Consistency of Interest and Perseverance of Effort

What it is

The two subscales come from Angela Duckworth's founding 2007 paper, where twenty-seven candidate items were cut to twelve, an exploratory factor analysis on half of a sample of 1,545 retained a two-factor oblique solution with six consistency items and six perseverance items, and the two factors correlated at r = .45. Confirmatory fit on the other half was poor and the paper reports it: a comparative fit index of .83 and a root mean square error of approximation of .11.

The sentence that justified using one total score is in the same paper, and it is the one the later evidence contradicts: that neither factor was consistently more predictive of outcomes than the other, and that in most cases the two together were more predictive than either alone.

The asymmetry, from the 2017 meta-analysis of 584 effect sizes across 88 independent samples and 66,807 people, is this. Against overall academic performance, perseverance correlates .26 and consistency .10, both on eleven samples and 5,221 people. Against high school grades, .29 against .13. Compared formally by the Meng, Rosenthal and Rubin procedure, the correlations differed significantly in all four cases tested. Credé, Tynan and Harms write that contrary to early assertions that both facets predicted success outcomes equally well, the perseverance facet exhibited much stronger relations with all academic performance criteria.

And the half a fair account must print: perseverance alone does add over conscientiousness, with increments in explained variance of .040 for overall academic performance, .085 for high school grades and .023 for college grades, surviving controls for conscientiousness and consistency together. The authors call that encouraging. Their own caveat on their own encouraging finding is that such incremental validity would be observed even if grit were simply a different manifestation of conscientiousness, which is what their other results suggest.

It replicates on more than four times the sample, which is the most important durability fact in the dispute. Lam and Zhou's 2022 cross-cultural meta-analysis covers 137 studies, 156 dependent samples and 285,331 people, giving overall grit with academic achievement at r = .19, perseverance at .21 and consistency at .08. Those figures reach this site through the vault's own earlier verification record rather than through a read of the paper, because the abstract could not be retrieved.

The two-factor structure is also attacked from the opposite direction, and the two attacks agree about the instrument while disagreeing about everything else. Jon Jachimowicz of Columbia Business School and colleagues argue in their 2019 PNAS reply that all perseverance items are positively worded and all consistency items negatively worded, which can produce a spurious two-factor structure. They then tested it: on an all-positive version of the scale with 958 participants the subfacets correlated at r = .95, with shared variance of .90 against average variance extracted of .55 and .47, so discriminant validity fails; on an all-negative version with 781 participants they correlated at r = .90. Their conclusion is that the Grit-S measures only perseverance and that it is most appropriate to collapse it into a single measure of that. Vazsonyi and colleagues reached the same artefact conclusion independently; that paper was not read and reaches this page through the reply's quotation of it.

The originators have conceded twice, and both concessions are theirs rather than extracted. Southwick, Tsay and Duckworth write in 2020 that streamlining the scale, and in particular omitting two items that explicitly noted a time frame of multiple years, may have been unwise, because what distinguishes grit from other goal-relevant traits is duration, and that the two subscales are more distinct in the original twelve-item scale at r = .47 than in the eight-item version at r = .57. And a 2021 commentary by Duckworth, Quinn and Tsukayama concedes that the two-factor hierarchical model cannot be distinguished from a model with two correlated subfactors and no higher-order factor. That commentary could not be retrieved and reaches this site through the vault's own record of 2026-09-24.

In effect

The two correlations that should not be averaged

The research case is the .26 against .10 pair, and the formal test that they differ. Averaging two predictors with that gap into one score is not a neutral summarisation: it dilutes the stronger one, and where the weaker one carries negative coefficients in a joint model it can do worse than dilute. A reader who wants the usable part of this literature should ask for the perseverance subscale and ignore the total.

Sticking with it, against never changing your mind

In ordinary language the two halves are different virtues. Perseverance of effort is finishing things and working through setbacks. Consistency of interest is not switching goals, which the scale measures with reverse-worded items about new ideas and projects distracting you from previous ones. Those are not obviously the same quality, and the evidence says the first predicts outcomes and the second barely does. The popular framing of grit as sticking with one thing for years has, on this evidence, picked the half that does not work.

What a school or employer should actually measure

The practical consequence is narrow and clear. If an institution wants a questionnaire that predicts completion, the best-supported item set is the perseverance subscale, and the established alternative is a conscientiousness inventory with a longer validation record. There is one strong independent result on the other side, and it belongs here: Sule Alan of the University of Essex, with Teodora Boneva and Seda Ertac, report randomised educational interventions in two independent elementary school samples with an effect of about 0.2 standard deviations on a standardised mathematics test two and a half years later, measured partly on an incentivised real-effort task rather than a questionnaire. That abstract is all this site has read of it, and it answers the charge that grit is not responsive to intervention.

What it does not say

It does not say perseverance adds nothing. It adds, on the critical paper's own regressions, and the authors call the finding encouraging.

It does not say the two-factor structure is established. One camp says the facets behave too differently to combine, the other says they are a wording artefact and there is only one, and both conclude the total score should not be used as it stands.

It does not say the originator defends the total score unconditionally. Her own 2020 paper says the facets are more distinct in the longer scale and that shortening it may have been unwise, and she states that she respects researchers' decisions to study the components independently.

It does not carry a magnitude from the trade literature. No effect size, correlation, odds ratio or coefficient appears in the book that popularised the construct.

Who developed it

The two subscales are Angela Duckworth's, Rosa Lee and Egbert Chang Professor at the University of Pennsylvania, from the 2007 paper with Christopher Peterson, Michael Matthews and Dennis Kelly. The asymmetry is Marcus Credé's first-authored 2017 meta-analysis at Iowa State University. The artefact argument is Jon Jachimowicz's at Columbia Business School, and the item-level factor work is Annette Ponnock's.

---