Sage Journals: Discover world-class research

Abstract

Chinese

In German cities, higher levels of education increase people’s propensity to cycle. However, it remains unknown whether this effect is restricted to certain contexts, such as cities with low or medium cycling rates, or whether it is a more universal occurrence. This paper develops and tests competing hypotheses on how the effect of education on cycling might depend on the overall cycling level: (a) educational inequalities in cycling could increase proportionally with the overall cycling level or (b) such inequalities might diminish in high-cycling cities because their advanced pro-cycling mobility cultures encourage cycling among all social groups. I analyse about 150,000 trips made by about 50,000 residents from 143 cities in the Netherlands and Germany using multilevel regression models. Results fall in between the competing hypotheses, meaning that the effect of education is similarly large in cities with low, medium, or high overall levels of cycling. Hence, there is no automatism in the sense that higher cycling shares in general will also imply greater cycling equity.

Keywords

cycling cycling equity mobility behaviour sustainability behaviour transportation urban mobility culture

Introduction

Cycling is increasing in many cities around the world and policy-makers greet this with enthusiasm (Buehler and Pucher, 2021; Hudde, 2022a; Lanzendorf and Busch-Geertsema, 2014). Cycling contributes to environmentally sustainable, liveable, and attractive cities with low noise-levels and clean air (Gehl, 2013; Gössling et al., 2016). Further, cycling is cheap, fast in urban settings, and good for physical and mental health (Götschi et al., 2016; Martin et al., 2014; Tranter, 2012; World Health Organization, 2019).

People with lower education might particularly benefit from this means of transport, as they have, on average, lower incomes and poorer health (e.g. Hendi, 2017; Zajacova et al., 2021). However, in several places of the Western world, those with lower education cycle less, and this gap is increasing (e.g. Buehler et al., 2020; Hudde, 2022a). Thus, an already underprivileged group benefits less from the advantages of cycling. This is an aspect of a lack of cycling equity and thereby transport equity in general (Litman, 2002). Policy makers consider educational differences in cycling behaviour as a mobility equity issue that requires targeted measures (BALM, 2023).

Social inequalities in cycling vary by geographical and historical context (Horton et al., 2016). For instance, in the United States and several European countries, cycling evolved from a bourgeois leisure activity in the 19th century towards a working-class means of transport in much of the 20th century (Horton et al., 2016; Oosterhuis, 2016). In post-World-War-II Britain and Germany cycling even turned into a stigmatised low-status activity for those who cannot afford a car and cycle out of necessity (Oosterhuis, 2016). A partly similar trajectory occurred in China, where the bicycle evolved from a luxury good of the upper class towards a mass means of transport. In contemporary Chinese cities, cyclists mainly have medium or low levels of income and education (Zhang et al., 2014). In Ghana, cycling is mainly limited to people with low levels of education who cannot afford to own a car (Acheampong and Siiba, 2018).

In several high-income cities and countries today, people with higher education, especially university degrees, cycle more (Buehler et al., 2020; Carse et al., 2013; Hudde, 2022a; Kroesen and Handy, 2014; Rachele et al., 2015). Evidence from German cities shows that the education–cycling link is not just spurious but prevails after statistically controlling for potential confounders or factors that shape the ease with which people could choose to cycle, such as trip distance, household situation, or city and neighbourhood characteristics (Hudde, 2022b).

However, it remains unclear to what degree the effect of education on cycling is a general occurrence in cities in Germany and other Western European countries today or to what degree it depends on the national and local context. Cycling rates vary dramatically between cities, even within the same region and country, and it may also be the case that educational differences vary between cities. For example, it could be that educational differences in cycling are larger in medium-cycling cities than in high-cycling cities. The large differences in medium-cycling cities could occur, for instance, if the cycling infrastructure is mainly restricted to more privileged neighbourhoods and if cycling is symbolically charged and thereby strongly related to social status. On the contrary, educational differences could be smaller in high-cycling contexts if they have quality cycling infrastructure spread out across all neighbourhoods, and if cycling is not symbolically charged but simply the normal thing to do for everyone.

This paper develops competing hypotheses on how the effect of education on cycling differs by the overall, macro-level cycling rate and tests them using information from around 150,000 short-to-medium distance trips made by around 50,000 individuals, from 143 cities in Germany and the Netherlands. Among Western countries, the Netherlands has by far the highest cycling rate with more than a quarter of all trips made by bicycle. In Germany, around one in ten trips are made by bicycle – much lower than in the Netherlands but still higher than in most other Western countries (Buehler and Pucher, 2021). However, in both countries, and Germany in particular, there is substantial variation in cycling behaviour between cities (Buehler and Pucher, 2021; Nobis, 2019).

Regression models analyse the interaction effect between macro-level cycling propensities and individual-level education on whether a certain trip is taken with the bicycle or another means of transport. The paper presents three sets of analyses. First are national-level analyses, comparing the effect of education on cycling in Germany and the Netherlands. Second are models that test whether the effect of education on cycling is dependent on the cities’ overall cycling level in Germany, the Netherlands, and both countries combined. Third and final are estimations of city-specific effects of education.

