Modelling scenarios in planning for future employment growth in Stockholm

.


Introduction
Through its long-term planning program, the City of Stockholm is evaluating how to best shape the geographical distribution of new office development.The City's goals include reducing the environmental impact of transportation, increasing accessibility, and combating segregation and social exclusion while accommodating anticipated future growth.Where workplaces are located will have an impact on both accessibility and equity in Stockholm by influencing who has access to those workplaces and shaping the commutes of eventual employees.Stockholm has developed a scenario planning exercise which considers three scenarios.The status quo development pattern is central and concentrated, with development hubs close to the city centre.The alternative scenarios are dispersed central development, in which many less-developed areas close to the central city are built up, and concentrated peripheral development, in which the city develops a few office park nodes in a polycentric pattern.
As part of the City's planning exercise, we model plausible futures for each of the three scenarios, exploring how the location of new workplaces impacts the daily activity and travel patterns of their workers.We use a modelling framework based on that of Naqavi et al. (2023), which incorporates a model for individuals' work location choices with a dynamic full-day scheduling model, SCAPER (Karlström, 2005;Jonsson et al., 2014;Blom Västberg et al., 2019).This paper presents an integrated application of the work location and dynamic scheduling models to explore the differences among our three scenarios for future office development patterns.We simulate the work locations and daily activity-travel behaviour of a synthesized Stockholm population under the assumptions of each scenario, then compare the groups of newly placed workers under each scenario.
To evaluate the results of our modelled scenarios, we ask three primary questions for analysis.First, who benefits most and least from each scenario?Using the logsum of our work location model as a measure of consumer welfare (measured as consumer surplus), we analyze how each scenario impacts patterns of welfare across socioeconomic and spatial groups.Second, who gets access to employment under each scenario?We examine the socioeconomic and geographic distributions of expected scenario workers to understand how the three scenarios will likely draw from different income groups and areas of the city.Finally, how does each scenario impact daily behaviour?We look at how Our theoretical conception of welfare comes from the microeconomic study of travel behaviour modelling.An individual's consumer welfare (consumer surplus) is informed by the quality of options available to them, with 'quality' measured by a utility function that accounts for (generalized) costs and individual-specific preferences.This conception of welfare is also often called accessibility: in our model, consumer welfare measures the individual's access to jobs and to the underlying opportunities for non-work travel provided by a particular work location.The theoretical foundation of consumer welfare, measured as consumer surplus, in discrete choice modelling was first developed by McFadden (1977McFadden ( , 1981)), Cochrane (1975) and Williams (1977); De Jong et al. (2007) and Karlström (2014) offer reviews.
While consumer welfare measures in discrete choice modelling are useful for equity analyses since they observe heterogeneous effects on individuals (Nahmias-Biran and Shiftan, 2016), their use in the analysis of distributional effects is limited, and indeed most transportation evaluation makes use of the cost-benefit analysis framework, which does not account for distributional effects (Van Wee and Geurs, 2011).More recently, several authors have made strides to rectify this deficiency, using the 'logsum' welfare output of activity-based models for equity analysis of policies including congestion charges (Kristoffersson et al., 2017), transportation demand management strategies (Hasnine and Habib, 2020), and shared automated vehicle mobility services (Ahmed et al., 2020).Ma and Kockelman (2016) use the welfare output of a home location choice model to investigate various transportation policy interventions.The methodological work this paper is based on, Naqavi et al. (2023), explored the equity-related impacts of car dependency and segregation in the case of a hypothetical relocation of a major hospital.
While our paper takes a novel approach to investigating the impact of different patterns of office development on welfare and equity, other fields have taken different perspectives on the issue.In particular, there is an emerging literature in regional economics on polycentrism and regional economic inequalities.Reducing regional inequality is a major policy goal in Europe (Meijers and Sandberg, 2021) and there is keen interest in policy options in the area.Empirical studies generally use aggregate regional data at the country level and have produced mixed evidence.Studies have shown a negative connection between polycentric regions and regional equality in China (Sun et al., 2019), ambiguous results in Czechia (Malỳ, 2016) and a positive link in Germany (Li et al., 2024).Our approach is not directly comparable to these studies as we investigate polycentric development at the level of a city using an agent-based approach; however, it offers insight into how polycentric or dispersed development impacts spatial equity through the choices of individuals in the population.

