Agent-Based Simulation to Inform Planning Strategies for Welfare Facilities for the Elderly: Day Care Center Development in a Japanese City

,


Introduction
. "The challenge of global aging, like a massive iceberg, looms ahead in the future of the largest and most a luent economies of the world." This quote from Gray Dawn: How the Coming Age Wave Will Transform America-and the World, published in , highlights the urgency of population aging (Peterson ). Economists have provided detailed analyses regarding how the aging of the population a ects health care costs (Sato ; Tabata ), capital flows and mobility (Dekle ; Guest ), labor supply, investment in human capital (Fougère et al. ) and fiscal sustainability (Faruqee & Mühleisen ). For social scientists, population aging brings immense challenges in health insurance reform, health care management, social security expenditures, and the exploration of aging-related policies (Ihori et al. ; Schulz et al. ; Davis et al. ; Jimeno et al. ; Van Sonsbeek ). Consequently, researchers have paid considerable attention to the economic problems and the health care and social policy needs resulting from population aging. However, as an important entity of complex urban systems, the spatial behaviors of the elderly in relation to urban planning also warrant investigation. The impact of population aging on transportation systems (Maoh et al. ) and mobility options in an aged society (Mercado et al. ) have been examined, as have the preferences and frequencies .
Simulation as a tool of analysis and prediction can be employed to establish a roadmap and explore possible solutions for the above questions. Previous literature on location models introduced a number of di erent approaches for locating facilities. Location-allocation is one approach for determining an optimal location for one or more facilities that will serve demand from a given set of points (Rahman & Smith ). Yang et al. ( ) employed an ordered logit model to investigate potential factors that contribute to hotel location choice. As a component of complex urban systems, the location of facilities for the elderly should consider the timely interactions between the aged population, facility owners, planners and governors. Agent-based models (ABMs) can better serve this demand by reflecting the evolutionary processes, such as demographics, in an urban space. Research using demographic process models has found that microsimulation of urban system evolution requires a synthetic population as a key input (D'Alessandro ; Puleston & Tuljapurkar ; Farooq et al. ). A variety of models and simulations have been employed to predict the scale, age structure, and demographic evolution of human populations (Frauenthal ; Ward & Weiss ; Fogarty et al. ; Shpak et al. ). In the field of public facility and health service, Davis et al. ( ) used a synthesized dataset to examine the e ects of policy options on health services under demographic aging. Mielczarek ( ) simulated hospital emergency services at the regional level using synthesized population data. Although the population was considered when planning public services in these studies, the decision-making processes for planning the locations of facilities were not combined with demographic aging processes or the operative management processes of relevant facility owners. Comparatively, powerful ABMs exploring the complexity of urban systems (Batty & Torrens ) and the individual decision-making processes of household agents, such as location choices for facilities, agent relocations, and agent shopping behaviors (Chen et  .
ABMs have been proven to be powerful tools in individual decision-making, and therefore, have been widely used for policy analysis and support in urban planning. However, few studies have focused on the e ects of ABMs on policy making and planners' use of simulation results determined by the models. Therefore, a synthesized modeling framework of the demographic aging process of a human population and agent-based behaviors for the planning strategies, including location planning of welfare facilities and DC center services, will provide a new perspective and add to the planners' knowledge. Planning compilation is not taken into account in our research. Instead, we focus on the simulation of possible planning strategies and how this can provide insight into planning issues. For this purpose, we developed an ABM focusing on DC center locations and made the assumption that the geographic conditions of urban space will not change. As a novel contribution to current research, we introduced this model to planning committees to illustrate its usefulness in assisting planners in acquiring information for DC center development. We predicted that our agent-based DC center location model would assist planners and governors in understanding the interactions between themselves, the elderly population, and facility owners. It is important for planners and policy makers to understand the relationships between each agent in a given situation. Therefore, we attempted to investigate whether planners and governors could e ectively use our simulation model to build knowledge for creating planning strategies.
. The remainder of this paper is organized as follows: the research methodology is introduced in Section Methodology, and our agent-based DC center location model is described in detail in Section Simulation Design and Dataset Preparation. In Section Model Application and Planning Knowledge, the initial conditions and simulation configuration of the model is described, and simulated results are compared with actual data from a Japanese city. Finally, in Section Conclusions and Future Work, the research results are discussed along with the final conclusions.