Theoretical considerations and previous research

The effect of education on cycling behaviour could differ between cities, depending on their mobility culture. Mobility culture is a holistic concept that includes all mobility-related social practices, norms, and discourses as well as urban planning and infrastructures (Götz and Deffner, 2009; Götz et al., 2016; Klinger and Lanzendorf, 2016; Klinger et al., 2013). Mobility cultures differ between countries and also between cities in the same country (Klinger and Lanzendorf, 2016; Klinger et al., 2013; Kuhnimhof and Wulfhorst, 2013; Tyndall, 2022).¹

Education may affect cycling behaviour through two groups of pathways. First are indirect effects, where education affects the probability of being in cycling-friendly circumstances, which then affects cycling behaviour (Elldér et al., 2022; Millard-Ball et al., 2022). For instance, people with higher education might more often live and work in cycling-friendly cities and near the city-centre. Second are more direct effects: even under otherwise similar circumstances, those with lower education are more hesitant to choose the bicycle. This could occur, for example, due to the symbolic meaning and social status associated with cycling, differences in health or environment attitudes, or because of differences in socialisation and habits or internalised norms resulting from that (Haustein et al., 2020; Hudde, 2022b). Elements of both these pathways could depend on cities’ mobility culture. That is, mobility culture and cities’ overall cycling levels might moderate the effect of education on cycling.

First pathway: Mobility culture and the distribution of cycling infrastructure. Whether and in which neighbourhood cities build cycling-friendly infrastructure is the result of political decisions and urban planning. Therefore, infrastructure has been described as a result of mobility culture, as a ‘physical extension of cultural priorities’ (Klinger and Lanzendorf, 2016: 246; see also Haefeli, 2005; Stone, 2009). A city’s mobility culture may thereby determine whether cycling-friendly infrastructure is distributed equitably across all neighbourhoods or concentrated in just some neighbourhoods, potentially those with a more highly educated population.

Previous research showed that cycling infrastructure may be more developed in areas where more highly educated people live (Böcker et al., 2020; Braun et al., 2019; Flanagan et al., 2016; Hirsch et al., 2017). It could be that cities with an advanced cycling culture have extended high-quality cycling infrastructure throughout the whole city, meaning that people also living in less privileged neighbourhoods have good conditions for cycling. However, access to cycling-infrastructure only explains a minor part of the effect of education on cycling. In Germany, the larger share of the effect of education remains when comparing people living in the same city and after accounting for the cycling-friendliness of the neighbourhood (Hudde, 2022b).

Second pathway: Mobility culture and the social meaning of cycling. A certain transport mode is more than just a tool to get from one place to another; it has a deeper underlying social meaning (Green et al., 2012; Guell et al., 2012). Mobility culture determines mobility-related social practices, norms, and discourses, and thereby also the symbolic meanings of cycling and other transport modes. Previous research on German cities shows that those with higher education tend to drive less and cycle more (Hudde, 2022b) – and these two modes of transport have strong symbolic meanings that are related to social status. Bicycles can symbolise pro-environmentalism and health-orientation, but also a lack of financial resources, that one has not ‘made it’ (Anantharaman, 2017; Frater and Kingham, 2020; Hudde, 2022b; Steinbach et al., 2011). On average, people with higher education face less danger that their surrounding sees them as poor or unsuccessful, that is, they suffer less from status anxiety (de Botton, 2004; Delhey et al., 2017). Without such status anxiety, they can profit from the fact that cycling is associated with environmental and health consciousness. People with lower education, on the contrary, might avoid the bicycle because it might signify that they have not ‘made it’ (or at least riding a bicycle may be seen as a missed opportunity to drive an expensive car and symbolise) (Hudde, 2022b).

Qualitative studies have found that the cycling–social status association is strong in some contexts where cycling rates are relatively low, such as London, Bangalore, and New Zealand (Anantharaman, 2017; Frater and Kingham, 2020; Steinbach et al., 2011). In countries or cities with very high cycling rates, cycling might simply be the normal thing to do. In the hypothetical extreme case where cycling is completely normalised, it is not symbolically charged, neither positively (environment, health) nor negatively (lack of money). Therefore travelling by bicycle would not just be for people who identify as ‘cyclists’, but for everyone, irrespective of a specific identity, political standpoint, gender, age, or social class (Aldred, 2015; Frater and Kingham, 2020; Haas and Hamersma, 2020; Heinen et al., 2010). Among Western countries, the Netherlands has the highest cycling rate (Buehler and Pucher, 2021), and qualitative studies report higher levels of normalisation than in other contexts (Aldred, 2015; Frater and Kingham, 2020; Pojani et al., 2017). However, reports from Amsterdam – probably the Western metropolis with the highest cycling rate (Buehler and Pucher, 2021) – cast doubt on the idea that bicycles will lose all their status-related symbolic meaning when cycling rates are high. Even in Amsterdam, cars remain a status symbol and the bicycle is associated with stigma among people with a lower social status (Buehler and Pucher, 2010).

Two competing hypotheses

Based on the theoretical considerations and previous research, I discuss two competing hypotheses for the link between a city’s overall cycling level and educational differences in cycling, visualised in Figure 1.

Figure 1.

Illustration of the two competing hypotheses.

