Use of Computational Tools to Support Planning & Policy by Johannes M. Bauer
0 likes211 views
The document discusses the role of computational tools and big data analytics in information and communication policy, highlighting their potential to address significant challenges while also acknowledging inherent limitations. Key topics include the need for dynamic systems approaches in policy-making, the drawbacks of big data in terms of bias and data interpretation, and the evolving landscape of advanced mobile broadband with implications for investment and regulatory frameworks. It further emphasizes the importance of integrating various analytical methods to improve policy design and to mitigate issues related to digital inclusion and platform power.
Use of Computational Tools to Support Planning & Policy by Johannes M. Bauer
- 1. Use of Computational Tools to Support Planning and Policy Johannes M. Bauer Michigan State University Quello Chair in Media and Information Policy IX. NIC.br Annual Workshop on Survey Methodology São Paolo, May 20-24, 2019
- 2. Four main messages 1. Information and communication policy faces grand challenges that require a dynamic systems approach to be addressed effectively 2. Big data analytics can contribute greatly to improve policy analysis, design, implementation, and monitoring 3. Because public policy often seeks to change the system it is interacting with, big data analytics also has limitations 4. Computational methods and integration with other types of social scientific analysis can help overcome these shortcomings
- 3. Plan for today • Big data analytics and ICT policy • Group discussion: succeeding in the race to advanced mobile broadband • Application #1: Overcoming digital divides • Case #2: Designing 5G markets to promote investment and innovation • Case #3: Mitigating the power of digital platforms • Recap and conclusion
- 4. Big data analytics and ICT policy
- 5. Grand challenges of information policy • Increasing digital literacy and reducing digital inequalities • Harnessing the benefits of next-generation technologies − Internet of Things (IoT), Internet in Everything − 5G wireless and ubiquitous connectivity − Robotics and pervasive automation − Big data analytics, machine learning, and artificial intelligence (AI) • Development of culturally and socially sensitive, transparent algorithms • Utilization of data while protecting sensitive data, privacy, and information security • Mitigating the power of digital platforms without harming innovation
- 6. Big data and policy • Policy as improving the performance of a given system (incremental policy) versus policy aimed at changing the structure and trajectory of the system (architectural policy) • Past behavior and policies may not be good predictors for the consequences of interventions. They also have limitations in deciding what should happen • However, big data is invaluable in building models to analyze options − Scenario building and system dynamic models − Computational, numeric models − Agent-based models, evolutionary models, genetic learning algorithms • Big data analysis is a complement to other methods and an important tool to improve policy design and implementation
- 7. Promises • More accurate data documenting availability and adoption of ICTs • Better understanding of ICT adoption, uses, and effects (e.g., Blumenstock et al., 2015) • Design of efficient policy responses (e.g., Twitter data on urban commute data, cyber-physical systems in smart cities) • Effective monitoring of policies • Examples of projects collected at https://unstats.un.org/bigdata/inv entory/ Improved commute statistics using social media data in Jakarta
- 8. Concerns • Big data changes the nature of “knowledge” • Claims to objectivity and accuracy can be misleading • Bigger data is not always better • Data may lose its meaning if taken out of context • Inherent biases of algorithms and machine learning • Proprietary data and algorithms create new digital divides • See boyd & Crawford (2012) • COMPAS − Widely used in U.S. courts since 2000 − Uses 137 features to determine risk of recidivism − Closer examination revealed strong biases against black men and in favor of white men • Virginia Eubanks (2018) − Automating Inequality examined algorithms intended to support programs fighting homelessness, child abuse, and determine eligibility for public assistance − Case studies reveal that they create “digital poorhouses”
- 9. Types of big data analytical methods Source: Sivarajah et al. (2017).
