The Great Britain Historical GIS

Ian Gregory and Humphrey Southall (University of Portsmouth, UK)

The goal of the Great Britain Historical GIS Project is to create a common framework within which the majority of spatially referenced official statistics published since the early nineteenth century can be integrated, locating this information in time and space. Fundamental to the project is the use of Geographical Information System (GIS) technology, which links the statistical data and the places they refer to. Geographical Information Systems model the world in layers or themes, and so enable us to integrate many datasets collected at different times and published using different administrative units into a common structure. The historian, unlike the contemporary researcher, has limited control over the data they have to work with; but the GIS creates new and powerful ways of manipulating available data to answer previously unanswerable questions.

Many writers in recent years have attempted to provide a definition of GIS. Perhaps the most appropriate definition for this paper is that provided by Goodchild

From this it can be seen that a GIS is in many ways a database management system with a spatial component, usually a computer mapping system, added to it. A GIS therefore contain two types of information: attribute data which are the statistical data, and spatial data providing locations, usually in the form of points, lines and areas (or polygons as they are more correctly called). To create a historical GIS a third type of data, temporal data, needs to be added to this basic model.

Our overview of the historical GIS begins, as did the project, with the attribute database. Almost all of these data came from statistical tables published in nineteenth and early twentieth century reports, and the left hand column in the vast majority of these tables contained the name of some administrative unit, or sometimes a less formal place-name. Computerisation of such material is time consuming but raises few technical issues. However, interpreting the resulting assemblage of information was problematic because so much of the date related to geographical entities which either no longer existed or which had changed their boundaries significantly over time.

One solution to this was to build place-name gazetteers, enabling us to calculate totals for counties and regions from data for low-level units whose names meant little to us. However, the great strength of the data we had assembled was that it provided information not on regions and counties but on sub-county units; on the localities within which people in the past lived their daily lives. In the period before the 1914-18 war, much of our data concerns Poor Law Unions or Registration Districts, areas that were designed to represent a market town and its hinterland. From the First World War to the 1970s a more detailed system of local government districts was used which distinguished urban areas from their hinterlands.

In order to exploit the richness of spatial detail within the database, we began work on the boundary change GIS described in the third section of this chapter. The initial GIS was quite modest, mapping just Unions and Registration Districts, but enabled us to develop a methodology for systematically representing changing boundaries within a computer. This was a crucial methodological breakthrough as it allowed us for the first time to take a dataset from any date and link it to an accurate representation of the administrative units that existed at that date. Creation of this initial system attracted an extensive network of collaborators who supplied additional attribute information, helped fund the project, but wanted a record of changing parish boundaries; about 15,000 units as compared to c. 630 Unions/Districts. We have also been extending the system to include the c. 1,500 Local Government Districts. At the time of writing, parish mapping has been under way for three years but the end is in sight and we hope to be able to present the completed system in August 2000. Work will then proceed to Scotland.

The final section of this chapter outlines the use of the GIS, describing straightforward use for mapping but also outlining the more complex methodologies we have developed for relating datasets gathered for quite different geographies, and for constructing time series for consistent geographical units when the reporting units are constantly changing. This last ability is probably the most important outcome of the project, enabling a wide range of long-term statistical analyses which were previously impossible. One major application is to accurately trace the history of local mortality decline in Britain between the mid-nineteenth century and the present: until now, almost all such research ends at 1911 because the reporting units then changed!

The GBH database holding our attribute data is a large Oracle relational database. What matters is its content and what follows is inevitably simplified; for full details, see our web site:

The origins of the GIS project lie in Humphrey Southall's doctoral research, concerned with the geography of economic distress in pre-1914 Britain, and the origins of the north-south divide which has been a key element of the country's economic, social and political geography throughout the twentieth century. From the end of the First World War onwards, economic distress has been defined primarily in terms of unemployment, and this has been measured through the centrally-managed National Insurance system, created in 1911, under which relief payments to the unemployed were a direct cost to government. Southall's research included an analysis of statistics from the locally-managed and funded Poor Law system, but focused on the records of the mutually-funded and managed welfare benefit systems operated by pre-1914 trade unions. Many historians have used the national totals of the unemployed extracted by the Ministry of Labour and its predecessors, but the figures for individual branches in the original union reports were completely unexplored.