The first hypothesis is that the effect of education on cycling is positively and proportionally associated with the overall cycling level. The higher the overall cycling level, the larger the gap between the educational groups. For instance, suppose a country where those with higher education have a 50% higher probability of travelling by bicycle. With a proportional effect, this 50% higher probability would show at all city-specific cycling levels. For a city where half the population has a tertiary degree and where the overall cycling level is 12.5%, this would mean that those with lower education travel by bicycle for 10% of trips and those with higher education for 15% of trips. In a city where the overall cycling level is double, this would mean that each group’s level is also doubled and that those with lower education cycle 20% of trips and those with higher education 30% of trips.

The second hypothesis is that the effect of education on cycling is diminishing at higher values of overall cycling levels. For instance, the association could be curvilinear, inverted U-shaped (as displayed in Figure 1). In low-cycling cities, people with high and low levels of education are similar and cycle rarely. In medium-cycling cities, people with high and low levels of education are dissimilar whereby those with high education cycle frequently but those with low education cycle relatively rarely. In high-cycling cities, people with high and low education are relatively similar again, and both cycle frequently.

If cycling infrastructure was (a) more comprehensive and equitably distributed and, most of all, (b) more normalised and less symbolically charged in the high-cycling cities, it would contribute to a pattern similar to the one displayed in the right-hand panel of Figure 1: educational differences in cycling diminish in cities with very high overall levels of cycling. If the contrary was the case and cycling infrastructure was inequitably distributed and bicycles charged with status-related symbolic meaning also in high-cycling cities, the effect of education should be proportional to the overall cycling rate.

Applied to the context of the Netherlands and Germany, where cycling rates mainly range between medium-low and high, the central observable difference to test the competing hypotheses lies in the comparison of medium- and high-cycling cities. The empirical analyses will test the observed association between the overall cycling rate and the effect of education, but they cannot disentangle the two proposed pathways of infrastructure and social meaning.

The question of which of these hypotheses is empirically supported is relevant for policy. Supporting the second hypothesis would mean that cities with high cycling mode shares also achieve cycling equity, and that cities with lower cycling mode shares could simply follow their lead to achieve both high overall cycling mode shares and cycling equity. However, if the first hypothesis is true, it means that social inequalities in cycling are largest in high-cycling cities. The greater the overall cycling level of a city, the greater also the difference in cycling behaviour by education group. This would mean that to achieve cycling equity, cities need cycling policies that are more targeted towards the less privileged than is currently the case in most high-cycling cities.

Data, methods, and analytical strategy

Data sources

I harmonise data from Dutch and German national travel surveys. This has two advantages over single-country analysis. First, it allows analyses at the country level: how do educational gaps differ between the medium-cycling case of Germany and the high-cycling case of the Netherlands? Second, for analyses at the city-level, it increases the range of observed cycling levels. The data analysed contain detailed information on all trips that individuals made on a specific reference day and city-specific information on people’s place of residence. Both surveys are conducted about evenly throughout the year. Data on Germany comes from the large-scale, nationally representative study Mobility in Germany 2017 (Nobis and Kuhnimhof, 2018). Data on the Netherlands comes from the 2018 and 2019 waves of the Dutch national travel surveys, the Onderzoek Verplaatsingen in Nederland (Centraal Bureau voor de Statistiek (CBS) & Rijkswaterstaat (RWS), 2018, 2020). Two survey waves are used to roughly match the sample size of the German survey data and the 2020 data are avoided because patterns in this year may be distorted by COVID-19 effects.

Sample selection

The sample covers all trips reported by adults who live in medium-sized and large cities (≥50,000 inhabitants). Cycling patterns strongly differ between rural and urban areas, and it is mainly cities that have recently seen relevant increases in cycling (Hudde, 2022a). Further, sample sizes for smaller towns or rural areas are small and estimates of local cycling levels for these areas would be noisy. By definition, respondents who took no trips during the reference day are not part of the analytical sample. In line with previous research on educational differences in cycling, the sample is restricted to people of typical working age (age range 25–64, to avoid results being excessively driven by people still in education or by educational health-differences among the elderly) and to trip distances between 0.5 and 7.5 km (the range most suitable for cycling, Malone, 2016; this distance group covers 52.8% of Dutch trips and 59.5% of German trips). Business trips are excluded because, for these, people usually cannot choose the mode of transport themselves (15.1% of trips in the German subsample and 4.5% in the Dutch subsample). People with missing or invalid information on education, household composition, or trip details are dropped from the analyses (Germany: 2.2%; Netherlands: 4.3%). In the last step, to avoid excessively noisy estimates at the city-level, the sample is restricted to cities with at least 250 observed trips (this reduces the German sample by 11.4% and the Dutch sample by 2.6%).

The final analytical sample covers 148,683 trips reported by 51,411 individuals from 143 cities (Germany: 88,298 trips of 31,156 individuals from 68 cities; Netherlands: 60,385 trips of 20,255 individuals from 75 cities).

Analytical approach and methods

All regression models predict whether the means of transport for a certain trip was the bicycle (=1) or any other means of transport (=0).

Linear probability models (LPMs) are chosen because they are generally more robust than logistic regressions, especially for testing interaction terms, and the coefficients are easier to interpret (Battey et al., 2019; Hellevik, 2009; Mood, 2010). A disadvantage of LPMs is that these models can produce predictions that are below 0% or above 100%. However, the advantages of LPMs weigh particularly strong for the analyses here, which focus on estimating interaction terms (Mood, 2010). In both data sets, certain cities are oversampled and, therefore, the models include survey weights at the city-level. The data have a hierarchical structure – trips are nested in individuals who are nested in cities – so multilevel regression models with random effects at the individual and city level are chosen.

