Using Twitter data to map local commuting patterns

One of the early areas of focus of the NEXUS project has been to look at the extent to which Twitter data can be used to accurately map commuting patterns (and what types of biases etc. need to be corrected before this mapping can really work). As part of this Graham has set up an interactive map (below) which allows us to visualise the relationship between twitter data and commuting in the UK, at the local authority level.

Click on the map and a local authority will be selected (highlighted in blue). The map on the left shows where people who work in that local authority live. The middle map shows where people who live in that local authority work. On the right, you can see our estimation of the middle map, based on one year’s worth of Twitter data, The more the right hand map looks like the one in the middle, the better job Twitter is doing at predicting commuting data.

A few notes about the map:

  • Tweets are from 2015 but the commuting data is from the 2011 census – so that’s a first source of inaccuracy
  • We assign users to living and working locations on the basis of where they send the most and second most tweets from
  • Brighter reds indicate a stronger connection (these are normalized to be specific to each local authority)
  • Doesn’t work with touchscreens yet (sorry!)

Findings? We can see there is some crossover, i.e. some signal in the data, which is really interesting. We can also see a few obvious sources of inaccuracy: big cities tend to be picked up in the Twitter map even if they have little to do with commuting for that area, some holiday destinations (e.g. Cornwall) appear to have the same problem. In general Twitter maps seem to have more local authorities lit up than the ones in the commuting data. One of the next aims for the project is to keep working on the way we produce these estimations to see if better ones can be brought out of the data.