The Government “Hybrid” Postcode Model

Summary

The Government proposed postcode model of “ABC-123” to be centred about 200 English language “post towns” is incapable of providing any better than a generously estimated average 0.49km² precision - which does not meet the DCENR’s own specification of a postcode identifying a group of 40 to 50 addresses.

At this, two addresses in the same postcode could be nearly 1 km apart as the crow flies. This is bigger than most villages and large enough to encompass the heart of many small towns. In rural areas it is a bit better - not quite as bad as the traditional vague townland address but still not accurate enough to overcome the rural addressing problem in Ireland without the addition of house numbers. It copies an unwieldy, outmoded, unsuitable UK postcode system.

The proposed code will provide less location information than the ordinary address.

It will be useless to emergency services. An Post do not need nor want such a code. The code will not help in verifying addresses. A non-uniform code will require a database-lookup approach to conversion to and from mapping locations - imprecise as they will be. It will also, no doubt, necessitate a new quango just to maintain and revise the code. Its built-in linguistic discrimination is yet another blow to the Irish language and bilingualism.

Some simple adjustment to the format, however, could remove the language bias, and extend the usefulness of the postcode to nearly 90m by 80m; and with the addition of a checksum character made self-validating and avoid transcription errors. However, this would still place it many times less accurate than alternative codes already in existence.

Some of its failings have already been addressed by the Joint Committee on Communications, Energy and natural Resources:

“Ireland is the only country in the EU that does not have postcodes and the opportunity arises for us, with proper planning and preparation to have the best postcode system.

At a time of severe economic recession and high unemployment it is important that the postcodes project offers measurable benefits and is delivered at a reasonable cost.

… we recommend that the option of a postcode system should be based on a unique identifier system in view of its range of benefits using the best, most up to date technology.

We do not believe that a system based on clusters of 40-50 properties as proposed by the Government is appropriate for the current and future needs of the smart economy.

A unique identifier system provides for a speedy response by emergency services, provides greater efficiency and interaction with GPS technologies and does not require change in names of townlands, etc, has low maintenance costs, supports spatial planning and the delivery of health services in particular.”[1]

Ian Spillane, 2012, openpostcodeireland at gmail dot com.


DCENR Description

The Government’s postcode model is described as follows on the DCENR website:

“The framework for the postcode system we plan to adopt is the 6 digit alpha numeric model, publicly available and accessible model as recommended by the National Postcode Project Board in 2006. The model will be capable of being further refined into a location-based code.

The country would be divided into approximately 200 post towns. Within each post town there would be groups of approximately 40 to 50 properties. The postcode would have the structure ABC 123 in its numeric code, the first three characters representing the post town, the second three representing the group of properties in which the particular building is located.”

http://www.dcenr.gov.ie/Communications/Postal/Postcodes.htm

Also, according to an article on the 4th of January, 2011, in the Irish Times, the then Minister proposed a so-called “hybrid” system:

"Mr Ryan said one of the most important features of the system would be “memorability”. In practice that means a location will be identified by its initials, in a manner roughly analogous to the vehicle registration system.

The system will also retain elements of the existing Dublin post codes. The current postcode of Dublin 7 could begin D07 and continue with numbers that pinpoint the location to a particular property. Similarly, the postcode for addresses in other areas will contain letters that readily identify the area, followed by a series of numbers that pinpoint the property. Possible permutations might include: GLY (Galway); CK (Cork); and KKY (Kilkenny).

A departmental briefing paper refers to two models – a “postal sector” model and a location model – being combined. The postal sector model would divide the country into post towns (identified by letter) and each post town into groups of approximately 40 to 50 properties (identified by number). There would be approximately 200 post towns.

The paper says the model is capable of being refined into a location-based code – in other words identifying each individual property within a post town. It says this hybrid model will provide the basis for the procurement for a national postcode system."

Analysis

This design needs analysis as although specifying an exact code format the DCENR provide no support nor rationale for the format. One fears a postcode system squeezed into an ill-conceived notion of what a postcode could be - without any mathematical basis or testing of the possibilities.

As specified by the DCENR: “The postcode would have the structure ABC 123 in its numeric code”. However apart from painting just a picture of a postcode format - the structure has a number of key mathematical limitations.

“‘ABC’ … representing the post town”

Firstly, “the first three characters representing the post town”: three alphabetic characters from “AAA” to “ZZZ” give exactly 26x26x26 different possibilities: which is 17,576 different codes. If applied to the island of Ireland[2] at its most efficient and suitable as a location code it would describe a grid of rectangles 132x132 or similar[3] - about 3.5km by 2.7km each[4]. To be safe from inadvertently spelling unwelcome words in three characters it would be sensible to remove vowels. This gives 9,261 codes; a grid of 96x96; of about 5.1km by 3.2km.