Before the city-level analyses, I compare the effect of education on cycling between the two countries. I run regression models with all control variables (as described below) and an interaction term between the level of education and the country (Netherlands versus Germany). I then calculate the predicted bicycling probabilities (margins) for each country and educational group at the global averages for all control variables.

The main analyses test how the effect of high education on cycling depends on the overall cycling rate in the specific city. The central predictor is the interaction term between individuals’ level of education and linear and square term of the general city-level cycling rate. These analyses proceed in two steps.

I estimate the overall cycling level for each city. Intuitively, this is the average between the cycling-probabilities of people with and without tertiary education, for an average trip. To estimate these values, I run regression models that include all control variables, and additionally dummy variables for each city. I then calculate the predicted cycling probabilities (margins) for each city and the two educational groups, at the global averages for all control variables. Finally, I calculate the average between the values of the educational groups. The advantage of this procedure over simply comparing the average bicycling rates across cities is that it adjusts for compositional differences across cities (e.g., some cities might have higher bicycling rates because they have younger or more highly educated inhabitants).

I run the main multilevel regression models with interactions between individual-level education and the linear and square term of the overall cycling level of the city. This step is done (a) for both countries combined and (b) for both countries separately, in order to test whether patterns differ between Germany and the Netherlands. To test whether the effect of education is proportional to the overall cycling level (hypothesis 1) or diminishing at higher values of overall cycling (hypothesis 2), the figures plot the proportional effect as a reference line. This proportional effect is the relative difference between the predicted probabilities of cycling for people with and without tertiary education at the average value of all control variables.

As final analyses, I calculate the effect of education on cycling for each city individually. This shows how cities vary around the overall trend. I run regression models with all control variables and an interaction term between individual-level education and the dummy-variables for each city. Then, I compute the marginal effect of education for each city.

Measures

Dependent variable: dummy for cycling-trip. The outcome variable is whether the bicycle was the main means of transport for a specific trip (=1) or whether it was any other means of travel (=0).

Main predictor: tertiary education. This variable indicates whether an individual has a tertiary degree (‘higher education’) or not (‘lower education’). Such a binary measure allows for a clear and intuitive display and interpretation of results. It is further chosen because data on the educational level of respondents is limited in the travel surveys and because more complex measures would make harmonisation across the countries difficult. Systems of education for universities and universities of applied sciences (i.e., institutions oriented less towards fundamental research and more towards applied research and practice) are comparable across Germany and the Netherlands. Other forms of education, such as secondary schooling and vocational training, differ more across countries (see the Online Supplement for more details including analyses of all educational levels).

In the analytical sample, 55.1% of German respondents and 55.6% of Dutch respondents have a tertiary degree, which includes degrees from universities and universities of applied sciences. The restriction to residents from medium-sized and large cities results in an analytical sample with a higher level of education than the general Dutch and German population.

Covariates at the trip level: distance, purpose, and timing. To allow for a flexible function of the distance–cycling association, I use categories for each kilometre, except for the lowest and highest categories, which are 0.5–1 km and 6–7.5 km. The trip purpose is divided into three groups: (1) work and education, (2) shopping, errands, etc., including accompanying trips, and (3) leisure. The time variable is a categorical variable that distinguishes between four categories of trip start time (early morning, 5:00–7:59; daytime, 8:00–18:59; evening, 19:00–21:59; night, 22:00–4:59).

Covariates at the individual and household level: gender, age, activity status, household income, and household composition. Age is coded with a category for each five-year interval. Activity status is coded with five categories: working, student/in education, homemaker, retired, and ‘other’, which includes unemployment. Both countries’ data sets include pre-defined variables for household equivalence income. The Dutch variable has 10 categories, which are grouped into five categories to match the German variable. A variable with four levels captures the household context (living alone; living with one or more other adults; living with younger children; living with older children).

All these covariates enter the model with dummy variables (n–1 dummy variables for a variable with n categories).

Results

The effect of education in Germany and the Netherlands

In regression models with interaction and non-linear terms, single coefficients are hard to interpret. Therefore, the results section shows plots with predicted probabilities and marginal effects (marginsplots) and the full regression tables are provided in the Online Supplement.

Figure 2 plots the results from the country-level analyses. Those without tertiary education in German cities take a bicycle for a certain trip with a model-predicted probability of 14.7% and those with tertiary education with a probability of 22.5%. The difference between the groups is 7.8 percentage points (22.5% minus 14.7%) or 53.1% (22.5% divided by 14.7%). In the Netherlands, cycling rates are much higher and those without tertiary education take a bicycle for a predicted 30.2% of trips whereas those with a tertiary degree take it for 38.5% of trips. In absolute terms, the educational difference is even slightly larger in the Netherlands (8.3 percentage points), but it is considerably smaller in relative terms (27.5%), because both educational groups cycle more in the Netherlands. These findings show that educational differences in cycling behaviour exist, even in the country that has the highest cycling rates and presumably the highest degree of cycling normalisation among Western countries (Buehler and Pucher, 2021).

Figure 2.

Predicted probabilities of cycling by education and country. 95%-confidence intervals are plotted.

