Notes on Experimentation, Variables, and Causal Inference

Experimental Design and Variable Concepts

  • Experimental thinking and lurking variables

    • The import of an experiment is to control for or account for all possible lurking variables you could conceive of. This helps isolate the effect you’re trying to study.
    • Caution: experimentation is not a panacea; it’s a powerful tool but has limits depending on design, implementation, and feasibility.
    • In the discussion, there’s an idea of using experiments with varying conditions (e.g., in a firefighting scenario) to observe outcomes under different treatments.

  • Pedagogical approach to the topic

    • The plan is to open with an activity rather than starting with abstract definitions right away.
    • Instead of jumping straight to definitions, the instructor will use an example first and then highlight the definitions in red inside the body of the activity so students can see what the terms mean in context.
    • This approach helps connect theory to concrete illustrations and keeps students engaged by showing practical relevance.

  • Core concepts: variables, axes, and terminology

    • In a typical data display, you have an x-axis and a y-axis.
    • In algebra class, the x-axis variable is called the independent variable and the y-axis variable is called the dependent variable.
    • In statistics class, the x-axis variable is called the explanatory variable and the y-axis variable is called the response (or dependent) variable.
    • The central question is: can we determine if X causes Y? This is the “eye on the prize.”
    • In the face of lurking or confounding variables, what can we do to bolster the claim that X causes Y?
    • A scatterplot is a common visualization: you plot the independent/explanatory (X) on the x-axis and the dependent/response (Y) on the y-axis.

  • Mathematical and conceptual framing

    • Without considering other variables, a simple model can be represented as:
    • Y=f(X)+εY = f(X) + \varepsilon
    • Here, $X$ is the independent/explanatory variable, $Y$ is the dependent/response variable, and $\varepsilon$ is the random error term.
    • When lurking variables are considered, a more complete model may include them explicitly:
    • Y=f(X,L)+εY = f(X, L) + \varepsilon
    • where $L$ represents one or more lurking (confounding) variables that might influence $Y$.

  • Goals of the activity and causal inference

    • The overarching goal is to determine whether X has a causal influence on Y, not just a correlation.
    • To strengthen causal claims in the presence of potential confounders, researchers consider experimental design features such as randomization, control groups, and replication (though such specifics aren’t detailed in the excerpt, they are implied by the discussion of lurking variables and the need to bolster causal inference).

  • Example references and context from the transcript

    • An example scenario referenced involves firefighters and an accompanying scatterplot idea; the instructor mentions sending firefighters under different conditions to observe outcomes, illustrating how an experiment might be structured to assess causality.
    • This serves as a concrete illustration of how you might think about X (an experimental condition or treatment) and Y (an observed outcome).

  • Group activity logistics

    • Students are organized into groups to work on the experiment-related activity.
    • The password for accessing the activity is toil (lowercase).
    • The class is encouraged to stay focused on the main objective (causal understanding) while engaging with the activity.

  • Summary of key takeaways

    • Experiments are used to control for lurking variables and to strengthen causal claims, but they are not a universal remedy for all research questions.
    • In algebra, X is the independent variable and Y is the dependent variable; in statistics, X is the explanatory variable and Y is the response.
    • The main question is whether X causes Y, and the design should aim to mitigate confounding factors to support that claim.
    • Visual tools like scatterplots help illustrate relationships between X and Y and guide thinking about causality.

  • Notation and formulas to remember

    • Basic relationship (without confounders): Y=f(X)+εY = f(X) + \varepsilon
    • With potential confounders: Y=f(X,L)+εY = f(X, L) + \varepsilon
    • Conceptual mapping:
    • X→YX \rightarrow Y (causal direction to investigate) with potential lurking variables $L$ affecting the relationship.

  • Quick reminders

    • Group work and access: password is toil (lowercase).
    • The wave of the lesson emphasizes connecting concrete examples to definitions and focusing on whether a causal link can be established, not just observed.