Promoting active mobility: Evidence-based decision-making using statistical models

Abstract

Shifting traffic to active transport modes (eg. walking/cycling) poses one of the most promising ways of tackling the persisting challenges that arise from motorized traffic. However, planning and policy making in walking and cycling domains is frequently impeded by a small-scaled and heterogeneous political landscape that rarely acts based on evidence thus limiting cost-effectiveness and target achievement. This paper proposes a largely data-driven planning approach that builds upon aggregated statistical models explaining walking and cycling modal shares. In addition to investigating a comprehensive set of influencing factors in relevant fields such as environment, climate, infrastructure or demographics, we bring attention to the role of political and administrative commitment in aggregated modal share modeling. Results suggest that our holistic approach is feasible both methodologically and in terms of its applicability in planning practice. As a first step towards evidence-based decision making the incremental effects of individual planning measures can be simulated and thus be used to rank options according to their effectiveness. Another outcome lies in the data-driven identification of spatial target areas for specific agenda setting in terms of awareness, mobility behavior, infrastructure, settlement structure and other planning-relevant domains.

Full text

Promoting active mobility: Evidence-based decision-making using
statistical models
Roland Hackl a, Clemens Raffler a
*
, Michael Friesenecker a, Hans Kramar b,
Robert Kalasek b, Aggelos Soteropoulos b, Susanne Wolf-Eberl c, Patrick Posch c,
Rupert Tomschy d
atbw research GesmbH, Schönbrunner Straße 297 1120 Vienna, Austria
bUniversity of Technology Vienna, Augasse 2-6 1090 Vienna, Austria
cResearch & Data Competence, Wiedner Hauptstraße 39/Hofgebäude 1040 Vienna, Austria
dHERRY Consult GmbH, Argentinierstraße 21 1040 Vienna, Austria
Abstract
Shifting traffic to active transport modes (eg. walking/cycling) poses one of the most promising ways of tackling
the persisting challenges that arise from motorized traffic. However, planning and policy making in walking and
cycling domains is frequently impeded by a small-scaled and heterogeneous political landscape that rarely acts
based on evidence thus limiting cost-effectiveness and target achievement. This paper proposes a largely data-
driven planning approach that builds upon aggregated statistical models explaining walking and cycling modal
shares. In addition to investigating a comprehensive set of influencing factors in relevant fields such as
environment, climate, infrastructure or demographics, we bring attention to the role of political and
administrative commitment in aggregated modal share modeling. Results suggest that our holistic approach is
feasible both methodologically and in terms of its applicability in planning practice. As a first step towards
evidence-based decision making the incremental effects of individual planning measures can be simulated and
thus be used to rank options according to their effectiveness. Another outcome lies in the data-driven
identification of spatial target areas for specific agenda setting in terms of awareness, mobility behavior,
infrastructure, settlement structure and other planning-relevant domains.
Keywords: active mobility; walking/cycling determinants; political commitment to set actions; statistical
modeling; GIS; regional policies; transport policy; transport planning; evidence-based planning; settlement
structure; accessibility; social milieus; operationalization.
Disclaimer: This paper combines the work presented at REAL CORP Conference 2017 (see Hackl et al. 2017)
and the content submitted to Transport Research Arena 2018 (see Hackl et al. 2018). It extends the previous
work through additional analyses on political and administrative commitment and provides a combined
presentation of both pedestrian and cycling statistical models.
* Corresponding author. Tel.: +43-660-260-1870.
E-mail address: c.raffler@tbwresearch.org

1. Introduction
1.1. Active mobility planning in Austria – a story of great plans and small steps
Since the beginning of mass motorization, the growing shares in motorized traffic pose a serious challenge in
Austrian as well as international transport planning. Problems that arise from the negative external effects
generated along the development path of today’s transport system (Merki 2008) present themselves as air-, noise
and other environmental pollution, negative health effects and high accident rates as well as specific urban
challenges pertaining to a lack of space (Perschon 2012). The numerous downsides of motorized traffic have
been often discussed, see Buchwald and et al. (1993), Banister (2008), Knoflacher (2013) or Cervero (2013) and
highlight the importance of a change in transport planning paradigms.
In contrast to ongoing motorization, the international focus of traffic planning has gradually shifted towards
active modes which are often labelled as a sustainable basis for modern transport systems (see Vandenbulcke et
al. 2008; Rietveld and Daniel 2004; Lovelace et al. 2017). They feature a set of desirable ecological (resource
neutrality, zero-emission, downsize of land consumption), economical (reflecting an indirect net product for
Austria of up to 882.5 Mio.€, equivalent to 18.328 full-time jobs) (BMLFUW 2009) and socially (positive
impact on health) sustainable properties (Meschik and Traub 2008).
In quantitative terms the current challenges for the Austrian transport system regarding mode-choice become
apparent when looking at the recent mobility surveys. Comparing 2013/2014 ‘Österreich Unterwegs’ (BMVIT
2016) total shares in motorized traffic with the preceding survey in 1995 reveals that private car shares (as a
driver and co-driver) increased by 6.6% during this 19 year period. This amounts to 57.1% of total motorized
traffic in 2013/2014 whereas pedestrian traffic shares dropped substantially from 26.9% to 17.4%. Cycling,
featuring the lowest shares with 5.3% and 6.5%, more or less stagnated
1
.
In order to mitigate the above negative external effects a number of policy papers with respect to active travel
have been drafted in Austria. In general it can be stated that on both national and federal levels there is a policy
emphasis on cycling traffic. The masterplans for cycling (BMLFUW 2011; BMLFUW 2015b) aim at increasing
bicycle shares to 10% in 2015 and 13% until 2025. In contrast to the quantitative goals in cycling, the current
pedestrian masterplan (BMLFUW 2015a) doesn’t feature a target pedestrian share. Instead its focus lies on
suggesting dedicated actions including the prioritization of pedestrian needs in newly planned settlements, better
infrastructure, awareness raising, coordination between planning and administrative bodies, monitoring of modal
shares, safety measures and in-depth mobility research. A recent momentum regarding political commitment to
walking originates from the international Charter for walking (walk21 2017). However, the strategic documents
on national and federal levels don’t bear legal obligations for local planning agencies to actually enhance
walking or cycling quality. While it is popular amongst local decision makers to commit themselves to the
improvement of walkability in cities and villages, there is yet no quantitative evidence that this commitment
actually results in increased pedestrian shares (rather, pedestrian share dropped significantly in the last two
decades). Nonetheless some federal plans set even higher goals such as increasing active modal shares to 40% by
2035 in Carinthia (Carinthian government 2016).
One of the key drivers of engaging in developing evidence-based planning approaches in this research project is
a general lack of quantitative information on the effectiveness of different (infrastructure) measures in Austria.
There is no systemically collected data that allows for determining the quantitative impact of planning measures
on walking or cycling shares. Hence there are no hard and fast rules for identifying the most effective measures.
This is a major drawback when it comes to agenda-setting and decision-making in active mobility planning,
particularly under difficult general conditions (e.g. restricted budgets, conflicts of interest with other transport
modes, etc.).
On top of that, the decentralized structure of the Austrian planning system somewhat impedes coordinated action
and the implementation of the federal strategies and funding programs (see fig. 1). Since most of the legal
competence for infrastructure planning is in the hands of the 2100 Austrian municipalities, high level
masterplans cannot impose actions at municipality levels where responsibility to build infrastructure for active
modes resides for the most part. Due to a fragmented local political landscape and the sheer number of decision
making bodies, the political commitment and/or willingness to invest into active mobility measures can be
characterized as rather heterogeneous. While there are generally accepted technical planning guidelines and
recommendations (Meschik and Traub 2008) or the generic RVS guidelines for non-motorized traffic suggesting
minimum technical standards they are not legally binding (FSV 2019). Hence it is difficult to identify an
Austrian-wide political common sense relating to cycling and walking planning.
1
For comparability those figures refer to workdays in autumn. The 2013/14 values for an average day of the week amount to 60.5%
(motorized transport), 17.8% (walking) and 6,4% (cycling), respectively.

Fig. 1 Structure of the Austrian bicycle planning System (Raffler 2016)
Planners are confronted with increasingly difficult general conditions: to achieve the best outcome in terms of
increasing pedestrian and cycling shares, i.e. modal shift effects while at the same time being quite limited
financially. The main problem to be addressed is “that investments in cycling promotion are currently not always
put into action where they may be most expedient, but there, where local political will is the highest” (Raffler et
al. 2019) which is even more true for walking. This is exemplified looking at the federal state of Upper-Austria
and comparing planning actions (measured as number of cycling projects funded by the Austrian federal ministry
of environment – BMLFUW – klimaaktiv program) and respective cycling shares at municipality level: By
looking at Figure 2 it becomes clear that current agenda setting and investment into cycling promoting measures
does not reflect actual performance in terms of cycling shares. In addition to the different levels of political
commitment, the weak relationship shows that the simple rationale of investing into any projects in order to
boost non-motorized modal shares does not duly account for the complexity behind active mode choice and its
driving forces.
Fig. 2 Comparison between the number of number of klimaaktiv projects for the promotion of cycling shares
In a nutshell, a general lack of knowledge about cause and effect patterns between active modes and their drivers
in the respective contexts of user groups and local settings is currently impeding agenda setting and planning
actions aiming at creating a substantial modal shift towards active travel.