The effect of education by cities’ overall cycling level

The bottom panels of Figure 3 show the distribution of respondents by the cycling levels of the cities they live in, for both countries combined and for each country separately. The top panels show the model-predicted cycling probabilities of the two educational groups and the middle panels plot the effect of education on the predicted cycling probability, which equals the difference between the two curves in the top panels.

Figure 3.

Top panels: predicted probabilities that a trip is taken by bicycle. Middle panels: the effect of education on the probability of cycling. Bottom panels: density plots of cities’ overall cycling levels. The ‘overall level of cycling’ represents the probability that an average trip is made by bicycle, averaged over people with and without tertiary education. The dashed lines in the middle panels represent what the curve would look like if the effect of education was proportional to the overall cycling rates.

For both countries combined, we see that those with higher education are more likely to cycle at all levels of cities’ general cycling level. The middle left panel shows that the effect of education is slightly positively associated with the overall cycling-level in the city. In the middle panels, the dashed lines serve as reference lines and represent what the curve would look like if the effect of education was proportional to the overall cycling rates. If the effect curve was less steep or inverted U-shape, it would suggest that cycling was relatively less education-dependent in high-cycling cities. Indeed, the estimated effect-curve is somewhat less steep than the reference line. At lower (higher) overall cycling levels, the estimated effect of education is greater (smaller) than proportionality would predict. However, the confidence interval mainly includes the reference line, especially for higher overall cycling rates.

Next, let us look at both countries separately. For Germany, the model estimates a positive effect of tertiary education on the cycling probability at all levels of the overall cycling rate. The predicted effect of education at the German average level of cycling level is plus 8.0 percentage points. At one standard deviation below the overall cycling level, the predicted effect is plus 6.3 percentage points (difference to the effect at the average: p ≤ 0.05). At one standard deviation higher than the average, the predicted effect is plus 8.1 percentage points (difference to the effect at the average: N.S.). There is a slight hint towards a curvilinear association, meaning that educational differences tend to be biggest in cities with medium cycling rates. At relatively high cycling levels, the estimated effect-curve differs from the proportionality reference line. However, as the histogram in the lower panel shows, there are relatively few observations from cities with such high cycling rates.

In the Netherlands, there is a positive association between the cycling-level in the city and the effect of education. The effect of education is plus 8.1 percentage points at the Dutch average level of overall cycling level, plus 2.8 percentage points (difference to the effect at the average: p ≤ 0.05) at one standard deviation below, and 11.4 percentage points (difference to the effect at the average: p ≤ 0.1) at one standard deviation above. This shows that, in the Netherlands, there is some evidence that the effect of education is larger in cities with a higher overall cycling rate. Indeed, the estimated effect-curve looks very similar to the proportionality reference line.

The effect of education on cycling, estimated for cities separately

Figure 4 shows the effect of education on cycling for each city individually, showing how cities vary around the overall trend.² To avoid very noisy estimates, only cities with more than 1000 observations in the data are plotted. The size of the dot is proportional to the number of observations: the larger the dot, the more precise the estimation.

Figure 4.

Effect of tertiary education on the probability that a certain trip is taken by bicycle, estimated for each city separately. Only cities with ≥1000 observations at the trip level are displayed. Dutch cities are displayed with orange (lighter) dots, and German cities with grey (darker) dots. The size of the dots is proportional to the number of observations from that city. The ‘overall level of cycling’ represents the probability that an average trip is made by bicycle, averaged over people with and without tertiary education.

The figure shows substantial variation around the general trend. Among cities with the same overall cycling level, the effect of education is small in some, and large in other cities. This variation is particularly large in the Netherlands. In some high-cycling cities, like Groningen or Haarlem, the effect of education is moderate, at 10 percentage points or lower. In Utrecht, the effect of education is estimated at 17.2 percentage points and in Amsterdam the effect is even at 25.4 percentage points – the highest value of all cities.

In sum, it shows that (1) cycling is positively affected by level of education in almost all cities, (2) the effect is small in some and very large in other cities, but (3) this between-city variation is largely not a function of the cities’ overall cycling levels.

Discussion

People’s level of education affects their cycling propensity and this paper examined whether this effect differs, depending on the overall cycling levels of the cities they live in. Building on the concept of mobility culture, I developed two main pathways through which the overall cycling rate might affect educational differences. First, high-cycling cities with an advanced pro-cycling mobility culture could have extended quality cycling infrastructure across the city, including less privileged neighbourhoods. Second, cycling may have become so normalised in high-cycling cities, that norms, discourses, or status-related symbolic meanings no longer pose a barrier to those with lower levels of education. If one or both of these considerations were true to a substantial extent, the effect of education on cycling should be curvilinear: smaller in high-cycling cities than in medium-cycling cities. Otherwise – if cycling infrastructure was inequitably distributed and bicycles charged with status-related symbolic meaning also in high-cycling cities – the effect of education should increase with the overall cycling rate. Hence, the competing hypotheses predicted either an effect that is proportional to the overall cycling rate, or an effect that diminishes at high levels of overall cycling.

Results at the country-level showed that the effect of education in Germany and in the Netherlands is about equally large in absolute terms, but smaller in the Netherlands in relative terms. This shows that educational differences in cycling behaviour exist even in the Netherlands, the country that has the highest cycling rates and presumably the highest degree of cycling normalisation among Western countries (Buehler and Pucher, 2021).

