Causal Inference for Policies, Interventions and Experiments

SOST30172 Causal Inference for Policies, Interventions and Experiments 20 Semester 2 Undergraduate Level 3 Eduardo Fe Rodriguez Unit coordinator Social Statistics 1 FHEQ level (Framework for Higher Education Qualifications) ' Last part of a Bachelors ' European Credit Transfer & Accumulation System Rating : 10.0 Researchers, government, policy takers, business leaders and people in general are motivated by "causal questions" of the type "Does X cause Y" (e.g does policing reduce crime? Do minimum wages increase unemployment? Does a new educational innovation increase educational achievement? Does a new policy reduce waiting lists in hospitals? Does expenditure in marketing increase sales? Dos affirmative action reduce discrimination?). Standard statistical methods, regardless of their complexity, cannot answer these questions on their own and a new set of statistical tools are needed.This unit introduces the modern methods of causal inference. You will learn Rubin's Potential Outcomes framework, and how to use this framework to clarify what data can tell you about a causal effect of interest. You will learn various methods to estimate causal effects from observational and experimental data. Critically, you will be able to gain a deep understanding of the role that different assumptions play in determining what one can learn from data regarding causal questions. The skills you can acquire in this course are applicable to explore causal questions and undertake policy evaluation in a myriad of fields, including economics, criminology, sociology and politics, development, medicine, epidemiology or psychology, to mention but a few.    When trying to make sense of the world, people are generally motivated by causal questions. Researchers, policy takers, business leaders and people in general want to know what would happen to a person or a group of people when one changes their environment (through a new policy, innovation or intervention). Examples of causal   questions include:  <ul> <li>Does policing reduce crime?  </li> <li>How does parental absence affect child and adolescent development?  </li> <li>Does expenditure in marketing lead to increases in sales?  </li> <li>Do new medical innovations improve health?  </li> <li>Do private or selective schools deliver better education than public schools?  </li> <li>Do minimum wages increase unemployment?  </li> <li>Which areas of affirmative action reduce discrimination?  </li> <li>Does confinement of populations or school closures help to stop a pandemic or is social distancing enough?  </li></ul>All these are cause-and-effect questions of the type: Does X cause Y?  Somehow surprisingly, standard statistical methods, on their own, cannot answer these questions, and a new set of statistical methods and ideas are needed. This problem arises because to estimate the effect of a policy or an innovation at a given point in time, we need to observe each person in a sample under both the new policy or innovation and in the absence of the new policy or innovation. In real life, we can only observe a person in only one state of the world at the time.  In this course, we introduce the statistical language and methods of causal inference. You will learn Rubin's Potential Outcomes framework, and how to use this framework to clarify what data can tell you about a causal effect of interest. You will learn various methods to estimate causal effects from observational and experimental data. Critically, you will be able to gain a deep understanding of the role that different assumptions play in determining what one can learn from data regarding causal questions. Overall, this module will help you to give well grounded, evidence-based and credible answers to causal questions using data.   Specifically, by completing this module, you will:  <ul> <li>Learn to estimate causal effects and answer causal questions in a rich variety of situations ranging from experimental settings to observational data from irregular assignment mechanisms  </li> <li>Understand the role played by assumptions in the identification of causal effects in different settings  </li> <li>Become acquaintance with a wide range of estimation and inferential methods for causal models in design based setting (e.g. instrumental variables, regression discontinuity, difference in difference) and model based settings (panel data, matching methods) as well as some more advanced techniques (principal stratification and partial identification)  </li> <li>You will learn to use the free software R to implement those statistical methods  </li></ul>The skills you will learn in this module are relevant in a vast range of fields spanning both sciences and social sciences and can be applied without modification to causal problems in economics, criminology, sociology and politics, development, medicine, epidemiology or psychology, to mention but a few areas. Therefore, upon successfully completing the module, you can apply the acquired skills to advice about causal questions to government or business leaders in any field, to work as an statistician or econometrician in a vast range of research think tank or a large corporations, or you could seek to expand your skill set in a post-graduate degree with a statistical or econometric component. Student should, at the end of this course, be able to:  <ul> <li>Select, among a pool of competing methods, those most appropriate to estimate the effect of a policy, experiment or intervention on an outcome of interest.  </li> <li>Implement the selected estimator using widely available software such as R.  </li> <li>Successfully write a report explaining and supporting the findings of their analyses.  </li> <li>Conditional on having a clear policy or research question, students will be able to design policies, experiments and interventions to estimate causal effects.    </li></ul> Teaching will be based on asynchronous lectures, regular exercises, and a weekly live session.  Please note the information in scheduled activity hours are for guidance only and may change.   0 Other 100% Written assessment 60% Set exercises 40%  Students will receive summative feedback which gives you a mark for your assessed work.The School of Social Sciences (SoSS) is committed to providing timely and appropriate feedback to students on their academic progress and achievement, thereby enabling students to reflect on their progress and plan their academic and skills development effectively. Students are reminded that feedback is necessarily responsive: only when a student has done a certain amount of work and approaches us with it at the appropriate fora is it possible for us to feed back on the student's work. The main forms of feedback on this course are written feedback responses to assessed essays and exam answers.We also draw your attention to the variety of generic forms of feedback available to you on this as on all SoSS courses. These include: meeting the lecturer/tutor during their office hours; e-mailing questions to the lecturer/tutor; asking questions from the lecturer (before and after lecture); presenting a question on the discussion board on Blackboard; and obtaining feedback from your peers during tutorials.   A previous course on statistical methods (e.g. ECON10072, SOST10062, MATH10282). Y Rosenbaum, P. (2017) Observation and Experiment: An Introduction to Causal Inference. Cambridge University Press.  Imbens, G. and Rubin, D. (2015) Causal Inference for Statistics, Social, and  Biomedical Sciences, An Introduction. Cambridge University Press.  Morgan and Winship, (2015). Counterfactual and Causal Inference: Methods and Principles for Social Science Research. Cambridge University Press.  Manski, C. (2007) Identification for prediction and decision. Harvard University Press.   Lectures 20 Practical classes & workshops 10 0 170