Urban development patterns and travel behaviour
Urban development patterns are closely linked with travel behaviour: commute mode choice, for instance, is influenced by the relative accessibility of transit vs. roads, the availability and cost of parking, and employees' distances to work (Aarhus, 2000).How these influences manifest in different development patterns is a topic of much debate, especially around concentration versus 'sprawl' in urban environments.In one camp, authors such as Ewing (1997) decry urban sprawl and advocate for more compact patterns of development that lead to more efficient and environmentally-friendly transport outcomes.Detractors such as Gordon and Richardson (1997) are bullish about decentralized development, arguing that firms and households will tend to strategically co-locate to reduce commuting distances in non-monocentric cities.However, empirical evidence shows that the actual relationship between where people live and work can bear little resemblance to the 'optimal' relationship from the perspective of minimizing commuting times (Giuliano and Small, 1993).Even where housing and jobs are colocated, the residents near those jobs may not be the ones working in those jobs, and households and offices may move for reasons external to employee commutes.
Following Tsai (2001) we will separate the compact/sprawl continuum into two distinct axes: mono-versus polycentric and centralized versus dispersed development.Polycentric cities have multiple business districts with substantially higher job density than the surrounding neighbourhoods.Most major cities start developing secondary employment hubs when it becomes impractical for many businesses to locate centrally due to, e.g., high commercial rents and congestion (Garcia-López and Muñiz, 2010).Research on the impact of polycentrism on travel behaviour has mainly focused on mode share and commuting.On mode share, studies have both found greater polycentrism to be connected with increased private car use (Schwanen et al., 2001;Wolday et al., 2019) and without a substantial mode shift to private vehicles (Bolotte, 1991;Kitamura et al., 2003); the difference may be due to investments in public transit infrastructure.On commuting times, Levinson and Kumar (1994) found average travel time to work remained constant with an evolution to polycentrism in Washington DC, Schwanen et al. (2003) found that polycentrism raised commute times in the Netherlands, and Jun (2020) found mixed impacts.
Many major cities are also undergoing processes of employment dispersion, likely due to improvements in private transport and communications which reduce the importance of agglomeration economies and thus the need for businesses to co-locate in concentrated areas (Garcia-López and Muñiz, 2010).A dispersed pattern of development is one in which there is a spread of employment opportunities across a regional area which is not concentrated into polycentric areas.However, empirical evidence suggests there may be downsides.In Italy, Veneri (2010) finds that dispersion is associated with shorter travel durations but less sustainable mobility patterns due to higher emissions.Susilo and Maat (2007) concur in a study of the Netherlands, finding that dispersion reduces car congestion leading to lower commute times than urban concentration, but reduces the competitiveness of public transit for commuting.
The work relocation literature offers empirical insights by studying individuals or groups who have recently changed work locations.In a literature survey, Maheshwari et al. (2023) documents substantial travel behaviour, sustainability and well-being impacts from the relocation of an individuals' workplace.Particularly relevant is that when workplaces S. McCarthy et al. moved from suburbs to the city centre, their employees had a drop in private vehicle use and and increase in sustainable modes, and vice versa (Yang et al., 2017;Cumming et al., 2019).The field has also investigated impacts of work relocation on non-work activities.Bell (1991) established that workplace change can impact the frequency and timing of performing different activity types, reporting that a change from city to suburban location in Australia saw a 10% reduction in activity participation among survey respondents.Sprumont and Viti (2018) investigated spatial distribution, showing that after a work location change, the spatial distribution of individuals shifted away from the old workplace and toward the new one.In households, the distribution of tasks can be influenced by relative work locations (von Behren et al., 2018); for instance, a household member whose work is conveniently located near a grocery store may have a greater responsibility for grocery shopping.