This research was undertaken in the late 1970s with limited computing facilities and no data entry assistance, so the main emphasis was on documenting the statistics and the underlying schemes, which were important but rarely acknowledged forerunners of National Insurance and the modern welfare state. Statistical analysis was limited to particular dates, and to computing unemployment rates for regions and main towns. However, between 1989 and 1991 the Leverhulme Trust funded a new project employing a research fellow, David Gilbert, and a clerical assistant. This enabled us to replace a collection of text files manipulated by Fortran programs with a modern relational database, and to gather far more systematic information from trade union reports and other records of economic distress, constructing a Labour Markets Database (LMDB) for Britain:

• Trade union branch returns: Unemployment and other benefit statistics bi-annually for the Amalgamated Society of Engineers from 1851 to 1912 (quarterly for 1851-72), and for the Amalgamated Society of Carpenters and Joiners for 1863 to 1912 (except 1870). These were the two largest national unions for most of the period.
• Government unemployment statistics: Pre-WWI local unemployment statistics for engineers, shipbuilders, printers and carpenters and joiners, mostly for 1902-14, compiled by the Board of Trade from trade union sources; and quarterly National Insurance unemployment rates by employment exchange, 1927-39.
• Poor Law statistics: The sources here are very extensive and our transcriptions are inevitably partial. At county level, we have bi-annual summary statistics for every year from 1859 to 1915, plus more detailed data for selected years in the 1860s and early 1870s. At union level we have bi-annual data for selected unions (all unions in London, Lancashire and Durham, plus other major towns) from July 1859 to July 1882, then January 1900 to January 1912; data for all unions for 1st January in 1907 to 1912; and for all unions from 1922 to 1930 for January. Lastly, following the abolition of the union system, we have data for counties and county-boroughs from 1931 to 1939. These holdings completely exclude Scotland.
• Small Debt statistics: Numbers of ‘plaints' for under £20 entered per annum in c. 500 courts, grouped by circuit; continuous series for 1868 to 1913, plus data for 1847, 1851-57, 1866 and isolated data for 1938. This formed the basis for collaborative research with Paul Johnson of the London School of Economics.
• Marriage statistics: Number of marriages in each Registration District in England and Wales by calendar year for 1841 to 1870 inclusive, plus annual marriages for RDs in Durham and Lancashire from 1871 to 1910. These data provide, perhaps surprisingly, a useful index of economic distress.
• Statistics of wages and hours of work: Wage statistics gathered by the Board of Trade Labour Department, covering 24 occupations for 87 towns (but not all towns appear for all occupations), from various start dates, the earliest being 1845, to 1906; and hours worked in coal-mining in 17-20 coal fields, monthly from 1900 to 1913.

A census of Great Britain's population has been taken every ten years since 1801, with the exception of 1941, and although it has been supplemented since c.1970 by several large official sample surveys it is still the most geographically-detailed source of information on Britain's people. The first four censuses were simple head counts, the forms being filled in by knowledgeable officials, most often the parish priest. From 1841 onwards, schedules were completed for each person, including information on age, sex, marital status, occupations and birthplace; and over time an ever-wider range of information has been gathered. Much historical research has been based on the individual information assembled into "enumerators' books", but these remain confidential for a hundred years. Our work aims to systematically link the present with the past, and therefore is based on statistics tabulated in the extensive published reports.

The earlier work on the LMDB included some census occupational statistics, and from 1995 onwards we have assembled further census data transcribed by collaborators; we acknowledge in particular the assistance of David Allan Gatley (Staffordshire) on mid-Victorian censuses and Daniel Dorling (Bristol) on 1951 and 1961. However, in 1998 much larger scale funding from the Economic and Social Research Council (ESRC) enabled us to employ a second team based at the Queen's University Belfast. The Belfast team use a specialised ProLector optical character recognition (OCR) system to digitise printed tables from the census reports. Most recent developments in OCR have emphasised the automated recognition of relatively straightforward office documents, but the ProLector system is highly trainable and performs well with difficult documents: small type faces, varying character shapes and faded ink. The system is still not perfect, but we employ low-cost Information Technology trainees to check and correct the output.

Even with this specialised team, our transcriptions from the census have to be selective, prioritising tables that recur from one census to the next and data for sub-county units, as listed below; many other tables are simple head counts for obscure and specialised administrative units, which we can easily compute from parish-level information once we know their boundaries, but we hope to obtain further funding to remedy the other gaps. Current work is limited to Great Britain, but the Belfast team had earlier completed very extensive transcriptions from the Irish censuses and we hope one day to add mapping of Irish boundaries. Coverage of Scotland is generally limited to data for counties, cities and large burghs.