- 10. 10 Informing all stages of policy development Examination of policy options • Conflicts between instruments • Indirect effects • Likely effects on static and dynamic performance goals • Possible undesired effects • Costs and benefits of intervention Policy implementation Monitoring, review, analysis of deviations between goals and outcomes Policy adaptation (termination, modification, …) Analysis of status quo, performance gaps, definition of policy vision
- 11. A digitally connected world • Characteristics of our connected life − “Exponential technologies” accelerate diffusion and widespread use − Near-ubiquitous connectivity amplifies interdependencies in work, innovation, political movements, environment, …) − As digital technology permeates nearly in all aspects of life and work, it increasingly becomes a “black box” for users (many features are hidden and/or unknown) • Recurring features of dynamic, adaptive systems − Interdependencies create new non-linear dynamics (e.g., network effects, contagion, “butterfly effect”) − Systems may have multiple equilibria (“attractors”), each representing different performance attributes − Initially small differences may lead to major differences in outcomes and path dependencies (e.g., Schelling, 1969, racial segregation model) − Diversity increases the performance and resilience of a system up to a point (but excessive diversity may eventually reduce them again)
- 12. Policy in dynamic adaptive systems • The dominant view: public policy as control − Abba P. Lerner, The Economics of Control, Macmillan, 1944 − Government can influence outcomes of economy to enhance welfare • Governance: limits of the state, policy process, and policy-makers − Policy makers face many limitations such as incomplete information, feasibility constraints, self-interested players (e.g., Dixit, 1996) − Other forms of governance (e.g., networks such as IETF, IGF), voluntary coordination (e.g., 3GPP, W3C), and emergent norms are critically important • Bottom-up policy for our connected world: active laissez-faire − Markets need appropriate rules and policies to work well − David Colander & Roland Kupers, Complexity and the Art of Public Policy, Princeton University Press, 2014
- 13. Many model-thinking (Page, 2018) • Applying multiple lenses increases our understanding • Examples include − Network models − Diffusion and contagion models − Game theoretical models − Path dependence models − System dynamic models − Threshold models with feedback − Collective action problems − Rugged-landscape models • Can contribute to better policy Transforming data into wisdom From: Scott E. Page, The Model Thinker, What You need to Know to Make Data Work for You, New York: Basic Books, 20018, p. 7.
- 14. Group discussion Succeeding in the race to advanced mobile broadband
- 15. Promises of advanced broadband • Advanced broadband connectivity promises innovative services for consumers, support for the Internet of Things (IoT) and seamless specialized services for sectors such as manufacturing, transportation and health care. • Advanced wireless services (e.g., LTE, 5G) will constitute an integral part of the future gigabit communication network infrastructure. Its technical attributes, such as high bandwidth and low latency, will enable a wide range of innovations. • Worldwide, countries are positioning themselves to take advantage of 5G services but policy models vary widely from hands-off entrepreneurship (e.g., U.S.) to regulated competition (e.g., EU) to state-led rollout (e.g., China)
- 16. Concerns about infrastructure rollout • High capital requirement (e.g., antennas, backhaul, spectrum, rights-of- way) • High innovation potential but not safe use cases yet, hence revenue potential uncertain Capex for wireless networks (2019-2020 estimated), based on GSMA market information
- 17. The Interamerican Advanced Wireless Task Force • Founded to utilize big data to design effective policies for advanced wireless markets • Hearing with major stakeholders (network operators, consumers, industry users, …) • All made recommendations on universal connectivity and which market design would be most conducive to achieve it • Your tasks: − Review the handout and the recommendation by the player assigned to you − Discuss the questions with your group − Write important points on the posters − Determine one of more speakers and share your findings with the other participants
- 18. Approximate timeline • Review the handout (5 minutes) • Discuss the questions (10 minutes) • Put your key points on the posters (5 minutes) • Report back to all (10 minutes)
- 19. Application #1: Overcoming digital divides
- 20. The “homework gap” challenge • “Homework gap” refers to disadvantages of students from kindergarten to high school (K-12) who do not have sufficient access to Internet connectivity and hence fall behind in school • One of the digital divides • Extent of problem is not well known, although there is reason to believe that existing data grossly underestimate it Divergence between public data and actual uses Source: https://www.govtech.com/biz/Microsoft-Speeds-Show-Broadband-Use-Is- Far-Lower-than-Access.html