Scenario development
Located in south-eastern Sweden, the Stockholm metropolitan area is the largest in Scandinavia and is expected to grow from around 2.4 million to 3.4 million people by 2050 (Stockholm, 2018).Stockholm is situated where Lake Mälaren to the west of the city meets the Baltic Sea to the east, creating a city spread across islands and peninsulas with a north-south travel bottleneck through the city centre.The city has historically developed in a monocentric pattern, with a central and concentrated business district which attracts commuters from a wide surrounding area.The metro network has a hub-and-spoke layout, with all lines connecting at the central downtown station, as shown by Fig. 1a.The road network has greater east-west and north-south connectivity, as Fig. 1b shows.Stockholm has had a cordon-based congestion charge since 2006 that surrounds the central city as shown on the map.
The City Planning Office of Stockholm City is exploring whether the city's historical centralized and concentrated development pattern remains optimal for Stockholm.Throughout the 20th century, Stockholm's land use planners played a leading role in the development of Stockholm city, deciding on where and how development should happen through comprehensive and detailed plans (Zakhour and Metzger, 2018).Swedish urban planning practice contains strains of thought, from the ABC City movement in the 1950s1 to today's ideas around sustainable urban development, which argue for closer co-location of workplaces and homes Bäckström (2014).While recent decades have seen a swing toward a developer-led approach to growth, there is evidence that Stockholm is once more moving back to the "active planning tradition" (Zakhour and Metzger, 2018), in part by considering citywide planning initiatives such as the one this research contributes to.
The Planning office is conducting a scenario planning exercise using the scenario-axis technique ( Van der Heijden, 2005).They have identified two axes on which patterns of future office development may change: centralization (central ↔ peripheral), which reflects how close new developments are to the centre of Stockholm, and concentration (concentrated ↔ dispersed), which reflects whether there are fewer, larger developments or more smaller ones.While office development patterns are ultimately decided by firms' decision-making, international evidence shows that planning decisions are important to geographic employment growth rates (Shearmur and Alvergne, 2003), making the outcome of this planning process an important factor in the development of Stockholm's future development.
For each quadrant, City planners selected areas of the city which reflected that development pattern; Fig. 2 shows these areas arranged on the axes.Using their professional judgment, the planners selected areas with reasonable development prospects: for example, available space and reasonable access to transit.The scenarios were constructed with some sense of what was economically possible in the next few decades, but are not meant to be reflections of market demand.The scenarios are: • Status Quo: reflects the historical centralized/concentrated development patterns.This scenario includes large new developments in Norra Djurgårdsstaden, a short distance to the east of the city centre, as well as a substantial development in the Söderstaden (Gullmarsplan and Globen) neighbourhoods just south of Södermalm and moderate development in Ulvsunda to the northwest and Södertäljevägen to the south.As the centre is effectively saturated with workplaces, all these developments occur in areas close to but outside the historical core of Stockholm.The fourth quadrant is a pattern of decentralized and dispersed growth, where larger numbers of smaller new developments would flow into communities farther from the city centre.As this pattern would run against prevailing economic trends toward both centralization and concentration, the City has ruled it out as a plausible scenario.
Next, City planners constructed numerical scenarios to assign an amount of new floor space for office development to the selected communities, considering both the scenario's position on the axes and the capacity the areas have for new development.They converted these into a number of new potential jobs for each area using the average floor space per worker.The scenarios reflect approximately 35,000 new jobs each, which are distributed as shown in Fig. 3. Finally, we distributed the new jobs into the broad occupational categories used in our model using a simple distribution of 80% 'business', 10% retail and wholesale shops, and 10% non-assigned other categories.This was determined from an empirical observation of the distribution of occupations at the Kista business park, which we anticipate to be roughly similar to the new jobs created in our scenarios; an important assumption as it underlies the connection with agent income levels in our model.We take our definition of 'business' from the national Swedish transportation demand model; it combines several occupational categories from the Swedish national statistics agency, including real estate, business activities, finance, transport, storage, and communication.

