Lecture 05b Activity

Unit B · Chapter 05

An in-class activity. Nothing to turn in and no answer key – this one needs other people, which is why it happens in class rather than at home.

The idea. The same correlation can be a good result, a useless one, or a disaster. Being assigned to defend one you did not choose is the fastest way to find out what you were smuggling in.

Format. About 10 minutes · pairs, then whole class · no laptop needed.

1. Your assignment (2 minutes)

Each pair is assigned one of these four studies. All four report \(r = .30\), \(p < .05\).

A. A 15-year study of 40,000 people: childhood adversity and adult depression.

B. An undergraduate project, \(n = 45\): hours of sleep and exam score.

C. A new 4-item screening tool against a full clinical diagnosis.

D. Two administrations of the same questionnaire, one week apart.

Your job is to argue that yours is the most useful result in the room. You do not get to pick.

2. Build your case (5 minutes)

• What does the .30 buy the researcher in your study, concretely?
• What is the strongest attack another pair will make on you, and what is your answer?
• Find the pair with the weakest study and prepare one sentence against them.

3. The floor (3 minutes)

One sentence each, in order A, B, C, D. Then one round of rebuttals. Then the room votes on a ranking.

If you have more time: take a correlation you believe from your own area. Your partner proposes three third variables that could produce it with no direct relationship, and you say which could be ruled out by measuring something and which only by designing the study differently.

If you were not in class

A and D are the informative ones; B and C are the problems. An \(r\) of .30 is a respectable effect for a life-course risk factor, and with \(n = 40{,}000\) it is precisely estimated. It is a poor screening tool – you would misclassify constantly. And it is a disastrous test-retest reliability: the questionnaire is mostly measuring noise, and every effect you estimate with it will be attenuated, which is Chapter 13’s subject. Study B’s interval runs roughly from .01 to .55, so the honest summary is “somewhere between nothing and a lot.”

Whoever drew D usually discovers mid-argument that they are defending the worst result in the room, and that is the useful moment. The four studies share a number and share nothing else; every reaction the room has comes from context the number does not carry. That is the argument for never reporting a correlation without an interval and without saying what it is a correlation for.