An intelligence test asks how well a student can reason under test conditions. A final grade asks something broader and, in some ways, more mundane: what the student learned, whether assignments were completed, how regularly they worked and how they performed across months of ordinary school days.
That distinction sits underneath Angela Duckworth and Martin Seligman’s much-cited 2005 finding. Among eighth-graders followed from autumn to spring, self-discipline predicted report-card grades more strongly than IQ.
This is one study, not settled consensus. It followed pupils at one American public magnet school, and it found predictive associations. It did not prove that self-discipline caused higher grades or establish a permanent ranking between character and intelligence.
The headline comparison came from the second cohort
The paper in Psychological Science actually contained two longitudinal studies. The first followed 140 eighth-graders. A replication conducted the next school year involved 164 students and added three things the sharper comparison required: a group IQ test, a behavioural delay-of-gratification task and a questionnaire about study habits.
Silicon Canals previously looked at the first cohort through the tension between immediate comfort and longer-term performance. The sample distinction is worth making explicit: the direct self-discipline-versus-IQ result in the title comes from the 164-student replication.
The researchers measured self-discipline in the autumn, then collected final grades and several other school outcomes in the spring. This temporal order is useful because the predictor came before the outcome. It is still an observational design from one school, not an experiment in which students were randomly made more disciplined.
Self-discipline was a composite, not a pep talk
Duckworth and Seligman did not reduce self-discipline to one student answering, “Yes, I work hard.” Their composite combined student self-reports, ratings from parents and teachers, hypothetical choices between a smaller reward now and a larger reward later, and an actual delay-of-gratification choice. Study 2 also included questions about study habits.
That multi-source design is one reason the study remains interesting. A student might present themselves one way while a teacher or parent sees a different pattern. Combining the views can reduce the noise of any single report.
IQ, meanwhile, was measured with the group-administered Otis-Lennon School Ability Test. The measurement asymmetry deserves attention. Self-discipline was a broad composite made from several methods and observers; IQ came from one instrument. The discipline measure may therefore have captured a stable pattern especially well. Teacher ratings of discipline may also partly reflect the same classroom behaviour that later shaped teacher-assigned grades.
Nor was this the later, broader construct of “grit” for which Duckworth became famous. The 2005 study focused more narrowly on delaying gratification, regulating behaviour and following through on schoolwork.
What “more than twice as strongly” meant
In the 164-student cohort, the correlation between the composite self-discipline score and final grade-point average was .67. The correlation between IQ and final GPA was .32. Those figures describe how closely each measure moved with grades; neither says how much one factor caused the other.
When IQ and self-discipline were entered into the same regression model, the standardised coefficient was .65 for self-discipline and .25 for IQ. That is the basis for saying discipline predicted final grades more than twice as strongly. The association also remained after the authors accounted for first-marking-period grades, earlier achievement-test scores and measured IQ.
Self-discipline predicted six of eight academic-performance variables significantly better than IQ. These included selection into the school’s competitive high-school programme, attendance, homework time, less television viewing and beginning homework earlier.
But one result stops the story becoming “IQ does not matter.” On the spring standardised achievement test, self-discipline correlated .43 and IQ .36; the difference was not statistically significant. Intelligence was much closer when the outcome was a common test rather than a report-card grade.
Grades measure execution as well as learning
A later paper by Duckworth, Patrick Quinn and Eli Tsukayama examined that split directly. Across two prospective studies, self-control better predicted changes in report-card grades, while IQ better predicted changes in standardised achievement-test scores.
The difference makes sense once we look at the jobs these measures perform. A standardised test asks students to solve the same set of problems under similar conditions. A report card accumulates many different judgements: tests, assignments, projects, deadlines, classroom conduct and sometimes participation. It records knowledge, but also the repeated delivery of work inside a system.
A broader PNAS analysis of what grades and achievement tests measure made a related point using multiple datasets. Both cognitive ability and personality predicted educational outcomes, while grades captured personality-related behaviour more strongly than test scores did.
That turns the original study into a claim about institutions as much as individuals. Change how a school calculates grades and the predictive balance could change too. A course dominated by one final examination may reward a different combination of abilities from one built around weekly assignments, attendance and long projects.
A later study did not reproduce the hierarchy
In 2021, Alexander Vazsonyi and colleagues reported a two-year study of 589 sixth- and seventh-grade pupils. Both self-control and intelligence predicted later teacher-rated academic competence and school-reported grades, but intelligence was the significantly stronger predictor.
That result does not make the 2005 correlations vanish. It shows why the ranking should remain attached to its sample, measures and school context. The later study involved younger pupils, a larger group, different measures and a longer follow-up. Those differences can matter.
The broader literature still supports the idea that non-cognitive traits contribute to academic performance. A 2022 meta-analysis covering 267 independent samples and more than 400,000 students found conscientiousness was a robust predictor even after cognitive ability was accounted for. But conscientiousness is broader than the self-discipline composite used in 2005, and a meta-analysis showing an additional contribution is not the same as declaring it universally more important than intelligence.
The useful claim is narrower and less moral
The 2005 paper supports a modest point that intelligence tests can miss: intellectual ability does not automatically turn into consistent performance. A student can understand difficult material and still lose marks through missing work, irregular study, late starts or avoidable absences. Grades can reflect that gap between capacity and execution.
It would be a mistake to turn this into a character verdict. Behaviour is shaped by schedules, household stability, resources, expectations, teacher relationships and the way a school organises assessment. Naming the predictor “self-discipline” does not make those surrounding conditions disappear.
I have written before about routines reducing the number of difficult choices people repeatedly face. That lens fits this evidence better than a sermon about trying harder. What looks like discipline from a distance may partly be a reliable structure that makes follow-through easier.
Duckworth and Seligman captured something real in their 164 students: the habits visible across a school year carried information that one ability test did not. The durable question is not whether discipline defeats intelligence. It is what any grade, test or performance system is designed to notice, reward and leave invisible.