Modelling approach
To answer the questions set out in Section 1, we model the behaviour of Stockholm County residents, simulating their decisions around work location and travel behaviour.We adopt a modelling approach which considers related choices across two planning horizons: the longer term of work location choice, and the short term of daily activity and travel behaviour.We use a modelling framework developed by Naqavi et al. (2023) which treats the work location and daily activity choice as two levels of a nested logit model.Nested models are a common way to pass information between different levels of decision-making within discrete choice models (Ben-Akiva and Bowman, 1998).In our model, the longer-term choice of work location uses information about how each work location would impact the agent's daily activities, and the shorter-term daily activity decision is made with reference to the chosen work location.See Fig. 4 for a graphical overview of the modelling approach.
Spatial equity is a key consideration to our modelling approach.As we are investigating distributional effects at the intersection of socioeconomic status and geography, our models are designed to be sensitive to the differences in the choice of location by income level-both at the higher level of work location and the lower level of daily activity location.Our models differentiate between low (under 28,000 SEK/month), medium (28,000-80,000 SEK/month) and high (above 80,000 SEK/ month) income individuals as described below.
As  and for a recent example see, e.g., Nurul Habib (2018).SCAPER is based in both the economic discrete choice theory of random utility models (McFadden, 1974) and the time-geography of Hägerstrand (1970).In it, agents are forward-looking, making sequential choices while aiming to maximize utility across the whole day.The model endogenously respects time and space constraints, producing realistic and behaviourally realistic travel behaviour patterns.This allows changes in the transportation system or land use to impact agents' choices in multiple ways, including activity participation, mode choice, and departure times.As well as work activities, SCAPER considers non-work activities that workers might perform on a work day, meaning that our measure of accessibility used by the work location model encompasses the agent's full day, a unique feature among similar models (Naqavi et al., 2023).SCAPER has been developed by Karlström (2005) The top level of the model, which decides the agent's work location, is constructed similarly to that of Naqavi et al. (2023) but with reestimated parameters shown in Appendix A. This level considers two factors: the potential benefit of each work location to the agent's daily activities (SCAPER accessibility), and the opportunities available in that work location across several occupational categories (Employment logsum).To reflect divisions in job opportunities by socioeconomic status, we estimate different model parameters for each income group.This sensitivity to income is important in modelling our office development scenarios, as we find the 'business' occupational category our scenarios are targeting is more important to higher-income agents compared to other occupations.
We apply our model to a 100% scale synthetic population of the workforce (1.1 million agents) in the greater Stockholm region in 2040.The synthetic population respects the aggregate spatial distributions of income and other demographic variables projected for Stockholm's Regional Development Plan (Stockholm, 2018), the region's best understanding of how it will develop over the next few decades.While the synthetic population is representative of the real population when aggregating over traffic analysis zones or socio-demographic groups, it does not represent real individuals.We also use expected spatial job data from the Plan as our baseline scenario for comparison.Our scenario jobs are added on top of this baseline.
The model's geographic unit of analysis is the traffic analysis zone, which represents a few blocks in dense urban areas, a local neighbourhood in suburban areas and a settlement in rural areas.We use the same zone system as Sweden's national travel demand model, SAMPERS2 ; our modelling area comprises 1375 zones.To simulate the integrated work location and scheduling models at such a large scale, we use importance sampling of zones within SCAPER, described by Saleem et al. (2018).The network levels of service in our model are exogenous inputs.For travel times, we take calibrated outputs from the base SAMPERS model.That model distinguishes between peak and non-peak travel times for car and transit modes; we use its peak travel times between 6:30 and 9:00 and between 15:30 and 18:00 in our model.Travel times for bike and walk modes are based on network distance between zones with a constant speed of 15 km/h for biking and 4 km/h for walking.Travel costs for car include a distance-based component and Stockholm's congestion charges when agents travel across the cordon (see Fig. 1b).Transit travel cost reflects Stockholm's flat-fare system with discounts for monthly card holders, and the model assumes no cost for walking or biking.
We run the work location model on the full synthetic population for each policy scenario and for the baseline scenario.From each model run, we obtain information about each agent which can be aggregated in many ways for analysis.Specifically, for each agent we obtain the expected utility value as a logsum (McFadden, 1998).Comparing this measure for each scenario with the baseline scenario serves as a measure of improvement in welfare and a proxy for change in consumer surplus.(See, e.g.De Jong et al. (2007) for a review of logsums in evaluation.)We also obtain the probability that the agent will have one of the newly added scenario jobs, allowing us to build up a picture of what we expect the cohort working in each scenario's added jobs to look like, from income distribution to car ownership to home locations.
To generate and analyze activity and travel behaviour, we then make a specific choice of work location for each agent.We do this over the whole synthetic population for each of the three scenarios, choosing a work location from the scenario zones according to that agent's probability of working in that zone conditional that they are a scenario worker, i.e. that they take one of the jobs created under the scenario.Using SCAPER, we then generate daily activity patterns for all agents.For analysis, these travel patterns are weighted by the probability that the agent takes a new scenario job.In this way, we obtain a properly weighted picture of how we expect scenario workers to behave using the whole synthetic population.This approach reduces simulation error compared to the naive approach of using only 35,000 selected scenario workers for analysis.In generating travel activity patterns, we ensure that all agents perform a work activity in their chosen work location; a Fig. 4. Overview of our modelling approach showing the inputs, modelling processes and outputs for each agent.The synthetic population is made up of households with home locations and household and individual characteristics.The models, shaded in gray consider the agent's home location, whether they have access to a vehicle, their income level, and the presence of children in the household.
choice to work remotely is outside the scope of this project.Finally, for geographical analysis each home and plan location was assigned coordinates uniformly randomly from the appropriate zone; this prevented spikes in the kernel density estimations used in analysis which would have occurred by using zone centroids.

