Causality is the relationship between a cause and an effect, such as smoking and lung cancer. However, establishing causality is not always easy, especially when you cannot run controlled experiments. In this article, you will learn how to use natural experiments and quasi-experiments to infer causality from observational data.
Natural experiments are situations where an external factor, such as a policy change, a natural disaster, or a historical event, creates a random or quasi-random assignment of treatment and control groups. For example, the 2016 US presidential election was a natural experiment that affected the mental health of many Americans. By comparing the changes in mental health outcomes before and after the election across different groups of voters, you can estimate the causal effect of the election on mental health.
Whether natural or Quasi - These experiments are non random in nature and don’t classify participants (samples) into control & experiment group making it rather difficult to establish causality. The major limitation in inferring causation from these experiments is the presence of unmeasured confounding. There is a set of methods designed to control confounding is based on instrumental variables. interpretation of Instrument variables estimates depend on STRONG assumptions, the plausibility of which must be considered with regard to the causal relation in question. However, they are a useful methods for researchers, and if used with care they can provide info that may help with a question that may not be obtainable in any other way.
Intuitively we could need a more cleaver view on sort of structural inference and systemic hypothesis : systemic as a hierarchized tree of conditioning, mixed small circuits of interacting causes, for if with physics terminal scientific data. For some causal networks, could test if microcircuits composing the system, properly formulated (in our mind : logical - physics of objective fact - mathematical). Sort of classifications within pannel of data (commonly some 40 - 60 variables - parameters recorded). Care we have a special view on structural system : having to show interdependance. Not sure automated methods clear with that are use : we are more an epistemologist.
Quasi-experiments are situations where you can create or exploit a non-random assignment of treatment and control groups based on some criteria, such as age, location, or time. For example, you can use a regression discontinuity design to compare the outcomes of students who barely passed or failed a standardized test, assuming that the test score is the only factor that determines their treatment status. By controlling for other confounding variables, you can estimate the causal effect of passing or failing the test on the outcomes.
In order to analyze natural and quasi-experiments, you need to employ statistical methods that can account for potential discrepancies between the treatment and control groups, and isolate the effect of the treatment from other factors. Difference-in-differences is a common method that evaluates changes in outcomes between the treatment and control groups before and after the intervention, assuming that the trends in both groups are parallel in the absence of the intervention. Matching pairs each treated unit with a similar untreated unit based on observable characteristics, then calculates the average difference in outcomes between them. Instrumental variables utilize a variable that affects the treatment status but not the outcome as an instrument to estimate the effect of the treatment on the outcome.
Natural and quasi-experiments have some advantages and disadvantages compared to controlled experiments. For instance, they can overcome ethical and practical limitations, such as consent, cost, or feasibility. Moreover, they can capture real-world settings and behaviors, increasing the external validity and generalizability of the results. Additionally, they can provide new insights and opportunities for causal inference that would otherwise be unavailable. On the other hand, they can suffer from selection bias, confounding, or reverse causality, which can invalidate or weaken the causal claims. Furthermore, they can be sensitive to the choice of methods, assumptions, and data quality, and difficult to replicate or verify. Finally, they are subject to alternative interpretations or criticisms.
Causal inference is a complex and evolving field that requires both theoretical and practical knowledge. To hone your skills, you should become familiar with the fundamentals of causal inference, such as potential outcomes, counterfactuals, and identification strategies. Additionally, you should read and critique papers that use natural and quasi-experiments, evaluating their validity, reliability, and relevance. Furthermore, it is important to practice applying different methods and tools to analyze natural and quasi-experiments while communicating your findings and limitations.