• Age structure: Every census since 1851 has tabulated the population of sub-county units by sex and age, using five-year age bands. The database includes Registration District statistics 1851-1911, Local Government District statistics 1911-1961, and we are adding Registration sub-Districts 1851-1881. When linked with mortality data these data enable us to calculate standardised mortality rates and, by comparing the actual number in a given age/sex cohort with the expected number based on the previous census and inter-censal deaths, an estimate of net migration flows.
• Occupational statistics: Information on occupations was reported by the census from 1841 onwards, and we have computerised sub-county tables from all censuses from then until 1951, with the current exceptions of 1881 and 1891. However, there is considerable variation both in the areas for which data were listed and the system of classification used.
• Migration/Birthplace: The completed database will include transcriptions of the birthplace tables, generally giving county of birth.
• Parish-level population: Only limited information was generally tabulated parish-by parish, generally numbers of men and women, and numbers of houses. However, the sheer number of units means these are very large tables. We have complete transcriptions for 1801, 1831, 1851, 1871, 1881, 1891, 1911, 1931 and 1951. As discussed below, the information these tables provide on population plays an important role in our re-districting procedures.
• Other: Coverage of other themes is more selective, but we have information on housing and over-crowding for 1901, 1931 and 1961 by local government district; industrial structure, social class, household structure and educational attainment for 1951; and a transcription of the 1851 religious census. Much of this information has been supplied by collaborators and we would be pleased to hear of other census data which might be added.

Much information on numbers, ages and causes of deaths has been assembled from collaborators at the ESRC Cambridge Group for the History of Population and London's Centre for Metropolitan History. However, we have recently obtained substantial funding from the ESRC and the Wellcome Trust to systematically computerise sub-county mortality statistics from 1851 to the 1970s, again working with the Belfast team. When these projects are complete we will hold:

• Age- and sex-specific mortality, by decade: For Registration Districts 1851-1910, then by aggregates of local government districts 1911-30.
• Cause of death, by decade: Again, for Registration Districts 1851-1910, then by aggregates of local government districts 1911-30. The pre-1900 data come from an earlier project directed by Robert Woods of Liverpool University.
• Annual mortality: we are computerising the annual mortality counts for local government districts in the Registrar General's Statistical Reviews 1921-73.
• Quarterly mortality: Much of our new funding is to computerise the hitherto little-studied Registrar General's Quarterly Returns, which between 1869 and 1911 tabulated numbers of deaths in each Registration sub-District (see below) in each three month period, identifying infant deaths and deaths from the major epidemic diseases. We will be adding annual counts for sub-Districts 1851-68.

We also have some limited information on levels of sickness, as recorded among the memberships of the Steam Engine Makers' Society 1835-76 and the Ancient Order of Foresters 1872-80; the latter data were supplied by James Riley of Indiana University.

While our holdings of census, mortality and unemployment statistics are intended to become comprehensive, other datasets results from specific research projects or collaborations. These include a database covering all strikes in the Board of Trade/Ministry of Labour strikes register for every fifth year from 1903 to 1938, and another constructed from Burdett's Hospital Yearbooks by John Mohan (Portsmouth). We have access to a large body of electoral statistics for Britain.

We have been exploring the potential to linking the GIS to various pre-census surveys of Britain, although most such datasets are limited to England. In this connection, we have samples datasets supplied by collaborators, generally limited to one or two counties:

• 1086 Domesday Book (John Palmer and George Slater, Hull).
• 1291 Taxatio (Manchester Taxatio Project/Jeffrey Denton, Manchester).
• 1334 Lay Subsidy (Robin Glasscock, Cambridge).
• 1524 and 1525 Lay Subsidies (John Sheail, English Nature, and Richard Hoyle, Central Lancashire).
• 1662 to 1688 Hearth Tax (Margaret Spufford, Roehampton Institute).
• 1676 Compton Religious Census (Keith Snell and Alisdair Crockett, Leicester).
• 1801 Crop Returns (Michael Turner, Hull).

More extensive work with these datasets will require significant further funding, but their existence draws attention to the range of systematic spatial data that exists for medieval and early modern England, generally at parish level.

Whatever the source, information in our statistical database relates to geographical entities, and most information relates to formally defined areas. While data for regions and perhaps counties can be interpreted using the researcher's ‘mental map' of Britain, most of our information relates to sub-county units and can be given meaning only through formal linkage to locational information; or put more simply, by being mapped. An enormous range of units might be mapped; the 1871 census report lists information for:

However, most of these units had specialised functions. In practice, two or three principal geographies can be distinguished, as shown in figure 1:

• Poor Law Unions/Registration Districts (PLUs/RDs) (and sub-Districts): Poor Law Unions were created by the Poor Law Amendment Act of 1834, and England and Wales was divided into about 630 unions by commissioners on the basis of market towns and their spheres of interest. When civil registration of births, marriages and deaths was introduced in 1837, the Registrar General used almost the same set of areas as Registration Districts (our work records the differences) but also sub-divided them into about 2,000 Registration sub-Districts. Unions and Registration Districts declined in importance in the late 19^th century, but disappeared only in 1930.
• Local Government Districts (LGDs): Starting in the 1870s, but especially as a result of the Local Government Acts of 1888 and 1894, approximately 1,500 LGDs were formed. These were also aggregations of parishes and were usually formed by designating urban parishes as County or Municipal Boroughs, or Urban Districts, and designating the remainder of each Poor Law Union as a Rural District. The LGD system was extensively revised in the mid-1930s and abolished by the 1972 Local Government Act.
• Constituencies: Electoral geography was distinct, based on Parliamentary Constituencies revised by infrequent Boundary Commissions. We aim to include these in the longer term.