City-level results differ when studying the countries separately. For German cities, results are weakly in line with the hypothesis that cycling becomes more normalised in high-cycling cities, whereby the effect of education is smaller in high than in medium-cycling cities. However, the decrease in the effect of education is not statistically significant and predicted only for very high levels of cycling, which are only reached in a small number of cities. Contrary to that, results from the Netherlands are more in line with the proportionality-hypothesis which predicts the greatest effects of education in those cities with the highest cycling rates. Consequently, when both countries are studied together, results fall in between the two hypotheses. Estimations of city-specific educational effects reveal remarkable variation around the overall trend. The greatest effect of education is estimated for Amsterdam, one of the metropolitan cities with the highest overall cycling rates in the world.

The analyses presented here have limitations. First are two types of data limitations. With almost 150,000 trips from around 50,000 individuals, the sample is very large at the individual level. However, for multilevel analyses, the number of clusters also matters for precision, and there are ‘only’ 143 cities with relatively few observations in some of these cities. This leads to relatively large standard errors and confidence intervals shown in Figure 3. Further, the data are cross-sectional and allow testing whether the effect of education differs between cities with different overall cycling levels. With large-scale longitudinal data, one could test within-city changes and whether the effect of education increases or decreases when the cities’ overall cycling level changes. Further, the data lack some variables that could be relevant. For instance, there is no valid information on migration background or ethnicity in the German data (cf. Haustein et al., 2020). Finally, these analyses do not distinguish between conventional and electric bicycles. The share of electric bicycles continues to rise, which might also affect social inequalities in cycling.

The city-level analyses revealed a puzzle that calls for future research: Some cities have much higher cycling inequality than others, but the overall cycling level and the theoretical hypotheses developed here cannot explain this. The data do not point to any obvious or general explanatory patterns. Of course, some of this heterogeneity will be a simple matter of randomness. This mainly concerns the little dots in Figure 4, that is, those cities with estimates based on comparatively small samples (cf. Figure S1 in the Online Supplement). However, there are also some larger outlier-dots, where estimates are based on fairly large samples.

Patterns among Dutch cities could indicate greater educational differences in larger cities. The estimated effect of education is particularly large for the three biggest Dutch cities, Amsterdam, Rotterdam, and The Hague (based on more than 1500 respondents per city). However, there is no similar pattern in Germany. Educational differences in the largest cities, Berlin, Hamburg, Munich, and Cologne, are around the national average. Whether we consider size, economic structure, cityscape, or population composition, it seems difficult to find a characteristic that Amsterdam, Rotterdam, and The Hague have in common that is not also shared by Berlin, Hamburg, Munich, or Cologne. In Germany, the estimated effect is particularly large for Karlsruhe and Heidelberg (based on around 400 respondents each). These two cities have much in common – they are affluent, mid-sized university towns in southern Germany – but the same description also applies to, for example, Freiburg and Erlangen, where educational differences are average. Overall, this intriguing heterogeneity calls for more in-depth research, such as comparative case studies of selected cities. Such analyses could dig deeper and inspect a complex interplay of various factors.

Cycling is environmentally friendly, low-cost, and healthy. In short: it is beneficial to individuals and society as a whole. However, these benefits remain underexploited and unevenly distributed across social strata if cycling is much more prevalent among those with higher levels of education. Further, wide public support for the pro-cycling policies that are currently being implemented in cities around the world will only be reached if all social strata profit from it (Buehler and Pucher, 2021; Lanzendorf and Busch-Geertsema, 2014).

A central finding from this study is that education affects cycling in almost all Dutch and German cities, including high-cycling cities. This means that there is no automatism in the sense that higher cycling shares in general will also imply greater cycling equity. Several cities – such as Freiburg and Erlangen in Germany or Amsterdam and Utrecht in the Netherlands – have achieved high overall cycling rates, but they have not achieved cycling equity. Hence, cities that want to reap the benefits of cycling and achieve cycling equity, where people from all educational levels are pedalling alike, will need to do more than just follow the suit of Freiburg, Erlangen, Amsterdam, or Utrecht.

Supplemental Material

sj-docx-1-usj-10.1177_00420980231172313 – Supplemental material for Have cycling-friendly cities achieved cycling equity? Analyses of the educational gradient in cycling in Dutch and German cities

Supplemental material, sj-docx-1-usj-10.1177_00420980231172313 for Have cycling-friendly cities achieved cycling equity? Analyses of the educational gradient in cycling in Dutch and German cities by Ansgar Hudde in Urban Studies

Footnotes

Author note

All replication code is available here:

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research,authorship,and/or publication of this article.

Funding

The author(s) received no financial support for the research,authorship,and/or publication of this article.

ORCID iD

Ansgar Hudde

Supplemental material

Supplemental material for this article is available online.

References

Acheampong

Siiba

(2018) Examining the determinants of utility bicycling using a socio-ecological framework: An exploratory study of the Tamale Metropolis in Northern Ghana. Journal of Transport Geography 69: 1–10.

Aldred

(2015) A matter of utility? Rationalising cycling, cycling rationalities. Mobilities 10(5): 686–705.

Anantharaman

(2017) Elite and ethical: The defensive distinctions of middle-class bicycling in Bangalore, India. Journal of Consumer Culture 17(3): 864–886.