Methodology
Committee for sharing planning information for DC center development . We used Kanazawa city, Japan, as our research area for theoretical application. Between and , the housing department of Ishikawa Prefecture conducted a "Living Home Protection Plan" project for the elderly. The urban planning laboratory of Kanazawa University, a national university in Ishikawa Prefecture, made a preliminary study of the project. A planning committee was organized, and some of its tasks included predicting population change and planning the possible location of DC centers. Committee meetings were held monthly. The committees were composed of four o icers from the local government and six planning experts (planners) from Kanazawa University, some of whom were local consultants while others were Kanazawa University students. Decisions made by this committee would become preconditions of the planning compilation for the local government. Our research group at the Urban Planning Laboratory of Kanazawa University decided to prepare an ABM for simulating and visualizing potential DC center demand until . This proposal, based on expert experience and o icial documents such as statistical records, was approved and accepted by the committee to replace conventional planning strategies. The research approach and datasets were readily accessible to the public, enabling open access to the results. .
The following three teams on the committee took di erent roles in the planning process: a simulation team, an elderly people team, and a DC center team. The simulation team was in charge of carrying out simulations, collecting information, and compiling the necessary datasets. The remaining local consultants and student members attended the meetings and assisted the teams. The elderly people team conducted a field survey to collect information on the daily life of the community elderly and details regarding their income, housing, health status, health insurance and use of transport. The DC center team investigated current services, including the location and operation of DC centers, using field surveys and interviews with DC owners. The work of the di erent teams was then interconnected through an ABM. Figure represents, at the bottom level, how the actual dataset was employed to represent Kanazawa city to enable planners to connect their knowledge to the simulation. At the second level, the four members of elderly people team were responsible for supplying data and documents related to the facility-using behaviors of the elderly. This information was connected to the work of the DC center team to support the simulation team in developing the model. ). These are always configured as parameter conditions, and researchers can check parameter sensibility to understand the impact of di erent scenarios. Our work attempted to investigate whether planners and governors could use our simulation model to build knowledge for creating planning strategies. As few future scenarios had been designed, we only simulated the near future based on current trends and DC center demand based on committee cooperation.

.
The committee acted as a forum for members to share their ideas and build new knowledge. In a conventional committee, discussion of DC center planning is done only a er field surveys of community members and the elderly have been carried out. These investigative activities were also conducted by our committee, with the relevant results used to define the parameters of the model. In contrast to conventional committees, we focused on future trends of the elderly population and DC center demand on the urban space with these trends becoming new knowledge for planners, which is not possible in a conventional committee that does not use model simulation. As shown in Figure , the elderly people team supported the work of the DC center team with knowledge of existing DC services and locations, operation modes, etc. The simulation team designed the DC center location simulation model based on knowledge shared by the other two teams. We believed that knowledge of the evolution process and the interaction mechanisms between agents and the urban space shared by the simulation team would be helpful for planners in understanding the model. The simulation model and results were discussed by the committee to further improve knowledge and understanding. The framework shown in Figure represents the committee as a platform for knowledge-sharing between members.
Creating planning knowledge and developing planning strategies using agent-based modeling .
Building a framework for knowledge sharing is one concept, but learning from and building new knowledge is another. The understanding of relevant planning issues di ered among the committee members based on Figure : Committee meeting framework and responsibilities of each committee team Figure : Knowledge sharing within the DC center planning strategy process di erences in educational background, experience, and responsibilities. Researchers typically use model simulation to support decision-making in planning; however, this is a "black box" in the planning process. Whether planners could readily understand the meaning of or trust simulation outputs remained questionable. Our work focused on changing this black box situation, and, in order to accomplish this, a decision-making process for model development, simulation implementation, and planning strategies that involved the entire committee was necessary. At its core, our decision-making process involved all committee members helping build knowledge on their current work. The framework of this process, which was organized into two parts-collaborative decision-making by the committee, and simulation and visualization of planning strategies -is shown in Figure  . . Figure , objectives for the DC center development strategy are defined through the committee and its consultation with stakeholders. The defined objectives are then taken as reference values for the simulation parameters. The data and parameters form the simulation input, and a er background modeling, the outcomes are revealed. These outcomes were DC center locations, the elderly, and their use of DC centers. Based on this output, the planners and the committee discussed and attempted to understand the future spatial demand for DC centers. The agent-based simulation could be represented as a large open box into which the committee could put their knowledge, data, and concerns. By sharing this information with the simulation team (as shown in Figure ), they could contribute to model development. Basic knowledge could be built from the model with the opportunity to understand the logic relationship between model input and output. Based on the knowledge acquired from the agent-based simulation, the planning strategy could then be shared by the committee (represented by the dotted line in Figure ). Background simulation was organized through this ABM and is introduced in the next section. Figure : Framework of the decision-making process for the DC center development strategy

