation on his web site.) I defined cartogram parameters manually, and used 6000 points for the first grid, 1024 for the second, and 3 iterations. As a starting point, I used the county shapefiles from the US Census Bureau. For the population, I used the Census Bureau intercensal estimates.
Are the population estimates you used to make the cartograms accurate?
Not very, unfortunately. The intercensual estimates are just an interpolation between the two every-ten-years census figures on either side. They are okay for a rough estimate, but they won't show dramatic changes (like the post-Katrina exodus from Louisiana) accurately.
How do you make the map tiles?
I use a framework that I wrote myself in PHP. You can read more about the technical details in a blog post, Optimizing Map Tile Generation.

For this map, I actually pre-rendered complete (static) tiles for zoom levels 0-5, just because there are so many and you can whip through them so quickly by clicking on the county graph or by hammering on the "Next Month" or "Previous Month" buttons. Tiles at higher zoom levels have the geometric skeletons pre-rendered, but the colouring is done on-the-fly.

Why does the name Newman sound vaguely familiar in the context of cartograms?
Probably because he did the Purple America elections cartograms.
How did you make the graphs of the unemployment by county over time?
Using the R statistical package. It has really great graphing tools.