BALM (2023) Förderaufruf im Förderprogramm “Nicht investive Modellvorhaben Radverkehr”. Bundesamt für Logistik und Mobilität. Available at: https://www.balm.bund.de/DE/Foerderprogramme/Radverkehr/NichInvestiveMassnahmen/Foerderaufruf/foerderaufruf_node.html (accessed 9 February 2023).

Battey

Cox

Jackson

(2019) On the linear in probability model for binary data. Royal Society Open Science 6(5): 190067.

Böcker

Anderson

Uteng

, et al. (2020) Bike sharing use in conjunction to public transport: Exploring spatiotemporal, age and gender dimensions in Oslo, Norway. Transportation Research Part A Policy and Practice 138: 389–401.

Braun

Rodriguez

Gordon-Larsen

(2019) Social (in)equity in access to cycling infrastructure: Cross-sectional associations between bike lanes and area-level sociodemographic characteristics in 22 large U.S. cities. Journal of Transport Geography 80: 102544.

Buehler

Pucher

(2010) Cycling to Sustainability n Amsterdam. Sustain 21.

Buehler

Pucher

(2021) International overview of cycling. In: Buehler

Pucher

Cycling for Sustainable Cities. Cambridge, MA: The MIT Press, pp.11–34.

10.

Buehler

Pucher

Bauman

(2020) Physical activity from walking and cycling for daily travel in the United States, 2001–2017: Demographic, socioeconomic, and geographic variation. Journal of Transport & Health 16: 100811.

11.

Carse

Goodman

Mackett

, et al. (2013) The factors influencing car use in a cycle-friendly city: The case of Cambridge. Journal of Transport Geography 28: 67–74.

12.

Centraal Bureau voor de Statistiek (CBS) Rijkswaterstaat (RWS) (2018) Onderzoek Onderweg in Nederland - ODiN 2018 [Data Set]. The Hague: DANS

13.

Centraal Bureau voor de Statistiek (CBS) Rijkswaterstaat (RWS) (2020) Onderzoek Onderweg in Nederland - ODiN 2019 [Data Set]. The Hague: DANS

14.

de Botton

(2004) Status Anxiety. London: Penguin Books.

15.

Delhey

Schneickert

Steckermeier

(2017) Sociocultural inequalities and status anxiety: Redirecting the Spirit Level Theory. International Journal of Comparative Sociology 58(3): 215–240.

16.

Elldér

Haugen

Vilhelmson

(2022) When local access matters: A detailed analysis of place, neighbourhood amenities and travel choice. Urban Studies 59(1): 120–139.

17.

Flanagan

Lachapelle

El-Geneidy

(2016) Riding tandem: Does cycling infrastructure investment mirror gentrification and privilege in Portland, OR and Chicago, IL? Research in Transportation Economics 60: 14–24.

18.

Frater

Kingham

(2020) Adolescents and bicycling to school: Does behaviour setting/place make a difference? Journal of Transport Geography 85: 102724.

19.

Gehl

(2013) Cities for People. Washington, DC: Island Press.

20.

Gössling

Schröder

Späth

, et al. (2016) Urban space distribution and sustainable transport. Transport Reviews 36(5): 659–679.

21.

Götschi

Garrard

Giles-Corti

(2016) Cycling as a part of daily life: A review of health perspectives. Transport Reviews 36(1): 45–71.

22.

Götz

Deffner

(2009) Eine neue Mobilitätskultur in der Stadt–praktische Schritte zur Veränderung. In: Verkehr

Bundesministerium Für

Stadtentwicklung

Bau-und

(ed.) Urbane Mobilität. Verkehrsforschung Des Bundes Für Die Kommunale Praxis, Direkt, vol. 65. NW-Verlag: Bremerhaven, pp.39–52.

23.

Götz

Deffner

Klinger

(2016) Mobilitätsstile und Mobilitätskulturen–Erklärungspotentiale, Rezeption und Kritik. In: Schöller

Canzler

Knie

Handbuch Verkehrspolitik. Wiesbaden: Springer, pp.781–804.

24.

Green

Steinbach

Datta

(2012) The travelling citizen: Emergent discourses of moral mobility in a study of cycling in London. Sociology 46(2): 272–289.

25.

Guell

Panter

Jones

, et al. (2012) Towards a differentiated understanding of active travel behaviour: Using social theory to explore everyday commuting. Social Science & Medicine 75(1): 233–239.

26.

Haas

Hamersma

(2020) Cycling Facts: New Insights. The Hague: Netherlands Institute for Transport Policy Analysis | KiM.

27.

Haefeli

(2005) Public transport in Bielefeld (Germany) and Berne (Switzerland) since 1950: A comparative analysis of efficiency, effectiveness and political background. European Journal of Transport and Infrastructure Research 5(3): 219–237.

28.

Haustein

Kroesen

Mulalic

(2020) Cycling culture and socialisation: Modelling the effect of immigrant origin on cycling in Denmark and the Netherlands. Transportation 47: 1689–1709.

29.

Heinen

van Wee

Maat

(2010) Commuting by bicycle: An overview of the literature. Transport Reviews 30(1): 59–96.

30.

Hellevik

(2009) Linear versus logistic regression when the dependent variable is a dichotomy. Quality & Quantity 43(1): 59–74.

31.