Simulation Design and Dataset Preparation
Agents . As discussed previously, the simulation team suggested that the DC center location model should be a spatiallyoriented ABM with the DC center location process a demographic aging one based on interaction between agents and the urban space. The agents are the entities involved in the DC center location process. The urban space (i.e., land units) is the space where these agents live and perform di erent activities. Our research identified the following four agents involved in the mechanism of DC center location: household agents, DC center agents, planner agents, and local government agents. Each agent had di erent attributes and exhibited di erent but interactive behavior during simulation. They were all living in the identified urban space, which was divided into the following three areas: Downtown (CCA); Urbanization Promotion Areas (UPAs); and Urbanization Control Areas (UCAs).

Household agent .
Planners need to understand the spatial and temporal demographic aging process and the locations of the elderly in order to plan the locations of DC centers. As an agent, a household is a coherent unit of simulation in the DC center location model that makes decisions as a single entity. This single entity is assumed to consist of a family of one or more people. The household agent has attributes such as age, marital status, number of members, savings, income and health status. The household agents in our simulation followed the lifecycle process presented in Figure as the demographic aging process, with a house location model designed in consideration of the aging process to simulate the locations of the elderly. We divided the total lifecycle of a household into five stages. As shown in Figure , in the first stage ("Be a baby"), a new person is created. Married couples raise children, who "Grow up" into independent individuals. A er several years of independent, single life, the individual marries and enters the third stage ("Married"). Although some people do not enter the third stage, as represented by the dotted line in Figure , everyone enters the fourth stage ("Grow old"). In this stage, the need for care services among the elderly household agents becomes apparent; however, our research only The committee members were planners and o icers; however, they were not "planner agents," a separate role that plans DC center locations. The planner agent controls total demand and locations of future DC centers based on the total elderly population and land use zones.

DC center agent
.
The DC center agents in our research were responsible for constructing and managing DC centers within the planning period between and , based on the location of the elderly who are expected to use these services. DC center agents use attributes such as spatial location, service type, capacity for visitors, and expenditure and revenue, drawing on the knowledge of the DC center team. DC center agents make decisions on the location and the daily operations of DC centers. Simulated locations and their operational situations were important variables for the planning strategy of our committee.

Government agent .
A government agent di ers from a planner agent. In Japan, the elderly are covered by national health insurance when using health care facilities. However, the care levels and types of services elderly persons require are determined by the government based on an evaluation of their health status. In our model, "government agent" refers to the individuals who implement these evaluations and determinations.
Interactive Roles of Agents .
Agent roles and activities were determined by the committee through discussions of available information. For example, for DC center agents to establish center locations they followed the plans provided by the planners. According to information from the elderly people team, elderly agents who want to use DC centers apply for a care Figure : Interactive roles of agents in the DC center location model level certificate from the local government. A er their application, the government agent estimates the health status of the applicant based on eight determinants, such as diet, mobility, toileting procedure, and so on, and issues these certificates to elderly agents. A er receiving their care level certificate, elderly persons attend a chosen DC center a specified number of times per week. Finally, DC center agents keep attendance records, calculate revenue, and compare revenue with expenditures to assess financial viability. Figure demonstrates these determinants in further detail. The next section discusses choice of DC center agent location, care level certification by the government agent, use of the DC center by the elderly agent, and work of the DC center agent, based on the committee findings.