All three geographies used parishes as building blocks, except in urban areas where an entire town might be a single parish but sub-divided for vital registration and elections. The parish was an ancient ecclesiastical unit, but the modern Civil Parish system was created only in the 1860s, i.e. after the first censuses and after the New Poor Law, and then very extensively modernised by the Divided Parishes Acts of 1876 and 1882. All three geographies were organised at a higher level into counties, but these are a source of considerable confusion. Firstly, PLUs/RDs aggregated into Union or Registration Counties which often differed substantially in area from the Ancient Counties of the same name, while LGDs aggregated into Administrative Counties which were generally closer to the Ancient Counties. Secondly, while most counties existed as Ancient, Registration and Administrative Counties, albeit with differing boundaries, the lists of units varied with the Soke of Peterborough, East and West Sussex and so on existing as distinct administrative counties.

However, the largest single problem is not apparent from examining names of districts that appear in our statistical tables, or even a single set of maps: the boundaries of Registration Districts, LGDs and the individual parishes were all subject to constant change. As discussed below, the census reports and other statistical sources include lengthy but little read lists of these changes. However, systematic and accurate interpretation of the main statistical tables requires not merely that we read these lists; we must establish what effect each change had on our map. Given the tens of thousands of boundary changes that occurred, the only possible way of assembling this information is within a computer, via a Geographical Information System.

Geographical Information Systems technology has developed to meet the needs of marketeers and utility companies, not academic research, and available commercial software makes little provision for incorporating a time dimension. Some historical mapping projects have side-stepped this problem by creating a sequence of static maps, and this could have been done for Britain if we were only interested in the boundaries as they existed at particular dates, such as those of the censuses, or if we were interested only in boundaries which changed only through infrequent revisions, as with the constituencies used in parliamentary elections. However, we needed to study bi-annual Poor Law data and three-monthly mortality statistics, and the boundaries of Registration Districts and LGDs were subject to a constant trickle of small changes.

Our approach was therefore based on a single GIS ‘master coverage' within which all features, both boundaries (‘arcs') and units (‘polygons'), were date-stamped with a start-date and an end-date. This enabled us to deal with the four main types of boundary changes:

• Transfers: Where part of a unit is transferred from one unit to another.
• Name changes: Where a unit changes name but its boundaries are unaffected.
• Mergers: Where a unit is abolished and its component parts are allocated to another unit or units.
• Divisions: Where a new unit is created from parts of an existing unit or units.

Although some changes were more complicated, such as an existing area being abolished and a new area with new boundaries created at the same time, all types of change can be built up from these four elements.

We might have implemented a date-stamped dynamic GIS by writing a bespoke temporal GIS system from scratch. The dangers of this approach are well-illustrated by the Great American History Machine, who followed this approach. Their aim was to include all US census data and their associated boundaries in a single software package, but this important resource has never been properly published essentially because of delays in removing relatively minor ‘bugs' from the software. We therefore based our system on existing GIS software, the well-known ArcInfo package, with bespoke software being written in Arc Marco Language (AML) to manage changes over time. One particular strength is that ArcInfo provides comprehensive tools for integrating its spatial data with attribute data held in an external database management system such as Oracle.

Like most GIS software packages, ArcInfo stores its data in distinct coverages. Each coverage holds information on a single theme or layer of data. A coverage will usually consist of either points, lines (technically termed arcs) or polygons. A point coverage might represent churches or hospitals, while an arc coverage could represent railways or rivers. In both cases the spatial features would have attribute data linked to them. The data model ArcInfo uses to represent polygons consists of a combination of points and arcs. Arcs represent the boundaries of the polygon, with their attributes holding information about the boundaries, to represent the polygon itself points (termed label points) are used and data attached to the label point are allocated to the entire polygon. This feature is crucial to the way that temporally referenced spatial data can be stored as it allows us to store the temporal data, in the form of date-stamps, about a unit (attached to the label point) separately from the temporal data about its boundaries (attached to the arcs). We can therefore store the changing state of an administrative unit for many dates in a single coverage, termed a master coverage, and extract an accurate representation of the unit for any required date. The date-stamping for the four types of changes listed above can then be incorporated as described below:

1. Transfers: Two arcs (or sets of arcs) are used in the master coverage, one representing the pre-change boundary and one the post-change boundary. The pre-change boundary has an end-date set to the date of the change, the post-change boundary has its start-date set to the date of the change. Label points are not affected. The figure below gives the example of a central area being transferred from ‘Anarea' to ‘Elsewhere' on the 1^st September 1894. Prior to the change the boundary between the two areas is arcs 1, 2 and 4, after the change arc 2 is replaced with arc 3 while all other arcs, and the label points remain the same. Label points are represented by crosses and identified by letters, while arcs are identified by numbers. Features in existence when a unit was formed are date-stamped 0/0/0000, while those in existence when the unit was abolished are date-stamped 0/0/5000.
   