- 21. Big data complemented by surveys • Numerous initiatives to measure network speeds (e.g., Akamai, Ookla, …), each with unique strengths and weaknesses, and examine their social and economic effects • However, network access and quality is only one among many factors shaping digital divides • Just examining network access/speeds samples on the dependent variable (does capture variations of access but not those without access) • Michigan Moonshot Project (Merit Network + Quello Center + MLab) − Crowdsourced network quality data to overcome inaccuracies of existing, public domain broadband maps − Paper-based survey in schools across the state (pilot study in 202 classrooms in three school systems across the State of Michigan) − Allows granular understanding of problem and the targeting of remedial measures (e.g., subsidies, PPPs) to specific locations and populations − Unique ID allows linking information to other databases but numerous challenges to protect identity of participants
- 22. 22 Survey in a box Turnkey Kits Leadership & Admin Introduction & Overview Parental Letter Teacher Instructions • Video/lesson • Paper in-Classroom survey (MSU) • Homework Assignment Instructions (Merit) • Student Key Collection Instructions
- 23. 23
- 24. Develop citizen-science/crowd-sourcing techniques to assess the “homework gap” in a more granular manner Share information statewide and become active nationally Foster public-private partnerships Broadband EDU Series Establish community connectivity teams to provide expertise in data analysis, broadband technologies, financing, sustainability, project management and network construction Assist in navigating community planning grants through state or philanthropic means; help communities acquire one-time construction subsidies Phase I Phase II Contribution of big data analytics
- 25. Application #2: Designing 5G markets to promote investment and innovation
- 26. Promises of 5G connectivity Source: https://www.cablelabs.com /insights/cable-5g-wireless- enabler • Part of future seamless gigabit network infrastructure • Enables numerous new services for consumers, businesses, government
- 27. The 5G value system Vertically integrated players Applications/services layer Logical/development layer Network layer Partially vertically integrated players External coordination costs (ECC) ECC ASPi ASPi Development platforms Pure MNOs ASPi C1 C4 C5 C6C2 C3 Source: Bauer & Bohlin, 2019 Areas of concern are: • Access bottlenecks − Fixed network backhaul − Rights of way − Data • Coordination costs − Transaction costs − Adaptation costs • What is the appropriate role of policy and regulation?
- 28. Can/should policy support 5G? • Should access to network transportation services be regulated? − Mobile virtual network operator (MVNOs) access to MNO networks (e.g., regulated reference offer) − Access by application and service providers (ASPs) and content providers (CPs) to networks and end users (e.g., mobile network neutrality) • Should public policy mitigate the market power of digital platforms and facilitate coordination among players? − Interoperability (e.g. open and transparent standards) − Open application programming interfaces (APIs) • How should access to resources be organized? − Spectrum management (initial allocation, secondary markets) − Rights of way (outdoors antenna locations, access to buildings) − Data about network, users, … • How can public interest goals (e.g., universal coverage, service, public interest innovations) be supported?
- 29. Emerging 5G market designs • Regulated competition (e.g., dominant in European countries) − Ex ante regulation to neutralize market power and dominance, typically after detailed examination of market structure and conduct − Might use backhaul regulation, mobile virtual network operator (MVNO) access, regulation of rights of way (ROW), network neutrality • Policy-push (e.g., some Asian and a few European countries) − Proactive policy intervention to accelerate infrastructure rollout and service innovation − Typically includes infrastructure rollout targets (possibly public investment), mandatory MVNO access, open network provisions (e.g., mobile network neutrality, mandatory open APIs), open data, industrial policy programs • Entrepreneurship (e.g., United States) − Strong reliance on private sector and entrepreneurship to advance 5G rollout and innovation − Minimal ex ante regulation and intervention; market failure to be addressed by competition policy (and possibly ex post regulation)