200 Post Towns

However, the Government has in mind using just “approximately 200 post towns”. One has no idea where the number 200 comes from. There were 612 census towns for the 2002 census. (Indeed using 612 would greatly enhance the precision of the code. But “approximately 200” it is for now.)

If the island was divided equally into 200 non-overlapping rectangles that would divide the country into a grid similar to 14x14 - each section 33.3km by 25.5km. And this would be the maximum precision.

Overlapping Circles

The most efficient use of the code is to divide the country equally by a 14x14 grid; the Government, however, has in mind that “permutations might include: GLY (Galway); CK (Cork); and KKY (Kilkenny)” - so it is intended to centre post towns about named places. Though we must also assume that the model requires that “post towns” are well defined areas which can be logically subdivided, as stated “the model is capable of being refined into a location-based code – in other words identifying each individual property within a post town”.

This is where the postcodes and maths part company.

To centre individual location bubbles around these post towns, at it’s most efficient, means to draw a circle around them. However, 200 non-overlapping circles in no way fill a geographical area - no matter where they are placed. Lots of places are not covered - while even the most lonely address in the country needs to be covered by a postcode.

In any case, circles are unworkable as a basis for coordinate geometry within the post town and will just assume a larger encompassing square if the next part of the post code is to work and provide a number for the “approximately 40 to 50 properties (identified by number)”.

Overlapping Rectangles

200 rectangles will also overlap and if needed to cover specific post towns may be of varying size. Again, to complete the postcode and provide postcode areas within each town, every post town may have to have different sized area codes - since every post town is to be divided into “the second three [digits, ‘123’] representing the group of properties”. To overcome this codes could have to be of a different length in different post towns (not unlike UK postcodes). Any location elements, to provide the “hybrid” ability, will be incredibly poor at any accuracy - wasting much of their mathematical potential in overlapping regions and maybe enforcing a lowest common denominator of scale.

Tighter Republic

However the code is not lost yet, we can salvage some room. For example, the proprietary loc8code slices the island into a grid 4x5, but slightly offset and excluding the top-left area of sea to minimise the area covered. Each square is said to be about 80km by 90km which gives a total area of 136,800km².[5] (The calculations thus far have relied on an area 166,892km² of a rectangle covering each extremity of the island and islands.)

Since the Government proposal has settled on the notion of 200 post towns and the code has capability for many many more - it is safe to assume that these 200 post towns can be in the Republic only without removing the possibility of an All-Ireland postcode in the future.

The crude ⅚ calculation[6] would imply 114,000km². It is hard to estimate without actually mapping a tighter grid: the land mass of the Republic alone is said to be 70,273km²[7] but no grid reference can be that tight around the boundaries. A more detailed grid will lie somewhere between the two measures.

The OSI Discovery[8] series attempts a tighter division of the country for its maps. It covers the Republic in 71 maps, about 1,200km² each: an estimated area of 85,200km² - though with significant overlap and missing some areas covered in OSNI maps for the North.

A tighter grid is always possible. But the more refined you make the grid the less geolocation calculated your grid becomes and the more database-lookup the postcode becomes. But for the sake of argument let us settle on 85,200km² as perhaps even quite an ambitious estimate of mapped area of Republic.

A grid larger than 200 elements will exceed the post town specification - although the “ABC” code itself is capable of many more.

Post towns would now be an estimated 433km² - about 21km by 21km.

Irregular Sized Areas

With differently sized, irregularly placed, post town regions, these regions will however by definition in places have to exceed the minimum size of regular divisions of 21km by 21km.

While crucially, there will then be no logical way to determine from a location on a map to which post town it belongs - whether it is on the outskirts of one neighbouring “town” or another or part of one overlapping grid or another. All postcode software will have to access a database to determine the postcode from a location or vice versa the location from a postcode.

Small Areas

The OSI[9] already has a defined Electoral District (ED) and Small Area (SA) division of the country. “There are 3,441 ED boundaries which contain an average of 1,144 people, in each. There are 19,760 Small Areas Boundaries (SA), which fit neatly within the Electoral Districts (ED) Boundary dataset. They were created in relation to the amount of households within each region to give areas of approximately equal populations. Each Small Area contains an average of 90 households to maintain privacy and data protection.”

The proposed code of has capacity to encode these SAs and it could be arranged with a new layer of Post Towns to group together EDs.