   
2. Name changes: Two label points are used in the master coverage, one with the name of the pre-change unit and one with the name of the post-change unit. Apart from the fact that label points are used rather than arcs these are handled in the same way as transfers. Arcs are unaffected. The figure below gives the example of ‘Oldname' being renamed ‘Newname' on the 10^th April 1932.
3. Mergers: These affect both label points and arcs. As part of a merger a unit is abolished; this is represented by a label point in the master coverage being given an end-date of the date of the change. By definition boundaries will also be affected. In the simplest form of merger the whole unit becomes part of an adjoining unit, in this case the arc representing the boundary between these two units needs an end-date of the date of the change. In more complicated mergers several units will gain territory from the area that is being abolished but the principle remains the same; several arcs are given end-date of the date of the change. A merger, therefore, has both label points and arcs being date-stamped with end-dates but no features are given start-dates. In the example below, ‘Oldplace' is abolished and entirely merged into ‘Anarea' on the 1^st December 1903. Only the label point representing ‘Oldplace' and arc 1 need date attributes to encode this change.
   
4. Divisions: These are similar to mergers in that both label points and arcs are affected however the features are given start-dates as a new area and new boundaries are being created. No features are given end-dates. In the final figure, ‘Newplace' is created entirely from ‘Anarea' on 1^st Jan. 1963. This is the opposite of the previous figure: the label point representing ‘Newplace' (point ‘b') and arc 1 are the only features affected.
    

For ease of data management the decision was taken that the three major types of unit would be stored in separate ArcInfo master coverages. This involves some duplication of data but greatly simplified the construction process and also has the advantage of speeding up the extraction of data, particularly for Unions/RDs and LGDs, by reducing the volume of data that features are to be extracted for. The minor differences between Unions and RDs were handled using attributes. Two fields were added to the attribute tables of both the arcs and label points for the Union/RD coverage. One was used to flag whether the area was a Union, and the other whether it was an RD. In this way only the appropriate feature could be selected.

The attributes for the LGD system were very similar to those for the Union/RDs. Attributes specific to Unions/RDs were not needed, while a new attribute was added to the label point attribute table indicating the type of area a label point represents: County, Municipal, or Metropolitan Boroughs, or Urban or Rural District. Where a unit changed its status, such as a Municipal Borough becoming a County Borough, a second label point was needed, as with name changes, but with the type field rather than the name field differing.

This section focuses first on map-based sources of boundary information, and then on textual sources describing boundary changes. A single series of maps had to be used as the core source for digitising to which the boundaries for other dates could then be added. The choice of which series of maps to use for core digitising was obviously of key importance: it had to be for a suitable date, at a scale that would allow complete digitising of the country, and based on a projection system that would allow counties to be joined together easily. It was decided to use a series of half-inch to the mile (1:126,720) Ordnance Survey (OS) county administrative diagrams published between 1906 and 1910 for the following reasons:

• The half-inch scale is a sensible compromise between the need for detail and the need to be able to cover the country quickly
• Unlike earlier series, they use a standard projection system for the whole of England and Wales.
• They were produced at the time of the change-over from Unions and Registration Districts to LGDs and show boundaries for both types of areas.
• They are over 50 years old and therefore completely free of Ordnance Survey copyright, important because the UK national mapping agency operates on commercial principles and imposes high licence fees on any resources incorporating their mapping.

Figure 3 shows an example from this series. The figure uses grey-scales however the original series was published in colour. The thick light line near the edge of the figure shows the boundary of Newport Poor Law Union, named in large letters. Both the boundary and the letters are dark red on the original. The darkly shaded unit (pale red on the original) in the centre of the map is the County Borough of Newport and the lightly shaded units (beige) are Urban Districts. Although there are no Municipal Boroughs in this example they are shaded pale blue on the series. The remainder of the Union, as bordered by the thick dark boundaries (dark blue), has been divided into the Rural Districts of St. Mellons and Magor, although this is slightly unusual. The thin lines (again red) mark parish boundaries, showing clearly that the larger areas are aggregations of these.

