Monday, 18 May 2015

How much do the Dutch really cycle ? How is it is measured ? Which are really the top ten cycling cities of the world ?

Lists are popular on the internet. As a result, there are often attempts to make lists which rank such things as cycling cities. Such lists are always false. There is no common methodology between different countries and so there is no reliable way to make a ranking. In reality it's quite difficult even to pin down the "correct" figure for one city in one country, let alone to find comparable figures for a range of cities in different countries. I've intended to write about this problem for several years but a recent online discussion led me to a Dutch language article about methodologies for measuring modal split in the Netherlands which made a very good start towards an explanation so I asked the Fietsberaad if I could translate it. Beneath the article you'll find an additional summary from me.

Do Appingedammers make 18%, 30%, 38% or 53% of their journeys by car?

Modal split figures show the relatively popularity of different transport modes. Modal splits can be measured in a number of different ways. The modal split usually shows the proportions either of kilometres travelled or of journeys made as car drivers, car passengers, by train, bus / subway or tram, bicycle, moped and by walking. The statistics often make use of OViN, a statistics from the BCS, based on a survey of a representative sample of residents of the area being measured. The accuracy of this technique is known to have limitations.

In this article we give a picture of the differences that exist in the modal split figures due to:
  • the different units used to measure the modal split
  • the method in which modal split is measured
  • the different groups or time periods which the modal split covers.
  • the error margin.

Units
The modal split is often given as a single figure, an unambigious concept. However there are several types of modal split in circulation, amongst which the most used are:
  • the modal split determined from number of journeys per mode
  • the modal split according to the number of kilometers traveled per mode.
In our example, we'll look at the modal split of Appingedam (a town of 12000 people in the province of Groningen). These are the figures from OViN 2010-2013 for the number of journeys made per mode in Appingedam:
Car driver
Car passenger
Train
Bus/Tram/
Subway
Bicycle
Walking
Total
28%
14%
2%
1%
33%
22%
100%

But by kilometres travelled the numbers look like this:
Car driver
Car passenger
Train
Bus/Tram/
Subway
Bicycle
Walking
Total
53%
28%
4%
2%
9%
4%
100%





Because journeys by car, train, bus, tram, and subway are on average longer than journeys by bicycle or by walking, their share of the modal split becomes when figures are given per kilometre travelled rather than per journey.

Method
There are many methods to gather data to calculate a modal split. For example through counters on the street or through surveys. in the future there will perhaps be more use made from other sources of information, for example details from mobile telephones or from public transport subscriptions cards.

When there is no data available from local sources, use is often made of the OViN data (onderzoeks verplaatsinsgeddrag in Nederland - Dutch travel behaviour survey) or one of its predecessors (OVG, MON). This is a survey by the CBS using a representative sample of the Dutch population. The data is collected in such a way that modal split figures can be given. As an example, here are the CBS figures for Rotterdam:


Number of journeys per person per day

Car driver
Car passenger
Train
Bus/Tram/
Subway
Bicycle
Walking
Total
OViN
2010-2013
0,57
0,31
0,07
0,26
0,46
0,65
2,31
modal split
25%
13%
3%
11%
20%
28%
100%
MON
2004-2007
0,66
0,34
0,07
0,28
0,49
0,61
2,46
modal split
27%
14%
3%
12%
20%
25%
100%












From these figures we draw a conclusion that the modal split changes very little and that the number of cycling journeys has stayed the same.

The local government's own figures show a different picture. For 15 years, Rotterdam has counted cycle usage at fixed points in the city and found consistent growth in cycling while the car traffic in and around the inner city has stayed the same.

Figure: Development in car traffic around cordons (1986=100)
(Binnenkordon = inner city area)
source: Rotterdamse Mobiliteitsagenda 2015-2018
Figure: Cycling intensity around the fixed counting points (work days, saturday, sunday):
bron: Rotterdamse Mobiliteitsagenda 2015-2018
Rotterdam therefore concludes:

Cycling traffic around the city centre of Rotterdam has grown by around 60% in the last ten years.