Choice of location of DC centers
. According to the law for opening DC centers in Japan, the DC center agents follow the decision tree shown in Figure to select a location. A DC center agent must first specify the location of elderly persons in each cell who wish to attend a DC center. They then calculate demand for DC centers within each cell in di erent urban areas. This demand indicates the number of elderly persons who want to attend DC centers in an area, within a radius of the maximum service distance. The DC center agent then compares the demand in each cell with a threshold to set that location as a potential site. A minimum demand threshold of elderly persons is required to build a DC center in one cell. The following equation is used for calculating demand in one cell: where D j is demand at cell j, a i is the number of elderly persons who wish to attend a DC center in cell i, d ij is the distance between cell i and cell j, and S max is the maximum service distance.
. Next, the DC center agent evaluates the appropriateness by factors such as land use zone and distance from a hospital. The probability of whether a cell will be chosen for building a DC center is then calculated. In this paper, we applied logistic regression (Long et al. ) to calculate this probability, as represented by the following equations: ( ) where S t ij is suitability of cell (i;j) at time t, S 0 is a constant that is one of [-; ], P t g is the initial probability at time t, P t gmax is the maximum value of P g during each step, β 1 ,β 2 , . . . , are coe icients in logistic regression, are spatial attributes of land use zones in cell (i;j), a is the di usion coe icient in the range of ( -), P av is the average probability between cell (i,j) and its four cell neighbors, and Pf is the final probability. .
Finally, the DC center agent chooses the site with the highest probability as the location of the DC center.

Certificate of care needs for elderly agents .
When government agents issue certificates to elderly agents, interaction occurs. Government agents make decisions specifying care levels for elderly agents based on their health status, as shown in Figure . As shown in Figure , we designed a simulation process to mimic the behavior of the government agents who issue certificates to elderly persons. First, elderly persons who want to attend DC centers submit their health status documents to the government agent, choosing scores for each determinant shown in Figure . The government agent then calculates a health score for each elderly agent by summarizing the scores selected by that elderly person. Finally, to specify the care level, the government agent compares the health scores with the classification table of care levels shown in Table . JASSS   Elderly agents follow the decision tree shown in Figure to select which DC center to attend. We assumed that the distance from an elderly agent's house to the DC center was the only criterion; that is, they will choose the nearest center. However, they may be asked to move to another center if their primary choice is full. This module also represented interaction between elderly agents. An elderly agent who has attended a DC center may introduce that center to their neighbors and ask them if they would like to attend the DC center together. In our model, elderly agents living in an area within a -meter radius, which corresponded to two cells in the model, of those who have attended DC centers were considered neighbors. In the DC center location model, DC centers do not operate indefinitely. Instead, DC center agents decide on the continued operation of centers based on revenue. This process within the model was based exactly on that used in reality. The model examined the behavior of DC center agents in recording the number of elderly agents who attended the DC centers according to di erent care levels. They also calculated revenue for one year and compared this to expenditures in order to ensure that DC centers could continue operating. If the revenue of a DC center is lower than its expenditure, the center will go bankrupt and be eliminated from the urban space. We suggested the following equation for calculating revenue: where R is revenue for one year, N i is the number of elderly persons according to care level who had attended a DC center, P i is the price di erence according to care level, F i is frequency of attendance at a DC center, is the number of weeks in a year, and ra is the price ratio between the three DC center types with three di erent values (ra = 1; ra = 1.1; ra = 1.5); i = [ ; ] corresponds to the following seven di erent care levels: Support (S ), Support (S ), Care (C ), Care (C ), Care (C ), Care (C ), and Care (C ). .
In summary, the simulation model was designed based on the knowledge and suggestions of the committee. This made the model less of a black box to planners during the simulation. To ensure DC center management and changes to households in the simulation matched real changes in Kanazawa, we developed a validation process for the model by comparing simulated results with actual census survey data and DC center location information.