1.2. Research as evidence based-role models: Concepts, methods, findings and their application
International research on active mobility planning includes a broad spectrum of papers aiming at remedying the
above planning problems by suggesting decision support to planners and political stakeholders. This builds upon
the concept of evidence-based planning which represents the main rationale for planners and policy makers
(Faludi 2006). This approach originates from Western-European planning culture (Davoudi 2006) and its
influence can be found in papers from the UK and the Benelux-States (see Parkin et al. 2007b, Vandenbulcke et
al. 2008, Rietveld and Daniel 2004). The paradigm is to better understand the factors that influence the
respective mode-choice which then can be used as scientific evidence in the planning process. This is
particularly true for research in the cycling domain and was best reflected by Heinen et al. (2010): “In order to
be able to develop sound policies that encourage cycling, it is essential to understand what determines bicycle
use.“ The main rationale behind investigating the determining factors of active travel is to reveal the relevant
mechanisms planners may need to address in order to positively influence the development of active modal
shares (Parkin et al. 2007b). Another key advantage of a solid evidence base on active mobility is that funding
activities can be focused on where they will provide the biggest return in terms of modal split increases, hence
tackling the problem of uncoordinated (or even ineffective) initiatives (Raffler et al. 2019).
When looking at latest research, a great number of hypotheses have been proven in order to assess the direction
and influence of determinants on active modal choice. Those can be roughly categorized into three groups of
influencing factors (Heinen et al. 2010):
1. determinants that can directly be influenced by planners: urban form, density, landuse mix and
infrastructure for active mobility (infrastructure type such as on- or off road cycling infrastructure, existing
infrastructure for motorized traffic and overall traffic organization)
2. determinants that can indirectly be influenced by planners: socioeconomic/sociodemographic mix in a city
or neighborhood depicting age, gender and income structures, predominant social and environmental
psychology such as social norms towards or against active mobility as well as social milieu mix which
largely constitutes on basis of the former factors
3. determinants that cannot be influenced by planners: Those are mostly geographical preconditions such as
climate, weather, topography which either limit or encourage walking and cycling (Raffler et al. 2019)
There have been intensive discussions on assumptions about the influence of compact and dense urban
structures, its landuse mix (Pucher and Buehler 2006) as well as microscale design of urban form (Rybarczyk
and Wu 2014). Although most research confirms the hypothesis that compact urban form (Saelens et al. 2003,
Pucher and Buehler 2006, Parkin et al. 2007b) and a heterogeneous landuse mix (Cervero and Duncan 2003)
encourages bicycle usage due to better accessibility and shorter trip length, there are also studies showing
insignificant results regarding density (Cervero and Duncan 2003, Winters et al. 2007).
Another core topic discussed by transport researchers and traffic planners concerns the best type of infrastructure
for active mobility, especially regarding the choice between different cycling infrastructures (off- on-road
cycleways and cyclepaths) (Garrard et al. 2008, Akar and Clifton 2009, Winters and Teschke 2010, Caulfield et
al. 2012). From a user perspective, infrastructure preferences vary among different groups of cyclists: For
example, Garrard et al. (2008) showed empirically that off-road cycling infrastructure is preferred by female
cyclists. Unlike for example Irish policy (Caulfield et al. 2012), Austrian cycling policy does not impose any
legally binding regulations regarding the type of infrastructure (BMLFUW 2015b). Yet Austrian planning
guidelines suggest using driving speeds and traffic volumes as parameters when deciding on cycling
infrastructure type (FSV 2019). However, in in real world planning processes the discussion frequently doesn’t
revolve around the type of infrastructure but whether to build any cycling infrastructure at all.
International research on the influence of social and environmental behavior comprises small-scale detailed
surveys on the relation between individual social background and attitudinal characteristics towards cycling (eg.
Guell et al. 2012, Souza et al. 2014). Another set of studies uses quantitative data at a larger scale in order to
investigate lifestyle types and their respective mobility behavior (van Acker 2015). In the Austrian context, the
latter approach has been complemented by the concept of SINUS social milieus which originated form market
research (Sinus Markt- und Sozialforschung GmbH 2019). As a theory-based construct, this approach clusters
society into different milieus according to personal attributes, attitudes, personal aims and the rejection of certain
goals in life (Dangschat et al. 2012). A German study investigated into the likelihood of riding a bicycle from the
perspective of 10 social milieus in 2017 (Sinus Markt- und Sozialforschung GmbH 2017a): The results show that
open-minded and financially well-off groups have a higher propensity to use the bicycle as traditional groups
and/or financially weak social milieus. In addition to that, the individual drivers for bicycle use differ among
groups. While open minded people use the bicycle for daily trips, the social milieu ‘performers’ cycle as outdoor
and leisure activity. Summing up, datasets on social milieus pose a new and promising way of incorporating
meaningful determinants for walking and cycling that previously have been neglected.

In contrast to cycling, papers on pedestrian traffic generally do not focus on planning support, rather they tend to
have a health science perspective (eg. Leslie et al. 2005; Cerin et al. 2009; Verhoeven et al. 2016). Hence they do
not explicitly propose results that are designed for the use in active planning processes; nevertheless they have an
indirect implication for planning activities.
From a methodological point of view, influencing factors can generally be identified by performing simple,
mono-causal correlational analyses (Leslie et al. 2005) or by setting up more sophisticated models using
regression techniques. There are two main approaches (Aoun et al. 2015; Parkin et al. 2007a):
1. Aggregated models estimate walking or cycling modal shares from census or survey data at the
administrative level of municipalities or origin-destination flows. Those models don’t reflect individual
behavior but investigate the impact of a local administrative area’s properties (infrastructural,
socioeconomic or social) on active mode shares. Most approaches achieve this by estimating the impact of
the respective municipal configurations applying slightly adapted OLS regression techniques as shown by
Rietveld and Daniel (2004) (2004), Parkin et al. (2007b), Vandenbulcke et al. (2008), Pucher and Buehler
(2006) and Cerin et al. (2009). Due to the aggregated perspective, those models are sometimes used in
strategic planning contexts. The models can then be used by planners and decision makers to assess the
impact of certain planning actions by consulting the raw model equation.
2. The second approach comprises disaggregate models which reflects the probability of choice to participate
in active mobility at the individual level. The individual’s preferences to walk or cycle are obtained by the
collection of data on stated- or revealed individual preference. For statistical modeling, mostly binary
logistic regression approaches are used. The practical application of these models is to explore and identify
individual properties (e.g. socioeconomic status) that influence active mode choice and therefore encourage
bicycle promotion among specific groups. For examples, see Wardman et al. (2007) or Heinen et al. (2013).
1.3. Knowledge gaps and practical problems
Regarding national and international research, our literature review revealed that there still exist three major
knowledge gaps relating to the application and extension of aggregated mode share models:
(1) As described above, a large amount of research focuses on decision support for bicycle planning and
therefore narrowing the view on active mobility which naturally includes pedestrian traffic. Hence existing
research should be extended beyond health science into transport research and planning in order to identify
possible countermeasures remedying declining pedestrian shares based on solid evidence (see 1.1: drop of
9.5% in Austrian pedestrian shares between 1995 and 2013/14).
(2) General critique on active mode share modelling approaches focuses on the lacking representation of so-
called ‘soft’ factors (in contrast to ‘hard’ factors, such as topography, infrastructure or accessibility). As
Heinen et al. (2010) point out with regard to poorly reflected psychological factors (eg. personal attitudes
towards active modes) we suggest to extend this critique to the omission of political/administrative
commitment factors as those are frequently cited as being crucial for a successful promotion of non-
motorized modes. This assumption is backed up by oral evidence sourced from local planning stakeholders
– and therefore considered a relevant point of critique. Despite the efforts of preceding models to
investigate into the effects of “policy” (Rietveld and Daniel 2004; Pucher and Buehler 2006) with the help
of proxy-variables such as gasoline prices per litre, cycling fatality rate (Pucher and Buehler 2006),
parking costs, network speed or voter-proportion of certain political parties (Rietveld, 2004), research on
comprehensive proxy variables that reflect political will in a more realistic way is missing. While there is a
great number research investigating British, Dutch, Belgian or American contexts of active travel, there are
no such comprehensive models for Austria which potentially lowers the success rate of national planning
activities. In addition to that we intend shedding new light on the discussion of social attitudes as an
determinant of active mobility, as new datasets describing social milieus became available in Austria.
(3) Austrian national decision support approaches are currently somewhat limited both in their thematic and
spatial views on active mobility as there exist only two approaches in federal states Vorarlberg and Tirol
building upon accessibility analyses (Verracon GmbH 2016; Tyrolian Government 2014). Accessibility
may be one of the most important determinants (largely reflecting neighbourhood form and settlement
density) but it is safe to say that it doesn’t amount to the only relevant factor of cycling and walking.
Moreover, such mono-causal approaches may neglect other important (nested) co-influencing factors that
need to be addressed by evidence-based policy-making. This issue calls for novel approaches aiming at
supporting the national active mobility planning in a systematic and holistic manner. Precisely, there is a
lack of scientific evidence to prioritize investments for walking and cycling at the municipal level. A first
approach for investment prioritization has recently been presented for bicycle measures by Raffler et
al.(2019): Built on the hypothesis that investment in cycling is most expedient in areas where cycling is
least physically exhausting, a prioritization technique for investment into bicycle traffic based on regression
residuals has been presented. However, the fact that this approach only considers the physical determinants

