Predictors of academic achievement

Last Updated on February 20, 2022

The purpose of this post is to cite predictors for academic achievement. I will focus on merely listing effect sizes rather than synthesizing the research. Note that when I say “predictor”, I’m referring to variables that are statistically associated with academic achievement. So this post does not establish that any of the variables below have a causal influence on academic achievement. Undoubtedly, some of the variables below do have a causal influence on academic achievement, but many of the variables do not. In order to determine whether a given variable has a causal influence, more analysis must be performed to determine if the association between that variable and academic achievement persists after controlling for confounding (as I’ve explained in this post).

The post will be split into sections. The first section will report correlation coefficients between various predictors and academic achievement (e.g., the correlation between intelligence and grades). The second section (TBD) will report standardized mean differences in achievement between pairs of groups (e.g., difference in mean grades between students with ADHD vs students without ADHD). The last section will cite sources which do not belong in the prior sections which provide useful data on predictors of academic achievement. I’ll try to focus on meta-analyses or very large samples when possible, though sometimes smaller samples will be provided if I cannot find a better sample reporting on the variables of interest.

The effect sizes will be reported in tables where each row contains the variable(s) in question, followed by the average effect size (r or d), the number of independent effect sizes (k), the number of subjects involved (N), and the source. A “Notes” column will sometimes be included indicating where to find the effect size in the source and any other useful information.

Correlation coefficients

This section will list correlation coefficients between academic achievement and various predictors. Some background on correlation coefficients is given here. Note that some of these correlation coefficients have been corrected for range restriction. Range restriction corrections are required when the sample is restricted to limited range of values for the predictor variable in question (e.g., the range of SAT scores for college students is much narrower than the range in the general population). When this occurs, I will list the corrected correlation in the r column and I will list the uncorrected correlation in the “Notes” column.

Academic achievement (primary school only)

This table lists predictors of academic achievement in primary school. Primary school corresponds to elementary school (grades 1 to 5) in the United States. All of the effect sizes are reported using grades as the measure of academic achievement. The correlation between grades and intelligence was corrected for range restriction (RR). The best predictor is intelligence, with all of the personality traits having moderate to strong predictive power as well.

Intelligence0.457118,584Roth et al. (2015)Table 1; uncorrected r = 0.40
Agreeableness0.3083,196Poropat (2009)Table 2
Conscientiousness0.2883,196Poropat (2009)Table 2
Openness to new experience0.2483,196Poropat (2009)Table 2
Parental SES0.2346Harwell et al. (2016)Table 2
Emotional Stability0.2083,196Poropat (2009)Table 2
Extraversion0.1883,196Poropat (2009)Table 2

Academic achievement (secondary school only)

The table lists predictors of academic achievement in secondary school. This corresponds to middle and high school (grades 6 to 12) in the United States. Some studies report effect sizes specifically for middle school (MS) and/or high school (HS). The studies use a variety of measures of academic achievement, including grades (Roth et al. 2015, Poropat 2009, Westrick et al. 2015) or combinations of grades and standardized achievement tests (Harwell et al. 2016). Again, the best predictor is intelligence, followed conscientiousness and various measures of parental SES. The predictive validity for all of the personality measures decreased significantly compared to the primary school data.

Intelligence (HS)0.587115,427Roth et al. (2015)Table 1; uncorrected r = 0.46
Intelligence (MS)0.547549,771Roth et al. (2015)Table 1; uncorrected r = 0.46
Conscientiousness0.213531,980Poropat (2009)Table 2
Parental income (HS)0.2015,606,273Westrick et al. (2015)Table 3; sample = ACT-test-taking population
Parental SES (HS)0.1941Sackett et al. (2009)Table 8; uncorrected r = -0.01
Parental SES (HS)0.1646Harwell et al. (2016)Table 2
Parental SES (MS)0.16104Harwell et al. (2016)Table 2
Openness to new experience0.122525,909Poropat (2009)Table 2
Agreeableness0.052425,488Poropat (2009)Table 2
Extraversion-0.032525,648Poropat (2009)Table 2
Emotional Stability0.012425,495Poropat (2009)Table 2

Academic achievement (K-12)

The table lists predictors of academic achievement in pre-college schooling (grades K-12). The studies used a variety of measures of academic achievement, including grades (Roth et al. 2015) or a combination of grades and standardized tests (Harwell et al. 2016, Castro et al. 2015, Wickersham et al. 2021, and Bücker et al. 2018).

