| marp | true |
|---|---|
| theme | gaia |
| paginate | true |
| style | |
| math | mathjax |
--
- What is about social system?
- Social system focuses on humans and their interactions, rather than nature.
- You can check definitions of society by Tominaga (1995) to consider social system
- There are interactions or communications among members.
- Members of societies last for some time; they don't dismiss soon.
- Members are organized.
- There are boundaries deviding people into members and non-members.
- Why do you model social systems?
- Some social systems have problems or unknowns.
- Such problems and unknowns are often tough to solve or analyze due to complexity and uncertainty of the social systems.
- Social system modeling enables you to deal with the comlexity and uncertainty and tackle the problems and unknowns in a better way.
- Identify problems in society.
- Define a social system.
- Describe concepts and structures of the social systems with natural langages, equations, or figures.
- Program the conceptual model into computerized model with computers and programming languages.
- Evaluate reliability of the models.
- Set parameters of the computerized models.
- Conduct scenario analyses.
- Resolution (解像度) of model as a hint to build conceptual models
- How detailed can you describe social systems?
- There are mainly three types of social model depending on resolution.
- It doesn't matter which type of model you select itself.
- Objectives of cocial system modeling guide you in decisons on resolution of models.
- Resolution guides you in choices on reliability evaluation of model.
| Resolution | Description | Objective | Reliability evaluation |
|---|---|---|---|
| Abstract model | Simple models with a few essential structures | - To comprehend micro-macro link mechanisms - To build theories - To obtain insights |
- Acceptable micro behavior and holistic features of system |
| Middle-range model | More realistic models with more parameters, describing specific situations | - To build prototypes to support decision-making - To comfirm or generate stylized facts |
- Consistency with stylized facts |
| Facsimile model | Most realistic models with less universality, describing very specific situations | - To conduct scenario analyses | - Survey on agents' behavior model - Consistency with real data |
- Segregation (分居) model by Schelling in 1971
- There are 2 types of agents on cells.
- One type of agent on a cell stay there when it finds fewer another type of agent on Moore neighborhood (ムーア近傍) than threashold.
- One type of agent on a cell moves to another cell randomly when it finds more another type of agent on Moore neighborhood than threashold.
- A simple model with just one parameter (thereshold) describes how and why people live with similar ones.
- Variation types for box office (興行収入) of film market proposed by Ainslie et al. in 2005
- There are two stylized facts (定型化事実) in film market.
- In general, stylized facts are phenomena which are observed with reproducibility, with or without explanation of the mechanisms.
- Blockbuster-type has an early and high peak.
- Sleeper-type has a slow starting and a late and short peak.
- The model focus on film market but does not mention any specific work.
- The model helps confirm and analyze stylized facts, and helps creates prototypes to support decision-making on advatisement and marketing.
- Analyses on COVID-19 by SIR model by Kartono et al. in 2021
- SIR model is a very general model but it is applied to a specific epidemic and specific countries at specific period in this study.
- The model includes parameters validated by real data.
- The model help forecast long-term trend of COVID-19 in specific countries.
- What type of model do you build in your research?
- Pick up one research project of yours to consider which the project is social system modeling or not, refering to the difinitions of society on slide #3. Otherwise, show the reasons why your project is not social system modeling.
- Classify your research project into abstract model, middle-range model, or facsimile model.
- Consider how you would build the other types of model with the same research theme.
- An example is shown below.
- Abstract model: Environmental action by penalty and reward
- Middle-range model: Analysis on GHG emission mitigation in international maritime transport focusing on the relationship between carbon levy (炭素課金) and freight rate (運賃)
- Facsimile model: Evaluation of the environment regulations for net-zero GHG emissions by 2050 in international maritime transport (Original reserach project)
- It is important to evaluate how reliable your models are.
- There are mainly three perspectives to evaluate models, which is called VV&A.
- Verification (正当性の評価): How right do your computerized models work?
- Validation (妥当性の評価): How right do you build conceptual models?
- Accreditation (実行承認性の確認): How robust are your models for practical use?
- Do your computerized models reflect your conceptual models and your intentions?
- In other words, do not your computerized models have any bug?
- Verified models are ones without bugs.
- Difficulties of verification
- It is unclear what kind of results indicates correct behavior of models.
- Bugs often result from logical errors of models.
- It is challenging to determine parameter setting methods.
- How to prevent and remove bugs from computerized models?
- Make comments on codes.
- Introduce test codes into functions or functional units.
- Refer to development processes in Information System field.
- Are your models able to achieve the modeling objectives?
- You often hear that your models are arbitrary or not realistic, but...
- Realistic models do not always represent they are able to achieve the modeling objectives.
- Realistic models are often as complex as real world, which makes analyses tough.
- To put it boldly, useful models are abstract and simplified arbitrarily, according to the modeling objectives.
- Models for social simulations should be validated by the correspondence with their modeling objectives.
- How to validate models?
- See if your models generate similar results to real data.
- See if your models reproduce phenomena explained by accepted theories, models, or stylized facts.
- Tips for building validated models
- Adopt plural (多元的) models rather than unitary models to model complex realities.
- Remember modeling objectives.
- Are your models robust enough to be applied to practical use?
- How to accredit models?
- Commit to real social issues and give simulation results back to organization or people concerned.
- It is especially important to have authorized managers (decision takers) accredit models as well as those who face the issues on spot.
- Models would not be used in the real world without accreditation by such decision takers.
- What are methodogies for accreditation?
- Methodogies for accreditation are still underdevelopment.
