Locations of Markets in English Market Towns, 1813. Constructing a Dataset

2.1 First pass

[5]On pages 1–86 of Owen, the author provides a county-by-county listing of towns holding markets and fairs in a standardised format. For example, under Buckinghamshire the entry ›Amersham** (26). Whit-Monday, Sept. 19, sheep. T‹ denotes that the town of Amersham elects two Members of Parliament (MP) (two asterisks), lies 26 miles by road from London, holds sheep fairs annually on Whit-Monday (seven weeks after Easter) and on September 19, and has a weekly market on a Tuesday. For my purposes I wanted to record the name, county, distance, and number of MPs, and also represent the location of the town on a map. Although a town has a spatial extent, determining historic boundaries for some 700 towns would have significantly increased the time taken, so I chose to represent each town by a simple point marker. One obvious choice for the location of the marker is the centre of the town, but many of these towns have changed size over the centuries, so the modern centre of the town does not necessarily coincide with the historic one. This discrepancy may prove significant when comparing to census or taxation data. However, since each town in the dataset hosted a market, and the location of the market as of 1813 (the publication date of Owen) is a well-defined quantity, I chose to use this instead. I entered data using the free Mapbox Studio Dataset Editor (hereafter the Mapbox Editor) as this application supports place name search, point marker types, multiple properties per marker, and the ability to export data in GeoJSON format.‍[5] I followed the algorithm in Figure 1 in deciding where to add a marker to the map.

Fig. 1: Algorithm for determining where to add a marker to the map. [Philip Allfrey 2025]

[6]In the ideal case, if an entry in Owen had at least one letter denoting a market day, I searched for that town in the Mapbox Editor search, clicked on the correct result to zoom the map, then clicked ›Add to Dataset‹. However, not all entries were this straightforward. Wales has a different heraldic tradition to England, so I excluded towns in Welsh counties from my research.‍[6] In a handful of cases the entry states something like »fortnight markets«‍[7] or »large market on Thursday«‍[8] rather than giving a day of the week. These towns were included in the dataset. Due to the quality of the printing or the digitisation, the town names are often hard to read. In these cases I performed a Google search for my best guess at the name, together with the county as given in Owen, plus the phrase ›market town‹, e. g. ›Annerstam Buckinghamshire market town‹. This was usually sufficient to find the correct town (Amersham in this example), either via Google’s autocorrect ›Showing results for…‹ feature, or a search result for a Wikipedia article beginning ›X is a market town in Y county‹, which I judged to be a sufficiently reliable source on such points of fact. When searching for the town in the Mapbox Editor I often encountered zero or multiple results with the correct name and county. The latter sometimes occurred for the town and an eponymous region, e. g. ›Bedford, Bedford, England, United Kingdom‹ and ›Bedford, England, United Kingdom‹. In this case I chose the more specific location to add to the dataset. In the other cases of ambiguous or non-existent results, a check in Wikipedia was sufficient to identify the correct town; I used the coordinates from the Wikipedia article to locate the correct place within the Mapbox Editor, and manually added a point to the dataset by clicking with the point marker tool.

[7]Once the marker was added I used the property panel in the Mapbox Editor to enter the name, county, distance from London, and number of MPs (if any) for that town. If the modern name of the town was different from the name used in Owen, I added the modern name in an alternate_name property. This affected 163 of 698 towns. Of these approximately 55 % comprised simple addition or deletion of letters or punctuation, representing changes in spelling or pronunciation over time, e. g. Ashborn → Ashbourne, Culliton → Colyton, Hales-Owen → Halesowen. A further 40 % involved adding or removing a descriptor, e. g. East Dereham → Dereham, Lyme → Lyme Regis. The remaining 5 % are given in Table 1. In the special case of London which had several markets, I added the name of the market (Smithfield) in the alternate_name field.

[8]For the few cases where Owen did not provide a distance from London for the town, I used the distance given in W. C. Oulton, Traveller's Guide, or English Itinerary, rounded to the nearest mile.‍[9] If this also did not give a distance, I determined a value by finding the nearest town on a route radially outward from London for which Owen provided a distance, then adding the modern driving distance between these two towns as suggested by Google Maps, rounded to the nearest mile.

