Cartographic Poverty - the grounded truth
Posted by DeBigC on 29 October 2020 in English (English). Last updated on 30 October 2020.Background
The following assertion was made by a blogger around 18 months ago:
"..mappers as a whole are mapping more features (buildings, houses, libraries, schools, green spaces etc) within the wealthier areas of the Dublin region than in the poorer areas." {53degrees}
It was made here and in a follow up here
Reading the two posts together with the twitter promotions for them it is clear that the hypothesis rests on the perception that OSM contributors, specifically those contributing to OSM around Dublin, suffer from some kind of collective unconscious bias, to the extent that they have neglected the deprived areas of Dublin and mapped the affluent areas to a higher level of completion. This assertion, if well founded, would reveal that not only does the map of Ireland, specifically Dublin have gaps, - but that these gaps are patterned in such a way as to further exclude and marginalise the residents of these areas and the second blogpost makes no doubt of this consequence.
We have a great name for the band!: "The Spurious Correlations"
The author refers to 3 or 4 examples of poor areas, sparsely mapped, compared in turn to affluent areas mapped to a higher extent using visual assessment. There is no transparent methodology or apparent logic for these comparisons other than they were what the author appears to want to choose. Perhaps they are conveniently selected to suit the conclusion.The author repeated the claim on twitter [now deleted] on the 23rd October 2020, presenting the evidence of these three or four comparisons as conclusive proof of a “startling lack of completion of working class areas in osm, a citizen science project”. Some charts were shown, demonstrating a very selective method of counting tags. Whether this represents traces of a deeper investigation remains unclear. This needs to be tested here and now.
Transparent Method
An inquiry with some more rigour than the two blogposts is much needed now. The first step is to refer to the deprivation index available here, and identify 10 outlier electoral divisions that are the most affluent, and 10 outlier electoral divisions that are most deprived1. If the assertion holds weight it should be detectable at the outlier values.
The ED boundaries were downloaded from the CSO website and the deprivation index was downloaded from the Irish Government’s official data portal and matched to these boundaries to enable comparison2.
The OSM data was extracted as a .pbf tile from Geofabrik on the evening of the 23rd October and parsed in a PostgreSQL database to exclude areas outside the boundaries of the electoral divisions. Three queries detected the three types of object geometry which is what the blogger’s assertion rests on and simply counts them. A fourth query detected In the case of ways and relations it included those that were partly inside the boundary3.
The calculations are simply the aggregate count of all types of objects, further aggregated into an affluent total and deprived total, and then divided by the size of the area to fairly reflect the fact that the areas vary from 1.06 square km (Avonbeg) to 5.96 square km (Priorswood).
The Result
Let’s get down to it…
The first result produced is a count of all objects in the 20 extreme affluent/deprived areas, grouped into those categories, aggregated and normalised by the area size of the total electoral division area. Ultimately this counts and compares the density of mapped objects in the two groups of areas
Affluent EDs objects per Sq KM; | Deprived EDs objects per Sq KM; | |||||
---|---|---|---|---|---|---|
696 | 1,173 |
This result disproves the blogger’s conclusion about deprived areas being neglected by mappers. I will provide a separate diary post showing changeset dates to demonstrate that from 2016 a big upturn in mapping of Dublin’s deprived areas took place, and a rationale for why that began to happen, but that is for another day. And now for the tagging depth result…
Affluent EDs Tag3+ objects per Sq KM; | Deprived EDs Tag3+ objects per Sq KM; | ||||||
---|---|---|---|---|---|---|---|
652 | 186 |
This shows that the affluent areas, though having less object geometry are tagged in somewhat more depth and detail. The figure may be skewed by two of the Affluent EDs which has peculiar characteristics, these are Mansion House and Rathmines East. The former has several city centre streets with specialised Government and Commercial services, however the latter is mainly suburban with a few shopping districts. This does support a totally new hypothesis of more nuanced mapping effort in Affluent areas, but more nuanced does not mean more “complete” as the original blogger’s claimed. The ages of the changesets suggest the original object geometries were typically provided around 2010-2012, and more osm contributors using on-the-go editors like StreetComplete, Vespucci, maps.me and OSMand.
Conclusion
These results, despite the finding of more detailed tagging in the affluent areas, do not support the main hypothesis that Dublin’s socially deprived areas have been deprived of mapping. It would appear that the mapping for these areas is at a higher level of completeness and will be easily converted into tagging depth. There is nothing in the original idea that provides much insight into either the OpenStreetMap community and its mapping preferences, neither is there much to increase the cause of Dublin’s deprived neighbourhoods. A further diary entry will deal with some of the extraneous claims and wrong assumptions made, but that is for another day.
Footnotes
-
This is an outlier comparison often used in large datasets to eliminate “middle to middle” statistical noise. Twenty areas were chosen to increase processing speed. ↩
-
Neither dataset was uploaded to OpenStreetMap since they do not have compatible licence. The analysis was done in QGIS. ↩
-
DeBigC is grateful for the support of Rusty Broderick, who mentored me through PostgreSQL six years ago, and patiently ran the database queries to correct counts ↩
Comment from Lineo on 30 October 2020 at 03:32
A well written piece. Well done!
