OAs and SOAs - Stick or Twist?

The development of sensible geographical units which describe the area that they are trying to represent was one of the huge benefits of the 2001 Census from research pioneered by David Martin and Stan Openshaw. They allow the comparison of different areas knowing that the results you get are not biased by the areas having a big difference in the underlying population of the different areas.

Being at a finer scale has allowed much more detailed analysis to be carried out. I would argue that the principal of these geographical units should not be dropped for the next Census. They are a great idea and seeing that individual records can never be released, the best compromise available.

Changing the definition of Output Areas or Super Output Areas will annoy any researcher who wants to carry out time series analysis. However, areas do change over time, urban decline or new housing developments as a result of government planning. Where significant change has occurred, surely these areas should be re-computed so that at the next Census, the geographical units best describe the areas they are meant to represent. This will give researchers or anyone who is interested in using these data the idea that they can sensibly analyse these data.

Wards are a throw back to the dark ages when geographical units were created for ease of use rather than any idea of allowing sensible analysis to be carried out. Super Output Areas are far more sensible units to use instead and can be used as compariable units. The Index of Multiple Deprivation is created at this level. Other data sets like mid-year population estimates are now produced at this level. Government policy seems to be releasing data sets at this level at quite regular intervals (biannual). With wards consistently changing for non-geographical reasons these are all strong arguments for using Super Output Areas, and ignoring Wards.

It seems to me that the most important feature of Output Areas for 2011 is that they are for reporting of output from the 2011 census data. To restrict them to a geography that will be 10 years out of date by the time the new census data starts to become available and 20 years by the time the data is replaced, undermines the value available from the census. Very little other official data is available at OA level, so time series analaysis is restricted anyway, for that it would be better to allow the OA's to be redrawn but restrict them to middle level SOA boundaries.
I would also say that this is why I totally agree with David Martin's comments that the use of ward boundaries cannot be seriously, considered they are simply too unstable.
I am also concerned that there does not appear to be details of how OA change will be allowed under the 'stable' option and whether there is a commitment to keep average OA size the same. Will OA's that are now too small for disclosure control be added to one other neighbouring OA (increasing the average size of OA's) or split between several (in which case what will be the basis for doing this)? Similarly if an OA grows and is now double the limit will two OA's be created by splitting the original? Whilst what happens if two neighbouring OA's grow, but not sufficiently to split will they be sub-divided to create a third?

The debate about geography is always interesting but in the drive to achieve the 'purest' statistical geography people tend to forget that a key purpose of creating data for small geographies is to influence where and how services are delivered and in doing so to make a difference to people's life. Hence there has to be a connection between the geography units at which data is OUTPUT and the areas at which services are DELIVERED. Wards may be imperfect from a theoretical point of view (and I agree with much of what is said) but in many local authorities and PCTs they are used as basis for service delivery and democratic accountability through ward co-ordination arrangements etc. The need to monitor the effectiveness and impact of these services means that ward level geography is, and should continue, to be a cornerstone of ONS policy.

Personally I'm delighted to hear that ONS are finally considering prioritising geographical comparability over time ahead of capturing "this year's" ward geography. (In 2001, of course, the decision was made the other way around.) Given where we are at, keeping the current SOAs for mid-layer 2011 outputs is the obvious way to go - if nothing else, it allows a more direct hook into the wealth of non-census information now made for these boundaries. Probably the same arguement applies for 2001 OAs, although I would be interested to see how many have to be changed/merged in order to keep populations above disclosure thresholds - or to reflect the building of new roads that cleave existing 'neighbourhoods' in half. The downside of going with SOAs and OAs is that, over time, they will become an increasingly irrational basis for an output geography due to population and other changes (ward bounardy changes, road and house building etc) that undermine the rationale for their original creation. For this reason, I still continue to believe that the 'real' solution to all of these problems is to adopt a grid-based geography, with sub-division/ aggregation of grid cells over time to keep cell populations only just above threshold, and with look-up tables (and utilities) provided to allow users to derive best estimates of their desired output geography(ies). A helpful spin-off is that the digitised boundaries of a cell-based geography would be cheap and easy to create and disseminate without the current troublesome copyright issues that attach to OA/SOA boundaries. From this point of view, going with SOAs for 2011 merely puts of the 'inevitable' for another few years...

As with all geographies whoever conceives it thinks it works. Rarely does it end up going ahead without interventions that go against the ethos of the original idea thus compromising the results from the beginning.

There are two points I'd like to make are.
1. Measuring change over time is an admirable aim, but what are you trying to measure?
• the change in land usage or
The main components here are land type, tenure and building types.

• the change in the composition in people.

Land doesn't move but people do. If you are measuring the change of people over time and not using linked records you are likely to be measuring different people using different snapshots in time.

Areas of deprivation can be measured and re-measured, most research points to a churning effect in undesirable areas with one group of residents being replaced by another group. An increase in affluence allows movement from these undesirable areas, to better areas.

As areas around water, transport infrastructure and inner cities increases in desirability the people who are in these areas change.

Mixed tenure developments suffer from similar issues with the less desirable housing usually being provided as affordable housing.

2. How much emphasis should be put on historical boundaries?
Are we more concerned in the here and now or historical aspect?
Should we not invest in algorithms to convert old data to new boundaries, rather than locking ourselves into something that might have been good when conceived but doesn't fit now.

I think that there would be less resistance in identify people from historical data that current.

There's absolutely no doubt in my mind that a stable approach to geography would have a huge beneficial impact on official statistics generally, and their use across the entire population.

I spent more than two years working at ONS, leading web development efforts. We were ambitious, and I don't regret that: but every time we worked up an idea to make statistics more accessible to the general public, someone would throw the geography spanner into the works, and things would become too complicated to ever deliver. (If you need a case study, just look at Neighbourhood Statistics - sorry folks.) I felt it was becoming a stock excuse, and it made people start projects with an expectation of failure.

My memory may be failing me here, but weren't OAs and SOAs conceived with the intention of being a consistent future-proof standard? I have been out of the stats business for some time, so I don't know what has happened in the meanwhile.

But I do know one thing: a bad standard is still much, much better than no standard at all.

A stable geography supports analysis over time which is a good thing. However the super output areas do not always match the areas we wish to analyse and that can cause a problem so should there be a redraw?
I think there are two points to consider
(1) If SOAs were redrawn now to reflect local communities how long would it be before there were calls to redraw again because communities had changed?
(2) If it were possible to produce statistics for specific communities from SOAs would people be more inclined to forget to take into account other factor such as year on year variability and changes to the underlying population when making comparisons?
I think the current boundaries should be kept - what may be needed is more help and guidance for people who are unfamiliar with them and so still want figures for administrative areas such as wards.
I look forward to seeing what other people think.