of cycling (hilliness and/or distance) constitutes a research gap in Austrian decision support. Put differently,
mono-causal approaches need to be extended by including other determinants of active mobility.
1.4. Aim of this research: a deeper understanding of spatial variation of active travel
In face of the various challenges limiting success for active mobility planning in Austria (heterogeneous political
commitment, lack of knowledge about cause and effect patterns, uncoordinated actions), this paper aims at
providing the scientific basis in order to tackle some of these problems. At the same time we aim to shed light on
the internationally existing knowledge gaps (lacking consideration of walking, measurement and inclusion of
political/administrative commitment in aggregated mode choice models as well as their application in real-world
planning).
We build a comprehensive aggregated modeling framework for active travel modes (one each for walking and
cycling) with a spatial focus on Upper-Austrian municipalities. The models examine the cause and effect
patterns behind the regional variation of active travel shares by investigating the quantitative links between
active mobility and spatial, infrastructural and social influences. Based on a dedicated pool of hypotheses we
devote special attention to the transparent operationalization of influencing factors including new variables
reflecting local social milieus and political commitment to promote active mobility. As a complementary
analysis we correlate the proxy variables on political commitment with an index on self-assessed local
commitment (sourced from municipal administrations through an accompanying online-survey) in order to
quantitatively assess the relationship between subjective and objective willingness to support active travel.
With this we aim at answering questions such as whether or not there are widely applicable generic concepts for
increasing active modal shares or whether plans rather need to be custom-made for each municipality. In the
latter case we aim at guiding planning by identifying the most promising fields of action and target population
groups as well as estimating the potential effects of the planned measures in the respective contexts. Summing
up, this paper aims at (1) providing the first holistic evidence-based approach to aid active-mobility planners in
achieving the best outcome considering their somewhat limited budget, (2) contributing to the internationally
scarce research on the determinants of walking and (3) shedding light on the operationalization and use of proxy
variables operationalizing political commitment in aggregated models.
2. Data and Methods
2.1. Data
A crucial step predating the model building process is the choice and collection of appropriate data for the
aggregate mode choice models. The current section describes the details behind data collection and refinement
which includes the preparation of traffic survey data, conduction of a supplementary survey on self-evaluated
importance of active modes by municipal representatives (eg. mayors, administrative staff) and acquisition of
quantitative data describing influencing factors on active mode shares.
Traffic survey data:
In order to operationalize active modal shares at the local level of municipalities we first acquired data from the
traffic survey of the state of Upper Austria which was conducted in October 2012 (Government of Upper Austria
2014). Although this decision narrows the spatial focus of our research through the exclusion of the other eight
federal states from the analysis, Upper Austria is one of the few states that features nearly every element of the
heterogeneous Austrian spatial structure (eg. alpine regions as well as rural forests, hills, urban and semi-urban
zones) and provides a reasonable sample-size of municipalities. Also Upper Austria currently holds a unique
position as it is the only Austrian federal state to provide complete data on modal shares on the municipality
level.
Modal shares for a total N of 444 municipalities are based on person-specific trips (specified by mode and trip
purpose): numbers of trips were projected and statistically weighted in order to correct for sample bias. Active
mode shares were calculated as the respective proportions of walking and cycling trips and the total number of
reported trips per municipality. In order to secure a sound 95% confidence interval of the modal shares, we
excluded municipalities where the number of interviewed persons was less than 200. (see Table 1, filtered).
Also, we used the unweighted number of reported trips to weight cases (municipalities) when calibrating the
regression models so to give relatively more weight to more robust values in the outcome variable. Those actions
do not harm the models’ representativeness but rather remove modal share values based on a weak empirical
foundation. Related issues pertaining to the confidence levels of the non-motorized modal shares impeded a
further differentiation of the models in terms of trip-purposes. Though initially planned this would have asked
for substantially larger sample sizes of the mobility survey and was therefore skipped. The descriptive statistics
presented in table 1 show that Upper Austrian walking shares at the municipality level are considerably larger

than cycling shares.
Table 1. Descriptive variables of the Upper-Austrian active mode shares by municipality
mode
model-type
N [municipalities]
mean [%]
min [%]
max [%]
SD [%]
walking
unfiltered
444
11.46
0.71
32.48
4.8
filtered
338
12.21
3.32
32.48
4.83
cycling
unfiltered
444
3.55
0
21.4
2.70
filtered
338
3.88
0.25
17.47
2.6
Walking and cycling shares range between 0.7% and 32% and 0% and 21%, respectively. They exhibit a
substantial right skew resulting from a far from normal distribution
2
. Removing municipalities with less than 200
surveyed persons reduces N by 106 while slightly increasing the mean values for both modes. Also, some of the
extreme values on the outer limits of the distribution have been excluded due to the filtering. Following the trend
in the national travel survey, walking shares in Upper Austria are generally higher than cycling shares (no zero-
share municipalities, higher maximum and mean values) but feature a similar statistical distribution.
Datasets on active travel determinants:
Alongside these traffic and survey data, numerous additional data sources have been tapped in order to form
model covariates describing local spatial, infrastructural and socioeconomic properties. Datasets range from
spatial information from the national Graph-Integration-Platform (GIP) and OpenStreetMap (OSM), digital
elevation models and population density rasters, demographic and socioeconomic data by Statistics Austria as
well as data on social milieus from INTEGRAL Markt- und Meinungsforschung. Weather and climate-related
information was sourced from ZAMG (Zentralanstalt für Meteorologie und Geodynamik). Some data was
directly obtained from the thriving Austrian OpenData initiative (data.gv.at) or representatives of the Upper
Austrian state administration.
Supplementary survey on political will:
Pertaining to defining proxy variables reflecting political commitment towards active mobility within a model
framework we conducted an online survey among Upper Austrian administration representatives on the
municipal level. The survey aimed at collecting the local importance of bicycle traffic by self-assessing
questions like ‘How important is bicycle traffic for your municipality?’, or ‘Is there a dedicated budget for
bicycle infrastructure?’ following a simple grading system. Owing to the thematic focus of the Upper Austrian
provincial government the survey had a focus on cycling. The collected data resemble an empirical picture of the
perceived preconditions and self-assessed efforts related to actions to promote cycling. In order to increase the
response rate the web link for participating in the survey was distributed by a well-known sender (Government
of Upper Austria) to all 444 municipalities shortly after the Summit for Cycling event in Linz, Upper Austria.
The relatively high response rate of 54 percent (242 cases) proves that this approach has been successful.
However, it was not possible to directly include the survey results in the statistical models due to the missing
municipalities.
2.2. Methods
Our research approach was guided by structuring the model building procedure in five major steps (see fig. 2).
Driven by literature, we first (1) specified three groups of influences that are known to have an effect on non-
motorized modal choice: spatial, environmental and climate, infrastructural and demographic/socioeconomic
(including commitment of the communal decision-makers) influences, hence to manage and structure a
potentially vast amount of covariates. In a second step (2), we formulated hypotheses on the expected impact
direction and strength of theoretical indicators that could be assigned to one of the three factor groups. The third
working step (3) was focused on the operationalization (data acquisition, geospatial and mathematical modelling,
econometric techniques) of the variables that built upon the data sources in the previous section. Figure 3 lists
the main highlights of the third working steps output variables. The fourth and last step of analysis (4)
constituted the statistical inference process and the formulation of multivariate regression models to predict non-
motorized modal shares as the outcome variable on the municipality scale.
While building the statistical models (see figure 3), we conducted a supplementary work-step (5) measuring the
correlational relationship between the proxy variables reflecting political and administrative commitment and an
affinity index (affinity to promote cycling) constructed from various items included in the above self-assessment
2
The mean modal shares among Upper Austrian municipalities should not be mixed up with the overall Upper Austrian shares which
amount to 14.6% for walking and 5.1% for cycling, respectively.

