Skip to main content
Social Sci LibreTexts

5.2: Okay, can you break down the reasons that correlations do not prove causation?

  • Page ID
    10345
  • \( \newcommand{\vecs}[1]{\overset { \scriptstyle \rightharpoonup} {\mathbf{#1}} } \) \( \newcommand{\vecd}[1]{\overset{-\!-\!\rightharpoonup}{\vphantom{a}\smash {#1}}} \)\(\newcommand{\id}{\mathrm{id}}\) \( \newcommand{\Span}{\mathrm{span}}\) \( \newcommand{\kernel}{\mathrm{null}\,}\) \( \newcommand{\range}{\mathrm{range}\,}\) \( \newcommand{\RealPart}{\mathrm{Re}}\) \( \newcommand{\ImaginaryPart}{\mathrm{Im}}\) \( \newcommand{\Argument}{\mathrm{Arg}}\) \( \newcommand{\norm}[1]{\| #1 \|}\) \( \newcommand{\inner}[2]{\langle #1, #2 \rangle}\) \( \newcommand{\Span}{\mathrm{span}}\) \(\newcommand{\id}{\mathrm{id}}\) \( \newcommand{\Span}{\mathrm{span}}\) \( \newcommand{\kernel}{\mathrm{null}\,}\) \( \newcommand{\range}{\mathrm{range}\,}\) \( \newcommand{\RealPart}{\mathrm{Re}}\) \( \newcommand{\ImaginaryPart}{\mathrm{Im}}\) \( \newcommand{\Argument}{\mathrm{Arg}}\) \( \newcommand{\norm}[1]{\| #1 \|}\) \( \newcommand{\inner}[2]{\langle #1, #2 \rangle}\) \( \newcommand{\Span}{\mathrm{span}}\)\(\newcommand{\AA}{\unicode[.8,0]{x212B}}\)

    Okay, can you break down the reasons that correlations do not prove causation?

    Yes, let’s start with a consideration of a typical bivariate concurrent correlation between two variables, and let’s pick variables that tap constructs we think could be causally connected, say, teacher involvement and student engagement (see Figure 23.1). Let’s say that in this research we get a robust correlation between good measures of both variables. Why can’t we conclude that teacher involvement influences student engagement? There are two main reasons. First, as also shown in Figure 23.1, the connection between these two variables could be due to a reciprocal causal effect, in which student engagement influences teacher involvement. This direction of effects is conceptually plausible, since more engaged students may attract more positive teacher attention and interaction whereas students who are more disaffected may lead teachers to withdraw from interactions or treat students more harshly. Of course, the correlation could be due to both feedforward (i.e., teachers’ influences on students) and feedback (i.e., students’ influences on teachers) effects.

    Insert Figure

    The second possibility is that there is no causal effect (forward or backward) between these two variables because they are both actually produced by another cause all together (the ominously named “third variable”); because they are both effects from the same cause, they covary, but the covariation is not causal, so it is called spurious. In our example, also shown in Figure 23.1, we selected students’ gender as our third variable because gender it is a plausible cause of both variables—girls are more engaged and teachers show more involvement with them whereas boys are less engaged and teachers show less warmth toward them. In this scenario, as in all other scenarios involving concurrent bivariate correlations, there are a very large number of third variables (alternative causes) that could be in play—it could be achievement (engaged students perform better in school and teachers attend more to high performing students) or social class or student sense of relatedness—as well as a large number of third variables that we can’t immediately imagine.

    So does this example tell us something about the problems we face in trying to get causal information out of naturalistic studies?

    Yes. In trying to extract causal information from a situation in which we just observe what is going on (i.e., a naturalistic design), we have two general problems, as well as a set of special issues that we have created for our own selves by choosing to belong to the relationalmeta-theory-club. The two general ones are: (1) causes naturally happen to people who were different before the causes ever landed on them, so we have to distinguish the pre-existing differences that may have attracted the causes to specific people from the effects of the causal experiences themselves; and (2) our potential causes naturally come in clumps and so we need to peel them apart in order to distinguish the active causal ingredients in these tangles.