Distribution of welfare benefit
We first look at our model's evaluation of how each scenario improves welfare across Stockholm's population.As discussed above, the expected utility of our work location model is a measure of consumer welfare and a proxy for consumer surplus.The nested logit structure of the model combines the opportunities agents have for finding work and the benefit to their daily activity patterns through the SCAPER accessibility measure.
Fig. 5 shows the average improvement in welfare per capita by zone due to the new work opportunities in each scenario, showing where the new workplaces will make the most difference given existing levels of accessibility and work opportunities.The darker green areas around the sites of new developments show the highest welfare improvement, but the new jobs have impacts beyond the local neighbourhoods where they are located.We can interpret the darker places as 'desire' areas where it would be most beneficial to live to take advantage of the new scenario jobs.The three scenarios show substantial differences in pattern.The Status Quo scenario (Fig. 5a) stands out as the most different of the three.Its main welfare benefit is clustered in the northeast of the city, being highest in Norra Djurgårdsstaden and having a tail of improvement out to the island of Lidingö to the northeast.In the Distributed Development scenario (Fig. 5b), the benefit is more distributed across the areas in the southwest of the city and in the municipalities to the south.As the scenario developments are more tightly concentrated in the Peripheral Hubs scenario (Fig. 5c), there are also more concentrated peripheral welfare benefits, especially in Älvsjö and Skärholmen in the southwest.In all scenarios, little welfare improvement is seen in the northwest and southeast areas of Stockholm City.
There is also low welfare improvement downtown due to an abundance of existing job opportunities there, which highlights a key relationship between welfare improvement and Stockholm's geography.In all scenarios, the locations of new jobs have a higher influence on welfare in the direction away from downtown.This is because areas closer to downtown have better access to downtown jobs, and therefore new jobs on the periphery make less difference to the prospects of people living there.In contrast, people living further from the central city have worse access to downtown jobs and will gain more from jobs which are closer to them.This relationship can also be seen in the difference in welfare improvement around peripheral locations where many jobs already exist and where the scenarios would establish relatively new office areas.In the Peripheral Hubs scenario, for example, the Kista neighbourhood in the north is an existing office park, and the scenario would add only a fraction of the existing jobs in that area.In Skärholmen in the southwest, the scenario would quadruple the existing job opportunities in the neighbourhood.Consequently, the welfare improvement in the areas around Skärholmen is distinctly higher than in the areas around Kista.
The distribution of welfare benefit by income group is shown in Fig. 6.The three scenarios do not differ much in the overall pattern of how they distribute consumer surplus to income groups; in all three, high-income agents derive the most benefit and low-income agents the least.This is an expected result since our work location model parameters showed that office development jobs were more important to the higher income groups.There are differences in the scenarios within groups, however.The Peripheral Hubs scenario results in the lowest median welfare improvement for each income group, and the Dispersed Development scenario results in substantially higher variation in welfare improvement across groups.These patterns are linked to geography as seen in Fig. 5: the highest welfare improvements accrue to individuals living near the new workplaces, especially away from central Stockholm.Spreading out development means that more residents are close to some of the new jobs, resulting in a longer tail of positive welfare differences.

Profile of scenario workers
We now consider the impact that the scenarios' expected welfare benefits will have on who ends up working in the jobs created under each scenario.For this analysis, we weight the entire synthetic workforce by the probability our model calculates that they will be a scenario worker.
As expected given the differences in spatial distribution of welfare improvement, the three scenarios differ substantially in the distribution of scenario workers' home locations.Fig. 7 displays heatmaps with our model's expected distribution of residences for each scenario.These follow the overall patterns of population density shown in Fig. 7a, weighted in the direction of the additional jobs.In the Status Quo scenario (Fig. 7b), adding the majority of new jobs in Norra Djurgårdsstaden leads to a high concentration of workers living in and around the city centre.The more distributed work locations in the Dispersed Development scenario (Fig. 7c) are reflected by a more evenly distributed pattern of home locations, without the concentration of agents in the inner neighbourhoods around downtown.The Peripheral Hubs scenario (Fig. 7d) is the most outward-oriented of the three scenarios but concentrates its jobs into three large areas of development.This is reflected in agents' home locations, which are more concentrated around the peripheral job areas.In the latter two scenarios, as most of the jobs are located in southwest Stockholm, the residential locations are also skewed toward the southwest.
Differences among the three scenarios in workers' home locations correspond somewhat to the scenarios' placement on the axis of concentration/dispersion of development.In both scenarios with concentrated development, the home distributions also have higher concentrations in the neighbourhoods surrounding the large work developments; in the Dispersed Development scenario home locations are The income distribution in each scenario, shown in Fig. 8a, reflects the distribution of welfare benefit by income group seen above.The three scenarios all have roughly the same proportion of workers by income group.All scenarios under-represent low-income agents and over- represent high-income ones since the office jobs targeted by the scenarios are linked with higher incomes.Fig. 8b shows the proportion of car ownership by scenario, which is more influenced by geography.Like home location and income distribution, differences between car ownership in the model results are driven by the varying scenario probability weights.While all scenarios have higher car ownership than Stockholm on average, the Dispersed Development and Peripheral Hubs scenarios show substantially higher levels of car ownership than the Status Quo scenario.Since car ownership is correlated with distance from home to the city centre and the more dispersed and decentralized scenarios draw a higher proportion of their workers from outside the centre, we should expect them to own more vehicles.The new workplaces in these scenarios are also more accessible by car as it is not required to drive through downtown Stockholm to reach Kista or Älvsjö, for example.