Hendi

(2017) Trends in education-specific life expectancy, data quality, and shifting education distributions: A note on recent research. Demography 54(3): 1203–1213.

32.

Hirsch

Green

Peterson

, et al. (2017) Neighborhood sociodemographics and change in built infrastructure. Journal of Urbanism International Research on Placemaking and Urban Sustainability 10(2): 181–197.

33.

Horton

Rosen

Cox

(2016) Cycling and Society. London: Routledge.

34.

Hudde

(2022a) The unequal cycling boom in Germany. Journal of Transport Geography 98: 103244.

35.

Hudde

(2022b) Educational differences in cycling: Evidence from German cities. Sociology 56(5): 909–929.

36.

Klinger

Kenworthy

Lanzendorf

(2013) Dimensions of urban mobility cultures – A comparison of German cities. Journal of Transport Geography 31: 18–29.

37.

Klinger

Lanzendorf

(2016) Moving between mobility cultures: What affects the travel behavior of new residents? Transportation 43(2): 243–271.

38.

Kroesen

Handy

(2014) The relation between bicycle commuting and non-work cycling: Results from a mobility panel. Transportation 41(3): 507–527.

39.

Kuhnimhof

Wulfhorst

(2013) The reader’s guide to mobility culture. In: Megacity Mobility Culture. Institute for Mobility Research Berlin: Springer, pp.55–64.

40.

Lanzendorf

Busch-Geertsema

(2014) The cycling boom in large German cities—Empirical evidence for successful cycling campaigns. Transport Policy 36: 26–33.

41.

Litman

(2002) Evaluating transportation equity. World Transport Policy and Practice 8(2): 50–65.

42.

Malone

(2016) Smart choices for cities – Cycling in the city. Civitas Policy Note. Brussels.

43.

Martin

Goryakin

Suhrcke

(2014) Does active commuting improve psychological wellbeing? Longitudinal evidence from eighteen waves of the British Household Panel Survey. Preventive Medicine 69: 296–303.

44.

Millard-Ball

West

Rezaei

, et al. (2022) What do residential lotteries show us about transportation choices? Urban Studies 59(2): 434–452.

45.

Mood

(2010) Logistic regression: Why we cannot do what we think we can do, and what we can do about it. European Sociological Review 26(1): 67–82.

46.

Nobis

(2019) Mobilität in Deutschland − MiD. Analysen zum Radverkehr und Fußverkehr. Berlin, Germany: infas Institut für angewandte Sozialwissenschaft GmbH.

47.

Nobis

Claudia

(2019): Mobiltät in Deutschland-MiD Analysen zum Radverkehr und Fuverkehr Studie von infas, DLR, IVT und infas 360 im Auftrag des Bundesministeriums für Verkehr und digitale Infrastruktur (FE-Nr. 70.904/15). Bonn, Berlin. www.mobilitaet-in-deutschland.de

48.

Nobis

Kuhnimhof

(2018) Mobilität in Deutschland – MiD Ergebnisbericht. Studie von infas, DLR, IVT und infas 360 im Auftrag des Bundesministers für Verkehr und digitale Infrastruktur (FE-Nr. 70.904/15). Bonn. Available at: www.mobilitaet-in-deutschland.de

49.

Oosterhuis

(2016) Cycling, modernity and national culture. Social History 41(3): 233–248.

50.

Pojani

Bakija

Shkreli

, et al. (2017) Do Northwestern and Southeastern Europe share a common “Cycling Mindset”? Comparative analysis of beliefs toward cycling in the Netherlands and the Balkans. European Journal of Transport and Infrastructure Research 17(1): 25–45.

51.

Rachele

Kavanagh

Badland

, et al. (2015) Associations between individual socioeconomic position, neighbourhood disadvantage and transport mode: Baseline results from the HABITAT multilevel study. Journal of Epidemiology and Community Health 69(12): 1217–1223.

52.

Steinbach

Green

Datta

, et al. (2011) Cycling and the city: A case study of how gendered, ethnic and class identities can shape healthy transport choices. Social Science & Medicine 72(7): 1123–1130.

53.

Stone

(2009) Contrasts in Reform: How the Cain and Burke years shaped public transport in Melbourne and Perth. Urban Policy and Research 27(4): 419–434.

54.

Tranter

(2012) Effective speed: Cycling because it’s faster. In: Pucher

Buehler

(eds) City Cycling. Cambridge, MA: MIT Press, pp.57–74.

55.

Tyndall

(2022) Cycling mode choice amongst US commuters: The role of climate and topography. Urban Studies 59(1): 97–119.

56.

Urry

(2004) The ‘System’ of Automobility. Theory Culture & Society 21(4–5): 25–39.

57.

Walks

(2018) Driving the poor into debt? Automobile loans, transport disadvantage, and automobile dependence. Transport Policy 65: 137–149.

58.

World Health Organization (2019) Global Action Plan on Physical Activity 2018-2030: More Active People for a Healthier World. Geneva: World Health Organization.

59.

Zajacova

Grol-Prokopczyk

Zimmer

(2021) Pain trends among American adults, 2002-2018: Patterns, disparities, and correlates. Demography 58(2): 711–738.

60.

Zhang

Shaheen

Chen

(2014) Bicycle Evolution in China: From the 1900s to the present. International Journal of Sustainable Transportation 8(5): 317–335.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

2.96 MB

0.00 MB