2.2 Locating the market

[9]Initially I had accepted the location of the town provided by the ›Add to dataset‹ button from the Mapbox Editor search results, with the intention of moving this point to the location of the market in the second pass over the data. However, at the time of initial data entry (2018) using this button had the side effect of adding information from Wikidata to the properties for that marker. For consistency with the markers I placed manually (for ambiguous or non-existent search results in the Mapbox Editor) I chose to remove the extra Wikidata information by opening the GeoJSON panel in the Mapbox Datset Editor and deleting the relevant lines from the properties object. As the first pass progressed it became apparent that it was more efficient for me to place all markers manually, rather than clicking ›Add to Dataset‹ then deleting Wikidata information. Consequently, I decided to change my process to locate the market before placing the marker, as shown in Figure 2.

Fig. 2: Algorithm for identifying the location of the market within the town. [Philip Allfrey 2025]

[10]In the ideal case I used the ›Standard Satellite‹ background style in the Mapbox Editor to locate the geographic centre of the town by eye, and find a large open space adjacent to the main street, or a wide or boat-shaped street nearby. If such as place was named ›Market Place‹, ›Market Square‹ or ›Market Hill‹, I took this as sufficient evidence of the market location. Roads named ›Market St‹ or (in Devon and Cornwall) ›Fore St‹ were in general not the site of the market, but the road leading to it, so further information was required to locate the market (see below). If the aerial photograph showed areas with a highly uniform street pattern, I excluded these as being likely mid-19^th century or later developments, and attempted to find the historic core of the town at the centre of the remaining urban area. If I could not find the historic centre by eye I used the road names, looking first for ›High St‹, and failing that, for road names referring to compass directions (e. g. ›North St‹), neighbouring towns (e. g. ›York Rd‹), or older words for road (e. g. ›Hungate‹). If this was not sufficient, particularly for towns which have significantly expanded since 1813, I applied the above algorithm to the 19^th century Ordnance Survey maps as digitised by the National Library of Scotland.‍[10]

[11]For the 353 towns where I could not definitively identify the market from cartographic evidence, I consulted other sources to find the location. For 80 % of these towns I was able to find a map or verbal description of the market location in reports produced for the relevant county councils by professional archaeologists and historians, or in published academic studies such as the Victoria County History series.‍[11] For a further 9 % I was able to find the market location from local history or museum websites, historical texts digitised by Google Books, or in four cases, a Wikipedia article. For the remaining 11 % I was unable to find, or find confirmation of, the location of the market; for these towns I added a location_uncertain field with a value of 1 to the marker properties, and placed the marker in the centre of the High St (or equivalent), near a crossroad, if any. When searching these texts if the location of the general market was different to that of the livestock market, I chose the general market location. In the small number of cases where the town had two general market locations, I chose the larger or more easily identified.

[12]At the end of this pass, I exported my dataset as a GeoJSON file from the Datasets page within Mapbox Studio.

2.3 Second pass

[13]For greater accuracy in placing the marker at the market location, I wanted to visualise the position obtained from the first pass on the 25-inch Ordnance Survey Maps from the late 19^th / early 20^th century. These often provided a clearer picture of the extent of the market place, as they predate more than a century of subsequent urban development and demolition of features such as market crosses. The National Library of Scotland ›Side by Side viewer‹ website in which I consulted the OS maps does not offer the functionality of importing a list of points, nor of marking a point of interest, so I wrote a browser extension to facilitate this comparison.‍[12]

[14]My extension loads the GeoJSON file from the first pass and inserts a panel at the top of the ›Side by Side viewer‹ page with a county selector dropdown, and ›Previous feature‹ and ›Next feature‹ buttons to allow navigating through the points in this file (see Figure 3). When a new town is selected via this navigation, the maps zoom to that location, a marker is shown at the coordinates entered during the first pass, and the data from Owen entered as properties of the marker appear in the top panel. At the start of each session I needed to manually select the maps shown in the ›Side by Side viewer‹. I usually chose ›OS 25 Inch, 1892–1914‹ for the left-hand pane as it had the highest resolution, though sometimes the earlier ›OS Six inch, 1830s–1880s‹ had useful information. For the right-hand pane I used ›MapTiler Satellite Hybrid‹. During the course of this project the National Library of Scotland released a new version of the ›Side by Side viewer‹, which is incompatible with my browser extension, so that as of the time of writing (December 2024) it is necessary to click the link to use the former ›Side by Side viewer‹.