- Generally, modelers design hypothetical methods based on knowledge and methodogies in relevant fields according to social issue, discuss the results, and give feedback to the methods.
- Accreditation is deeply connected with comminication with people concerned with social issues, so it is also effective to build models with such people together.
- The main challenge for building models is setting parameters.
- There are mainly six ways to set parameters of model.
- Collect and refer to real data.
- Conduct sensitive analyses (感度分析).
- Conduct inversive simulations (逆シミュレーション).
- Conduct virtual grounding.
- Estimate parameters by Bayesian network.
- Adopt unknown or uncertain parameters as scenario -> Go to Chapter 6
- Models contain many parameters, but some of which are tough to set.
- Because there is no data to be referred.
- Because you cannot observe nor measure the parameters.
- You can prepare many options for such parameters to conduct simulations with and pick up ones which provide intended outputs.
- How to set parameters by sensitive analysis?
- Select parameters for the analysis.
- Analyze the relationships and influences of the parameters.
- Parameters can be set independently if the parameters affect each different objective functions.
- Parameters should be determined in order of influence if the parameters affect same objective functions.
- Prepare options for the parameters and objective functions.
- Conduct simulations to pick up a set of options which optimize the objective functions.
- Too many parameters make it tough to analyze the relationships and influences of the parameters nor conduct sensitive analyses one by one.
- In that case, inversive simulations are adopted as a solution for an inversive problem.
- A representative approach for inversive simulations is genetic algorithm (遺伝的アルゴリズム), which is a general method to search a huge solution space efficiently.
- How to set parameters by genetic algorithm?
- Select parameters for the analysis.
- Create a set of genes corresponding to the parameters.
- Prepare objective functions.
- Conduct simulations with the genes.
- Update the genes.
- Grounding is a method to collect data for modeling by questionnaire or survey.
- Virtual grounding is another method of creating virtual world to have participants make decisions there and collecting data for modeling...
- when questionnaires and surveys are not available.
- when modeling situations that rarely occur.
- when modeling situations threatening the participants.
- How to set parameters by virtual grounding?
- Extract qualitative characteristics of agents from a terget system to determine the behavior structures of the agents with accepted existing models, such as multinomial logit model (多項ロジットモデル).
- Collect participants to make decisions in the virtual situation.
- Estimate the values of parameters based on the data with methods corresponding to the models adopted at Step 1, such as maximum likelihood estimation (最尤推定).
- You can refer to the textbook for an example of modeling visitors' behavior at Disney Land by virtual grounding.
- Bayesian network is a probabilistic graphical model that represents a set of variables and their conditional dependencies via a directed acyclic graph (有向非巡回グラフ).
- When behavior models have multiple behavior causes and sets of the causes determine the behavior probabilistically, Bayesian networks are helpful to get the probalities of the behavior.
- Bayesian networks also help identify the best relationships among the behavior factors.
- How to set parameters by Bayesian network?
- Take surveys on a terget domain.
- Create a behavior model for agents.
- Generate a hypothetical Bayesian network out of the behavior model.
- Collect data with questionnaires.
- Construct a Bayesian network and determine the behavior model.
- Estimate the values of parameters.
- You can refer to the textbook for an example of modeling insurance market by Bayesian network, as it is too long to talk here.
- Social systems must have uncertainties.
- You cannot remove all uncertainties from social simulations.
- The biggest objective of social simulation is showing possible futures under uncertainty, not showing one only result in future.
- Scenario analyses (シナリオ分析) help visualize bunches of possibility (可能性の束) and support future decision-making under uncertainty.
- There are mainly three ways to analyse simulation results.
- What-if analysis (What-if分析): Analyze results with defferent inputs or conditions.
- Uncertainty analysis (不確実性分析): Analyze impacts of uncertainties on results.
- Hypothetical scenario analysis (仮説シナリオ分析): Analyze what hypothetical scenarios generate current social systems.
- Conduct simulations with different inputs or conditions to identify the impacts of the scenarios on the results.
- There are two types of what-if analyses.
- Try various inputs with conditional variables fixed.
- I call them case studies in papers.
- An example of this kind of what-if analysis is trying verious GHG measures (対策) under specific future prices of methanol and ammonia.
- You can analyze what inputs are effective on specific conditions.
- Try various conditions with inputs fixed.
- An example of this type of what-if analysis is trying verious future prices of methanol and ammonia under a specific GHG measure.
- You can analyse on what conditions specific inputs are effective.
- Conduct simulations with uncertainties and with no/few uncertainties to identify the impacts of the uncertainties on the results.
- You can compare results when indentifying multiple uncertainties in your models.
- It is meaningful to see if specific uncertainties in models affect significantly the simulation results or not.
- If the uncertainties affect significantly the results, you can discuss how important it is to mitigate the uncertainties.
- Otherwise, you can discuss that the uncertainties are not so important for the scenario analysis.
- You can refer to the textbook for an example of analysing achievement evaluation system (業績評価制度) for companies by uncertainty analysis.
- In many cases, you build a model that reproduces a target social system, conduct simulations with verious scenarios, and analyse their possible futures.
- On the other hand, hypothetical analyses allow you to input hypothetical scenarios in past and to simulate current social systems.
- Hypothetical analyses enable you to discuss which hypothetical scenario reproduces a target social system and to identify its mechanisms, processes, and conditions.
- Note that hypothetical analyses focus on mechanisms, processes, and conditions generating current social systems, rather than possible futures.
- You can refer to the textbook for an example of discussion on the causes why you stand on left/right and walk on right/left on escalator by hypothetical analysis.