- 30. 30 Complexity of non-linear dynamics Investment/innovation decision layer A (e.g. MNOs) Investment/innovation decision layer B (e.g., ASPs) OpportunitiesB AppropriabilityBContestabilityB OpportunitiesA AppropriabilityAContestabilityA Coordination cost Complementarity + + – – +/– + + +/– + + Regulation • Horizontal (backhaul, roll-out, …) • Vertical (MVNO, net neutrality, …) Rate and direction of sector investment and innovation Source: Bauer & Bohlin, 2019 + … variables move in same direction, – … variables move in opposite direction, +/– … ambiguous
- 31. Positive and negative feedbacks Mandated MVNO access Incentives for MVNO innovation Incentives for MNO innovation + – Rate and direction of innovation + + + Overall effect? 31 +
- 32. 32 Regulation does not control, but „tune“ the system Investment, inovation (incentives) No access regulation Strict access regulation Applications and services investment Total investment, innovation RL R* RU “Workable” regulation Acceptable performance Network investment, innovation
- 33. Granular data and scenario analysis allow exploring possible futures Source: Batrouni et al. (2018)
- 34. Scenarios for 5G capex in the EU-14 and U.S. 0 5000 10000 15000 20000 25000 30000 35000 40000 45000 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 Capex in US$ million EU14 US 0.0 20.0 40.0 60.0 80.0 100.0 120.0 140.0 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 Capex per PoP, US$ EU14 US Source: Bauer & Bohlin, 2018
- 35. Contributions and limits of big data analytics • Contributes to building better models of the underlying non-linear dynamics − Establish directionality of relations − Calibrate effect sizes − Better and more accurate measures of outcomes • Allows better monitoring of effects of policy changes − Provisions for data collection and data sharing needed (e.g., data trustees) − Needs to be complemented by analytical, causal models of underlying processes • Can provide limited advice as to how policy changes will affect overall system and what the preferred course of action is
- 36. Application #3: Mitigating the power of digital platforms without harming innovation
- 37. Increasing concerns about platform power • Economic and technological attributes of digital markets contribute to enormous market concentration − Supply-side economies of scale and economies of scope − Demand-side economies of scale (network effects) − Positive feedback effects in platform ecosystems − Winner-takes-all dynamics • Concerns include − Ability to manipulate digital markets (e.g., design algorithms that favor own products/affiliated partners) − Increasing control over data that are not accessible to competitors (but an essential input to competitors) − Ability to influence and potentially manipulate information flows with broad effects on political system
- 38. Policies to mitigate platform power • Network neutrality − Strict neutrality: each datagram needs to be treated alike (“a bit is a bit is a bit”) − Weak neutrality: differentiation of quality of service (QoS) allowed, but needs to be done in non-discriminatory fashion • Open network requirements − Interconnection and interoperability − Open data, open APIs, open algorithms • Vertical (structural) separation − E.g., Open Reach in the United Kingdom − Platform providers (e.g., networks, providers of logical platforms) cannot also be present in applications, services, and content markets • Competition policy measures − E.g., EU cases against Google; U.S. cases against Microsoft, Facebook − Often works slow and faces tremendous challenges producing compelling evidence
- 39. Contribution of computational methods • All these policies generate differential, positive and negative feedback effects on players in the digital ecosystem (see above) • Net effects depend on initial conditions, strength of interdependencies, and behavior of players • Difficult, if not impossible to model analytically (non-linearities, multiple equilibria (“attractors”), fast-paced change) • Some challenges can be overcome by computational methods − System dynamic simulation models (e.g., Sterman, 2000) − Generative social science, especially agent-based models (e.g., Epstein, 2006)
- 40. Agent-based modeling (ABM) • Simulate the actions and dynamic interactions of agents (individuals, groups, organizations) with the goal to assess their effects on the system • Components include − Agents that act based on rules (ranging from simple to rich specifications) − Relationships between the agents − An environment in which the agents interact • Computational modeling frameworks (e.g., Swarm, NetLogo, RePast, AnyLogic) allow analyzing the emergent properties of such interactions • Allow examining the consequence of changing the rules and/or structure • Models need to be verified (testing whether the model works correctly) and valid (checking whether the right model has been built) • Examples at http://ccl.northwestern.edu/netlogo/models/
