Traffic flow maps

Traffic Diagram of London, by Ludwig Karl Hilberseimer, 1944. Ludwig Karl Hilberseimer Archive, Ryerson and Burnham Archives, The Art Institute of Chicago.

A tweet by artist, designer and developer Jill Hubley drew my attention to a traffic flow map of London, created by Bauhaus architect Ludwig Karl Hilberseimer. The map shows the number of buses passing through London’s central arteries in one hour. «The traffic diagram of London shows both the typical congestion in the center and the lack of transportation facilities at outlying points,» he commented. Hilberseimer thought the solution to this transportation problem was to decentralise the city, by creating satellite cities with a population of at most 100,000.

Hubley has been tweeting numerous historical traffic flow maps, including a beautiful 1944 map showing transport along the waterways of Belgium and the Netherlands. What struck me about the Hilberseimer map is its similarity to a series of maps in a traffic study published by the city of Amsterdam in 1976. These maps were discovered by Marjolein de Lange of cyclists’ organisation Fietsersbond, and have been reproduced in the book Bike City Amsterdam she wrote with Fred Feddes.

From: Voorontwerp verkeerscirculatieplan Amsterdam, 1976. Photo Marjolein de Lange

The Amsterdam maps illustrate how cycling had declined in Amsterdam between 1961 and 1971, and how rising car use had created a congestion problem. It wasn’t until later that the city developed measures to promote cycling, as analysed in Bike City Amsterdam. I tried to create a 2016 version of the cycling map using Fietstelweek data, but it should be noted that the cycling routes of Fietstelweek participants may not be representative of overall bicycle traffic in Amsterdam.

Compared to Hilberseimer’s map, the maps created by the city of Amsterdam have a very clean design: all cartographic details that do not represent traffic data have been omitted. And then there’s the elegant legend. Hubley has tweeted a Swedish traffic flow map from 1977 with a similar type of legend, as well as a map of Florida from 1952, a map of St. Paul - Minneapolis from 1949, and a 1945 map of eastern Germany with a horizontal version of the legend. (Update - interesting variant on this 1963 Lincolnton map; plus see this 1921 map of Seattle.) I wonder whether earlier examples exist.

Were the flow maps in Amsterdam’s traffic circulation plan inspired by Hilberseimer’s Traffic Diagram of London? Possibly, but Hilberseimer wasn’t the first to create a traffic flow map. In fact, both Amsterdam’s map makers and Hilberseimer are indebted to a map created a century before Hilberseimer’s map, by the French civil engineer Charles-Joseph Minard.

Carte de la circulation des voyageurs par voitures publiques sur les routes de la contrée où sera placé le chemin de fer de Dijon à Mulhouse (source), by Charles-Joseph Minard, 1845. Reproduced with the kind permission of the Ecole nationale des ponts et chaussées.

In her book The Minard System, visualisation strategist Sandra Rendgen comments:

In this revolutionary map, created in the middle of a debate about where to project the railroads between Dijon and Mulhouse in eastern France, Minard analyzed the street traffic on preexisting roads in the region.

Apparently, the map was so influential in shaping the debate that a fake copy was made ‘in an attempt to prove another route to be more promising’.

Rendgen describes how Minard initially created bar charts to represent traffic along segments of a route. At some point, he decided to project these graphs onto a map, which resulted in the creation of the flow map. Over time, Minard’s flow maps gained in complexity, as he used colour to represent different types of data. Minard is sometimes credited with inventing the flow map, but Rendgen points out that the design was possibly invented more or less simultaneously in Ireland, France and Belgium.

Minard’s charts and maps often contain detailed descriptions of the data and methods he used. He collected data from a range of sources, and emphasised that graphs should accurately represent the data. On the other hand, he was willing to sacrifice geographic detail or accuracy for clarity. Rendgen points to the ‘clean and minimalist aesthetics’ of his work, devoid of decorations or other clutter. It is no wonder that Edward Tufte, the renowned proponent of clutter-free data visualisation, described Minard’s work as an example of ‘graphical excellence’ (in The Visual Display of Quantitative Information).