municipality survey. This was done in order to safeguard the relevance and adequacy of the selected proxy
variables.
The following section gives a brief overview by determinant category over the 700+ variables that were gathered
and computed. We will describe variables listed in figure 3 in more detail as those were built devoting advanced
methodological attention in ways not yet presented by international research on active mode share modeling. The
main tools that were applied in the variable-forming process include GIS (ArcGIS, QGIS and PostGIS) as well
as the statistical software package SPSS for the data management, processing, testing and inferencing model
formulations.
Fig. 3 Overview of model building workflow
Spatial and environmental determinants:
When looking at the variable configurations of existing research, spatial and environmental determinants have
always been a core element when investigating active mode choice (eg. Parkin et al. 2007b; Vandenbulcke et al.
2008). This group of factors includes determinants that can be characterized as slow-changing factors: settlement
structure and various ‘static’ environmental characteristics that cannot directly be influence by planners (relief,
climate). Secondly, a broad set of accessibility indicators resembling determinants that can be (directly or
indirectly) influenced by planning decisions:
(1) The calculation concept for the accessibility covariates follows an extended version of the density-based
intra-zonal and external distance estimation approach by Kordi et al. (2012): Local/regional walking,
cycling and driving distances and times from the cells of a 250m population density grid to different
categories of trip destinations (e.g. health services, social infrastructure, shopping) were computed using
network analysis. In order to obtain a single aggregated accessibility value per municipality we used the
population of the origin raster for the calculation of a weighted mean of all possible route configurations in
a municipality. The population density raster provides the necessary information for the calculation of a so
called degree of affectedness (DOA) of municipal population by mode specific accessibilities (see fig. 2)
and was also utilized for other environmental variables.
(2) The second focus of the accessibility analysis was the reflection of mode specific characteristics of
accessibility. Our approach integrates attributes from different data sources (number of lanes, lane-speed,
cycling-infrastructure from GIP and OSM). To further reflect realistic impedance for both active modes in
an alpine country like Austria, we considered street-slopes according to a 10m digital elevation model that
was geographically matched to the street network. We calculated the arithmetic ratios between non-
motorized travel time and its motorized counterpart in order to realistically capture the rationale behind
travel-time acceptance for motorized households. Due to limited degree of data completion in the Austrian
road graph at the time of writing this paper, it was not possible to include detailed information about
different types of cycling infrastructures, such as off- or on-road cycling infrastructures and cyclepaths.

Instead, a classification of graph edges was conducted in order to distinguish between edges that are
especially suitable for cycling (low velocity, declared as bicycle infrastructure in either OSM or GIP) and
edges where motorized traffic is dominating. By calculating the proportion of suitable cycling infrastructure
along the shortest paths in the accessibility analysis, it was possible to account for the local infrastructural
situation despite being limited from the data perspective. Access to public transport was operationalized by
calculating the mean distances to stations by using raster based cost-path analysis. This type of routing
algorithm works with traversable raster surfaces rather than road graphs. The choice for this approach was
specifically related to the fact that a traversable raster surface is more suitable to model the walking or
biking accessibility to public transport: They better reflect the nature of a pedestrian’s pathfinding to
stations as routing is not bound to discrete graph-edges.
(3) Weights were attached to the trip-destinations of routes in order to take into account the relative importance
of a destination (visiting frequencies for doctors, hospitals, pharmacies, schools, grocery stores,
supermarkets, administrative facilities) and size of their target groups. This was accomplished by adding a
literature-based list of demand factors and empirical findings on target groups.
(4) Approaches related to the above step (1) of the accessibility analysis were applied for the operationalization
of environmental variables. Examples are determinants that reflect climate (e.g. number of snow cover
days, frost, etc.) or topography as a DOA of local population. Although topography is considered as a
negative impedance in active travel – especially in cycling (Raffler et al. 2019) – it can also be interpreted
as scenically valuable. We therefore included measures by applying state-of-the-art slope- and ruggedness-
index analysis through the use of GDAL algorithms.
Infrastructural determinants:
Determinants reflecting infrastructural conditions play a crucial role in this research, as they can be directly
addressed by planning actions and local/regional development plans. We calculated measures describing the
local topology of the road network following the approach of Tresidder (2005): Those are represented by
municipal Intersection Density (arithmetic ratio between connecting nodes and the total municipal network
length/settlement area) as well as the Connected Node Ratio (arithmetic ratio between the number of connecting
nodes and all network-nodes in a municipality). Those variables describe the permeability of the municipal road
network as these can influence the enjoyment and comfort of local active trips through more direct routes.
Walkability and bikeability reflect mode-specific time advantages as well as the convenience and scenic quality
of cycling routes. In order to operationalize these features we included the density of cycle tracks, the share of
traffic-calmed streets or the density of traffic accident hotspots at the road graph level. This was particularly
challenging as the data sources GIP and OSM comprised unstable and incomplete information on cycling and
walking infrastructure at the time of the data acquisition.
Demographic and socioeconomic factors and political/administrative commitment:
Demographic and socioeconomic factors have frequently been a major point of discussion in the context of
research on non-motorized traffic (Goodman 2013; Heinen et al, 2010). Therefore we extracted several variables
from census-based surveys which include aggregate measures on demographic structures, household structure
(eg. mean household size), age groups, education, car ownership or purchase power per person/household. A
more sophisticated view on local mind sets was provided by variables on social milieus (local shares of SINUS-
milieu groups) that cluster population according to lifestyles and attitudes (milieus include conservatives,
hedonists, or performers, etc.). In the specific context of this research, we aimed at extending the existing efforts
of operationalization of the local active travel mode culture and the local commitment to support active modes
among decision-makers. One approach consisted of collecting information on the municipality’s membership in
federal or state-level initiatives such as at the Upper Austrian cycling promotion programme
(fahrradberatung.at, bicycle coaching initiative for municipalities), Klimabündnis Austria (an organisation
promoting climate protection) or the number of projects realized in the Klima-aktiv programme (climate
protection initiative of the Austrian BMLFUW). This information was used to calculate workable variables such
as number of years since first assignment or simple 0/1 dummy variables. A second approach was based on
including election results on municipal and state level elections. In this context it shall be noted that past political
commitment may have implicitly manifested itself in kind of actually realized infrastructure projects or
awareness-raising projects in favour of active travel modes whereas the above variables describe the current
local ‘climate’ for active travel modes and potential for its promotion in the near future.
Regression model
We derived multivariate regression models aiming at identifying the relative importance of the determinants on
the spatial variation of both active travel modes at the scale of Upper Austrian municipalities. The outcome
variable (share of walking/cycling trips in all trips in a municipality) and the regression coefficients comprise the
matrix of the municipalities’ characteristics in the independent variables or covariates. The final set of

independent variables was derived iteratively from a pool of 700+ candidate variables adopting a hierarchical
scheme of model selection and a set of complementary tests and procedures. As sample size is relatively small
(338 municipalities after applying the filter) the possible number of predictor variables is somewhat limited.
However, with up to 17 variables in the pedestrian model and up to 22 predictor variables in the bicycle model,
the upper value following Green (1991) in minimum sample size is 234, which is well exceeded. As a guiding
principle we were aiming at combining several individual variables to form combined indicators (eg. the
composite accessibility or landscape scenic quality variables) wherever feasible in order to reduce the number of
covariates while increasing their explanatory power. Starting off by testing the inclusion of a basic set
determinants (largely based on previous research) which were force-entered into the regression model we
continued to include thematic sets of additional variables in stepwise modes (both backward and forward) in
order to check for incremental improvements by adding new predictors to the equation. Each step was checked
in terms of theoretical plausibility and accompanied by applying statistical tests (e.g. checking for
multicollinearity or suppressor effects) so not to leave crucial modelling decisions to purely statistical criteria or
let them be unduly influenced by random sampling variation. To quote and example some variables on adverse
weather conditions (eg. number of frost days or rain days) had to be removed from the models as they exhibited
considerable correlation with each other. For each model variant we tested for autocorrelation (independent
errors) and heteroscedasticity – both tests signalized their absence.
3. Results
Table 2 and 3 show the main statistical results for the pedestrian and cycling models. The outcome variable is the
respective modal share in Upper Austrian municipalities. The determinant variables are labelled according to
their respective factor group (see 2.2) by prefixes ENV, INF or POP, respectively. It shall be noted that while
from a planner’s viewpoint the focus is clearly on variables that can actually be influenced by planning actions
(e.g. relating to infrastructure, behaviour, awareness) it is nonetheless crucial to include other variables in order
to cover all relevant determinants as comprehensive as possible and to control for the respective effects while
explaining the corresponding variance proportions in the regression outcome variable. Omitting these controlling
covariates one would run the risk of falsely attributing non-related parts of variance in outcome y to planning-
relevant variables while they are in fact due to other factors (potentially non-controllable by planners such as
weather or topology). In order to duly compare the effects of individual determinants on active mobility shares
one should consult column β containing the standardized coefficients.
Pedestrian Model
Overall, the model on walking shares explains 77.5% of the variance in pedestrian shares among Upper Austrian
communities (R2=0.775). The large positive value for the composite variable on walking accessibility to various
POIs confirms the hypothesis that compact settlement structures and relative proximity to basic amenities is a
key requisite for walking (ENV_composite_acc_pot_walking). This is also partially reflected by the effect of
‘INF_share_urban_environment’ expressing the share of land use category ‘urban’ along the municipal road
network indicating that denser environments are in general more pedestrian-friendly. In a similar vein the share
of out-commuters in the local workforce exhibits a negative effect of walking shares. The positive sign of the
climate variable ‘ENV_no_snow_cover_days’ indicates a potential swap of modal choice during the snowy
season as the same variable shows a negative sign in the cycling model. A part of regular cyclists is switching to
walking mode in case weather conditions appear unsafe or discomforting for cycling. With respect to policy
relevant factors the weighted (according to type of PT) distance to public transport access points is an important
predictor suggesting that the availability of adequate public transport is encouraging walking when controlling
for other relevant factors.
Table 2. Pedestrian model coefficients, standardized coefficients, t-statistic, significance and
correlation with pedestrian share (**p<.001, *p<.005)
Variable
b
ß
t-statistic
p
correlation with y
Constant
-0.338
-
-248.841
0.000
-
ENV_composite_acc_pot_walking
0.0022
0.594
316.332
0.000
0.730
ENV_no_snow_cover_days
0.0008
0.272
308.277
0.000
0.032
ENV_landscape_scenic score
0.0089
0.054
72.098
0.000
0.177
INF_distance_PT_weighted
-2.02E-05
-0.127
-172.589
0.000
-0.210
INF_share_connected_nodes
2.5619
0.066
55.570
0.000
0.523
INF_share_urban_environment
0.0758
0.313
181.514
0.000
0.680
INF_relative_prob_accidents
-0.2328
-0.008
-11.471
0.000
0.016