Travel and activity behaviour
Last, we consider the daily activities and travel that we expect to see from agents working in our scenarios.For these results, for each scenario we have assigned all agents a work zone from the scenario according to the probability that they worked in that zone, conditional on them working in the scenario.SCAPER produces full-day activity plans and we analyze the results over all agents, weighted by the probability that the agent is a scenario worker.In this way we remove some of the variability that would come from choosing a specific group of workers to analyze, and instead obtain results indicating the overall behaviour given our expected distribution of scenario workers.The scenarios show quite different commuting patterns, as shown by the mode share in Fig. 9 and by travel time distribution in Fig. 10.
The Status Quo scenario has a much lower share of car commuting than the other two scenarios, with a car mode share of 34%.Agents in this scenario are more likely to take public transit to work, with a 45% share.On the other hand, both Dispersed Development and Peripheral Hubs scenarios have a substantially higher share of car commuting (49% and 50% respectively) and lower public transit shares (33% and 35% respectively).Car ownership is not the only factor; indeed, the Status Quo scenario has slightly higher car ownership than the region overall, but a car commuting share of 4% lower than the regional average.Fig. 10, which shows travel times by mode for each scenario, helps us understand why.In the figure, the distribution of car commute times for the Status Quo scenario is substantially higher than in the other two scenarios.This is a result of Stockholm's transport geography.In the Status Quo scenario, with most new jobs in Norra Djurgårdsstaden, driving would take many agents through the city centre, causing a slower commute; alternatively they can choose to take public transit, which is generally optimized for travel to areas close to downtown.As well, agents living outside the city centre would pay the congestion charges, providing a cost disincentive to use a private vehicle.On the other hand, new jobs in peripheral neighbourhoods such as Skärholmen are easily accessible by car but more difficult to access by transit.
Total private vehicle kilometers travelled (VKT) follows a similar pattern to mode share.The Status Quo scenario results in a daily average VKT of 13.4 km per scenario worker for all activities.In comparison, Dispersed Development results in 19.3 km, Peripheral Hubs results in 19.4 km per capita per day, and the average Stockholm worker travels 15.5 km in a car.The increased commute time seen in the Status Quo scenario in Fig. 10 is not a result of longer distances but of slower traffic.
SCAPER allows us not only to examine commuting characteristics but also to explore what agents are likely to do outside of work hours.These non-work behaviours have an impact on an individual level, for example on well-being, and also at a societal level, such as through shopping habits.Here, we look at agent behaviour that has economic consequences: their shopping and leisure trips.Shopping includes both regular shopping such as for groceries and special occasion shopping such as for furniture.Leisure includes activities such as playing sports, going to a movie, or eating out at a restaurant.Together, these can be considered to be consumer activities.
Fig. 11 shows how the scenarios impact agents' shopping and leisure activities.For the Stockholm region as a whole, the highest concentration of consumer activities are performed in downtown Stockholm and the neighbourhoods around it.The Status Quo scenario does not show much change from the overall patterns, with slightly higher concentration of consumer activities to the east side of downtown, toward the new jobs in Norra Djurgårdsstaden, and slightly lower on the west side away from these jobs.The Dispersed Development and Peripheral Hubs scenarios show the largest difference to the overall pattern.Agents working in these scenarios are less likely to perform consumer activities in central Stockholm and more likely to do these activities in the south of Stockholm, near where they work (and in large part, live).Concentrations of activity can be seen around the neighbourhoods in which jobs are added.This is partially because agents may do some shopping or a leisure trip near their work, and partially because a higher concentration of agents are expected to live in the areas nearby.