Intelligence0.54240105,185Roth et al. (2015)Table 1; uncorrected r = 0.44
Parental income0.2617Harwell et al. (2016)Table 2
Parental SES (Composite)0.23113Harwell et al. (2016)Table 2
Parental expectations0.228Castro (2015)Table 2
Parental occupation0.2150Harwell et al. (2016)Table 2
Parental SES (white students only)0.2175Harwell et al. (2016)page 12
Depression-0.1922Wickersham et al. (2021)
Parental education0.1748Harwell et al. (2016)Table 2
Reading with children0.174Castro (2015)Table 2
General parental participation0.1737Castro (2015)Table 2
Subjective well-being0.1664Bücker et al. (2018)Table 2
Parental home resources0.1684Harwell et al. (2016)Table 2
Parental style0.1314Castro (2015)Table 2
Parental SES (black students only)0.1275Harwell et al. (2016)page 12
Teacher efficacy0.109814,215Kim et al. (2018)Table 1
Parental supervision of schoolwork0.0218Castro (2015)Table 2
Parental participation in school activities0.0121Castro (2015)Table 2

College admissions tests

This table lists predictors of scores on college admissions tests. Because there are multiple college admissions tests (the SAT and ACT), the first column will list both the predictor and the specific test or subtest. Unsurprisingly, the best predictors of scores for a given test are the scores from other tests. The next best predictors of test scores are measures of cognitive ability, measured as either g or Raven’s Advanced Matrices. High school GPA and parental SES are also strong predictors of test scores. Finally, aside from openness to new experience and verbal test scores, personality measures seem to be weak predictors of test scores, although this may be due to range restriction due to the elite sample (UC-Davis undergraduates).

SAT and ACT0.921103,525Dorans (1999)page 2
ACT math and SAT math0.891589,753The College Board, ACT (2018)page 7
ACT Reading/English and SAT Reading/Writing0.881589,753The College Board, ACT (2018)page 7
g and SAT0.821917Frey and Detterman (2004)Table 2
g and ACT0.7711,075Koenig et al. (2008)Table 2
Raven’s Advanced Matrices and ACT0.751149Koenig et al. (2008)page 158; uncorrected r = 0.61
Raven’s Advanced Matrices and SAT0.721103Frey and Detterman (2004)page 376; uncorrected r = 0.48
HSGPA and ACT0.5816,625,660Westrick et al. (2015)Table 3
HSGPA and SAT0.53171223,858Westrick et al. (2019)Table 5
HSGPA and SAT0.53110151,316Kobrin et al. (2008)Table 4
Parental SES and SAT0.4241Sackett et al. (2009)uncorrected r = 0.22
Parental income and ACT0.3416,257,643Westrick et al. (2015)Table 4
Openness to new experience and SAT verbal0.20110,497Noftle and Robins (2007)sample = UC-Davis undergraduates
Conscientiousness and SAT math-0.07110,497Noftle and Robins (2007)^
Neuroticism and SAT math-0.07110,497Noftle and Robins (2007)^
Agreeableness and SAT math-0.06110,497Noftle and Robins (2007)^
Extraversion and SAT math-0.06110,497Noftle and Robins (2007)^
Neuroticism and SAT verbal-0.05110,497Noftle and Robins (2007)^
Openness to new experience and SAT math0.05110,497Noftle and Robins (2007)^
Agreeableness and SAT verbal-0.03110,497Noftle and Robins (2007)^
Extraversion and SAT verbal0.02110,497Noftle and Robins (2007)^
Conscientiousness and SAT verbal-0.01110,497Noftle and Robins (2007)^

College GPA

The following table lists correlations between college GPA and various predictors. All values are from Richardson et al. (2012). Note that many of these correlations are likely underestimates due to no controls for range restriction.