This would not however be a hybrid system and would not facilitate the addition of a location code as per “the model is capable of being refined into a location-based code – in other words identifying each individual property within a post town”.

SAs and EDs have to be redrawn periodically to accommodate changing demographics. Postcodes would require ongoing management and would change regularly as divisions were redrawn.

They would also not be capable of reaching the Government’s stated target of “approximately 40 to 50 properties (identified by number)”.

Names

Aside from the maths - this proposed scheme using the first three characters of which “permutations might include: GLY (Galway); CK (Cork); and KKY (Kilkenny)” has no uniform logic to decide on a code. Indeed, even in the example given by DCENR they have broken their own schematic and given Cork a shorter two-character code - which will come as a lovely surprise to any software programmed to expect the “ABC-123” format.

In English, just looking at the top 100 sized towns in Ireland[10] - 2 begin with “ATH”, 4 with “BAL”, 2 “BAN”s, 5 “CAR”s, 2 “CLO”s, 4 “DUN”s, 2 “ENN”s, 5 “KIL”s, 2 “KIN”s, 2 “MAL”s, 3 “NEW”s, 3 “POR”s, 2 “ROS”s, and 2 “TRA”s. We would find the same clashes no matter what rule we applied to Irish town names. (And that’s just the top 100 - not 200 as proposed).

Moreover the use of GLY-123 for Galway addresses is nothing but a complete waste of coding. We already have gone to the trouble of writing Galway in the address. “GLY” is just redundant. While the government believes that it makes the code easier to remember and is similar to car registration codes, it fails to notice that cars can leave their registration county. It is actually extra information that a car has a C reg while driving through Dublin. It is no extra information to say that an address in Cork has a CK postcode. It is strictly pointless and inhibits any code’s use to verify an address.

The Government are also entirely underestimating the ability of the Irish people to remember their own postcodes. Surely this cannot be a problem. We’ll all only have one code to remember. If we are very lucky: two. It must be said that we have no problems remembering our phone numbers which have no relationship to us at all, nor car registrations, even where the car has an original registration outside our own county!

Indeed it will be to the code’s benefit that other people’s postcodes are not trivial to remember - it makes the code an essential point of accuracy, a verification of address information. PPS numbers are a good example. They are not trivial to remember. They are an essential point of accurate precision information for State services. We are forced to not approximate people’s PPS numbers and only ever purposefully record exact numbers.

Language

And yet perhaps one of the biggest problem is not even maths - it is language. English is not the only official language of the State; it is not even the first official language of the State.[11] To enshrine English, as proposed, into the postcode is to make a complete mockery of any bilingual aspiration - especially when there is no necessity and ample reason not to use language based codes.

We’ve already had a modern example of places changing their names. In 1994 the Government establish new counties: Dún Laoghaire–Rathdown, Fingal, and South Dublin. There’s nothing to say town name changes won’t happen in the future.

“‘123’ … representing the group of properties in which the particular building is located”

The proposed code continues with three numerical digits: “the second three representing the group of properties in which the particular building is located”.

1000 Postcodes

A set of digits from 000 to 999 gives us simply a very small set of a 1,000 different codes to work with. Within a “post town” (as designated by the first three characters of the code), this number will describe only a crude 31x31 grid: which will describe areas of 0.49km² about 0.7km by 0.7km.

40 to 50 Properties

Even according to the Government’s own very odd low-definition aim for a postcode of “approximately 40 to 50 properties”, the area that the final “ABC-123” code could describe will be significantly larger - a generous estimate of 0.7km by 0.7km with some properties within the same postcode being nearly 1km apart as the crow flies. In urban areas many times more properties with be covered by each postcode. A postcode including Dublin City Hall could strength from the river to as far St. Patrick’s Cathedral and Stephen’s Green. While it is entirely possible that this estimate will have to be multiples higher.

In rural areas it a bit better - not quite as bad as the traditional vague townland address but still not accurate enough to overcome the rural addressing problem in Ireland without the addition of house numbers.

Indeed, while the format plans for just 200 post towns it is an incredible waste of coding to describe the town with an “ABC” code capable of 17,576 different entries while leaving the precision of identifying an address within an area no smaller than 21km by 21km to a coding only capable of a 1,000 entries. Even without vowels, three letters would be over 9 times more precise than 3 digits. If anything the code should at least have the format "123-BCD" not “ABC-123”.

