More ESRI and Twitter

A while ago I posted about a map from esri.com that incorporated social media such as Twitter with more traditional sources of fire information.  My conclusion was that the social data was not that helpful.

ESRI has just put out another set of maps, this time on acceptance same-sex marriage.  The first map uses a proprietary demographic model to try to rate how likely a given county is to favor same-sex marriage.  As far as I can tell, this model isn't particularly based on, say, polling data or election results from ballot measures, but more on factors like how urban or rural the county is, how many people have gone to college and so forth.

The second map shows state laws.  It corresponds roughly with the first map.  With the interesting exception of Iowa, the first map shows significant support where same-sex marriage is legal, as one would expect.

The third map is based on Twitter data.  It says there is strong support for same-sex marriage across the country, opposition in five states (including Minnesota, despite significant support in the populous Twin Cities), and very strong opposition in exactly one state: North Carolina.  Idaho, Wyoming and Vermont had insufficient data.

The support map shows North Carolina as fairly similar to neighboring Virginia, which the Twitter data shows as strongly supportive, and as more supportive than neighboring South Carolina and Tennessee. The Twitter data show both of those states as moderately supportive.

Clearly something is out of line here.  Two possible explanations:

• Concerning the overall map, the Twitterverse is not a representative sample.  Overall, Twitter traffic is much more supportive of same-sex marriage than the country as a whole.  This probably shouldn't come as a surprise.
• Concerning North Carolina, the Twitter data covers May 9 through June 30, 2012.  As the map explanation notes, North Carolina had voted on May 8 against a proposition supporting same-sex marriage, by a roughly 60-40 margin after a very intense campaign.  It would be interesting to know what portion of the Twitter traffic surveyed is from the immediate aftermath of that election.  My guess would be a large portion.

Along with the fire map I mentioned above and a study on Twitter rumors after the London riots, which was presented in the Guardian as confirming the notion that Twitter is a good, self-correcting source of information but in fact shows anything but, this is the third piece of evidence I've run across suggesting that you should treat any inferences drawn from Twitter traffic with a grain of salt.

Twitter may well be a great way to find out what people, at least those with Twitter accounts, are paying attention to at the moment, but it's risky, to say the least, to draw conclusions about objective facts from that.