The Census is not 100% Accurate

Judging by reading the #tcot twitter stream ("top conservatives on twitter") and sites like this one (claiming 50,000 hits in less than 2 days online), it appears rather than learning that reality > ideological fantasy, conservatives are intstead opting to double-down on ideological fantasy and are busy constructing an even more elaborate set of voter-fraud theories. 

The leading attempt to empiricize their claims revolves around locating counties which have more registered voters than the US census bureau indicates their population should be in its county level data set.

The first and most obvious thing to point out to conservatives is that the number of registered voters in a county is not the number of Obama votes or even the number of total votes recorded in that county as I doubt more than a tiny number of US counties see 100% turnout in any election.  Typical of the right wing noise machine, they appear to be deliberately conflating these things, such as this RedState headline "Colorado Counties Have More Voters Than People" - the article itself is discussing voter registration but the headline is pretty clearly mendacious, indicative that these counties had more actual votes cast than people living there.

Putting aside the sadly typical right wing disingenuity, the deeper flaw of all this is that they're basing the number of "people" in these counties on the US census, as if the US census was some kind of perfect count of population without any error, and completely accurate even 2 years after the actual 2010 census attempted to count everyone (something that no census actually achieves).  The Census is not 100% accurate even the year it is taken, and every year becomes less accurate.  It is simply insufficient to conclude that because a county has more registered voters than the census shows as population that this is indicative of anything untoward. 

The Census attempts to estimate population changes at State and County level over time between actual censuses, but of course these are estimates and subject to error.  According to research into past censuses, the average of these errors in small counties can be as high as 8.8% per year, and for all counties, averages 3.5% to 4.6% in error.  That smaller counties are more subject to drift over time may appear counter intuitive, as it should seem simplier to accurately count the number of people in a small county, but the drift is more significant because small events like a single factory opening or closing can have an outsized effect on small counties.  A county that includes a big city is going to "average out" events that cause people to leave the county with events that cause people to enter it.

Consider all the variables and sources of error:

• Some people really don't want to be counted in the census.  This includes Michelle Bachman and the paranoid right, who were particularly concerned about the 2010 census seeing as it was being run under President Obama. 
• People die, and while national death rates are easy to measure, there's no reason to assume people die evenly at the county level.  A county with lots of elderly people is going to have a lot more deaths than a young county.
• People have babies, again unevenly.
• People move internally, and the US census isn't notified
• Legal immigration - numbers of immigrants can be measured, but what counties they settle in, not so much
• The 2008 economic collapse and housing crisis has made for a much higher deal of internal migration and churn as a much larger than normal group of people were evicted and foreclosed, forced to move in with relatives and so forth. 

We can go on, but the point is that the census is only a tool.  It may be useful as a guide to compare its numbers against what various counties recorded for voter registration is suggestive but hardly conclusive.  In fact, it is highly likely that the county data is in some sense more accurate since it is gathered every year.

As an amusing coda, I looked into several of the counties that RedState fixated on in Colorado, and as predicted above, in every case, the counties were low population places (all below 10,000 by the 2010 Census), and in every case I checked, had 2012 results that closely matched the county's 2008 Presidential results as far as two-party breakdown:  San Juan 2008 (Obama 53, McCain 44), San Juan 2012 (Obama 52, Romney 42), Mineral 2008 (Obama 43 McCain 54), Mineral 2012 (Obama 45 Romney 45), Gilpin 2008 (Obama 59 McCain 38), Gilpin 2012 (Obama 57 Romney 40), Hinsdale 2008 (Obama 39 McCain 56), Hinsdale 2012 (Obama 38 Romney 59).  Some trail of fraud there!  Each county voted pretty much the same way it did in 2008, and in only 1 did Obama actually increase his vote share. Amusingly, RedState contacted local officials in these states and they all gave explanations for the disparity in their registration figures to the census that comport to what I said above - transient populations, vacation homes, students, errors in the census. 

Because I've already wasted more time on this than I care to admit, I will also point and laugh at this bit, which notes that Obama lost EVERY state that has a photo ID requirement.  Turns out there are a huge sample of four states with such requirements:  Kansas, Indiana, Georgia and Tennessee.  Lucky those voter ID laws were there to stop Obama from winning Kansas, a state that last voted for a Democrat in 1964.  Al Gore couldn't win Tennessee in 2000, but surely voter fraud would have given it to Obama.  Indiana and Georgia are a little more interesting, in that Indiana did vote for Obama (barely) in 2008, but per polling no one expected him to win it in 2012.  Georgia was also suprisingly close in 2008, only going for McCain by 5% but there was no reason to think it could possibly have voted for Obama this time around.  Really at best these voter ID laws cemented a foregone conclusion by disenfranchising enough potential Obama supporters to merely pad Romney's margin in these states (but were probably consequential at the House and State level races).  The whole argument presumes that voter photo ID laws only prevent fraud that is presumed to exist.  But if they also (and more substantially) disenfranchise voters without easy access to photo ID, then they also succeed at keeping teetering "red" states more solidly red than they might otherwise be.