A recurrent theme in Minard’s explanatory notes is that he aimed to make relationships quickly apparent to the eye. One of these notes has an almost futurist sense of modernity to it: «The figurative maps are thoroughly in the spirit of the century in which one seeks to save time in all ways possible.»

One could argue that the 1976 Amsterdam traffic flow maps are true heirs to Minard’s approach, and especially to his first, monochrome flow map reproduced above. As Rendgen notes, Minard’s map is «extremely stripped down; it features barely any landscape details other than a network of local place names and rivers». Even those subtle geographical hints have been omitted from the Amsterdam traffic flow maps. Of course, this only works because of the very recognisable pattern of Amsterdam’s streets.

Scraping Airbnb

Airbnb is not exactly keen to share data that might help analyse its impact on local housing markets. In 2016, the Amsterdam Municipality decided to collect Airbnb data using a scraper - a computer programme that automates the job of retrieving information from web pages.

Amsterdam is not the only government to use web scraping. Increasingly, this technique is used to obtain data about topics ranging from consumer prices to jobs vacancy statistics and business data. Collecting data from the internet has advantages, but it also poses some challenges. It may be difficult to aggregate data coming from different websites, and data found online may not cover all aspects of a phenomenon you’re trying to understand (for example, not all job vacancies are published online). On a more practical level, your web scraper code may break when websites change.

In March 2017, Amsterdam reported that its weekly scrapes of major platforms like Airbnb required little maintenance. But last week, it sent a report to the city council describing how Airbnb has been making changes to its website - perhaps in an attempt to frustrate efforts to collect information about its business practices. Initially, Amsterdam’s digital surveillance department succesfully updated its scraper, but following new changes to the Airbnb website since May 2018, Amsterdam now appears to have given up on scraping Airbnb.

This made me curious about the technical characteristics of the Airbnb website. Here are some observations, based on an (admittedly superficial) examination:

  • The initial download of a web page isn’t the final version: after downloading, the contents of the page are dynamically altered using Javascript. For some purposes like navigating search results, you may prefer the final version of the page, which you can get using Selenium. Selenium would especially come in handy for interacting with the calendar to get availability and price information, which seems to be rather tricky.
  • Some details on listings only appear to be available in the Javascript code. You can find them using patterns like '\"lat\"\:(.*?),\"lng\"\:(.*?),'
  • Airbnb uses NGINX to control access to its website. If you request too many pages too fast, you’ll hit a rate limit and get an error page. I guess it should be possible to avoid the rate limit by adding pauses to your programme, but it may take quite some time to figure out how often and how long they should be.

While it appears that barriers to scraping the Airbnb website may be surmountable, it’s quite possible that I underestimate what this would take. If you’d actually build a scraper and would use it to frequently collect information about all local listings, all kinds of new problems might arise.

Meanwhile, other sources of Airbnb data are available. In a previous post, I used data made available by Tom Slee and by Murray Cox’ Inside Airbnb. Slee has since stopped updating his data, but Inside Airbnb is still active. As the Amsterdam Municipality notes in its report, Inside Airbnb has succesfully adapted its scraping technique each time Airbnb changed its website.

UPDATE 13 May - See comments on Twitter: Jens von Bergmann from Vancouver also has a scraper that is working. Following some requests, Tom Slee recently updated his scraper; his code is available on Github.

Visualising Amsterdam’s cyclists

Bike City Amsterdam, a new book by Fred Feddes and Marjolein de Lange, recounts how Amsterdam developed a cycling policy (more on the book below). An important source for the book is the archive of the Amsterdam branch of cyclists’ organisation Fietsersbond. In addition, traffic data was used to analyse trends.

An interesting dataset consists of counts of the number of cyclists, cars and other road users moving into and out of Amsterdam’s city centre, over the years 1980–2009. Most of the locations where traffic was counted are on the Singelgracht, which encircles Amsterdam’s city centre.

The data represents manual counts on a single day, between 7am and 7pm, of traffic in both directions.