POP_no_klimaaktiv_pop
0.0082
0.082
124.273
0.000
0.108
POP_share_pop_o_65y
0.0016
0.072
49.474
0.000
0.490
POP_share_pop_u_15y
0.0028
0.076
72.223
0.000
-0.398
POP_share_educ_university_lvl
-0.0012
-0.085
-44.637
0.000
0.379
POP_dummy_klimabuendnis
0.0037
0.027
37.668
0.000
0.256
POP_part_time_rate_men
0.0017
0.080
105.880
0.000
0.305
POP_share_milieu_adaptive_pragmatic
0.0047
0.212
169.723
0.000
-0.411
POP_share_milieu_post_material
0.0114
0.421
234.423
0.000
0.279
POP_share_milieu_traditional
0.0043
0.261
138.970
0.000
-0.269
POP_share_out-commute
-0.0005
-0.124
-89.936
0.000
-0.609
R
0.880
R2
0.775
R2adj
0.775
With respect to population and political/administrative commitment it can be concluded that certain features of
the local population (such as relatively high shares of both older and young people; ‘POP_share_pop_o_65y’ and
‘POP_share_pop_u_15y’) as well as relative high shares of specific social milieus (adaptive pragmatic, post-
materialistic or traditional) contribute to walking. In general terms milieu variables tend to have a substantial
explanatory power for walking shares. To a slightly lesser degree the same is true for the proxy variables
reflecting political and administrative commitment at the municipal level (‘POP_no_klimaaktiv_pop’,
‘POP_dummy_klimabuendnis’). Finally, the relative probability of accidents and the share of third-level
education in the local population have negative effects on walking shares while a high share of part-time
employment among the male workforce is positively impacting walking shares. Those findings are in line with
theoretical considerations (in particular when controlling for social milieus).
Note that some variables show a reversed sign in the regression model compared to the direct (zero-order)
correlation with the outcome variable. While this could be a potential causer for concern, it can be made
plausible by considering that the inclusion of other predictors controls for several effects that are confounded in a
zero-order correlation but split across dedicated covariates once they are included in the model. To quote an
example, the share of tertiary-level education among the local population has a positive zero-order correlation
with walking shares. However, once we control for attitudinal features through the inclusion of social milieu
variables (‘POP_share_milieu’, etc.) the impact of high education levels on walking shares reverses. By contrast,
traditional milieu shares are over-represented in settlement structures typically associated with low walking
shares (suburban regions, regions with agricultural land use, etc.). Once some of these effects are controlled for
(through the inclusion of composite accessibility variables), the model results show that – other things being
equal – attracting traditional population will actually help increasing the local pedestrian share.
Cycling Model
In total the model includes 22 predictors accounts for 71.9% of the cycling share among the municipalities
(R2=0.719). Like in the pedestrian model we included various variables controlling for static influencing factors
having substantial impact on cycling shares. As expected, negative impacts originate from hilliness within
settlement areas as well as the number of days with snow cover, which confirms the hypothesis on positive
influences on pedestrian shares when it is snowing. The scenic quality along the road
(variable‘INF_pleasant_green_roadside’) has a positive impact on cycling modal shares: this composite variable
measures the share of certain land-use categories that have been considered attractive along the road network.
High positive beta-values are displayed for the accessibility variables ‘ENV_ratio_accessibility_pot_cycle_car’
and ‘ENV_ratio_accessibility_prim_schools_walk_car’ reflecting the weighted POI-related accessibility ratios
between cycling and motorized traffic as well as the bikeability along routes to schools. This poses a strong
statement for urban planners: Better accessibility ratio and bicycle-friendly environments play key roles when
aimaing at a modal shift in favour of cycling traffic.
As expected, variables typically associated with conflicts between cyclists and motorized road users (eg.
‘INF_density_accident_hotspot’) or the prevalence of car-centred infrastructure
(‘INF_dummy_highway_access’, ’INF_share_roads_GT_60kmh’) indicate negative influences whereas the
provision of stationary cycling infrastructure generally encourages people to cycle
(INF_bike_racks_per_1000_pop). In this context the sign of the distance to highway access
(‘INF_minimum_distance_highway’) was not expected initially. However, on second thought this relationship
matches with properties of remote regions that are highly dependent on car while typically having bad cycling
accessibility. They therefore pose impedances to cyclists reflected by this measure of remoteness. Similarly to

the pedestrian model, the municipal share of out-commuters has a negative effect on cycling shares.
Table 3. Cycling model coefficients, standardized coefficients, t-statistic, significance and
correlation with cycling share (**p<.001, *p<.005)
Variable
b
ß
t-statistic
p
correlation with y
constant
0.145
-
226.077
0.000
-
ENV_no_snow_cover_days
-0.001
-0,482
-353.471
0.000
-0.317**
ENV_hilliness_settlement_area
-0.010
-0,440
-379.508
0.000
-0.495**
ENV_share_agri_areas
-0.042
-0.345
-195.986
0.000
-0.499**
ENV_ratio_accessibility_pot_cycle_car
0.001
0.350
229.645
0.000
0.612**
ENV_ratio_accessibility_prim_schools_walk_car
0.001
0.220
197,862
0.000
0.494**
ENV_bikeability_routes_schools
0.007
0.043
41.857
0.000
0.410**
INF_share_pleasent_green_roadside
0.033
0.349
229.723
0.000
-0.338**
INF_density_accident_hotspot
-1.157
-0.120
-117.715
0.000
0.322**
INF_dummy_highway_access
-0.004
-0.056
-58.815
0.000
0.106**
INF_minimum_distance_highway
-6.280E-7
-0.197
-172.320
0.000
-0.189**
INF_bike_racks_per_1000_pop
0.000
0.063
81.225
0.000
0.208**
INF_settlement_proportion
-0.005
-0.031
-20.921
0.000
0.496**
INF_minimum_distance_major_cycle_routes
-8.895E-7
-0.071
-72.060
0.000
-0.323**
INF_share_roads_GT_60kmh
-0.044
-0.090
-101.198
0.000
-0.030**
POP_share_out-commuters
0.000
-0.221
-169.553
0.000
-0.450**
POP_share_milieu_established
0.003
0.124
109.934
0.000
-0.378**
POP_share_milieu_performer
-0.007
-0.408
-187.035
0.000
0.321**
POP_mean_duration_work_commute
0.001
0.146
154.036
0.000
-0.183**
POP_dummy_klimabuendnis
0.003
0.043
52.905
0.000
0.154**
POP_years_participation_fahradberatung
0.001
0.101
120.091
0.000
0.329**
POP_share_workplaces_agri
-0.016
0.014
-47.462
0.000
-0.531**
POP_purchase_power_index_person
-0.001
-0.155
-97.576
0.000
0.146**
R
0.848
R2
0.719
R2adj
0.699
In terms of attitudinal variables ‘POP_share_milieu_established’ (local population share of social milieu
established) and ‘POP_share_milieu_performer’ (local population share of social milieu perfomers) have the
most significant effects on cycling shares. On average approximately 10% of the Austrian population belong to
the social milieu established. It represents the performance-oriented and success-oriented elite in middle age
groups. With other effects being controlled for in the model, a 1% increase of established milieu among the total
local population will increase the municipality’s cycling share by approx. 0.3%. Performers being the younger
part of the elite can be broadly characterized as being globally oriented, highly efficient, success-oriented with
comprehensive skills in IT and business (making up approx. 9% of the Austrian population). Model results
indicate that a 1% increase in performers population share will reduce the cycling share by -0.6%. Note that the
zero-order correlations with y show reverse signs for both milieus. This is again due to the inclusion of other
factors which explain large parts of the variance in cycling. Hence the milieu population shares explain unique
parts of the variance. Our interpretation is that both milieus share specific patterns of other mobility relevant
factors such as choice of residential location or purchase power. Once these variables are controlled for and the
other factors are kept constant, the coefficients for the milieu variables express the respective net effects while
other things are being equal. While performers have a tendency towards high performance recreational sports
they do not have environmentally conscious or cost-conscious mind-sets when it comes to everyday mobility
(Dangschat et al. 2012). German studies by Sinus Markt- und Sozialforschung GmbH (2017b) found similar
patterns of bicycle usage in the milieus of performers. Hence we even expect rebound effects on active travel
shares to be related with performers’ recreational behaviour (e.g. using the car to go to cycle routes). In Upper
Austria, established milieu shares are over-represented in settlement structures typically associated with low
cycling shares (suburban regions, regions with some agricultural land use, etc.). Also they exhibit above average
household sizes generally associated with below average cycling shares as well as above average income levels