Model Application and Planning Knowledge
. In the previous section, we discussed how the simulation model was designed using the knowledge of the committee teams. In this section, we attempt to assess the actual situation of Kanazawa city and investigate how JASSS, ( ) , http://jasss.soc.surrey.ac.uk/ / / .html Doi: . /jasss. Figure : The interface of agent-based simulation for DC center development in the Netlogo platform the committee planners used the simulation tool to develop their knowledge of agent behaviors and DC center demand for a DC center planning strategy. In regard to the validation of the agent-based simulation, we focused on system validation and system application. The interface of the system is showed in Figure . It is developed on Netlogo platform, which is authored by Uri Wilensky in .

Simulation configurations
Committee assumptions .
The data available from committee and the local government database were limited, so some assumptions were necessary for the simulation. In consensus with committee members, the following assumptions were therefore made based on the actual circumstances in Kanazawa city: . Statistical data in Kanazawa city provided by local government o icials indicated that the number of people moving in and out of the area has remained constant each year from to . Therefore, we assumed the migration population to be zero until , the final target year of the research project. Members recognized the element of uncertainty involved with this assumption and agreed to carry out the simulation under this simplified condition.
. DC centers are situated based on zone regulations on land use in Japan. This assumption was agreed upon as common knowledge between committee members.
. One tick in the simulation represents one year. The committee agreed that it was unnecessary to simulate too far ahead as the planning strategy is adjusted each year; therefore, one year was considered one step.
. We made some assumptions from information from the DC center team. First, elderly persons wishing to use DC centers will choose the nearest center if it is not full. Second, there are three types of DC centers with di erent capacities ( , , and users) indicating the maximum number of elderly persons that a DC center can serve at one time. Higher capacity DC centers incur higher expenditures. Finally, when calculating demand, DC center agents consider the number of elderly persons living within a . -km radius, which corresponds to cells, of the potential DC center location.

Parameters .
Land use zones and urban planning were based on Japanese law with no obvious changes until the year ; therefore, the only data changing over time is household data. To accommodate this, we used population data for Kanazawa city for as the initial population ( Figure ) of elderly persons, and began the simulation from the year with six existing DC centers.
. The elderly people team collected statistical data on Kanazawa city and conducted a sample survey on the housing situation and community activities of the elderly, as an investigation of the entire population of Kanazawa was not feasible. A random sampling procedure covering a total of , household agents living in the urban area was used. This number was based on the actual number of households in the year , and consisted of . % of the , households in Kanazawa (Figure ). Birth, death, and marriage rates, at . %, . %, and . %, respectively, were assumed to be similar to those during the period from to . To set the initial input data, we used the same ratio of elderly agents to total population and the same ratio of elderly persons who attended DC centers to the number of elderly persons as the actual data, at . and . , respectively. Therefore, we initially calculated , elderly agents, including elderly agents who attended DC centers. The details of the initial parameters for simulation are shown in Table . .
For the simulation assumptions and parameters, collecting data and reaching consensus through cooperation with other teams and local government o icers was straightforward, and therefore easily applicable in planning strategy practice.

Simulation results
.
We attempted to analyze the topics discussed in the planning committee meetings in order to view how planners understood and accepted simulation outputs. Model validation was useful for demonstrating to the planners that the simulation was technically sound. With acceptance from planners, it would be possible for planning strategies to be devised based on predicted results from future simulations.

Model testing and validation
. The simulation results showed the location of households and DC centers. The DC centers and elderly people teams doubted that the location of households and DC centers in the simulation outputs was reasonable.