Grade goal0.49132,670
High School GPA0.414634,724
Effort regulation0.35198,862
A level points0.314933
Test anxiety-0.212913,497
Emotional intelligence0.17145,024
Academic intrinsic motivation0.16227,414
Locus of control0.15132,126
Socioeconomic status0.152175,000
Academic integration0.131113,755
Openness to new experience0.095223,096
Social support0.09145,840
Sex (female)0.04216,176
Social integration0.031519,028
Institutional integration0.031819,773
Academic extrinsic motivation0.00102,339

The following table also lists correlations between college GPA and various predictors. However, unlikely the previous table, this displays correlations after correcting for range restriction. The best predictors are combinations of standardized test scores and high school GPA (HSGPA), followed by  test scores or HSGPA individually. Parental SES and letters of recommendation are comparatively weak predictors. Some studies focus on first-year GPA (FYGPA) whereas others use fourth-year cumulative GPA (Cum GPA).

SAT + HSGPA0.645556,939Mattern and Patterson (2006)Table 4; Cum GPA; uncorrected r = 0.46
SAT + HSGPA0.61171223,858Westrick et al. (2019)Table 5; FYGPA; uncorrected r = 0.42
HSGPA0.5850150,305Westrick et al. (2015)Table 5; FYGPA; uncorrected r = 0.47
SAT0.565556,939Mattern and Patterson (2006)Table 4; Cum GPA; uncorrected r = 0.37
HSGPA0.565556,939Mattern and Patterson (2006)Table 4; Cum GPA; uncorrected r = 0.36
HSGPA0.53171223,858Westrick et al. (2019)Table 5; FYGPA; uncorrected r = 0.33
ACT0.5150169,818Westrick et al. (2015)Table 5; FYGPA; uncorrected r = 0.38
SAT0.51171223,858Westrick et al. (2019)Table 5; FYGPA; uncorrected r = 0.32
Letters of recommendation0.2865,155Kuncel et al. (2014)Table 1
Parental income0.2450139,354Westrick et al. (2015)Table 5; FYGPA; uncorrected r = 0.12
Parental SES0.2241Sackett et al. (2009)Table 8; FYGPA; uncorrected r = 0.12

Grad School GPA

The following table lists correlations between various predictors and GPA in graduate school. The best predictors are various test scores followed by undergraduate GPA.

MCAT0.46284,706Kuncel et al. (2007)Table 1; uncorrected r = .32
GRE-Subject0.41222,413Kuncel et al. (2007)Table 1; uncorrected r = .24
LSAT0.4014222,218Kuncel et al. (2007)Table 1
MAT0.397011,368Kuncel et al. (2007)Table 1; uncorrected r = .27
GRE-Total0.3710314,291Kuncel et al. (2007)Table 1; uncorrected r = .25
GMAT0.35285,538Kuncel et al. (2007)Table 1; uncorrected r = .25
Undergraduate GPA0.30589,748Kuncel et al. (2001)Table 2; uncorrected r = .28
Letters of recommendation0.137489Kuncel et al. (2014)Table 1

School variables

The following table lists correlations between academic achievement and various school-level predictors. Some studies focused on standardized achievement tests (Armor et al. 2018, Bankston and Caldas 1998), whereas others used combinations of tests and grades (Holzberger et al. 2020).

Percentage of female-headed-0.35118,310Bankston and Caldas (1998)Table 2; sample = Louisiana 10th graders
Percentage of black-0.30118,310Bankston and Caldas (1998)^
SES composition0.3069Holzberger et al. (2020)Table 3; International sample
Mean level of school poverty-0.29118,310Bankston and Caldas (1998)^
School SES0.271500,000Armor et al. (2018)page 6; North Carolina 3rd to 8th graders
Mean level of parental education0.22118,310Bankston and Caldas (1998)^
Out-of-school activities0.1811Holzberger et al. (2020)^
Academic pressure0.1428Holzberger et al. (2020)^
Classroom climate0.1323Holzberger et al. (2020)^
Instructional practices0.1319Holzberger et al. (2020)^
School climate0.0846Holzberger et al. (2020)^
Material resources0.0852Holzberger et al. (2020)^
Personnel resources0.0337Holzberger et al. (2020)^