According to the traffic counts, cycling has grown significantly in the inner city. However, we would draw a totally different conclusion from looking at figures from OViN for the entire council area:



Number of kilometres travelled per person per day

Car driver
Cra passenger
Train
Bus/Tram/
Subway
Bicycle
Walking
Total
OViN
2010-2013
10,23
5,17
2,97
2,07
1,87
1,08
23,39
modal split
44%
22%
13%
9%
8%
5%
100%
MON
2004-2007
11,38
5,87
3,14
2,66
1,68
0,78
25,52
modal split
45%
23%
12%
10%
7%
3%
100%












According to the figures for kilometres travelled (table above) cycling kilometres have grown by 11% while the growth for all transport modes is 7%. This could indicate that people are now more willing to travel for longer distances. When we look at the error margins (see below) we find that these differences are not significant: but there could still be growth in the numbers of journeys or distance being made by bicycle.

The smaller city of Delft also shows a good improvement. According to the CBS figures, the share of journeys per bicycle has grown by 6% between 2004-2007 and 2010-2013. The bicycle share is therefore now 39% in Delft, placing it amongst leading cities such as Groningen, Zwolle and Leeuwarden. Considering the high proportion of public transport (10%) this is a good result. The share of car usage fell bv 4% over the same period. What's more, the error margins from the CBS data are relatively good (13% for bicycle to 45% for bus and tram), and Delft's cordon counts appear to confirm this result:

Figure: Cordon counts for car traffic in Delft, index 2002=100
(Delft = whole city, Buitenring = outer ring, Binnen ring = inner ring)


Source: Gemeente Delft
Groups and periods
In the previous examples, not only are the methods used different (cordon counting vs. surveys) but also the groups of people counted differ. Rotterdam has counted how many people pass a cordon around the inner cty while the OViN figures measure the entire city. We also need to consider the lengths of bicycle journeys being made: if cycling distances double then the change of an individual cyclist being counted can also double. It's important to plan in advance what we wish to measure. Visitors or only residents ? Which year ? Which period (rush hour, morning rush, whole day ?)

These distinctions naturally produce different results. These are the 2004-2008 figures for modal split for all traffic in, from and to Appingedam:
Car driver
Car passenger
Public transport
Bicycle
Walking
Other
Total
38%
17%
2%
22%
19%
2%
100%


While these are the figures for nearly the same period (2004-2007) but just for residents of Appingedam::
Car driver
Car passenger
Train
Bus/ Tram/ Subway
Bicycle
Walking
Total
30%
18%
1%
0%
24%
27%
100%

That cars are used more in the first example than the second can be easily explained in that fewer people make the longer journeys between towns by walking and cycling.

Another distinction is in the presentation of the figures. The second example separated train travel from other public transport.

The following figure shows modal splits reported by the city of ’s-Hertogenbosch in 2014, showing modal splits for 2004-2006, for 2010-2011 and their ambition for 2015. Note that car passengers and drivers are combined and walking is omitted:

Auto=car passengers and drivers combined, OV=public transport, Fiets=cycling. Walking share omitted (it's around 20%)

Below are figures from a different methodology for a slightly different period:


's-Hertogenbosch modal split by journey

Car driver
Car passenger
Train
Bus/Tram/
Subway
Bicycle
Walking
Total
2010-2013
33%
16%
3%
1%
27%
20%
100%
2004-2007
34%
18%
3%
1%
25%
19%
100%








modal split by distance travelled

Car driver
Car passenger
Train
Bus/Tram/
Subway
Bicycle
Walking
Total
2010-2013
51%
24%
12%
1%
9%
3%
100%
2004-2007
51%
28%
11%
2%
7%
2%
100%








Not only is the period different but also the data sources differ. The tables make use of the new set of OViN data which used the same methodology as in previous years. Walking was missed out from the city's own figures. It's a significant part of the total and missing this out made cycling, public transport and driving seem more significant than the reality.