Variable Precision

Then if we do attempt to make the code work at higher precision by concentrating detail on populated areas the corresponding dilution of precision in other areas will be enormous; for example, if we chose 50 urban “post towns” to have a quarter the standard area so that we could be a lot more precise about addressing within each it would double the size every other of the remaining 150 post towns. An urban “post town” might be roughly 10.4km by 10.4km; but it will make rural post towns 29.42km by 29.42km. And yet final property codes would still be only 0.35km by 0.35km in urban areas and increase to 0.99km by 0.99km in rural areas - which is quite some ground for a postcode.

Obsolescence

We might also ask, if the postcode scheme is to be based on current population centres; what happens in the future when demographics shift, when towns expand, when new towns appear? Will the postcode system have to be rewritten? Are we automatically envisaging yet another costly administrative quango to continually refine the postcode system to keep up with every new house built?

New addresses will appear. Indeed, new building, streets, and towns will always appear. One cannot assume that what is less dense at present will remain so.

Checksum

The proposed code does not include a self-checking mechanism such as that included in PPS numbers. Codes can not be validated.

Conclusion

The Government proposed postcode model of “ABC-123” to be centred about 200 English language “post towns” is incapable of providing any better than a generously estimated average 0.49km² precision. At this, two addresses in the same postcode could be nearly 1 km apart as the crow flies. This is bigger than most villages and large enough to encompass the heart of many small towns. In rural areas it is a bit better - not quite as bad as the traditional vague townland address but still not accurate enough to overcome the rural addressing problem in Ireland without the addition of house numbers. It copies an unwieldy, outmoded, unsuitable UK postcode system. The proposed code will provide less location information than the ordinary address. It will be useless to emergency services. An Post do not need nor want such a code. The code will not help in verifying addresses. A non-uniform code will require a database-lookup approach to conversion to and from mapping locations - imprecise as they will be. It will also, no doubt, necessitate a new quango just to maintain and revise the code. Its built-in linguistic discrimination is yet another blow to the Irish language and bilingualism.

It will achieve very little in rural places without widescale address restructuring; and in urban areas do absolutely nothing but needlessly join whole blocks of streets into a single code, less refined that the name of any city street itself. It will not easily be extended as a hybrid system to provide a location of single addresses as suggested: “the model is capable of being refined into a location-based code – in other words identifying each individual property within a post town”.

An area at least 0.49km² is nearly useless for anything but statistics and a general notion of an area. It is as large as villages and the centres of small towns. It will not locate specific properties nor small groups of properties nor help with emergency services.

Neither can the code be made to work successfully at higher precision by concentrating on populated areas and not spread over all possible land and islands without seriously abandoning rural usefulness. Every address needs to be covered by a system. New addresses will appear. Indeed, new building, streets, and towns will always appear. One cannot assume that what is less dense at present will remain so.

UK Postcode

An average area 0.49km² is a far cry from a comparable UK postcode which identifies small groups of houses or parts of streets.

Within each Post Town or delivery area in the UK the area is subdivided into a "9AA" format. This is a code over six times more dense than the "999" proposed for Ireland. Regardless of country size, the local area expected density has to be no different in the Republic as it would be in the North. Northern Ireland, by comparison, currently has 82 Postcode Areas or "Districts". To propose 200 by comparison for the Republic with each with a significantly less precise final postcode doesn't look anywhere near possible.

Beyond Postcodes

One has to wonder if anyone has ever thought this hybrid approach through, or even done the most basic thinking about the mathematics and coordinate geometry involved.

Apart from which, there is the obvious fact that to even copy the North of Ireland UK postcode would be to adopt an unwieldy, outmoded, unsuitable, and outdated system.

UK postcodes are based on towns and street addresses, Irish addresses however and Irish settlements are nowhere near as orderly. Rural Ireland doesn’t have street addresses and hardly has significant village and town settlement. We’ve been living in splendid isolation doted around the landscape in large blocks of townlands all this time. A postcode planner trying to fit these addresses into a town-street idiom won’t be long finding an infinity of impossibility.

An Post itself has statedly decided that it doesn’t need nor want a postal delivery code. Service providers, couriers, planners, statisticians, individuals, and emergencies services however are crying out for a sensible location based addressing system in Ireland.