and purchase power. Once we control for some of these effects, the model results show that attracting
established population will help increase the local cycling share.
The covariates on political and administrative commitment remaining in the model are
‘POP_dummy_klimabuendnis’ (1 if the municipality is a member of Klimabündis Austria, 0 otherwise) and
‘POP_years_participation_fahradberatung’ the number of years since the municipality first enrolled to the
fahrradberatung.at programme are significant in the model context and the related coefficients generally suggest
that political/administrative commitment in favour of cycling has a positive effect on the modal spilt share of
cycling trips. More specifically, for every year since the first enrolment to fahradberatung.at the municipality
gains a 0.11% increase in cycling share, i.e. after approx. 9 years of taking part in the initiative the cycling share
will increase by 1%. Given that the average municipal cycling share is at some 3.5% proves that the programme
does have an impact. In a similar fashion the enrolment to Klimabündnis will increase the cycling share by
0.22% constituting a one-time effect. It needs to be stressed here that these figures are incremental meaning that
they reflect the net effect of the respective predictor while all other variables are kept constant. In this sense
supporting planning actions affecting any of the other thematic areas will add up to a more pronounced increase
in cycling modal share.
Regarding the former, we tested additional variables for inclusion before committing to the final variant of the
model: the number of Klima-aktiv supported projects in the walking/cycling domain positively correlates with y
(number of projects: +0.144, no. of projects by municipal area: +0.267, both correlations are significant at a level
of 0.001). The subjective evaluation of the state administration on the municipal level of pro-cycling activity (on
a scale between 0 and 3; 3 is best) proved to be positively correlated with y (+0.226, significant at a level of
0.001). However, when controlling for the many other determinants affecting cycling modal shares those
variables turned out not to be significant in the regression model and have consequently been excluded.
Complementary analysis: correlation between subjective and objective commitment towards cycling
In order to deepen the analysis of proxy variables depicting commitment towards cycling among local political
or administrative representatives we analysed whether or not there is a correlation with the self-assessment of
these stakeholder groups. We used the surveys response data (see 2.1) for the calculation of a summed affinity
index of self-assessed political willingness to promote cycling. To solve the question whether the included proxy
variables (eg. ‘POP_years_participation_fahrradberatung’) actually reflect the self-stated commitment, we
applied correlation analyses between the above affinity index and the model proxy variables as well as the
cycling modal share (s. table 4)
3
. The resulting correlation coefficients are highly significant. The medium-sized
correlation (R=0.448) between cycling modal share and the affinity index underlines both the introducing
statement by Raffler et al.(2019) and the findings of the cycling model: A strong political or administrative
commitment towards cycling is key to achieve high local cycling shares; however it is not the only relevant
factor. This is also emphasized by the model outputs (see table 3) which list the proxy variables as relevant
determinants among other factors.
Table 4. Pearson correlation [R] between proxies and stated-political willingness
Affinity to cycling
R
p
Cycling modal share
0.448
0.000
Number of Klimaaktiv mobil Projects
0.408
0.000
Years since 1st enrolment in Fahrradberatung.at
0.384
0.000
Years since 1st enrolment in Klimabündnis
0.537
0.000
Further inspection of table 4 shows that there exists significant (p < 0.001) medium to high positive correlations
between the proxy variables and the surveyed affinity values. Bearing in mind that the affinity index is a highly
subjective measure, the strength of the relationship with the proxy variables (being based on objective data) is
quite high. This indicates that measurable variables such as enrolment in federal cycling promotion programs
actually reflect the self-assessed political will to promote cycling which is an argument in favor of calculating
meaningful proxy variables using existing data sources and including them in aggregated models.
Decision support and measure simulation
Our research is guided by aiming at outcomes that can actually be implemented in planning processes and
agenda setting. Thus we need to come up with approaches dedicated to translating raw statistical model results
into planning practice. As a first component of a decision support system for the Upper Austrian federal state
government we developed ‘active travel potential maps’ such as presented in figure 3 for pedestrian traffic.
Methodologically these maps are based on analysing the model residuals produced by the above aggregate
3
Due to the missing state wide coverage of the municipality survey it was not possible to directly test the affinity index in the context of the
cycling model.

models. As not the whole variance in modal shares could be accounted for in the models (and hence by the
covariates included therein) there are model residuals: a positive residual indicates that the subject municipality
has a higher modal share than expected given the local premises (environmental, infrastructural, social &
attitudinal, etc.). Reversely, a negative residual can be found in areas that could potentially achieve higher active
modal shares if they made best use of the local conditions. Put differently, they underachieve when it comes to
active mobility. As a planning tool the maps are currently being used as a means to support strategic decisions
related to the extension of the Upper Austrian cycling promotion programme fahrradberatung.at.
Fig. 4. Map of walking-model residuals and their statistical and spatial distribution
Negative residuals = large unused walking-potential | Positive residuals = best use of local walking conditions
From a planning perspective the areas displayed in red mark target municipalities that are likely to produce the
highest return on investment in terms of pedestrian modal shares while municipalities displayed in blue mark
target regions with a potential to balance out Upper Austrian disparities in walking shares (however, at the price
of reduced incremental return of investment). However, the actual choice of investment strategy is ultimately a
political matter rather than a scientific one: Both options are equally viable and highly dependent on the
respective political agenda. In this context we see two critical questions that need to be answered by decision
makers on the federal state level:
1. Should overall active shares of municipalities be levelled out among all communities?
2. Should measures to boost active shares focus on municipalities that already implemented a thriving
culture of walking and/or cycling?
The first option means investing in underperforming municipalities displayed in blue with a potential to balance
out Upper Austrian disparities in walking shares. Those municipalities exhibit lower walking shares than could
be expected when fully utilizing their respective local conditions (infrastructure, population, topography and
climate). In case decision makers choose to follow the second strategy, the recommendation is to invest in areas
displayed in red. Target municipalities are likely to produce the highest return on investment in terms of
pedestrian modal shares by building on an already established culture of active mobility.
Those strategic decisions pose a new scope for Austrian decision makers as quantitative measures of
municipality performance in the context of modal choice by means of statistical modelling haven't yet been
applied in Austrian active mobility planning. A second element of supporting decisions in active travel planning
comprises the simulation of potential measures in terms of their expected impact on pedestrian or cycling modal
shares prior to their implementation. This can support planning, e.g. by prioritizing potential measures subject to
their impact or target achievement. Measure simulation is methodically facilitated by using the above models,
adequately interpreting coefficients and entering modified values (according to the planning measure to be
assessed) into the model equations. It should be noted however, that the measurement scale and dimensionality
of the covariates as well as whether or not they are composite variables largely determine the way model