I was asked to think about a way to visualise this dataset, which posed an interesting challenge (and was a lot of fun to do). Below, I’ll discuss a few of the options we considered.

Radar chart

Given the geographical distribution of counting locations, it seemed to make sense to try a circular chart design. In fact, that idea had also occurred to the city’s infrastructure department. In a 2007 fact sheet, they used a radar chart (or cobweb chart) to visualise the Singelgracht bicycle counts.

Incidentally, they didn’t use the term radar chart, but called it a fan (waaier). They used a bicycle metaphor to describe how it works: «from the middle, the counting locations around the city centre are connected like spokes in a bicycle wheel».

The chart looks really nice, but this chart type also has a drawback: there’s an implicit suggestion that the area within the purple line represents the number of crossings, which is in fact misleading (see this article for a discussion of a similar problem). Another limitation is that the chart doesn’t show how bicycle traffic changed - although it would be possible to make a version with separate lines representing 1980 and 2009.

Radial lollipop chart

As an alternative, I created what I’ll call a radial lollipop chart (to my knowledge, this chart type didn’t exist yet). The chart library that I use, D3.js, doesn’t seem to have a method to draw the ‘spokes’, or at least I couldn’t find it. Therefore, I wrote a function that calculates the start and end points of the lines. I had long forgotten how to use sine and cosine, so I had to look that up. I’ve published the code here.

Here’s a radial lollipop chart showing how cycling has increased at virtually all the Singelgracht crossings.

And here’s one showing the opposite effect for cars:

I love it when a chart has data points that break out of the chart area - although this is perhaps a bit extreme. The outliers are due to the fact that a large share of car traffic uses the Wibautstraat - IJtunnel route. I could have changed the scale to include those outliers, but then changes on other routes as well as changes in bicycle use would have become much more difficult to discern.

Area chart

I rather like the radial lollipop chart, but it has a limitation: it shows changes between 1980 and 2009, but not when those changes happened. Car use started to go down before cycling really started to increase, but from the radial lollipop chart you couldn’t tell.

This is why the chart used in the book is an area chart, with colours corresponding to the broad geographical orientation of the crossings. Simple, but effective. And if you want to explore the details, click here for a draft version of the charts: bicycle, car.

About the book and exhibition

On 4 April, the Amsterdam branch of cyclists’ organisation Fietsersbond has handed over its archive to the Municipal Archive. Marjolein de Lange, who coordinated a volunteer project to prepare the archive, came up with the idea to use the material as input for a book - a project she carried out with author Fred Feddes.

This resulted in a very interesting book about activism versus cooperation; the place of cycling in urban planning; and how the magic power of Amsterdam’s cycling culture decided the epic fight for the right to cycle through the passage under the Rijksmuseum. The book, which contains a wealth of great photos; maps and posters, is a must-read for anyone interested in cycling, Amsterdam, or activist poster design. It’s been published both in Dutch and in English. There’s also an exhibition at the Municipal Archive (until 30 June, Vijzelstraat 32, access is free).

Using a jagged baseline to indicate a broken y-axis

In an article for the recently created Data Visualisation Society, R.J. Andrews suggests using a jagged baseline to indicate a broken y-axis (i.e., an axis that doesn’t start at zero). The idea - inspired by some beautiful charts dating back to WWI - is to suggest that the bottom part of the chart has been torn off. I like the idea - but I found it isn’t easy to implement.

Contrary to the view of some chart fundamentalists, using a y-axis that doesn’t start at zero can be perfectly ok in some situations. Still, one might want to alert the reader that the zero line is missing. One way is to add a little zigzag or some other symbol to the y-axis, as shown here. And then there’s Andrews’ suggestion to use a jagged baseline.

I tried to implement this in a chart that shows the number of flights at Schiphol Airport. For background: Schiphol has all but reached the cap of 500,000 flights per year, agreed on after negotiations between local residents and the aviation industry. There’s currently a heated debate on whether Schiphol should be allowed to grow further. Experts expect that maintaining the cap will result in more efficient use of the available slots (e.g. fewer short-distance flights, fewer low-cost flights, larger aircraft and fewer empty seats).