Improving the DCENR Format

  1. Naming post towns is both linguistically biased and wasteful in coding terms. Using a “123-BCD” format instead of “ABC-123” as proposed would number post towns - something not unlike the French Department codes. Using the alphabet for postcodes within post towns will instantly make the postcode 9 times more precise (that is excluding vowels to prevent unwanted words forming - base 21).
  2. Aligning the coding to “post town”s is wasteful - it doesn’t allow an efficient, straight-forward, logical calculation of postcodes from location; and it wastes coding potential by overlapping areas. Using the first 3 digits for a simple grid allocation much like the Irish Grid is preferable for efficient coding.
  3. A three digit number has factors of 25 by 40 which may be a suitable shape. A square grid is limited to 31 by 31 - which wastes nearly 4% of coding. But ideally if a postcode is to eventually create a location code it would be logical to have numbers of digits based on pairs of digits - latitude and longitude - two or four digits.
  4. Extending the second group of characters to include numbers to make base 27 - using the numbers 2-9 (0 and 1 can look like vowels) and the alphabet minus the vowels and “Y” to avoid spelling words, “S” which gets confused with 5 - is nearly twenty times more precise than using letters alone, e.g., “123-B2C”. This variation could be nearly as accurate as 90m by 80m. As stated by the Joint Committee “We do not believe that a system based on clusters of 40-50 properties as proposed by the Government is appropriate for the current and future needs of the smart economy. A unique identifier system provides for a speedy response by emergency services, provides greater efficiency and interaction with GPS technologies and does not require change in names of townlands, etc, has low maintenance costs, supports spatial planning and the delivery of health services in particular.”[12]
  5. Adding a checksum character to the end will add useful validation of codes entered, make the codes more useful and reliable, and avoid most transcription errors. e.g. “123-B2C-D”.
  6. An ideal would be at least “AAAA-BB-C”, where “AAAA” is an area code describing a location grid using the 27 characters “BCDFGHJKLMNPQRTVWXZ” and “23456789”, “BB” the final postcode using the same 27 characters, and “C” being a calculated checksum using the greater 33 characters “ABCDEFGHJKLMNPQRTUVWXYZ” and “0123456789”.

Alternative Systems

The ComReg report of 2005 stated that:

“There are a number of complex technical issues that need to be taken into account in designing postcodes. However, it is desirable to set out some general principles that underline the work being undertaken moving forward.

State Systems

GeoDirectory is an autonomous company under An Post with a location database of all Irish addresses.

OSI has a defined Electoral District (ED) and Small Area (SA) division of the country. “There are 3,441 ED boundaries which contain an average of 1,144 people, in each. There are 19,760 Small Areas Boundaries (SA), which fit neatly within the Electoral Districts (ED) Boundary dataset. They were created in relation to the amount of households within each region to give areas of approximately equal populations. Each Small Area contains an average of 90 households to maintain privacy and data protection.”

Both are capable of providing a traditional non-location based postcode immediately.  While the GeoDirectory is capable of converting every address into a geolocation calculated code such as those below.

Proprietary Systems

Loc8code Ltd. have an eight character closed-source "Loc8code" license calculated from OSI Mapping only which is claimed to be able to "define locations to an accuracy of +/-6 metres". The code has a built in checksum.

Go Code Ltd. propose a seven character closed-source "GoCode" stated to be "up to an accuracy of 5 square metres approx.". It does not include a checksum.

Proprietary systems are not free and open and have code structures would could be described as “hidden”.

Open Source Algorithm

The OpenPostcode, developed by the author, is Ireland's most accurate location code at 8 digits; while even at 7 characters at <3m average radial offset from a point - over 4 times more precise than Loc8code: free, logical, extensible, independent, convertible, mappable; simply calculated by formula (not by committee) - open to everyone.


[1] http://www.oireachtas.ie/parliament/mediazone/pressreleases/title-1733-en.html 

[2] As per Article 3.1. unification is a constitutional goal.

[3] Or any other grid based on the factors of 17,756.

[4] Height 466.7km (from 55.5N to 51.3N at 8.05W), width 357.6km (from 10.75W to 5.35W at 53.4N).

[5] https://www.facebook.com/photo.php?fbid=371772859518643&set=a.152747578087840.31508.139089036120361&type=1&theater 

[6] https://www.cia.gov/library/publications/the-world-factbook/geos/ei.html 

[7] https://www.cia.gov/library/publications/the-world-factbook/geos/ei.html 

[8] http://www.irishmaps.ie/productlistingM.cfm?SeriesID=11 

[9] http://www.osi.ie/en/alist/products-overview.aspx?article=c86d160d-cc84-4951-8517-0ee457e2175a 

[10] http://en.wikipedia.org/wiki/List_of_towns_in_the_Republic_of_Ireland_by_population 

[11] http://www.irishpressreleases.ie/2009/09/30/conradh-na-gaeilge-demands-irish-postcodes-for-an-irish-country 

[12] http://www.oireachtas.ie/parliament/mediazone/pressreleases/title-1733-en.html 

[13] http://www.comreg.ie/_fileupload/publications/ComReg0507.pdf