coefficients need be interpreted in this context. Single variables measured on a metric percentage scale are most
straightforward in terms of interpretation whereas composite variables or non-dimensional measurement scales
require some preparatory work when simulating planning actions. To quote an example, increasing the number
of bike racks at rail stations by 10 per 1000 inhabitants (currently amounting to a mean of 5 per 1000 inhabitants
in Upper Austrian municipalities) will increase the cycling modal share by 0.17% (amounting to approx. 3.5%
for the average Upper Austrian municipality) while other things are kept equal. Cutting the density of accident
hotspots in half (e.g. by investing in construction measures to defuse accident accumulation points) will increase
the cycling modal share by 0.22%. With regards to walking, the significance of interdepartmental action
becomes apparent, particularly relating to policies that affect the composition of the local population: Increasing
the share of post-materialists in the local population (e.g. by specifically attracting respective households through
housing schemes, city marketing, etc.) by 10% will increase the pedestrian modal split by 1.1%. Increasing the
population share of children below 15 years increases the walking share by 0.28%. By quoting these examples of
cross-sectoral planning measures we aim at pointing out that influencing modal shares is by far not limited to
traditional measures usually concerning infrastructure or accessibility, but can also be implicitly facilitated to a
great extent by the composition of the local population (which again is subject to the municipalities'
attractiveness for certain groups). The local share of part-time working relationships among employed men has a
positive impact on walking shares: increasing the share of this kind of jobs by 1% (e.g. by attracting appropriate
businesses, introducing new worktime schemes or attracting certain lifestyle types) walking shares increase by
0.17%. A potential explanation for this is the better temporal affordability for 'slower' modes of transport. As
another example, the composite walking accessibility variable coefficient can be generally interpreted as
increasing walking accessibility to relevant destinations will boost walking modal shares. However, since the
measurement scale is non-dimensional the exact impact simulation of improving accessibility needs to be based
on a re-calculation of the respective indicators after changing the path network and / or its attributes reflecting
the respective measures in the GIS model in a case-by-case fashion.
4. Conclusions and outlook
At this stage our work has demonstrated that aggregated statistical models for active travel modes are
methodologically feasible and that data-driven methods can actually be used to support planning and agenda
setting. First results prove that a considerable proportion of the observed variation in walking and cycling modal
shares can be explained by multivariate regression models including on a comprehensive set of covariates. In
accordance with the aims set in section 1.4, the added value of our research lies in the following results: (1) A
systematic approach to model active mode shares on a municipal level in Austria, therefore laying the
foundations for evidence-based decision making in walking and cycling domains as well as in other sectors with
relevance to mobility patterns; (2) Presentation of determinants on pedestrian modal shares, their strength and
direction of impact, in the context of transport planning; (3) The operationalization of proxy variables reflecting
the political will to promote cycling more realistically as well as the assessment of the correlational relationship
between those proxy variables with an affinity index generated from empirical data.
That being said, we are aware that considerable research tasks lie ahead. Aiming at making our approach highly
relevant for practical planning, widening its scope of application and improving the reliability of the model
results future research threads include both methodological aspects and developing implementation tools,
respectively. In terms of improving the models we aim at including additional predictor variables by forming
composite variables or factors, including data to operationalize on- and off road cycling infrastructure. Another
research goal lies in including non-linearity and saturation effects as well as in adding variables on
infrastructural qualities that were unavailable at the time of building the models. Broadening the statistical basis
by re-calibrating the model with data from other regions, both national and international (facilitated by a
dedicated data interoperability concept) is regarded key in terms of making the model results even more
generalizable and robust. In terms of transferring model results into planning practice we aim at developing a set
of tools including an expert system in order to make our approach workable for external experts in the planning
domain. This includes various interfaces for planning-relevant model input data as well as coherent ways of
presenting model outputs to the target groups.
Acknowledgements
This research was conducted in the framework of the ACTIV8! Project ('Aktive Mobilität effizient fördern‘) co-
financed under the Mobilität der Zukunft programme, ‚Personenmobilität innovativ gestalten,
Verkehrsinfrastruktur gemeinsam entwickeln’ (4th call). The authors are also grateful to the Office of the State
Government of Upper Austria, in particular to the Cycling Officer at the Department of the Overall Traffic
Planning and Public Transport, Christian Hummer for providing good advice and hard-to-find datasets as well as

for support during the survey. In addition we would like to thank the promoters of Radgipfel 2016, 2017,
Radvernetzungstreffen, AG Radbeauftragte and REAL CORP Conference 2017. Finally we thank Martin Eder
(BMLFUW), Holger Heinfellner (Umweltbundesamt) and Michael Bürger (Klimabündnis Tirol) for event
invitations.
5. References
Akar, Gulsah; Clifton, Kelly J. (2009): Influence of Individual Perceptions and Bicycle Infrastructure on
Decision to Bike. In Transportation Research Record 2140 (1), pp. 165–172. DOI: 10.3141/2140-18.
Aoun, Alisar; Bjorndstad, Julie; DuBose, Brooke; et al. (2015): Bicycle and Pedestrian Forecasting. State of the
Practice. Edited by Federal Highway Administration. edestrian and Bicycle Information Center. Chapel Hill.
Banister, David (2008): The sustainable mobility paradigm. In Transport Policy 15 (2), pp. 73–80. DOI:
10.1016/j.tranpol.2007.10.005.
BMLFUW (2009): Kurzstudie Wirtschaftsfaktor Radfahren. Die volkswirtschaftlichen Auswirkungen des
Radverkehrs in Österreich. Edited by BMLFUW - Bundesministerium für Land- und Forstwirtschaft, Umwelt
und Wasserwirtschaft. BMLFUW - Bundesministerium für Land- und Forstwirtschaft, Umwelt und
Wasserwirtschaft. Vienna. Available online at https://www.bmlfuw.gv.at/dam/jcr:c88ba588-5a4e-48a7-94ae-
2bfd88f0fb19/Studie%20Wirtschaftsfaktor_Radfahren.pdf, checked on 11/9/2017.
BMLFUW (2011): Masterplan Radfahren. Umsetzungserfolge und neue Schwerpunkte 2011 - 2015. Vienna.
BMLFUW (2015a): Masterplan Gehen. Strategie zur Förderung des FussgängerInnenverkehrs in Österreich.
Wien.
BMLFUW (2015b): Masterplan Radfahren 2015 - 2025. Wien.
BMVIT (2016): Österreich unterwegs 2013/2014. Ergebnisbericht zur österreichweiten Mobilitätserhebung
"Österreich unterwegs 2013/2014). BMVIT - Bundesministerium für Verkehr, Innovation und Technologie.
Vienna.
Buchwald, Konrad; et al. (Eds.) (1993): Umweltschutz - Grundlagen und Praxis. Bonn: Economica Verl.
Carinthian government (2016): Mobilitäts Masterplan Kärnten 2035. Amt der Kärntner Landesregierung.
Klagenfurt. Available online at https://www.ktn.gv.at/328812_DE-Dokumente-
Momak_Abschlussbericht_neu.pdf.
Caulfield, Brian; Brick, Elaine; McCarthy Orlad Thérèse (2012): Determining bicycle infrastructure preferences
- A case study of Dublin. In Transportation Research Part D: Transport and Environment 2012 (Vol. 17, Issue
5), pp. 413–417. Available online at https://www.sciencedirect.com/science/article/pii/S1361920912000363.
Cerin, Ester; Leslie, Eva; Owen, Neville (2009): Explaining socio-economic status differences in walking for
transport. An ecological analysis of individual, social and environmental factors. In Social science & medicine
(1982) 68 (6), pp. 1013–1020. DOI: 10.1016/j.socscimed.2009.01.008.
Cervero, Robert (2013): Transport Infrastructure and the Enviroment: Sustainable Mobility and Urbanism. In
2nd Planocosmo International Conference, pp. 1–20.
Cervero, Robert; Duncan, Michael (2003): Walking, Bicycling, and Urban Landscapes: Evidence From the San
Francisco Bay Area. In American Journal of Public Health 2003 (Vol 93, No. 0), pp. 1478–1483.
Dangschat, Jens; Segert, A.; et al. (2012): Der Milieuansatz in der Mobilitätsforschung. Vienna.
Davoudi, Simin (2006): Evidence-Based Planning. In disP - The Planning Review 42 (165), pp. 14–24. DOI:
10.1080/02513625.2006.10556951.
Faludi, Andreas (2006): Introducing Evidence-Based Planning. In disP - The Planning Review 42 (165), pp. 4–
13. DOI: 10.1080/02513625.2006.10556950.
FSV (2019): RVS 03.02.13 Radverkehr Februar 2014. Edited by FSV - Österreichische Forschungsgesellschaft
Straße - Schiene - Verkehr. FSV - Österreichische Forschungsgesellschaft Straße - Schiene - Verkehr. Vienna.
Available online at http://www.fsv.at/shop/produktdetail.aspx?IDProdukt=e283b686-009d-4cc8-bdf1-
8a06fd7fb2a7.
Garrard, Jan; Rose, Geoffrey; Lo, Sing Kai (2008): Promoting transportation cycling for women. The role of
bicycle infrastructure. In Preventive medicine 46 (1), pp. 55–59. DOI: 10.1016/j.ypmed.2007.07.010.