The indicators for the variables from Holzberger et al. (2020) are as follows (note different studies may have relied on different indicators):

  • SES composition: Several indicators were used, parental educational and occupational background, SES index (e.g. International Socio-Economic Index of Occupational Status [ISEI], Index of Economic, Social and Cultural Status [ESCS]), free or reduced lunch eligibility, and ethnic minority.
  • Out-of-school activities: offering school activities (e.g. science clubs, excursions, competitions) or programs in science and math.
  • Academic pressure: academic achievement expectations and emphasis on homework.
  • Classroom climate: learning climate in class (e.g. student-teacher or student-student relationships, and disciplinary climate) and limitations on instruction (e.g. academic abilities, backgrounds, or needs too diverse; disruptive or uninterested students).
  • Instructional practices: contrasted student-centered techniques with teacher-centered techniques.
  • School climate: problem behaviors (e.g. attendance, crime, violence, suspension), quality of relationships (e.g. student-teacher, teacher-teacher), school learning climate, and safety.
  • Material resources: a school’s material infrastructure at the school – in particular, the availability or shortage of adequate resources and information and communication technology (ICT) resources.
  • Personnel resources: the degree to which teachers were certified, whether they had a teaching degree or a Master’s in education, teacher experience, the number of teachers, and teachers’ readiness to teach.

State-level reading/mathematics scores

from McDaniel (2006).

State health0.75150
Percent low birth weight-0.71150
Percent whites not in public school-0.63150
Violent crime-0.58150
Percent no prenatal care-0.58150
Percent black-0.51150
State expenditure per student0.39150
Public/teacher ratio-0.38150
Percent Hispanic-0.34150
Gross state product0.28150
Percent Asian-0.27150

PISA reading scores

Predictors of PISA reading scores are all taken from Lee et al. (2019) which reported data on nearly half a million students in 65 countries. I used data specifically for the 2012 dataset. Here are the within-country correlations, which are obtained by calculating the correlations within each country and then averaging the correlations from each country:

Economic, social and cultural status (ESCS)0.351473,648
Parental occupational status (HISEI)0.311450,621
Home possessions (HOMEPOS)0.261479,807
Parental education in years (PARED)0.241473,091
Cultural possessions (CULTPOS)0.241471,357
Home educational resources (HEDRES)0.241477,772
Wealth proxy0.131479,597

Here are the between-country correlations, which are obtained by calculating the correlation between country-level scores and country-level predictors.

Home possessions (HOMEPOS)0.54165
Home educational resources (HEDRES)0.53165
Wealth proxy0.47165
Economic, social and cultural status (ESCS)0.45165
Parental occupational status (HISEI)0.32165
Parental education in years (PARED)0.26165
Cultural possessions (CULTPOS)0.07165

Notes on the variables used:

  • Parental occupational status (HISEI): Highest parental occupation
  • Parental education in years (PARED): Highest parental education
  • Wealth proxy: constructed based on students’ responses (yes/no) on the items of whether they had: a room of their own, a link to the Internet, a dishwasher, and a (DVD or some other types of) player. In addition, there were three items that would represent family wealth of a particular country as the country-specific wealth-related items, which also had the yes/no response options. Lastly, students were asked whether they had the following items at home: cellular phones, televisions, computers, cars and the number of rooms with a bath or shower, which were assessed on four-categories of “none, one, two, and three or more”.
  • Home educational resources (HEDRES): based on the students’ responses (yes/no) on the following seven items: a desk to study, a quiet place to study, a computer that students can use for school work, educational software, books to help with school work, technical reference books, and a dictionary.
  • Cultural possessions (CULTPOS): based on students’ responses (yes/no) on the following three items: classic literature, books of poetry, and works of art.
  • Home possessions (HOMEPOS): Number of books at home, CULTPOS, HEDRES, and Wealth
  • Economic, social and cultural status (ESCS): HOMEPOS, PARED, and HISEI

Other sources

This section lists some other sources that are useful in providing information on predictors of academic achievement. They are listed here instead of above either because they do not use an effect size used above (e.g., they report regression coefficients, etc.) or because I have not read through them to determine that they belong above.

