Creating Robust Small Area Population Estimates

Data nerds with long memories always remember challenges from the decadal round of censuses. Here in the UK the highlights have included. The first small area geographies for Enumeration Districts (ED) in 1991, which cost over £100,000 to license a year! The missing million from the 2001 Census (turns out more youngsters had scarpered to the EEC than expected). From 2011 we had the undercount controversy affecting specific Local Authorities. This year we face dealing with the challenges from a pandemic census.

The first small area results were released this month, I am not going to provide early analyses from these, that will be coming in the new Year as we get closer to a full set of the key univariate tables. Rather I thought it would useful to outline a couple of those pandemic challenges.

The biggest challenge relates to the simplest of all census counts. How many people live in each Output Area. Remember a full census is a snapshot of a particular week 18 months ago. A time of lockdowns, work from home orders, closed universities, bans on travel, a crazy housing market, and dealing with the premature death of 200,000 loved ones.

Based on some early work we did comparing population estimates to reported census populations we had a clue that somewhere between 500,000 and 1,000,000 seemed to be in the ‘wrong’ place. Based on survey data (and having lived through it), we believe.

  • Some people who had access to second homes in the country had left principal homes closer to their work in large cities.
  • Students at University Halls had been ordered to leave, some to travel ‘home’ some to fend for themselves.
  • Many international students had left the country to continue studies remotely from outside the UK.
  • People were (perhaps temporarily) moving out of cities to rent or buy properties in more affordable areas from which they could work from home.
  • The rental market was still facing all sorts of disruption including flows into and out of the short stay ‘airbnb’ economy.

Migration patterns were severely disrupted with many European workers returning to their country of origin either permanently or temporarily. Furthermore, the established patterns of new in and out migration had in affect been frozen for the previous 12 months, creating some ‘pent up’ migratory pressures.

It is worth remembering that the census asked us to fill in the questionnaire as things were at the time of completion. Respondents had to make assessments as to what they considered their usual residence, and provide details of other places they sometimes live. In addition to the usually resident population the householder also had to provide details of ‘visitors’ who were in the household but would be completing a census form elsewhere.

The five areas of disruption outlined will have been distinct data signatures. A key area of work is how we use the census to roll forward small area estimates to 2022, 2023 and beyond. In order for the census to act as the reliable ground truth to do that we need to have ways of quantifying at small areas the impacts of the disruption.

The data to do this fully have yet to be published. The ONS will be doing their own exercise to reallocate students from mainly parental properties back to their halls. But they won’t be doing the same for those living in rented accommodation, and the ‘flight to the country’ and the migration disruptions will need to be unwound or not. Some of the changes will have stuck.

In addition to the detailed census at Geolytix we will be making use of some very large scale administrative data sets to help us figure these questions out. These sources include but are not limited to.

  1. Land Registry data on housing stock and it’s sale and purchase.
  2. Aggregations of raw mobile phone geolocated activity logs.
  3. Aggregations of spending pattern data from debit and credit cards, both registered home areas and areas of spending are available.
  4. Anonymised loyalty card data aggregations from major retailers who saw peoples areas of shopping move around.
  5. Other large scale ONS population movement data including in and out migration surveys.
  6. Geolytix data around locations and size of university Halls, pre-Census OA level estimates of population, and updated place of origin for recent migrants.
  7. Previous census data (2011) to establish some baselines for non-standard accommodation types.

These disruptions will also potentially affect the demography of small areas. Was it that it was the more affluent who ‘fled to the country’ thus skewing small areas characteristics either permanently or temporarily? And did the international students come back? Or did the patterns post census stick? What was the balance between permanent house moves and temporary ones that have since unwound? There are other challenges around unoccupied properties, have all these voids since been filled. How much Airbnb stock has returned to the rental market.

As well as dealing with this disruption we will see the demographic changes caused by the tragic premature deaths of 200,000 mainly elderly residents. Whether these deaths ‘pulled’ forward need to change our ongoing assumptions around death remain unclear.

In short we are going to be busy early next year!


Blair Freebairn, CEO at GEOLYTIX

Photo by Pawel Czerwinski on Unsplash