Self-control, and what happened to the famous account

The course changes subject here, and it is worth saying what the join is.

Lessons 1 to 4 were about habit, which is the case where a behaviour has stopped needing a decision. This lesson and the next are about what happens when a decision is needed after all, which is most of life, and which is what people call willpower.

And there is a second reason this lesson exists in a course about habits. The best-known account of willpower didn't survive being tested properly, and watching that happen is worth as much as anything else in this course. You'll meet the resource model everywhere, including in the books that will tell you about habits.

What was claimed

The resource or strength model of self-control's easy to state, which is part of why it travelled.

On this model self-control is a limited resource that gets depleted by a period of exertion, and self-control failure is what happens when it runs low.¹ That is this course's wording rather than the paper's, because the research file does not hold the sentence verbatim. Use it, and there is less of it: resist a biscuit this morning and you have less left for the argument this afternoon.

The way it was tested is called the sequential-task paradigm. People do one task requiring self-control, then a second unrelated one, and the prediction is that the first makes the second worse. The state this is supposed to produce has a name: ego depletion.¹

And the model was not fringe. A 2010 meta-analysis of 198 published experiments reported the depletion effect as medium-to-large, at d = 0.62 with a confidence interval from 0.57 to 0.67.² Two of that meta-analysis's four authors are the proposing authors of the replication that follows, which is worth knowing before you read it as a hit job: the people who assembled the best case for the effect are the people who then ran the test it failed.

One thing to keep hold of before the rest of this lesson. The claim under test is about the mechanism: whether one act of control makes the next one harder. It isn't the claim that self-control matters. The replication's own abstract opens by granting the point: good self-control has been linked to better health, to more cohesive personal relationships, to success at work and at school, and to less susceptibility to crime and addiction.¹ That summary is this course's, not a quotation, for the same reason as above. None of that was ever in question, and it still isn't, and a reader who finishes this lesson doubting it has read it wrongly.

What happened to it

Three steps, in order, and each one tightened the method.

First, somebody asked what publication bias would do to that meta-analysis. Small studies tend to report larger effects, and the usual explanation, which this course is giving you as the standard account rather than as something it has verified,⁵ is that a small study finding nothing is less likely to be written up and accepted. Carter, Kofler, Forster and McCullough ran a series of meta-analyses correcting for exactly that.

Their conclusion, in their own words, and read the middle of the sentence: "We find very little evidence that the depletion effect is a real phenomenon, at least when assessed with the methods most frequently used in the laboratory."³

That qualifier is not decoration. They're saying something about a literature and a method, not about human beings.

Second, a registered replication. Twenty-three laboratories agreed a single protocol in advance, published it before collecting data, and ran it. Total participants: 2,141.¹

Third, the result. "Meta-analysis of the studies revealed that the size of the ego-depletion effect was small with 95% confidence intervals (CIs) that encompassed zero (d = 0.04, 95% CI [−0.07, 0.15]."¹

And the detail that does most of the work. The paper's own front matter records that the protocol was vetted by Roy Baumeister,¹ who originated the model being tested.

What the authors say their own result cannot settle

This is the part that almost never travels, and it is why the lesson's first exercise is to read it.

The replication has a limitations section running to pages, and the authors raise problems with their own study that a critic would have had to work to find.

The task may not have induced what it was meant to induce. Their depletion task "did not include an initial period in which individuals familiarized themselves with the no-depletion version",¹ and that period is supposed to induce the habitual response participants then override. They judge the omission unlikely to have been decisive, and both of those sentences are this course's summary of their argument rather than their words.

Or it may not have gone on long enough. "It is possible that the letter 'e' task was sufficiently arduous but not of sufficient duration to deplete individuals' self-control resources."¹ They raise it themselves and leave it open.

Notice what they're doing. They have a striking null result, and they spend pages on the reasons it might be wrong. That's what a paper looks like when the people who wrote it care more about the answer than about having been right.

Worked: watching an effect shrink

Set the three estimates out in order with what produced each. The table is this course's own assembly of figures from three papers; no single source presents them together.⁵

Read down the middle column rather than the right-hand one. Nothing about people changed between 2010 and 2016, so far as anybody knows. What changed was the method: from pooling what had been published, to asking what publication had filtered out, to fixing the protocol in advance and running it in 23 places at once.

That is the general lesson and it is bigger than this subject, and it is this course's reading of the three rows rather than a claim any of these three papers makes.⁵ An effect that shrinks as the method tightens is the signature of a literature rather than of a phenomenon. You will use it again.

What the model's defenders say

This course read no source written by a defender of the resource model, and that is a real limitation of this lesson: the 2010 meta-analysis was not opened, and everything above comes from papers arguing the other way. So take the defence from the fairest place available, which is the process model's own paper, whose authors say in their conclusion that they admire and have contributed to the evidence for the limited-resource view.⁴

Their footnote records what the model's defenders propose. The resource account, they write, "has been stretched to accommodate such results by suggesting that initial acts of self-control only partially deplete the resource and that ego depletion occurs because people are unwilling to draw further from their reserves".⁴ That is called an energy-conservation account, and it matters for reading the rest of this lesson: on that version, a person who performs badly on the second task has not run out, but is holding something back.