Note that totals again add up to 100%, even though the categories of "mopeds" and "others" are omitted. These are extremely small percentages and are often omitted because they affect the figures rather less than does omitting walking. The figure below shows the modal split for the entire country including mopeds (bromfietsen) at just 1% of the total:

Modal split for the whole of the Netherlands. Source: KiM, Mobiliteitsbalans 2013
auto=car, trein=train, fiets=cycling,lopen=walking,bromfiets=moped, overig=other


Error margins
To finish, we have error margins. Figures are often presented as if they represent an absolute truth, but that it very often not the case. Any sample or measurement will not only have some errors, but also they can never represent all journeys made. In sampling people typically aim for a 95% confidence interval. Such a confidence level indicates that we have a high degree of confidence that our result is relevant. It means that we are 95% certain that the true figure is within the intervals given. Note that is can say nothing about any individual sample nor can we say that 95% of samples are within the interval (the language of the Dutch original is slighty confusing).

Such an error margin also applies to the OVin figures. But how can you express that in modal split ? As an example again take the proportions of journeys per mode for Appingedam and take the minimum and maximum values according to the error margins:


Number of journeys per person per day - Appingedam

Car driver
Car passenger
Train
Bus/Tram/
Subway
Fiets
Walking
Total
OViN
2010 t/m 2013
0,72
0,35
0,04
0,01
0,85
0,56
2,54
Average modal split
28%
14%
2%
1%
33%
22%
100%
Relative error margin (95%)
25 %
36 %
139 %
340 %
24 %
34 %
14 %
All minima modal split
31%
13%
-1%
-2%
37%
21%
100%
All maxima modal split
27%
14%
3%
2%
31%
23%
100%
Car minima* modal split
18%
16%
3%
2%
35%
25%
100%



















*) In this table the modal split calculated for journeys by car is reduced by considering the effect of the error margins for other modes. There are many possible variations. This example is given to show the sensitivity of modal split data.

In Appingedam there are few journeys by public transport. This is also a small town and therefore the sample size is small. These factors result in a large margin of error. The relatively high error margin makes meaningful comparison with other towns difficult. For public transport it's impossible.

A large city like Amsterdam, which also has higher usage of public transport, gives us a larger sample and therefore much lower error margins for all modes. Therefore we have relatively trustworthy modal splits for this city:


Number of jouneys per person per day - Amsterdam

Car driver
Car passenger
Train
Bus/Tram/
Subway
Bicycle
Walking
Total
OViN
2010 t/m 2013
0,44
0,24
0,10
0,27
0,85
0,66
2,56
Average
modal split
17%
9%
4%
11%
33%
26%
100%
Relative error margin (95%)
7%
9%
11%
9%
5%
6%
3%
All minima
modal split
17%
9%
4%
10%
34%
26%
100%
All maxima modal split
17%
10%
4%
11%
33%
26%
100%
Car minima* modal split
15%
10%
4%
11%
33%
26%
100%




















*) Here the modal split is calculated for car journeys is reduced by adjusting all the other modes to their maximum values using the relative error margin. There are many possible variations as to how this could be done. the idea is to demonstrate the sensitivity of the figures to the error margins.


Under registration
Beside the inaccuracy due to sampling, there is also another source of error due to how correspondents to surveys under record their shorter journeys. For instance, people forget about the short walked journeys from the front door of their home to their car and between car parks and shops. From research is is become apparent that due to under-reporting figures for walked journeys should be 1.57 times higher than they are reported.

Conclusion
The examples given above demonstrate that there is no single modal split figure. Readers must always look at how the statistics were gathered and what they relate to. On the basis of the CBS figures shown above, we can answer the original question by saying that Appingedam residents make 18, 27, 28, 30, 31, or 38 % of their journeys as the driver of a car. Or by saying that 53% of the distance that Appingedammers travel is as the driver of a car.


Dutch data, Dutch modal split, my summary
All of the above relates to how the Dutch collect data about modal split in the Netherlands. As you see above, there can be significant variations in figures even from one methodology, but even here there are actually different methodologies which result in different results and those results can have different interpretations. As a result, different figures can be found for the same Dutch town, and this means it is difficult to reliably compare different towns. i.e. it's difficult to make sensible comparisons even within this country. However, at least within this country it is usually possible with a bit of effort to find data which has been collected in the same way in different places, though even when we are comparing Dutch cities with each other we must still be careful about error margins as they can be more significant than the apparent reported differences in modal splits.