Goodman, Anna (2013): Walking, cycling and driving to work in the English and Welsh 2011 census. Trends,
socio-economic patterning and relevance to travel behaviour in general. In PloS one 8 (8), e71790. DOI:
10.1371/journal.pone.0071790.
Government of Upper Austria (2014): Ergebnisse & Schlussfolgerungen der
oberösterreichischen Verkehrserhebung. Government of Upper Austria. Linz. Available online at
https://www.land-oberoesterreich.gv.at/Mediendateien/LK/PK_LH-
Stv._Hiesl__LR_Entholzer_20.1.2014_Internet.pdf, checked on 9/6/2017.
Green, S. B. (1991): How Many Subjects Does It Take To Do A Regression Analysis. In Multivariate behavioral
research 26 (3), pp. 499–510. DOI: 10.1207/s15327906mbr2603_7.
Guell, C.; Panter, J.; Jones, N. R.; Ogilvie, D. (2012): Towards a differentiated understanding of active travel
behaviour. Using social theory to explore everyday commuting. In Social science & medicine (1982) 75 (1),
pp. 233–239. DOI: 10.1016/j.socscimed.2012.01.038.
Hackl, Roland; Raffler, Clemens; et al. (2017): Measuring political commitment in statistical models for
evidence-based agenda setting in non-motorized traffic. In Manfred Schrenk, Vasily Popovich, et al. (Eds.):
REAL CORP 2017 Proceedings. REAL CORP Conference. 1 volume. Vienna.
Hackl, Roland; Raffler, Clemens; et al. (2018): What makes and breaks active travel? A statistical model for
evidence-based decision-making in transport policy for nonmotorized modes. Transport Research Arena 2018.
Heinen, Eva; Maat, Kees; van Wee, Bert (2013): The effect of work-related factors on the bicycle commute
mode choice in the Netherlands. In Transportation 40 (1), pp. 23–43. DOI: 10.1007/s11116-012-9399-4.
Heinen, Eva; van Wee, Bert; Maat, Kees (2010): Commuting by Bicycle. An Overview of the Literature. In
Transport Reviews 30 (1), pp. 59–96. DOI: 10.1080/01441640903187001.
Knoflacher, Hermann (2013): Zurück zur Mobilität. Wien: Ueberreuter.
Kordi, Maryam; Kaiser, Christian; Fortheringham, A. Steward (2012): A possible solution for the centroid-to-
centroid and intra-zonal trip length problems. In Multidisciplinary Research on Geographical information in
Europe and Beyond 2012, pp. 147–152.
Leslie, Eva; Saelens, Brian; Frank, Lawrence; Owen, Neville; Bauman, Adrian; Coffee, Neil; Hugo, Graeme
(2005): Residents' perceptions of walkability attributes in objectively different neighbourhoods. A pilot study. In
Health & place 11 (3), pp. 227–236. DOI: 10.1016/j.healthplace.2004.05.005.
Lovelace, Robin; Goodman, Anna; Aldred, Rachel; Berkoff, Nikolai; Abbas, Ali; Woodcock, James (2017): The
Propensity to Cycle Tool. An open source online system for sustainable transport planning. In JTLU 10 (1). DOI:
10.5198/jtlu.2016.862.
Merki, Christoph Maria (2008): Verkehrsgeschichte und Mobilität. Stuttgart: Hirzel Verlag.
Meschik, Michael; Traub, Robert (2008): Meschik_Planungshandbuch_Radverkehr // Planungshandbuch
Radverkehr. Vienna: Springer-Verlag/Wien.
Parkin, John; Ryley, Tim; Jones, Tim (2007a): Barriers to Cycling: An Exploration of Quantitative Analyses. In
Dave Horton, Paul Rosen, Peter Cox (Eds.): Cycling and Society. Burlington: Ashgate Publishing Limited,
pp. 67–82.
Parkin, John; Wardman, Mark; Page, Matthew (2007b): Estimation of the determinants of bicycle mode share
for the journey to work using census data. In Transportation 35 (1), pp. 93–109. DOI: 10.1007/s11116-007-
9137-5.
Perschon, Jürgen (2012): Nachhaltige Mobilität. Handlungsempfehlungen für eine zukunftsfähige
Verkehrsgeschichte. Development and peace foundation. Bonn.
Pucher, John; Buehler, Ralph (2006): Why Canadians cycle more than Americans. A comparative analysis of
bicycling trends and policies. In Transport Policy 13 (3), pp. 265–279. DOI: 10.1016/j.tranpol.2005.11.001.
Raffler, Clemens (2016): Untersuchung des Körperenergieverbrauchs als evidenzbasierter Ansatz zur
Unterstützung der Radverkehrsplanung. Betrachtung andhand österreichischer Berufspendlerdaten 1971 - 2001.
Master Thesis. University of Technology Vienna, Vienna. Fachbereich für Verkehrsplanung und
Verkehrstechnik. Available online at
http://repositum.tuwien.ac.at/obvutwhs/download/pdf/1529420?originalFilename=true.
Raffler, Clemens; Brezina, Tadej; Emberger, Günter (2019): Cycling investment expedience. Energy expenditure

based Cost-Path Analysis of national census bicycle commuting data. In Transportation Research Part A: Policy
and Practice 121, pp. 360–373. DOI: 10.1016/j.tra.2019.01.019.
Rietveld, Piet; Daniel, Vanessa (2004): Determinants of bicycle use. Do municipal policies matter? In
Transportation Research Part A: Policy and Practice 38 (7), pp. 531–550. DOI: 10.1016/j.tra.2004.05.003.
Rybarczyk, Greg; Wu, Changshan (2014): Examining the Impact of Urban Morphology on Bicycle Mode
Choice. In Environ Plann B Plann Des 41 (2), pp. 272–288. DOI: 10.1068/b37133.
Saelens, Brian; Sallis, James; Frank, Lawrence (2003): Environmental Correlates of Walking and Cycling:
Findings From the Transportation, Urban Design, and Planning Literatures. In The Society of Behavioral
Medicine 2003 (Volume 25, Number 2,), pp. 80–90.
Sinus Markt- und Sozialforschung GmbH (2017a): Fahrrad-Monitor Deutschland 2017. Ergebnisse einer
repräsentativen Online-Befragung. Sinus Markt- und Sozialforschung GmbH. Heidelberg. Available online at
https://www.bmvi.de/SharedDocs/DE/Anlage/G/fahrradmonitor-2017-ergebnisse.pdf?__blob=publicationFile.
Sinus Markt- und Sozialforschung GmbH (2017b): Wertemilieus und Radverkehr. Sinus Markt- und
Sozialforschung GmbH. Available online at https://nationaler-radverkehrsplan.de/sites/default/files/pdf/2017-11-
06_10-fahrradkommunalkonferenz_ag1_thementisch1_jurczok.pdf.
Sinus Markt- und Sozialforschung GmbH (2019): Sinus Milieus in Austria. Sinus Markt- und Sozialforschung
GmbH. Vienna. Available online at https://www.sinus-institut.de/sinus-loesungen/sinus-milieus-oesterreich/.
Souza, Adriana A. de; Sanches, Suely P.; Ferreira, Marcos A.G. (2014): Influence of Attitudes with Respect to
Cycling on the Perception of Existing Barriers for Using this Mode of Transport for Commuting. In Procedia -
Social and Behavioral Sciences 162, pp. 111–120. DOI: 10.1016/j.sbspro.2014.12.191.
Tresidder, Mike (2005): Using GIS to Measure Connectivity. An Exploration of Issues. School of Urban Studies
and planning. Portland. Available online at
http://www.web.pdx.edu/~jdill/Tresidder_Using_GIS_to_Measure_Connectivity.pdf, checked on 9/7/2017.
Tyrolian Government (2014): Radkonzept Tirol. Themenfeld A - Infrastruktur. Tyrolian Government. Innsbruck.
Available online at
https://www.tirol.gv.at/fileadmin/themen/verkehr/verkehrsdatenerfassung/downloads/RadkonzeptTirol_Bericht_
20141119.pdf, checked on 9/7/2017.
van Acker, Veronique (2015): Defining, Measuring, and Using the Lifestyle Concept in Modal Choice Research.
In Transportation Research Record 2015 (1), pp. 74–82. DOI: 10.3141/2495-08.
Vandenbulcke, Gregory; Dujardin, Claire; Thomas, Isabelle (2008): Cycling to work: Modelling spatial
variations within Belgium 2008, pp. 1–23.
Verhoeven, Hannah; Simons, Dorien; van Dyck, Delfien; van Cauwenberg, Jelle; Clarys, Peter; Bourdeaudhuij,
Ilse de et al. (2016): Psychosocial and Environmental Correlates of Walking, Cycling, Public Transport and
Passive Transport to Various Destinations in Flemish Older Adolescents. In PloS one 11 (1), e0147128. DOI:
10.1371/journal.pone.0147128.
Verracon GmbH (2016): Quaravo. Qualitätsbewertung des Alltagsradverkehrs in Vorarlberg. Edited by
Energieinstitut Vorarlberg. Energieinstitut Vorarlberg. Wien. Available online at
https://www.vorarlberg.at/pdf/endbericht_erreichbarkeit.pdf, checked on 9/6/2017.
walk21 (2017): Internationale Charta für das Gehen. walk21. Available online at http://www.walk-
space.at/wissen/Charter.pdf, checked on 8/28/2017.
Wardman, Mark; Tight, Miles; Page, Matthew (2007): Factors influencing the propensity to cycle to work. In
Transportation Research Part A: Policy and Practice 41 (4), pp. 339–350. DOI: 10.1016/j.tra.2006.09.011.
Winters, Meghan; Friesen, Melissa C.; Koehoorn, Mieke; Teschke, Kay (2007): Utilitarian bicycling. A
multilevel analysis of climate and personal influences. In American journal of preventive medicine 32 (1),
pp. 52–58. DOI: 10.1016/j.amepre.2006.08.027.
Winters, Meghan; Teschke, Kay (2010): Route preferences among adults in the near market for bicycling.
Findings of the cycling in cities study. In American journal of health promotion : AJHP 25 (1), pp. 40–47. DOI:
10.4278/ajhp.081006-QUAN-236.