Notice what it does to the tests. A conservation account expects performance to come back when there is a reason to spend, which is a different prediction from a store that is simply empty, and it is the version a defender will give you rather than the muscle. A reader who takes "willpower is a muscle" as the only thing the resource model ever said is arguing with the popular version rather than with the position.

Four things people get wrong about this

"Willpower is a muscle that runs out." The model that says so has lost most of the evidence that supported it, in the paradigm it was tested in.

"So willpower is a myth." Overshoots, and the replication's authors would be the first to say so. An interval spanning zero is not a demonstration of absence, and they list their own reasons the result might not be the last word.

"The replication proved self-control does not matter." It tested a mechanism, which is a narrower thing. The outcomes are in its own first sentence and were never at issue.

"Science reversed itself, so nothing is known." This is the reading to be most careful about, because it feels like sophistication and it's the opposite. What happened here is the process working: a claim was made, it was tested with better methods, the estimate did not hold, and the people involved published the result and their own doubts about it. A field that couldn't do that would be worse, not better.

Practice

Connections

Back. Lessons 1 to 4 were about removing the decision; this lesson is about the occasions when there is no association to fall back on. Lesson 2's habit of putting a figure back beside what produced it is the method of the worked table here, at a larger scale. Lesson 4's twenty-one days was a number with no study behind it; this is a number with 198 studies behind it that still did not hold, which is a different and more unsettling case. From earlier Core courses: Logic and Argument lesson 6 is what the predict block is built on, and Reading Well lesson 8 is what the first exercise asks you to do.

Forward. Lesson 6 is what is proposed instead, in its proponents' own terms, and why a less tidy account can be better. Lesson 7 puts the whole subject's numbers side by side and asks you to read a claim about your habits.

Go deeper

• A Multilab Preregistered Replication of the Ego-Depletion Effect (Perspectives on Psychological Science, 2016). This course has read the abstract and the limitations section. The publisher's page is paywalled and open copies are posted by several universities, so it is worth a search by title. Read the limitations beside the abstract, which is what the first exercise asks: it is a model of how to report a result you did not want.
• A Series of Meta-Analytic Tests of the Depletion Effect (Journal of Experimental Psychology: General, 2015). This course has read the abstract only, from a copy posted by one of its authors. The abstract alone is worth it for the qualifier in its conclusion, which is the thing that never travels with the finding.

Sources

1. M. S. Hagger, N. L. D. Chatzisarantis and 55 others, "A Multilab Preregistered Replication of the Ego-Depletion Effect", Perspectives on Psychological Science 11(4), 2016, pages 546 to 573. The abstract and the limitations section were read verbatim; the rest of the paper was not. Supports: the name of the sequential-task paradigm; k = 23 and N = 2,141; the quoted result with its confidence interval; the front matter recording that the protocol was vetted by Roy Baumeister; the quoted fragment about the missing familiarisation period; and the quoted limitation about the task's duration. The statement of the resource model and the summary of what good self-control has been linked to are this course's wording, not the paper's, because research/SOURCES.md does not carry those sentences verbatim, and the body says so at both.
2. M. S. Hagger, C. Wood, C. Stiff and N. L. D. Chatzisarantis, meta-analysis of ego-depletion experiments, 2010. Not opened by this course. The figure of 198 published experiments and d = 0.62 with a confidence interval of 0.57 to 0.67 is taken from source 3, which quotes it. Note that two authors of this meta-analysis are the proposing authors of the replication in source 1, which is worth knowing and is recorded on that paper's own front matter.
3. Evan C. Carter, Lilly M. Kofler, Daniel E. Forster and Michael E. McCullough, "A Series of Meta-Analytic Tests of the Depletion Effect: Self-Control Does Not Seem to Rely on a Limited Resource", Journal of Experimental Psychology: General 144(3), 2015. Abstract read verbatim from a copy posted by one of the authors; the rest was not read. Supports the quoted conclusion including its qualifier, and the 2010 figures it quotes.
4. Michael Inzlicht and Brandon J. Schmeichel, "What Is Ego Depletion? Toward a Mechanistic Revision of the Resource Model of Self-Control", Perspectives on Psychological Science 7(5), 2012. Read in part: the abstract, the overview of the process model, the motivation section, the conclusion and footnote 2. Supports: the existence of the process account, which lesson 6 takes properly; the quoted energy-conservation statement of what the resource model's defenders propose, which is in this paper's footnote 2 and is attributed there to Baumeister and Vohs, 2007; and the conclusion's statement that these authors admire and have contributed to the evidence for the limited-resource view. The defence in this lesson is therefore taken from a paper arguing against the model, which the body says, because no source written by a defender was read for this course.
5. The three-row table is this course's own assembly, stated as such where it appears. No source read for this course presents those three estimates together, and the reading of a shrinking estimate as the signature of a literature is the course's own inference rather than a claim any of these papers makes, marked inline where it appears. The account of why small studies report larger effects is also given as the standard explanation rather than as something this course has verified, and the body says so at the sentence.