What is good about (most) Dutch methodologies
Dutch methodologies as shown above (especially OViN) usually can be expected to take into account the modal split for the entire population being considered. By this I mean the entire town or city which is being considered and everyone of every age group within that location (group). They also usually take the entire year (i.e. winter and summer) into consideration (period). For instance, anonymous looking bicycle counters in the Netherlands are typically put in place for a year in order to gather counts which will not be unduly influenced by the weather on one day vs. another. These Dutch methodologies are used to build so accurate a picture as possible so that while the results cannot always be compared very meaningfully with other towns, they should be comparable over time within the same location. This makes it possible to build up a picture of progress over time.

Other statistics and why I criticize them
Unfortunately, statistics are sometimes gathered for entirely different reasons. For instance it's quite common to see published figures which are for commuters only. On some occasions this is because statistics are traditionally collected about commuters in that location but on other occasions these figures are used because it makes it possible to report higher and more impressive sounding figures than would be the case otherwise. The problem with counting only commuters is that only adults of working age (for whom subjective safety is relatively unimportant relative to children or older people) will be counted. This creates a false

Other ways in which figures can be collected so that they are not times you'll find figures quoted which relate only to journeys in part of a city. I've also seen examples of counts performed only on sunny days, or on peak days for student traffic, or where part day counts are extrapolated as if they relate to the whole day (all those things and more in one instance). Such figures are more applicable to use for politicians to placate local cyclists or to boast about what they claim to have achieved, or for marketing people to sell their city to cycling tourists.

Lastly, there's a tendency in some places to publish targets aggressively and it is sometimes the case that people go on to re-publish targets as if they have been met. Anyone can set a target. Achieving it is something quite different.

Whatever caused the confusion, presenting figures in a way which inflates them does not help to further understanding. It hides problems and does nothing to further cycling. Having figures from inconsistent methodologies or which do not cover the entire population makes it difficult to tell whether real progress is being made.

In the past I've criticized the use of modal split statistics gathered elsewhere and I'll probably continue to do so. For instance, I've written about how statistics from Cambridge, New York, London, and of course Copenhagen (multiple times) have been published and used in a way which confuses rather than enlightens.

Sadly, the Dutch don't always publish useful figures either. I've criticized figures from Amsterdam and Groningen in the past when walking was missed from their statistics because this has been used to present figures internationally which appear to show a higher cycling share than exists in reality. The figures produced by 's-Hertogenbosch used as examples in the article above would appear to be more of the same. They've been reported elsewhere as fact and indeed I once reproduced that claim myself.

We all need to be skeptical about claims, especially when they are made in a way which could be seen as promotional.

Why ranked comparisons of cities make no sense at all
Even modal-split data which appears to cover the same group and period in different countries, and which is gathered in a reliable and consistent manner will vary so much in methodology that comparisons made are largely meaningless. In reality, figures from different countries vary greatly in group and period and this makes the comparisons completely unreliable. As result, any ranking produced on the basis of the results of such a comparison will be meaningless.

While I have on occasion quoted figures from Dutch or other cities, you won't ever read a "top ten cycling cities" list on this blog. I've never ranked cities in this way because it wouldn't mean anything. Such a ranking this could only create a false impression of having more information than actually exists and this would mislead readers. People who produce lists of "top cycling cities" do so either out of ignorance about how these statistics cannot be compared or for commercial reasons. i.e. to sell something.

How much cycling is there in the Netherlands ?
The overall figure for cycling in the Netherlands as a percentage of trips has not varied appreciably in many years. According to the graph above, 27% of all trips in the Netherlands are by bicycle. This is the highest figure for any country in the world (certainly amongst relatively wealthy nations where people have a choice of transport modes). If we're interested in green modes of transport, we can add on the 16% of journeys made by walking and note that relatively prosperous Dutch people make a massive 43% of their journeys by human power.

Using a different methodology, the Flash Eurobarometer came to a similar conclusion about the overal cycling and walking share of the Netherlands, left, in orange.
This is great news, however we should always also recognise the flip side. If 27% of journeys are by bike and 43% of journeys are by genuinely green modes, that still means that 73% of journeys are not by bicycle and 57% are by motorized forms of transport. The graphs above show that the Dutch cover 73% of the distance that they travel by private car.

While what has been achieved in the Netherlands is wonderful, there is still more that can be achieved. What's more, what has so painstakingly been achieved could still quite easily be lost. There is no space for complacency.