Discussion and conclusions
We started by asking three questions to guide our analysis of each  scenario: who benefits, who gets access to employment, and what happens to daily behaviour?The results above offer a detailed answer to these questions regarding the workers we expect to see in each scenario.While the scenarios admittedly make only a marginal impact on employment in the City as a whole, the broader paths forward that they represent can help us draw some inferences about the impacts of how new office developments are distributed in Stockholm.First, we note that the model reports relatively similar outcomes for the Dispersed Development and Polycentric Hubs scenarios, with the greatest contrast being between the Status Quo scenario and the other two.This discussion concentrates on this larger distinction.
By income, the scenarios show little difference in who benefits and who has access to employment.The geographical distribution of jobs within Stockholm appears to have less impact on the scenarios' income distributions than the type of jobs on offer.In our case, the model finds that for new office jobs, high-income groups have the highest welfare benefit across all three scenarios, and are correspondingly overrepresented among scenario workers.
However, our model also predicts that the distribution of home locations for scenario workers depends heavily on the location of workplaces: while employees of any one area will reside in homes around the city, we expect a higher density of them to live closer to that area.This interacts with the model's observation about income to create equity considerations.If new office development is focused in the central city where there is a higher proportion of high-income residents, this new development may help reinforce the historical patterns of income inequality by drawing more middle-and high-income residents to nearby neighbourhoods.Our model shows that either a dispersed or polycentric pattern of development would draw a reasonable number of middle-and high-income individuals from neighbourhoods across suburban Stockholm, including historically low-income and vulnerable neighbourhoods.This is likely to increase the income mix in those areas, reducing a measure of spatial segregation.It is not within the scope of our model to understand how this will happen, but it could be a mix of current residents finding new, higher-paying office jobs and new, higher-income residents moving in.We note that this could cause a gentrifying effect in some neighbourhoods over time, however, this has not been seen in the areas around Kista where a relatively high-income employment hub exists now.
In Stockholm, the divide between the north and south parts of the city is particularly salient, as these regions are separated by water and only bridged by a few road and transit links.As Fig. 3 shows, downtown Stockholm-and correspondingly, most employment in the city-is in the north half of the city.The central-north and northeast parts of the city are historically high-income areas, whereas the southern suburbs have more lower-income residents.Low-income neighbourhoods such as Älvsjö have also been relatively disconnected from the transit network and currently have relatively few job opportunities.To ameliorate this divide, Stockholm City aims to bring more job opportunities to the southern half of the city; the scenarios we consider here show how either a dispersed or polycentric approach could help.
However, the answer to our third question places some of the City's objectives in conflict.Our model predicts that concentrated employment will have more environmentally friendly transport outcomes in the form of lower VKT and car mode share compared to a more distributed employment pattern.We have not directly modelled emissions here but VKT is a reasonable proxy.As the central city is well-served by transit and subject to congestion resulting in higher car travel times, it makes sense that encouraging more central employment also encourages more transit and active commuting and lowers VKT.It is possible that a polycentric model could be achieved without increasing average VKT with the right investment in public transit infrastructure, as in Bolotte (1991).Our model held the transit level of service fixed and did not explore the potential for upgraded service to peripheral hubs; however, based on the literature we suggest that a fairly major investment (such as a new metro line) would be necessary to avoid a shift toward car commuting.Exploring scenarios with different levels of transit service could be a useful avenue for future inquiry.
A unique advantage of the SCAPER model is that it considers non-work activities, both in how they impact work location and commuting decisions, but also as areas of inquiry in their own right.Our model shows that workplace location has a large impact on where individuals are likely to perform secondary activities; since individuals often run errands on the way to or from work, they tend to perform activities near their work location as well as their home location.Providing new jobs away from downtown Stockholm would also have the secondary effect of reducing its importance as a monocentric destination for shopping and other activities.There is also potential to create new combined shopping-recreation-employment hubs to take advantage of people's desire to perform activities in convenient locations near their work.Kista's mall offers an existing example, albeit one which is predominantly car-centred and not as transit-, walk-or bike-friendly as the City would like to see in future neighbourhoods.
Our modelling approach has a few key limitations.First, we use static levels of service for all modes, so the model does not account for how the scenarios would change traffic patterns (and thus travel times) in Stockholm.Due to the small size of these scenarios relative to Stockholm's overall employment, we think this impact would be marginal and not lead to important differences among the scenarios; however, for larger scenarios it will be more important to model how they would impact traffic patterns and congestion.Second, we model at the individual level and do not include household interactions such as dropping off children at school, which are often important in determining daily behaviour (Bhat and Pendyala, 2005;Picard et al., 2018).Finally, we do not model the interaction between work and home location choice, instead keeping home location fixed for agents across all scenarios and using this to determine work location.In reality, the situation is more complex, both for households who may choose their home location based on their work location(s) or vice versa, and for the locations of new residential developments.It is plausible that the different scenarios presented here would have differing impacts on developers' decisions on where to site new residences.In particular, adding substantial new jobs on the periphery of Stockholm City could help drive residential construction in neighbouring municipalities.Modelling this would require a much broader integrated city model which accounted for the decisions made by several types of decision-makers, and is outside the scope of this work.
Since the Covid-19 pandemic the future of work has become more uncertain, and in particular it is difficult to forecast the extent to which employees will work remotely years or decades into the future, and how these trends will impact companies' office development investments.However, Stockholm has seen a large rebound in the proportion of employees working from the office since 2020, and continued high levels of working from home may be detrimental to the city's goals to combat segregation and social exclusion.We recommend exploration of the potential varied impacts of remote work as an excellent area for future study.Aarhus (2000) notes the importance of ensuring that land use policy instruments are aligned with the key factors around travel behaviour, such as transit accessibility, in order to ensure that planning goals around sustainable mobility are accounted for in businesses' location decisions.This research contributes to our understanding of the potential impacts of diverging land use policies.
This paper has modelled and illustrated likely outcomes for three scenarios for new office developments by 2040 in Stockholm.While the results are specific to the greater Stockholm region, they also have relevance to other cities in similar situations, considering whether to continue historically monocentric employment patterns or develop more polycentric or dispersed office locations.The results demonstrate that the largest differences stem from whether development is concentrated near the central city, as in the status quo, or spread out into suburban areas.The model illustrates a tension between the goals of equitable access to employment and environmental benefit.To increase access to jobs for lower-income suburban communities and promote more income-mixing, new office job opportunities outside the centre should be pursued.However, locating jobs in peripheral areas makes them more easily accessible by car and tends to raise total VKT and car mode share for commuting trips.City planners should take this conflict into account when planning for future office development locations, and should consider whether there are ways to mitigate the tension between goals, for example through increased investment in peripheral transit infrastructure.

