Research Summaries: Establishing Causality Using Longitudinal Hierarchical Linear Modeling: An Illustration Predicting Achievement From Self- Control

Today’s research summary is slightly technical. It is based on this paper [pdf] by Angela Duckworth et al that shows a causal relation between self-control and academic achievement.

The Illustrated Sutra of Cause and Effect. 8th...

The Illustrated Sutra of Cause and Effect. 8th century, Japan (Photo credit: Wikipedia)

  1.  Some personality variables like self-control predict important life outcomes. It is well know that self-control as measured at age 4 (using the marshmallow test) can predict important life outcomes years later. However, prediction may not imply causality as a third factor may be responsible for causing both the phenomena under consideration.
  2. The test for causality is a) causal variable must precede the effect in time; b) the causal variable and outcome variable should be correlated; and c) any third party confound or variable should be ruled out. This is easy to achieve in  double blind randomized placebo controlled experiments, but personality traits like self-control are hard to manipulate as trait variables in experimental settings.
  3. Typically personality traits and their outcomes are studied using a longitudinal study design where changes in say self control at time T1 are correlated with outcomes like academic achievements at a later time T2, of course measure other confounding variables and factoring their effects; thus self-control, along with IQ, may be measured at the beginning of a school session and at the end of session the CGPA obtained will be used to find whether and how much self-control led to academic achievement. This however cannot establish causality in a strict sense as not all variables of interest can be identified and measured. Often the dependent variable (CGPA in our case) is itself controlled for to ensure that a higher CGPA at point T1 does not lead to higher CGPA at time T2 independent of self-control at T1.
  4. To take care of third party confounding variables, Angela et al used growth curve analysis with Hierarchical Linear Modelling (HLM). This involves taking multiple measures of say self -control at different times and also multiple measures of the outcome say CGPA. The independent variable is considered a time varying co-variate and used to figure the within-person relationship between the two variables of interest. Consider a between subjects confound like socio economic status (SES) that could potentially lead to different outcomes (CGPA) – if not controlled for the self control- CGPA relation arrived at by analysis of between subjects data might lead to erroneous conclusions. However, a stable thing like SES (which doesn’t change with time and is constant for an individual) will have no impact on the correlation or causal relation between how changes in self-control affect CGPA over time in the same individual.
  5. The direction of causality can also be ascertained by using HLM with reversed time lagged, time varying co-variates. What this means os that we can try to see of the causal arrow runs in other direction by taking measures of CGPA as predictor and  self control as outcome variable.
  6. In this study, self control was measured using self-report, parents and teachers ratings of students for four consecutive academic years (as they moved from fifth grade to eighth grade) using the Brief Self-Control Scale  ; CGPA was measured each year as the outcome variable. Self-esteem and IQ was also measured and so was gender, ethnicity etc.
  7. They found that  self control measured 6 months earlier predicted CGPA six months later; average self-control predicted the baseline CGPA as well as the slope of CGPA changes (how fast the CGPA increased or decreased over time). Howsoever, the reverse analysis whereby short term CGPA was used to predict self-control gave negative results thus establishing the causal direction.
  8. It was thus established that self-control does indeed cause or lead to higher academic outcomes like higher CGPA.  A limitation of the study was that a time varying third variable that increased and decreased in tandem with self-control can still account for the relationship between self control and academic achievement.

I liked the paper,  though its more methodological. You can find the full paper here [pdf].

Effecient Related Posts: