We have now put a little more detail around the outline model , in particular starting to consider the work required at each phase, including where specific experiments are required as well as starting to identify key data sources (and contacts) required for the work.
Pre-setup – Define scope
Data outputs: It’s going to be csv. Unless a really clever system comes along, we’re working on the assumption that this will be done in Excel. We want to build something which is replicable amongst a range of users, so if we want to avoid avoid programming languages if at all possible.
Demographic, Geographical and time-series scope:
Time Series: The time series are largely defined by the user need, in this instance we are starting off with shorter-term budget planning which sits in the 2-3 year timeframe. However we can anticipate longer-term demand for planning policy and economic development purposes, so we will look towards a 10 year period as an interim position.
Demographic: Overall volume, single year of age and sex are the minimum required data to support requested demand modelling inputs. These variables have the advantage of being replicated within published ONS population models and feel like a logical place to start.
Geographic: Historically local government wards have been the best spatial level to use as they contain a good level of local familiarity as well as being structured within ONS geographical structures, unfortunately we start this process part way through a local government boundary review which will culminate in the redrawing of local ward boundaries.
To that end we will look initially at using an LSOA level forecast. There are some initial concerns that this may create geographies too small for accurate analysis and so we may well review.
Phase 1 – Define a baseline
As we are using standard ONS geographies and categories in our scope it seems logical to use ONS mid-year small area population estimates as a baseline. 2016 are the most recently published. Let’s hope it’s all this easy…
Phase 2 – Remove “static” population
… it isn’t. Static populations are those which are unlikely to change significantly over time, e.g. an 18 year old student moving out of halls of residence is likely replaced by another 18 year old. Within B&NES these are fairly easily identified as our residential care home and student halls of residence.
There are a range of data which could be investigated as part of this.
- Care home beds – work is currently underway on a market position statement for adult social care, which could include this. – This data is not currently open, so we will investigate making it so. (This adds health & social care commissioners to our list of people we need to speak to).
- Student halls of residence: This is harder, We know that the ONS have been working with HESA on residency based student populations as part of their work to define an administrative data derived census. Existing data releases do not seem to cover this, so we may need to make a specific request, which could have licensing and re-use implications
Council tax records hold student discounts (add Council Tax to the list of people to speak to) at a property level, but this does not contain occupancy volume.
Population profile of static populations: We do not have consistent local data for either population regarding granular age and sex details. This will require more detailed conversations, but in absence an expected age profile could be calculated using 2011 Census tables.
Phase 3 – ‘Age on’ the population
Once we’ve removed our static population we can age on the remaining population. We will account for births, deaths and mortality later in the model, so this phase is a straightforward case of increasing age by a year for each year of the model.
Phase 4 – Natural Change
As defined through Births and Deaths: We have recently gained access to the national births and deaths databases, which mean that we can develop granular data outputs. These are not open data, but could be (no doubt pending formal sign-off from NHS Digital).
Deaths data will require an additional level of granularity due to the significant variation in death rates relating to both age and sex.
Experiment 1 – Can we derive reliable small area births and deaths models? will involve assessing the size and scope of births and deaths data, over time to see whether LSOA models can be developed without being constrained by challenges with small numbers.
Phase 5 – Migration
As this is going to be a households based model, we need to worry less about the quantum of in and out flowing migration (because new houses will either be filled by inward migrants
The scope of the project does not include the origin point of migrants, so the extent to which change is comprised of internal or international migrants is not of interest.
The main relevance of understanding migration will therefore be in relation to changing household composition. For example, to what extent does an area with minimal population change see older households leave and younger households move in.
Experiment 2 – Examine whether LSOA level population models can explain the nature of population change. This experiment will look at historic change in a number of sample LSOAs, some with minimal population growth and some with known housing developments.
Data for these historic models can be derived from historic ONS population estimates.
The local area has historically seen significant increases in the 18-24 poplation due to the rise in the student population of Bath and North East Somerset. It may be relevant to consider different scenarios for student growth.
Data for these models would be required from the Planning Policy team.
Phase 6.1 – New Developments – Market & Affordable
This phase will be the most significant in terms of introducing new” residents into the population model. As such it will be the area with the most risk in terms of statistical inaccuracy or error.
Although at its most basic level a crude occupancy rate can be determined and applied to a local area (on the hypothesis that people who move into an area are much the same as those who are already there), this seems highly unlikely to be the case in practice, particularly given the preferred spatial granularity of the project. As a consequence we want something more accurate.
Experiment 3 – Can we determine the likely nature and occupancy of housing developments – Using historic baselines determined in experiment 2, we can test the extent to which projected models would have been successful in forecasting historic growth.
Experiment 4 – To what extent are housing development forecasts successful in projecting actual developments – To test whether a sensitivity measure should be built into the model for this test, we can test projected against actual completions.
Data will be required from in-house Planning Policy teams on trajectories and completions. The accuracy of this element of the model will depend significantly on the nature and quality of this data.
Phase 6.2 – Care Homes and HMOs
As with the static population (see Phase 2) some new developments will have highly predictable and replicating population groups. The extent to which these can be defined will be dependent on the level of data held by planning policy teams.
Data will again be required from Planning Policy with regards the specific likely nature of developments on particular sites.
Phase 7 – Add static population back in.
Ensuring that we include the new static populations identified in Phase 6.2
Phase 8 – Apply constraining factors
Following development of the model and refinement of each phase, it may be necessary to consider whole population or targeted constraints, these will be identified and considered as the project develops.