Fig. 1 .
Fig. 1.Stockholm City and its major inter-city transportation networks.Land use is indicated with a figure-ground diagram; at the scale of the maps, darker gray areas represent greater density of built form.

Fig. 2 .
Fig. 2. Scenario axes showing planned developments.Each circle represents an area for development.Circles are scaled to show the number of new jobs in that scenario area.The City has ruled out the peripheral dispersed scenario as not plausible as a development path.

Fig. 5 .
Fig. 5. Map of per capita welfare differences for scenario workers by zone.Darker green areas show greater increases in consumer welfare due to the new scenario jobs.Scenarios are shown as circles with area proportional to the number of new jobs the scenario creates at each location.(For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 6 .
Fig. 6.Distribution of individuals' improvements in welfare by income group for each scenario.Boxes represent the first to third quartiles of the data with median shown as thick inner line.Whiskers extend to 1.5× the interquartile range or the extremum.

Fig. 7 .
Fig. 7. Heatmap of expected home locations for scenario workers.Home density index (HDI) is a measure of where agents live using kernel density estimation with a 1 km radius, where 100 is the highest-density part of Stockholm (~300 people/ha).Graphs (b) through (d) show home locations weighted by the probability that residents will have one of the new scenario jobs, and are displayed on a consistent scale at 1 / 20 of the overall population density.Scenarios are shown as circles with area proportional to the number of new jobs the scenario creates at each location.

Fig. 8 .
Fig. 8. Proportions of workers in each scenario by income and car ownership.

Fig. 10 .
Fig. 10.Commute travel time distributions by scenario and mode.Boxes represent the first to third quartiles of the data with median shown as thick inner line.Whiskers extend to 1.5× the interquartile range or the extremum.

Fig. 11 .
Fig. 11.Relative frequency of leisure and shopping (consumer) activities for scenario workers.Activity frequency index (AFI) is a per-capita measure of how likely agents are to perform consumer activities in the area; the most visited place has an AFI of 100.Activity distributions for scenario workers are displayed relative to the baseline distribution for all workers shown in (a).Scenarios are shown as circles with size proportional to the number of new jobs the scenario creates at each location.
• Dispersed Development: development remains relatively centralized but is more spread out over different neighbourhood of Stockholm.Under this scenario, the neighbourhoods of Brommaplan, Alvik, Telefonplan, Årstaberg, Årstafältet, Skärholmen, and Älvsjö all see levels of new office development not seen in the other scenarios.Norra Djurgårdsstaden, Södertäljevägen, and Ulvsunda are also developed in this scenario, but not to the extent of Scenario A.
• Peripheral Hubs: adopts a decentralized but concentrated approach, emphasizing development in three areas on the periphery of Stockholm City: Kista to the north and Älvsjö and Skärholmen to the south.Kista already has a peripheral node of office development which would be strengthened under this scenario, but the bulk of new development would go to the south, especially to the vulnerable community of Älvsjö.