There is a series of "Diagrams of Sanitary Districts" for the late 1880s showing Boroughs, Sanitary Districts, Unions, and Civil Parishes but at the smaller scale of four miles to the inch (1:253:440). Another source is a set of half-inch to the mile Index sheets to the 1:2,500 maps published around 1900 that show civil parish boundaries but as these are index sheets they are difficult to work with. The General Registrar Office (GRO) had access to a series of maps compiled from contemporary one-inch to the mile OS maps which were cut up into a separate map for each Registration District. The outline of each Registration District was highlighted in red with the boundaries of Sub-Districts added in pink and civil parishes in blue. These maps were used by the GRO for administrative purposes as they contain notes explaining boundary changes. They are now held in the Public Record Office (PRO) in RG.18/198-829. These provide information about boundaries for the late 1880s and early 1890s.

For earlier years a different set of maps was used. These were listed in the PRO Index volumes as showing the parish boundaries in 1870 and are in PRO class RG.18/1-73. They are a series of OS one inch-to-the-mile sheets, probably from editions published in the early 1850s, and were not cut up in the manner of the 1891 maps. They show Registration Districts and Sub-Districts, but the units shown probably do not date from 1870, in fact several of the boundaries of Registration Districts appear to be those of 1851. The fact that many of the boundaries have been updated by hand on these sheets suggests that they remained in use for a long period after they were published. Unfortunately, this series was only around 80% complete and where there were gaps yet another series of maps, this time in PRO IR.105/2-74, had to be used. These maps were produced quickly for the Registrar General to show civil parishes as reference material for the 1851 census. They are one-inch maps marked up with tithe districts on the sometimes incorrect assumption that these were the same as parishes.

A major problem with these two earlier series is that, due to the slowness of large scale surveying of northern England, only ancient parish boundaries are marked north of a line between Preston and Hull. As there were major differences between ancient parishes and townships, which became civil parishes in the north, this means that many boundaries were never accurately recorded. A second problem, particularly prevalent in the East Midlands, occurs where there were no tithe boundaries marked. In either cases, either the 1888 sanitary district maps or, in exceptional cases, first series six-inch maps were used. The final series of maps used were the least satisfactory. These were included as sketch maps of Registration District boundaries in the published reports of the 1851 census. Their scale, however, was only 12.5 miles to the inch (nearly 1:800,000) and the detail and accuracy provided was far more limited than any other sources used. These were only used in the creation of the Union/RD system when there were insufficient resources to visit the PRO on a regular basis.

Although maps are the most useful source of spatial information, they only provide snapshots of boundaries at certain dates. Researching the information required to create the date-stamps involved using textual description of boundary changes. These were available in similar formats for the three main types of units giving the name of the area loosing territory, the name of the area gaining territory, a description of the area affected, and an exact date of the change. In some cases data such as the population and area affected are also given. Information on changes to Registration Districts has generally come from the Annual Reports and Decennial Supplements of the Registrar General, changes to LGDs were taken from the Annual Reports of the Local Government Board and its successors, and changes to parish boundaries are included in the printed census reports from 1881 onwards. An example of these lists, taken from the Registrar General's Decennial Supplement for the 1890s is shown in figure 4. Changes from the other sources follow a similar structure.

Combining the map and text based sources provided the information to create the full time-variant GIS. Construction proceeded county-by-county in the following stages:

1. The county was digitised from the appropriate map from the 1906 to 1910 series and the relevant features copied to each master coverage. Each arc and label point was then given a start date of 0/0/0000 and an end date of 0/0/5000 on the assumption that it existed all the way through the period. After digitising, the resulting coverage was then transformed onto real world co-ordinates, a process known as geo-referencing. The maps used the Cassini projection and included no information on latitude and longitude, so the first step was to establish real world co-ordinates, i.e. British National Grid references, for at least four locations on each map sheet, based on features such as churches, stations, and railway junctions. These features were incorporated into the coverage as "tic points", point features used by ArcInfo to provide a spatial reference for an entire coverage.
2. Temporal information was then added by plotting out the boundaries of the county onto tracing paper, at a scale that allowed it to be laid over a series of maps from a different date. The maps could then be compared and information from the lists of boundary changes used. Where changes had occurred the alternate boundaries were traced and given a date from the textual sources. The tracings were then taken back to the digitising table, the new boundaries added to the coverage, and attributes added. In cases where a change could not be properly entered a note would be made in the accompanying documentation.
3. Once the county had been finished it could be joined to existing counties using basic ArcInfo functionality. The results then had to be rigorously checked. This could only be done by extracting that county for all significant dates and checking in particular that there were no "dangling nodes" (boundaries that failed to join any other) or polygons that had either no label points or more than one. In either of these cases the likely explanation was that there was an error and this would be rectified and the county re-checked. When a new county was added to the master coverage any adjoining counties would also have to be re-checked where the county boundaries had been affected by any new changes.

Although the boundary change GIS is a potentially valuable resource for administrative historians, its main function is to give meaning to our statistical database. The simplest application is, of course, the creation of choropleth and other thematic mapping, but we are developing more sophisticated methodologies, especially for the analysis of long-run change.