Figure : Distribution of created households in the NetLogo platform
For example, the simulated location of households and DC centers for di ered from actual locations in Kanazawa city for the same year. Household locations were di icult to present to the committee due to the large number of communities in the area. .
During meetings, we explained that the simulation provided an accurate distribution of DC centers in di erent urban areas and the corresponding number of elderly agents who attended the centers, but did not give detailed location coordinates of the centers. Our simulation results could predict trends in DC center development and the needs of the elderly for Kanazawa city, supplying planners with a reference point for DC center planning. Next, we validated the simulation results for the years through using actual census survey data for for Kanazawa city. We also showed predictions of the locations and numbers of DC centers in di erent urban areas and the numbers of elderly persons who would attend them in . This enabled team members to understand how model validation demonstrates the validity of the simulation results.

.
The entire model test for validation was repeated times using the value parameters listed in Table . One tick in the simulation represents one year, and ticks are simulated in each experiment. The results therefore represent the -year period from to . One simulation experiment took hours, so a total time of , hours was required to obtain the simulation results for all experiments. We explained the simulation results in detail so that team members could understand why the simulated and actual locations were di erent. Fig. shows the simulation results for the distribution of DC centers for . Small house symbols represent existing DC centers with the same coordinates as actual centers in . Large house symbols represent the simulated DC centers. The simulation results over the experiments provided an average of DC centers (Figure ), % of which were located in residential areas, % in commercial areas, and % in industrial areas ( Figure ), which are similar to the actual percentages of %, %, and %, respectively, for Kanazawa city. As planners tend to concentrate DC centers in residential areas, UCAs and UPAs are the focus, with most industry land use zones located in CCA. Therefore, the number of DC centers located in industrial areas is smaller than the actual percentage. .
To discuss the stability of the simulation results, we represented the changes in the results throughout the experiments (Figure and    .
Finally, the committee understood that locations could not be precisely simulated, but general trends can be checked across urban areas. They also understood that the location of the elderly population is the key factor in simulating DC center location. Despite their surprise that the simulation tool integrated many factors within one platform, it remained di icult for them to consider the result based on the agreed parameters in the previous step, as they had di iculty connecting these parameters to the simulation. They attempted to understand this method of location of DC centers, as it is a key topic in planning strategy, even if they were unable to perfectly understand the detailed mechanism of DC center demand simulation. Most importantly, they understood the steps in the simulation process and the relationship between each step as they shared information during the simulation design phase. Therefore, through the committee meetings, we were able to predict DC center locations to the target year of .
Forecasting DC center distribution to .
In the previous subsection, we discussed model validation and the statistical stability of our simulation results, which confirmed that the DC center location model was applicable and reliable for predicting DC center distribution in urban areas and the numbers of elderly persons using the centers, and suggesting that committee members could consider simulations as a possible planning strategies for future DC centers. In this subsection, we further implement our simulation model for the location of DC centers and predict the number of elderly persons who will use these centers up to . This experiment was also repeated times. The corresponding forecasts are indicated by Figures a to . The simulated results predicted DC centers and , attending elderly agents to . The number of DC centers and elderly persons who will use them in increased . -and . -fold, respectively, from .
. However, as shown in Figures and , the numbers of elderly persons at S , S , and C was almost zero. This result seemed unreasonable to the committee. We explained that the reason for this was probably because we had no dataset for the health status of the elderly, and we therefore had to simulate this for each elderly agent using the random process introduced in section . . . A reliable dataset would solve this problem. This confused the committee members, and suggested that if agent-based simulation is used for a planning strategy, a reliable and complete actual dataset relevant to the simulation inputs and model parameters is necessary. Otherwise, there is a risk that the committee may not fully trust the simulation, making an ABM less useful in planning practice. It was evident from the results in subsection Forecasting DC center distribution to that obvious predictions could be made to the year , i.e., that the number of elderly persons will continue to increase rapidly, and that DC center services need to increase accordingly; the committee reached consensus on this matter. Our simulation informed planners and governors that local governments must work on the planning of care facilities in the area. By , nearly four times the total number of DC centers in will be needed. The committee accepted our simulated prediction of this trend, but doubted our predicted results based on the relative amount of di erent care levels required to because we did not use an actual health status dataset in our simulation. Despite our explanation that the related health status parameters were repeatedly adjusted throughout the model validation process to fit the actual situation, the committee continued to express doubt.