Creating a jagged baseline is a bit of a hassle: you have to remove the regular baseline, move the axis labels down a bit and create a new, jagged baseline.

And then there are some design issues. Having the baseline and the ‘regular’ chart lines look too similar may cause confusion. In fact, all of Andrews’ examples have very pronounced chart lines, which are clearly distinct from the baseline. If you prefer a more subtle approach, another solution is to use a light colour for the baseline.

Then again, it also matters whether there are gridlines. After some experimenting, I think the jagged baseline only works well with gridlines added; without them it looks a little weird. But see for yourself if you agree.

I’ve written a Python script to download and clean Schiphol Airport traffic data; find it on Github.

Tags: 

How to investigate assets: lessons from The Wire

I’m rewatching The Wire. It’s a great series anyhow, but for researchers, episode 9 of the first season (2002) is especially interesting. It features detective Lester Freamon instructing detectives Roland Pryzbylewski and Leander Sydnor how to investigate the assets of drug kingpin Avon Barksdale.

They use microfilm instead of the Internet. They don’t have databases like Orbis, Companyinfo or OpenCorporates, and they don’t seem to calculate social network metrics. Yet the general principles behind Freamon’s methodology still make perfect sense today:

Start with the nightclub that Barksdale owns. Look up Orlando’s, by address, you match it, and you see it’s owned by - who?

Turns out it’s owned by D & B Enterprises. Freamon tells Prez to take that information to the state office buildings on Preston Street.

Preston Street?

Corporate charter office.

Corporate who?

They have the paperwork on every corporation and LLC licensed to do business in the state. You look up D & B Enterprises on the computer. You’re going to get a little reel of microfilm. Pull the corporate charter papers that way. Write down every name you see. Corporate officers, shareholders or, more importantly, the resident agent on the filing who is usually a lawyer. While they use front names as corporate officers, they usually use the same lawyer to do the charter filing. Find that agent’s name, run it through the computer, find out what other corporations he’s done the filing for, and that way we find other front companies.

This is pretty much the same approach you’d take when investigating shady temp agencies: trace connections via (former) shareholders, board members, company addresses and related party transactions. And, of course, try to figure out where the profits go.

On that aspect, Freamon also has some wisdom to share:

And here’s the rub. You follow drugs, you get drug addicts and drug dealers. But you start to follow the money, and you don’t know where the fuck it’s gonna take you.

Tags: 

Delete Facebook

This is becoming a bit of a tradition: me writing about people who make a New Year’s resolution to quit Facebook. The story is simple: around the turn of the year, there’s a peak in people googling how to quit smoking, but there’s an even larger peak in people trying to figure out how to delete their Facebook account.

But this year, the story is a bit more complicated (and more interesting).

Google Trends data isn’t available yet for the last days of the year, so there’s no new peak in searches for “quit smoking” yet. Other than that, the yearly pattern is dwarfed by a huge peak in search volume for “delete Facebook” in the week starting on 18 March. What happened?

The Guardian has helpfully created an overview of Facebook-related incidents during 2018; I’ve added a few stories that also seemed relevant (for sources, see Method below; thanks to Vicki Boykis for the suggestion to annotate the Google Trends chart).

No surprise: the largest peak in “delete Facebook” searches happened a few days after the publication of the Cambridge Analytica story on 17 March. The news resulted in a veritable #deletefacebook campaign, although according to Mark Zuckerberg, «I don’t think we’ve seen a meaningful number of people act on that.»

Arwa Mahdawi has argued that deleting your Facebook account isn’t a bad New Year’s resolution, even though it probably won’t change how the company operates: «Facebook’s abuse of power isn’t a problem that we can solve as individuals. Technology giants must be regulated.»

So how much impact did the controversy have on Facebook? One way to try and answer this is to look at the share price.

The pattern for Facebook is rather interesting. The share price dropped after the publication of the Cambridge Analytica story, but quickly picked up again. But then it took a plunge on 25 July, resulting in ‘the biggest-ever one-day wipeout in U.S. stockmarket history’.

One possible interpretation is that investors initially thought the Cambridge Analytica story wasn’t going to harm Facebook’s profits. But when Facebook published its Q2 earnings report, they were shocked to learn that user growth had stalled.

But the chart also shows that all major tech companies saw their share prices go down. This suggests there’s more going on than users leaving Facebook. In addition to broader economic trends, a likely explanation is that investors fear more government regulation of major tech companies in response to the controversies they are involved in (and also to their dominant market position). While this may not be the whole story, it does seem to support Mahdawi’s view about the key role of regulation.

Method

Note that Google Trends data should be interpreted with caution because Google doesn’t provide much detail on the methodology used to produce the data.

For periods longer than three months, only weekly data can be downloaded. For the 2018 chart I wanted daily data. As suggested here, I downloaded three-month batches with overlapping data and then used the overlapping dates to calculate a ratio to adjust the scales. Here’s the code:

import pandas as pd
import numpy as np
 
def stitch(df1, df2):
    df1.index = df1.date
    df2.index = df2.date
    overlapping = [d for d in df1.date if d in list(df2.date)]
    ratios = [df1.loc[d, 'delete facebook'] /
              df2.loc[d, 'delete facebook']
              for d in overlapping]
    ratio = np.median(ratios)
    for var in ['delete facebook', 'quit smoking']:
        df2[var] *= ratio
    df = pd.concat([df1, df2[~df2.date.isin(overlapping)]])
    return df
 
df = dfs[0]
for df2 in dfs[1:]:
    df = stitch(df, df2)

I used this Guardian article as my main source on Facebook-related incidents in 2018. I added a few from other sources: in April, Facebook announced 87 million people had been affected by the Cambridge Analytica scandal. Subsequently, it announced that it would notify people who had been affected. Dutch comedian Arjen Lubach organised a Bye Bye Facebook event (reminiscent of the 2015 Facebook Farewell Party). In September, Pew found that one in four Americans had deleted the Facebook app from their phone; and later that month a Chinese hacker threatened to delete Mark Zuckerberg’s Facebook account.

Tags: 

Who use the Dafne Schippers bicycle bridge

If all goes well, the Dafne Schippers bicycle bridge in Utrecht should reopen on Monday, after a short closure for maintenance. I have a special affinity with this bridge: it opened on the day I started working in Leidsche Rijn, west of the Amsterdam-Rhine Canal, and it’s part of my favourite cycle route to work.

Who else use this bridge? With the usual caveats, data of the Fietstelweek can provide some insights. The charts below show, for each direction of traffic, at what time cyclists use the bridges across the canal.

There’s a morning peak in cyclists crossing the canal from Leidsche Rijn (west) to the city centre (east), and a peak in cyclists going the opposite direction around 5 pm. This suggests that the bridges are popular among commuters from Leidsche Rijn. That doesn’t really come as a surprise: if you cycle to Leidsche Rijn during the morning rush hour, you ride past huge numbers of cyclists going in the opposite direction.

The map below shows the routes of cyclists using the bridges. From top to bottom: Hogeweidebrug (or Yellow Bridge), Dafne Schippers bridge and De Meern bridge.

It appears that many cyclists use the bridges to go to the area around Central Station. Users of the De Meern and Dafne Schippers bridges tend to use nice routes that converge along the Leidseweg. Users of the Yellow Bridge use the not-so-nice route along Vleutenseweg, or the slightly better route along the railway track.

Research has shown that cyclists don’t always prefer the shortest route to their destination; the quality of the cycle tracks also plays a role.

Yet the map suggests that many cyclists opt for the shortest route, even if a nicer alternative is available. For example, few cyclists from the northern part of Leidsche Rijn seem to use the Dafne Schippersbrug, or the route along Keulsekade (the latter avoids long waits at traffic lights).

See also this analysis by DUIC, which shows that the bridge is not only popular among cyclists, but also among runners, which is fitting given the name of the bridge.

Tags: 

Which Amsterdam neighbourhoods might qualify for an Airbnb ban

Last week, city council member Sofyan Mbarki (Social-Democrats) proposed a motion to ban holiday rentals in Amsterdam neighbourhoods such as the Haarlemmerbuurt, the Kinkerbuurt and the Wallen. A concentration of holiday rentals results in rising house prices, lower social cohesion, increasing pressure on the housing market and inequality, he argued. The motion has support from a majority of the council.

The city government is inclined to implement the motion, but alderman Laurens Ivens (Socialist Party) wants to study the legal aspects. He considers the neighbourhoods mentioned in the motion good candidates for a ban on holiday rentals, but he doesn’t rule out that other neighbourhoods may be selected.

So what neighbourhoods might qualify? One criterion might be Airbnb density, which is shown on the map below (for caveats see Method below).

Unsurprisingly, neighbourhoods with high Airbnb density overlap with areas where residents complain about holiday rentals: Centrum-West, Centrum-Oost, Westerpark, Oud-West/De Baarsjes and De Pijp/Rivierenbuurt (source).

Airbnb frequently claims that it contributes to tourist dispersion because many hosts are located outside the city centre. However, the map suggests that Airbnb is in fact heavily concentrated in neighbourhoods such as the Wallen, the Jordaan, the Pijp and the Kinkerbuurt. While some of these neighbourhoods are outside the city centre, the pattern appears to be concentration rather than dispersion.

While these neighbourhoods would be likely candidates for a ban on holiday rentals, Ivens may also want to anticipate future developments. A number of neighbourhoods still have a relatively low Airbnb density, but have seen their density double or even almost triple over the past three years: Transvaalbuurt, Hoofdweg e.o., Van Galenbuurt and Westindische Buurt.

UPDATE - It was rightly pointed out that Airbnb density partly reflects housing density. An alternative measure would be Airbnb relative to addresses or population. However, this would result in high values for some areas with low population density where holiday rentals don’t appear to be perceived as much as a problem as in some of the more densely populated areas.

See also:

  • Is tourist dispersion working? An analysis of Lonely Planet maps
  • Airbnb’s agreement with Amsterdam: some insights from scraped data

Method

Both Murray Cox’s Inside Airbnb and Tom Slee provide data collected by scraping the Airbnb website. While this data has some limitations, it’s probably the best publicly available data source on Airbnb. Since Tom Slee stopped collecting data last year, I used Inside Airbnb data for the current article. A discussion of methodological aspects related to that data is here.

In addition, I used land surface data from Statistics Netherlands (CBS). This data is for 2017.

I calculated an indicator for Airbnb density in the following way:

  • I assigned each listing to a neighbourhood (note that coordinates for listings aren’t 100% accurate as discussed by Cox);
  • For each listing, I calculated an indicator for the number of stays as: reviews per month (an indicator of the number of rentals) * the minimum length of stay (capped at 3 nights following this study) * the number of beds (an indicator for the number of guests, capped at 4 because that’s the maximum number of guests allowed by local regulations);
  • I summed that number for each neighbourhood and divided that by the land surface of the neighbourhood (ha).

Note that the indicator for the number stays will not be equal to the actual number of stays, for a number of reasons:

  • It’s possible that not all beds are occupied;
  • Not all guests write a review (Cox suggests the number of rentals could be twice as high as the number of reviews);
  • People may stay longer than the minimum number of nights;
  • Sometimes more than four people may stay in an Airbnb, despite the fact that that’s not allowed;
  • For some listings, the indicator could not be calculated because of missing data (about 11.4%).

According to Airbnb, the number of stays in Amsterdam is 2.5 million. Based on that number, the actual number of stays would be about 3 times as high as the indicator for the number of stays I calculated. Given the considerations listed above, that’s more or less what one would expect.

Python script here.

After the voter revolt: Collaboration in the Amsterdam city council

The 21 March city council election saw a bit of a voter revolt. Four new parties got elected onto the city council, thanks primarily to voters in the less affluent, peripheral parts of the city. The election outcome reflects Amsterdam’s social divide.

As a result, the composition of the city council changed considerably. So how are the established parties and the new parties getting along?

Before trying to answer that question, let’s have a look at collaboration in the previous city council. There was a left-wing majority in the council, but the government was relatively right-leaning. There was an effective opposition, with GroenLinks (Green Party) and PvdA (Social-Democrats) frequently collaborating to file motions and amendmends.

The chart below shows collaboration in the current city council. The city now has a more left-leaning coalition of GroenLinks, D66, PvdA, and SP. The pattern of collaboration has changed considerably.

The chart suggests that there are three clusters in the city council. One contains the coalition parties GroenLinks, D66, PvdA, and SP. The second contains right-wing / conservative parties VVD, CDA, FvD and PvdO. And the third contains DENK, BIJ1 and ChristenUnie. PvdD (Party for the Animals) appears to be a bit of an outsider by this measure.

Opposition

Are opposition parties able to exert influence, despite their divisions? An interesting measure is whether they succeed in getting proposals adopted despite a part of the coalition voting against. So far, this has happened twice.

One case was a motion from Diederik Boomsma (CDA), asking to provide parking permits to people who have a private garage but have turned it into something else. Coalition party GroenLinks voted against, arguing that people who have made the decision to use their garage for other purposes are now turning to the city to solve their parking problem.

The second one was a motion from Sylvana Simons (BIJ1) asking to the local government to support teachers in their fight for fair wages. PvdA voted against, arguing that the alderwoman had already taken a stand.

The motions can be downloaded here, and here’s a Python script to process them.

Tags: 

Trust instead of algorithms

UPDATE 11 March 2019 - A survey by Dutch trade union FNV found that 25% of participating municipalities apply algorithms to personal data to label welfare recipients as potential frauds. 9% said they hire a commercial organisation to do the analysis. FNV’s vice president Kitty Jong condemned the practice.

A number of Dutch cities have contracted a company named Totta data lab to predict which welfare recipients may have committed fraud (the cities were somewhat secretive about this approach, but newspaper NRC wrote about it last spring). Totta has trained algorithms on a considerable amount of personal data: 2 to 3 hundred variables over a period of 25 years.

Such analyses carry the risk that existing biases are reproduced:

Luk [A Totta spokesperson] says that in some municipalities more fraud is found among people who have a partner (e.g., they don’t report income), whereas in others it is people without a partner (failing to report they live together). «But it’s quite possible that only that group has been investigated and we build our algorithms on that.»

Luk says they sometimes add ‘deviant’ citizens to the suspects, apparently in an attempt to look beyond the usual suspects.

Another problem is the lack of transparency regarding how this type of algorithms work. Totta doesn’t disclose its algorithms because it wants to protect its business interests; further, it can be difficult to interpret and explain how algorithms work. As a result, the government is unable to explain what criteria it uses to prepare decisions that affect citizens. Recently, the Dutch Council of State expressed concerns over digital decision-making by the government.

Proponents of algorithms argue that they help to detect more fraud while reducing the burden for innocent citizens. In fact, there may not be such a clear distinction. The organisation of welfare agencies said that alleged welfare frauds are often people who mean no harm, but who get into trouble as a result of complex and ambiguous welfare rules.

Still, Amsterdam city council member Anne Marttin (VVD) finds the approach interesting. She asked if Amsterdam uses algorithms and data mining to detect welfare fraude. The answer is no. This is why:

The city government is aware of the use by other municipalities of algorithms and/or data mining to fight welfare fraud. The city does not use such instruments to deal with or prevent welfare fraud. […]

Our services for welfare recipients are based on trust. Further, the city government attaches great importance to the privacy of citizens and the way in which their data is used by the government, for example to develop algorithms. The city government thinks it’s very important that the use of data mining and algorithms doesn’t have a negative impact on the privacy and the legal protection of citizens.

Source (pdf)

Pages