Causal Reasoning in Real Life

This page contains a list of links to case studies that illustrate various aspects of making inferences about causality from statistical data. These case studies are intended to provide students with an opportunity to practice using the sometimes dry and technical concepts of this course to critically reason about scientific results reported in the brief and engaging journalistic format often encountered in a variety of media sources. The case studies are mostly drawn from online versions of standard news organizations, such as the British Broadcasting Corporation or The New York Times. Some of the case studies are journalistic style summaries authored by people who have worked on the Jcourse project. In a few instances, the case studies are entirely fictional, usually created to illustrate the amusing absurdities that can result from mistaking association for causation. All fictional case studies are clearly marked as such.

How this Index is Organized

The case studies in this index are organized according to two features: the concepts taught in the course that case studies illustrate and keywords that give some indication of the issues discussed in the case study. The keyword list is entirely open-ended, being limited only by the diversity of topics covered. The concepts, on the other hand are limited to seven general categories, which are listed below along with explanatory comments.

Concepts

Variables

Case studies falling in this category can be used to practice identifying the variables involved in the causal claim at issue in the case study. This concept tag is appended to every case study in this index.

Causation vs. Association

These are articles that can be used to help students grasp the difference between association and causation. Their general form is the following. The article reports that variable X was found to be associated with variable Y and then proceeds to conclude that X causes Y despite the presence of obvious, often more plausible, alternative explanations. Silliness is a big plus here. The more absurdly the article ignores obvious confounders or the possibility that it is Y that causes X rather than the other way around, the more apt it is to be classified under this heading.

Paradigm example: A study finds an association between having more sex and appearing more youthful. Conclusion: having more sex makes you look younger. The possibility that a more youthful appearance causes having more sex is not considered.

Relative Frequencies/Associations

These are articles that students can use as practice for ferreting out claims about relative frequencies from blocks of text. All case studies falling under this concept label contain sufficient information to assign an at least approximate numerical value to at least one relative frequency. Most commonly, however, cases studies in this category give numerical values for two conditional relative frequencies that make it possible to decide whether a pair of the variables of interest are associated.

Paradigm example: A study about the effectiveness of an early childhood education program on subsequent academic performance reports that 49.7% of those who participated in the program graduated from high school while the graduation rate was only 38.5% among those who did not participate in the program.

Causal Graphs

Case studies falling under this label contain causal claims that students can practice representing by means of causal graphs. Almost all cases studies in the index fall in this category.

Confounders

These are articles in which plausible confounders pose a problem for inferring a causal claim from associations drawn from non-experimental investigations. Such articles differ from those in the causation vs. association category primarily in virtue of being less silly and in that some of them actually describe efforts of the researchers to take some of the potential confounders into account, for example by conditioning on them.

Paradigm example: A study finds that depressed men are more likely to suffer from heart disease than those who are not depressed. The article lists several possible confounders that were taken into account, such as high blood pressure and smoking, although some seemingly obvious confounders, e.g. alcoholism, are omitted. Moreover, the possibility that illness is a cause of depression is not mentioned.

Experiments

Both randomized and non-randomized controlled experiments are included here. Articles in this category are intended to be useful in helping students understand how experiments differ from observational studies, as well as helping students to recognize the sorts of practical difficulties often arise in implementing an experiment and the strategies for dealing with them.

Paradigm example: A study to test whether the drug piracetam is effective in boosting the cognitive capacities of Down syndrome children concludes that the drug has no positive effect. Subjects received either the drug or a placebo. The article does not state whether treatment assignment was randomized, which it may well not have been given the small sample size (18).

Interventions

Every case-study listed under the "experiment" heading is also included in this category, though not vice versa. Besides clinical or laboratory experiments, interventions also include cases of policy initiatives aimed at combating some social or public health problem. Articles falling under this concept heading are intended to be useful for helping students to apply the abstract notion of an intervention to concrete cases, and for helping them to understand how real-life interventions can fail to live up to the ideal.

Paradigm Example: One study describes an effort of two Colombian cities to reduce astronomically high homicide rates by implementing a ban on carrying guns on days in which violence is thought to be most probable.

Overview || Concept Modules || Case Studies || Applets & Shockwave ||
System Requirements || Personnel || Related Links || Users