The initial justification for the historical GIS was the analysis of historical unemployment or poor relief statistics, and a conventional way of presenting such information is a choropleth map, in which areas with high percentages of unemployed workers or paupers have a darker shading than those with low percentages. Creating such a map with the GIS involves two stages: firstly, creating the appropriate coverage for the source and date; secondly, linking the statistical data to the coverage via a gazetteer.

The master coverages described above contain all the boundaries that existed at any time within the period covered, stored as lines or ‘arcs' rather than as polygons. To create a map which can be linked to data, we must:

• Specify a type of unit, a precise date and an area to be covered; for example, Poor Law Unions on the 1^st of January 1894 in East Anglia (Norfolk, Suffolk, Essex and Cambridgeshire). We have developed a ‘point-and-click' interface which, for example, permits users to select the date by moving on-screen sliders.
• Use this information supplied by the user to select firstly the relevant coverage and then, using arc and label point attribute information, the arcs and label points within that coverage which were in force at the relevant date.
• Create a new polgyon coverage from the selected arcs and label points. These polygons will be given names and other attributes using information from the label point attributes of the master coverage.

This process is computationally demanding, taking approximately two minutes for a Union/RD or LGD coverage on a Sun Sparc Station 20 with 133MHz processors.

The second stage is to link the newly created polygon coverage to the statistical data. Although ArcInfo comes with its own database management system, this has a number of limitations and all our statistical information is held in a separate Oracle relational database management system (RDBMS). ArcInfo includes a ‘Database Integrator' module providing access to data stored in an external RDBMS, so the remaining problem is not technical but substantive: our statistical tables are, as far as possible, direct transcriptions of historical sources, containing place-names rather than numerical identifiers; and place-names sometimes have complex relationships with places.

One problem is that a given place may go by a number of slightly different names: Newtown or Newton or Newton-le-Willows; St. Helen's or St Helens or Saint Helens. This problem is relatively easily dealt with by a gazetteer which matches the alternative versions of the name to a single standardised form that appears in the GIS (but it needs to include every possible way of writing St Helens!). The other problem is that a given name may be associated with a number of different places. This is not a problem with county names, affects a relatively small number of Registration Districts, and is widespread with parish names. In general, historical sources attempt to avoid ambiguity by mentioning higher level units, most often counties, but sometimes we need to guess the right place from the wider context, and parishes whose names are duplicated within a single county raise fundamental problems. Note that these two forms of ambiguity interact: ‘Newton-le-Willows' is an unambiguous location, but ‘Newton' is very common.

Building a gazetteer for Poor Law Union names, linking all the combinations of union name and county that appeared in the database with an unambiguous label used in the GIS, was a day's work. However, building a gazetteer containing all the variant names of the 15,000-plus parishes in England and Wales has proved a significant project in itself, but the resulting resource has a number of other uses. Firstly, it has to associate each parish with the higher level units it was part of; but this means the gazetteer can be used to assemble parishes together into maps of those higher level units. Secondly, it serves as a geographical thesaurus aiding resource discovery: users searching for information on, say, the county of Rutland can be directed to the town of Oakham, and vice-versa. Thirdly, the final gazetteer will necessarily contain a great deal of information about Britain's changing geography, especially when linked to our large database of changes, and could form the basis for an important reference tool in its own right.

One critical aspect of this is in mapping information for earlier periods. There is very little information about boundary changes before the mid-nineteenth century, and in fact most earlier mapping projects have used late nineteenth century base maps with much earlier sources; for example, Darby's well-known Domesday Geography used the 1888 maps discussed above to map 1086 data. However, we can greatly increase the usefulness of the GIS as a tool to aid in the interpretation of pre-1851 data by extending our gazetteers, to include earlier versions of parish names and, in particular, earlier hierarchies of units: the ecclesiastical system of deaneries, arch-deaconries and dioceses; and the ancient civil system of hundreds and ancient counties. The latter, however, is complex as hundred boundaries sometimes sub-divided parishes, while some early sources relate not to parishes but to vills or manors. We will be extending our gazetteer system by drawing extensively on F. Youngs' Local Administrative Units of England, the most authoritative source.

Any number of sample maps could be included to demonstrate the potential of our system, but figure 5 shows population growth or decline in Wales during the 1900s at parish-level.

The GIS can be used as a very powerful analytical tool in its own right, and we are investigating analytical methodologies of specific relevance to historical research:

• The construction of consistent time series when reporting units change: As already discussed, many demographic statistics were published for Registration Districts until 1911, for local government districts (County and Municipal Boroughs, Rural and Urban Districts) between 1911 and 1974, and are available for Districts and as unpublished Small Area Statistics since then; and each of these systems of units was subject to constant smaller-scale revision. This means that while comprehensive vital statistics have been gathered continuously since the introduction of civil registration in 1837, the subsequent 160 years are rarely studied as a whole and research on really long-run demographic change in Britain often ends in the 1840s, the principal focus being the ESRC Cambridge Group's work on parish registers. By overlaying boundaries for different dates it is possible to compute new series for a standard set of units; for example, we could re-compute mortality statistics from the whole period in terms of 1911 Registration Districts. There is nothing magical about this, and if all we have are Registration District boundaries and statistics we have to assume that their populations are uniformly distributed over their areas, even though we know most Registration Districts combined town and country. Adding parish-level population counts to our database, and parish boundaries to the GIS, greatly enhances the accuracy of our calculations. A quite separate problem, of course, is standardising the various classification systems used at different dates: for causes of death, for occupations and so on.
• Analysis of causal relationships where the different variables are recorded for different units. Today, government statisticians endeavour to use certain standard sets of units for gathering a wide range of information. In the past, a much wider range of units were employed and the historian often concludes that statistical analyses are impossible because explanatory variables are not available for the units they are interested in. A good example is research into electoral behaviour: voting statistics are available for Parliamentary Constituencies, but little other information is available for these units; in the mid-Victorian period, Registration Districts were used for reporting socio-economic data from the census. Political historians have dealt with this problem by massive aggregation, typically to county level, but the GIS can estimate socio-economic variables for constituencies, again by overlaying.
• Analysis of changing administrative structures. The GIS contains a mass of information on the evolving administrative geography of Britain, and we are exploring methods for summarising long-term trends in these systems. Similarly, we plan to add a mass of information from the various Parliamentary Boundary Commissions, and will be exploring the geographical impact of the widening franchise.

NB in all of this work, the principal output from the GIS is not a map but a new set of statistical data to be used in further analysis.

Lastly, while new statistical methodologies may advance academic knowledge, they influence popular understanding and political decision making rarely, slowly and indirectly. New methods for presenting our work, to students and to the ‘general reader', are just as important. Another part of our work has explored how best to present our work to a wide audience so as to develop understanding of how long-run socio-economic change and demographic processes have influenced the geography of modern Britain. In this research, we explored how best to combine established tools for cartographic visualisation, creating static images on paper, with multimedia technologies for the creation of animations and three dimensional landscapes. One aspect of this project was the creation of a complete on-line atlas, viewable on the web. This was based closely on an existing paper atlas, the Atlas of Industrialising Britain, and staying with the existing text limited the multimedia formats we could include:

The historical GIS has been developed, first in London and now in Portsmouth, as a research resource for use by the project team and their extensive network of collaborators. However, the operational GIS is not publicly accessible, and wider academic access is through our dissemination partners, who are funded by the Joint Information Systems Committee to provide a service to the UK higher education community (researchers from outside the UK are not automatically entitled to assistance, but are still advised to contact them, possibly through their own national data archives):

• The History Data Service is based within the Data Archive at Essex University, and is receiving all our statistical data. They have already established an on-line service from which registered users can download selected information from part of the database:

Obviously, having on-line statistics at one site and on-line boundaries at another means that at present users must request two separate downloads, and then carry out analyses and draw maps on their own computer. However, more sophisticated services including on-line gazetteers are being discussed.

It is hard in a single paper to present all aspects of a large project involving many staff at several sites, plus a wide network of collaborators. The large task of mapping changing parish boundaries is almost complete, but full exploitation of this resource will take even longer. What we hope is clear is how our core GIS containing administrative boundaries is not simply enabling various historical researchers to map their data, but is serving to integrate together a very wide range of information about Britain's past. The project will certainly provide academic researchers, and especially quantitative historians, with data permitting quite new analyses of truly long-term social, economic and demographic change. Our new challenge is how best to make this information available to many different communities of potential users:

• Local historians and perhaps family historians, contextualising their more personal researches. This means providing access from public libraries.
• School children. Under the English National Curriculum, all children aged 8 to 11 must prepare a ‘study investigating how an aspect in the local area has changed over a long period of time, or how the locality was affected by a significant national or local event or development or by the work of a significant individual', using information on ‘education; population movement; houses and housing; religious practices; treatment of the poor and care of the sick …'.
• Archivists and a wide range of archive users, needing placename authority lists and place-name thesaurii which are both more authoritative and more easily searchable.

The research described here covers Great Britain, and we hope to extend it to Ireland. However, we are also interested in linking together similar projects in other countries, partly to share experience but also to explore the potential for linking together national projects to create an electronic map of Europe. With financial support from the European Science Foundation, we are organising a specialist workshop in Florence in June 2000, and then reporting its conclusions at a workshop session within the International Congress of Historical Sciences in Oslo. Also in Oslo, we will be presenting some initial findings from our mortality research, and demonstrating the historical GIS in the exhibition area.