.
Our simulation results suggest that new DC centers will be mostly centralized in the CCA area, and the committee accepted this location trend. However, there was also consensus that the locations of DC centers in our simulation could not truly represent the future locations of new DC centers, and the simulation could only show the spatial needs of the elderly for DC centers based on the current situation. The simulation results can be used  as reference for the spatial planning of DC centers, but they cannot be used to replace the actual location trends of DC centers. Center locations may need to be adjusted based on factors such as market demand and competition. In summary, our simulation assisted the planning committee with information on the increasing numbers of elderly persons over a -year time period and their potential need for DC center services. Furthermore, we assisted the planning committee in recognizing the mechanisms of the demographic aging process and DC center demand, as well as related factors a ecting future DC center locations. Although the results could not be used directly for DC center planning, planners were provided with more comprehensive information for their planning strategies.

Conclusions and Future Work
. Based on our findings, conclusions can be drawn on two distinct aspects: agent-based simulation for DC center location planning in Kanazawa city, Japan, and the application of agent-based simulation to inform planners in planning practice. For the former, an ABM simulating the process of DC center development in a local Japanese city was introduced and illustrated. This model was developed based on the institutional mechanism of Japanese DC center development. We combined with our model with household lifecycle simulation to predict the future age structure of the population of Kanazawa city. A representative characteristic of this Japanese-style model is that it includes data on government-issued health certificates for the elderly and the supply of DC center services. With the government-issued health certificates, elderly persons, the elderly agents in the model, may use DC centers with their health insurance. Therefore, the agent-based DC center location model could competently simulate the entire Japanese DC center development process. .
The outcomes of the simulation experiments proved to be statistically stable, validated by census survey data from . The number of elderly persons who used DC centers in increased . -fold from , which was close to the simulated result of . -fold, confirming that this increase could be statistically predicted by our model. Our model was also able to predict the future distribution of DC centers in Kanazawa city and the number of elderly persons who will use them. The potential distribution of DC centers in di erent urban areas to suggests the need for di erent levels of DC center services. Compared with traditional methods of statistical analysis, our model used complex interactive behaviors based on existing DC center issues, providing planners with a more vivid visualized reference. .
Some consensus was reached by the committee based on our results. First, the total number of DC centers needs to increase in accordance with the increase in the elderly population, an observation that was evident from our simulation and useful to planners. The planning committee accepted our prediction of this trend for to , but doubted our prediction on the demand for di erent care levels due to the lack of an actual health status dataset for the elderly population in our simulation. Although we adjusted for this in the model validation process, doubts remained. Second, although the planning committee could accept location trends based on our simulation, they felt this could only be spatial demand based on the current situation, and that our simulation could not accurately represent the future location of new DC centers because their location may need to be adjusted based on market competition.

.
Although the simulation results were not used directly for DC center planning, they did assist the committee with information on trends a ecting the elderly and the need for DC center services in di erent urban areas. The modeling also helped the committee recognize the complex factors a ecting future DC center locations. Most of the doubts remaining among the committee stemmed from uncertainty regarding the reliability of the dataset used in the simulation, which lacked information on health status.

.
Within the planning process, our simulation could be used at the very beginning to provide planners with knowledge on the mechanisms of DC center demand and assist them in planning further strategies. Our findings suggest that it is necessary for planners and policy makers to consider demographic aging processes, operative management processes of facility owners, and other factors when considering the future locations of new DC centers. In addition, the design of an ABM should be based on relevant regulations and laws in order to reflect actual human decision-making processes, which represents one type of interaction between agents in a simulation. Furthermore, our results suggest that for an agent-based simulation to be useful and trusted by planners, a reliable and complete dataset is necessary.