Comment from FredrikC on 30 October 2020 at 18:15
I had an issue with the original causal link. There are other, more benign reasons why wealthier neighborhoods would have better OSM coverage than poorer neighborhoods. The most obvious comes from post [Maxims for Maximally Effective Mappers(https://www.openstreetmap.org/user/AntiCompositeNumber/diary/394620) which was posted immediately after this post. And I quote:
“2. Map only what you know.”
Mapping on OSM requires a certain amount of leisure time. In general, the poorer the individual, the more time they spend just making enough money to live on and the less time they have for leisure activities. Additionally, the poorer the individual, the less money they have for traveling to other places.
So I’d imagine that the individuals living in poorer neighborhoods don’t have as much time to map their own neighborhoods as wealthier individuals. And that once a neighborhood is completely mapped, the wealthier individuals will be traveling to other places to gather data more than the poorer individuals.
This could also explain why wealthier areas would have better coverage than poorer areas. Bias and opportunity are not the same thing.
Comment from DeBigC on 31 October 2020 at 16:48
@FredrikC That maxim cannot possibly contain the motivations of every mapper who has mapped Dublin. People move around the city of Dublin everyday, they have several areas that they will know just as well as what is immediately around them.
The problem with the “findings” of the blogger were that once we examine at the volume of contributed geometry we found the deprived areas doing significantly better - see the results of the extracts above. The only thing there is less of in the deprived areas is tagging depth, which like your point may require opportunity, knowledge, community support and I would agree that there is likely to be a big gap there.
Having said all this, there are huge gaps all over Dublin, all sorts of areas are not mapped out to completion with landuses and buildings. We simply don’t have enough people mapping, or at least not enough yet to be able to make solid pronouncements like that blog.
Comment from VictorIE on 1 November 2020 at 19:30
I think access to technology and education may be barriers. While basic mapping on OSM may not be difficult, it will be difficult to some who have, e.g. literacy difficulties or don’t know how to tell one type of building from another. And while many will have smart phones, mapping is easier on a computer with a full-sized keyboard and screen.
Another factor will be the number of Points of Interest (POIs) in an area. In suburban areas, there are simply fewer POIs in deprived residential areas than in affluent residential areas. That’s a problem with society, not OSM.
I tend to map what: * I know about and/or have an interest in technically, e.g. power network, speed limits, bus networks. * I know about and/or have an interest in geographically, i.e. where I live and visit. * I am capable of mapping. I don’t know how to map most types of relations. Mapping houses in housing estates bores me.
Comment from fpagenk on 8 November 2020 at 22:40
I’m not convinced by your quick conclusion. Apparently, object density is about 1.7 times as high in deprived areas and tagging depth is 3.5 times as high in more affluent areas. However, instead of investigating further, you brush away the higher tagging depth with a few sentences and decide that object density is the marker for mapping quality. Could it not also be possible that deprived areas have higher object density because they are simply more densely populated? In that case, the higher density of OSM objects would be due to a higher density of real features. Without trying to take that into account, the raw number of object density isn’t worth much. I’m not saying that the original assertion that wealthy neighbourhoods are mapped better is true, I don’t know that. But your analysis alone can not disprove it.
Comment from Smef09 on 9 November 2020 at 18:54
Thanks for doing this research. The result is a surprize because most ideas about the volunteered mapping would say poorer places are less able to map, but you do show how the poor places are more full of the buildings. When I started mapping my area was already mapped and I started on doing road side trees two miles from home :) …. my point that you shouldnt assume OS mappers only map what is beside them. Poor parts of Dublin have less points to map in my opinion. There are definitely less trees and less lighting, and somebody described it as society’s problem and it is.
Comment from DeBigC on 10 November 2020 at 23:06
@VictorIE I absolutely agree - I think that people from deprived neighbourhoods will not have the same opportunities, but my diary post wasn’t about that. I think there is a role for organisations that have access to such communities to ask experienced mappers to visit them. The fewer level 3 tags is also related to the low numbers of POI opportunities. I agree with you, poorer areas can be bland and devoid of street furniture, planting and infrastructure - creating the POI opportunity, and given what we know about tags those are more likely to be the item that attracts tagging depth.
@fpagenk I didn’t brushoff anything off actually, I went to extra lengths to count and write about things within the confines of a diary post - and I gave the tagging depth finding good coverage of the observed differences between deprived and affluent areas. By the way it is strange for you to suggest that I can compare things mapped against things that might exist in greater density when the map of Dublin is generally incomplete. I disagree with you about deprived areas drawing in more nodes due to density, the opposite is true since oftentimes public housing in Ireland is multi-level apartment based, which in most cases isn’t subdivided and mapped as a single building=residential. I wish you posted more because you don’t actually explain what you mean about this being insufficient proof/disproof. It almost sounds like you didn’t like that I did this diary.
@Smef09 I understand, there are many reasons why a mapper will not map what’s right in front of them, and I also think these maxims may have been things to do in the past, but as the map fills up more they become less relevant. I know mappers whose reach is vast within Ireland. You are right, trees are an item that can be vandalised, and oftentimes that is a feature of deprived neighbourhoods. Do you only map trees?