- 41. ABM applied to vertical integration • Kendall J. Koning, An agent model of vertical integration in tele- communications and content, Ph.D. Dissertation, 2017 • Models ICT value system as an interaction of heterogeneous agents (ISPs, consumers, content providers). Individual agents seek to optimize their profits (utility) following a decision heuristic and adapt their strategies based on the outcomes (“fitness”) • Over many generations, agent decisions improve and adapt to the parameter setting of the model, which include regulatory and policy dimensions (e.g., whether vertical integration is permitted, whether network neutrality is mandated) and economic/technical conditions • Policy parameters and environmental settings can be mapped to outcomes
- 42. Overall model structure Source: Koning, 2017Not examined in this model
- 43. The simplified agent model • Consumers optimize utility • Internet Service Providers (ISPs) and content producers optimize profits • Regulatory variables • Structural separation (Y/N) • Bundling allowed (Y/N) • Zero rating allowed (Y/N) • Ability of ISPs to charge content providers (Y/N) • Factorial design, model runs generate empirical data on outcomesSource: Koning, 2017
- 44. Relations among major variables Source: Koning, 2017
- 45. Selected findings for market concentration Source: Koning, 2017
- 46. Big data analytics and ABM • Big data analytics can help to build better agent-based models − Calibration of effect sizes − Assessing the validity of model (e.g., by replicating outcomes using past data) • Big data can improve monitoring of effects of policy changes − Provisions for data collection and data sharing needed (e.g., data trustees) − Needs to be complemented by analytical, causal models of underlying processes • ABM allows evaluating potential effects of structural policy changes on overall system and in selecting a preferred course of action • However, ABM needs considerable refinement before robust recommendations can be derived
- 48. Recap of main points 1. Information and communication policy faces grand challenges that require a dynamic systems approach to be addressed effectively 2. Big data analytics is a promising tool that can help improve policy analysis, design, implementation, and monitoring 3. However, because policy often seeks to change the system it is interacting with, big data analytics has limitations 4. Computational methods and integration with other types of social scientific analysis can help overcome these shortcomings
- 50. References • Batrouni, M., Bertaux, A., & Nicolle, C. (2018). Scenario analysis, from Big Data to black swan, Computer Science Review, 28, 131-139. • Bauer, J. M., & Bohlin, E. (2018). Roles and effects of access regulation in 5G markets. East Lansing, Michigan; Gothenburg, Sweden. Available at SSRN http://dx.doi.org/10.2139/ssrn.3246177. • Bauer, J. M., & Bohlin E. (2019). Horizontal and vertical regulation of 5G markets: Effects on investment and innovation. Work in progress, Michigan State University and Chalmers University. • boyd, d., & Crawford, K. (2012). Critical questions for big data: Provocations for a cultural, technological, and scholarly phenomenon. Information, Communication & Society, 15(5), 662-679. • Blumenstock, J. E., Cadamuro, G., & On, R. (2015). Predicting poverty and wealth from mobile phone metadata. Science, 350(6264), 1073-1076. • Colander, D., & Kupers, R. (2014). Complexity and the art of public policy: solving society's problems from the bottom up. Princeton, NJ: Princeton University Press. • Dixit, A. K. (1996). The Making of Economic Policy: A Transaction Cost Politics Perspective. Cambridge, US and London, UK: MIT Press.
- 51. References … • Epstein, J. M. (2006). Generative social science: studies in agent-based computational modeling. Princeton, NJ: Princeton University Press. • Eubanks, V. (2017). Automating inequality: How high-tech tools profile, police, and punish the poor. New York: St. Martin's Press. • Koning, K. J. (2017). An agent model of vertical integration in tele-communications and content, Ph.D. Dissertation, Michigan State University, December 2017. • Lerner, A. P. (1944). The economics of control: principles of welfare economics. New York: The Macmillan Co. • Page, S. E. (2018). The model thinker: What you need to know to make data work for you. New York: Basic Books. • Schelling, T. C. (1969). Models of segregation. The American Economic Review, Papers and Proceedings, 59(2), 488-493. • Sivarajah, U., Kamal, M. M., Irani, Z., & Weerakkody, V. (2018). Critical analysis of Big Data challenges and analytical methods. Journal of Business Research, 70, 263-286. • Sterman, J. D. (2000). Business dynamics: Systems thinking and modeling for a complex world. Boston, MA: McGraw-Hill.
- 52. Johannes M. Bauer Quello Chair in Media and Information Policy Chairperson, Department of Media and Information Michigan State University, East Lansing, MI 48824, USA bauerj@msu.edu | http://www.msu.edu/~bauerj | @jm_bauer 52