Fig. 3: Screenshot of the National Library of Scotland’s Side by Side map viewer, with my browser extension enabled. [Left panel: CC-BY (NLS). Right panel: OpenStreetMap]

[15]My algorithm for the second pass is shown in Figure 4. I worked alphabetically by county and town and compared the data from Owen with that entered in the first pass, as displayed by my browser extension. If there were any errors or omissions, I located that town within the Mapbox Editor, and corrected the data there. I did not re-export the data from Mapbox until the end of this pass. If the town did not have sufficient cartographic evidence for the market location, I searched for and added a reference as described in section 2.2.

Fig. 4: Algorithm for the second pass through the data. [Philip Allfrey 2025]

[16]For consistency I decided to place the marker at the focal point of the market place, if there was one (e. g. market cross or market hall), otherwise in the centre of the market place. For towns where I could not confirm the location of the market, I placed the marker in the centre of the main street, near a crossroads if any. If the marker from the first pass was not in the correct location according to these criteria, I moved the cursor to the correct place in the ›Side by Side viewer‹, and noted the latitude and longitude at the bottom of the screen. I transferred the decimal version of these coordinates to the GeoJSON panel in the Mapbox Editor.

[17]During the first pass I noticed that there was frequently a road named ›Silver Street‹ near the market. To tag these for further study I searched by eye for Silver Street after confirming the market location. I added a numeric silver_street property to the town in the Mapbox Editor, with a value of 1 if Silver Street led into the market or 0.75 if Silver Street connected to a road leading to the market. If I could not find Silver Street near the market location, I searched Google Maps for ›Silver St‹ plus the name of the town. If this returned a result elsewhere in the town, I entered a value of 0.5 for silver_street. Approximately 10 % of towns had a Silver Street.

[18]Finally, to allow this dataset to be more easily connected to other datasets, I changed the ID field for each marker from the default alphanumeric string to a standard identifier. Since these towns are all in the UK I chose the Ordnance Survey Open Names identifiers.‍[13] At the time of initial data entry (2018) the Ordnance Survey provided a Linked Data reconciliation endpoint. When queried with the name of a town, this service would return a list of possible matches. I clicked through for each result and consulted the map and data on the Open Names entry until I determined the correct identifier, making sure to check the place type to avoid selecting the identifier for a railway station with the same name as the town. When I had located the correct identifier, I copied and pasted it into the ID field at the bottom of the Mapbox Editor.

[19]By 2023 this reconciliation endpoint had been removed, and replaced with a download of the dataset in various formats. I downloaded the July 2023 version as a ZIP file containing one CSV for each square on the British National Grid. This was unwieldy to work with so I joined all 819 files into a single CSV file by using the Linux cat command within the Git Bash Shell on my Windows computer. Because this concatenated file contained over 3 million rows it could not be opened in Microsoft Excel. Instead, I opened the file with Modern CSV, sorted it by the LOCAL_TYPE column and deleted all rows which corresponded to names for non-populated places based on their value for LOCAL_TYPE (e. g. Railway Stations, Postcodes, Sections of Named Roads), and all rows where the place lay outside England. To reduce the need to scroll along the row when checking for the right identifier, I also removed extraneous columns.‍[14] The final spreadsheet contained 33382 rows and 18 columns, which I sorted alphabetically by the NAME1 column. To find the identifiers for the remaining market towns I used the Ctrl+F search feature within this spreadsheet. This turned out to be faster than the reconciliation endpoint because I could see the county without having to click through to another screen, and the false positives (e. g. railway stations) had already been removed.

[20]Once I had completed the second pass, I re-exported the data from the Mapbox Editor as a GeoJSON file.

2.4 Third pass

[21]Because I had entered data in multiple stages over several years, I wanted to do a sanity check to ensure there were no errors or missing fields. My first check was to zoom out the map in the Mapbox Editor to confirm that all the markers were within the boundaries of England. I carried out the remaining checks programmatically to avoid human error when scanning large amounts of data. Because the dataset was already in GeoJSON format, it was easiest for me to perform these checks in an interactive JavaScript session in a web browser. The abbreviated transcript of a JavaScript session below shows the commands I executed in the Console tab of the Developer Tools in Google Chrome. It was used to programmatically check for data errors. Lines starting with > denote input, lines starting with < denote output. Large sections of data have been replaced by an ellipsis ›(…)‹. The data structure of the GeoJSON file is described in section 3.

1. > const geojson = {...} // Copy and paste entire GeoJSON dataset
2. < {features: Array(698), type: 'FeatureCollection'}
3. > const features = geojson.features
4. > features.length
5. < 698
6. > const ids = features.filter(x => x.id.startsWith('http://data.ordnancesurvey.co.uk/id/')).length
7. < 698
8. > const properties = features.map(x => x.properties)
9. > const placenames = properties.filter(x => !!x.place_name).length
10. < 698
11. > const counties = properties.filter(x => !!x.county).length
12. < 698
13. > const distances = properties.filter(x => !!x.distance_to_london).length
14. < 697
15. > properties.filter(x => !x.distance_to_london)
16. < 0: {alternate_name: 'Smithfield', county: 'Middlesex', distance_to_london: 0, number_mps: 4, place_name: 'London'}
17. > const openNames = {...} // Copy and paste reduced OS Open Names in JSON format
18. < (33381) [...]
19. > let openNamesLookup = {}
20. > for(const place of openNames) {
21. openNamesLookup[place.NAMES_URI] = place;
22. }
23. > const names = features.filter(x => {
24. const id = x.id;
25. const openName = openNamesLookup[id];
26. return openName.NAME1 !== x.properties.place_name
27. && openName.NAME1 !== x.properties.alternate_name;
28. })
29. > names.length
30. < 3
31. > const mismatchedNames = names.map(x => {
32. return {
33. place_name: x.properties.place_name,
34. alternate_name: x.properties.alternate_name,
35. open_name: openNamesLookup[x.id].NAME1
36. }
37. })
38. > console.table(mismatchedNames)
39. (index) place_name alternate_name open_name
40. 0 'Sutton' 'Sutton Coldfield' 'Royal Sutton Coldfield'
41. 1 'Barnet' 'High Barnet' 'Chipping Barnet'
42. 2 'Sherburne' 'Sherburn in Elmet' 'Sherburn'

[22]I assigned the GeoJSON data to a variable, and extracted the features array, which contains an ID and set of properties for each market town. For every attribute I wanted to check I used the Array.filter function to return a new array satisfying a logical test (e. g. the attribute is a non-empty value). By comparing the length of the filtered array to the length of the features array I could determine whether there were any towns which were missing properties. Where the counts did not match, I inverted the Array.filter condition to display the affected entries, then added the missing information in the Mapbox Editor.

[23]In order to programmatically check that I had copied and pasted the correct OS Open Names identifier as the ID for each market town, I converted the reduced spreadsheet into JSON format, then constructed a lookup table from OS Open Names identifier to name.‍[15] Iterating over the features array I compared the place_name and alternate_name fields to the name from the lookup table. This highlighted cases where I had not copied the whole identifier, pasted the identifier from a previous town, or chosen the wrong identifier when using the reconciliation endpoint. I corrected the affected identifiers in the Mapbox Editor, then exported the data for a final time in GeoJSON format. After converting the dataset to CSV I uploaded both formats to GitHub and the Knowledge Commons Works repository.‍[16]