5 comments on Predictors of academic achievement

  1. For the college GPA correlates, are you aware of any studies examining whether chosen major/courses would affect these results? It seems Richardson et al. (2012) mostly relied on cumulative GPA, and from a very brief look at the references, I gather some studies in their meta-analysis focused on students from select majors and some didn’t.

    For instance, for intelligence, if assessing overall student body and if it happens that more intelligent students pick more difficult majors/courses, then that would distort the variance in GPA explained by intelligence, no?

    1. That could very well be an issue, but my guess is it wouldn’t distort findings too much since there aren’t large IQ differences by college major from what I’ve seen (though some of the data that I’ve seen on IQ by major is questionable/outdated). Also, there isn’t much variation in IQ in general among college students due to range restriction.

      Anyway, I’m not aware of any studies that take into account major when analyzing IQ-GPA correlations. The [closest study]( that I know of looks at SAT-GPA correlations by college majors. If you look at the SAT-GPA correlation within a specific major, they deviate a bit from the overall SAT-GPA correlation (overall correlation is 0.57, but the within-major correlation hovers between 0.42 and 0.63 depending on the major). If you wanted, you could calculate the weighted average of the within-major correlations to determine the SAT-GPA correlation after controlling for major, but I doubt it would differ too much from the overall correlation. Of course, this is SAT rather than IQ so there could be different findings if you did a similar analysis for IQ.

      1. Thanks a lot for the response. I’ll admit I’m coming at this from an IQ-skeptic perspective, which is why I find the Richardson et al. (2012) results intriguing.

        As an aside, did you catch that the purported IQ-Job Performance relationship looks like a total bust ( Seems like IQ experts had themselves fooled for more than half a century. Vindicates Richardson & Norgate (2015)( Very brief summary by Russell Warne here(

        1. I haven’t looked into this, but I have similar papers bookmarked for later. Quickly skimming through the Sackett paper, it seems like the current estimates of the meta-analytic IQ-performance correlations are lower than previously estimated due to (1) improvements over the conventional range restrictions corrections which turned out to be overcorrections (these improved techniques result in lower validities for most predictors, not just IQ), and (2) more recent data samples which show lower IQ-performance correlations than the older data samples (which the authors speculate are due to the decreasing role of manufacturing jobs and the increasing role of team structures).

          Regarding (1), I don’t have the technical expertise or domain knowledge to personally audit the credibility of their claims at the moment. The authors do refer to a series of authors (Oh et al.) who are critical of their argument ( It seems like there have been a few exchanges back and forth by Sackett et al. vs Oh et al. and Ones et al. (see papers in the sidebar here The latest paper in this debate was published just a few weeks ago. So if I were really wanted to get a full understanding of these criticisms, I would just wait for the debate to settle and read the exchanges from each party. More likely, I’ll just wait a few months/years and see where the expert consensus settles instead of trying to personally audit what seems like a controversy over technical matters (side point about my general epistemology: as a layman, I think the only reliable way to acquire knowledge of a technical field is to start by understanding the facts for which there is no controversy. Learn as many of these facts as you can and then maybe you can become qualified to start picking sides on contentious topics. Until then, I’ll just remain agnostic on the controversies).

          Regarding (2), they cite the following source as evidence of lower IQ-performance correlations in the 21st century: “A contemporary Look at the Relationship Between Cognitive Ability and Job Performance”. However, in the references, this is cited as a “poster” and I can’t find the source anywhere online. It would be interesting to see if the lower validities are due to real changes in the workplace vs better methodology, data collection, etc.

          Anyway, these factors imply that the more appropriate meta-analytic estimate of the IQ-performance correlation is 0.31 for the older data samples and 0.23 for more recent data samples. I’m not sure what you mean by “total bust”, but worse-case this just means that the IQ-performance correlation was large in the 20th century but merely modest in the 21st century (based on the differential psychology guidelines reported by which argue that 0.1, 0.2, 0.3 should be considered small, medium, and large correlations). Regardless, I’m going to withhold judgment for the reasons I gave earlier. Ideally, I’ll wait to see where the consensus shifts. Even if there is no consensus, the debate appears to be still ongoing, which means I need to wait for future replies even if I wanted to personally audit each party’s arguments in detail.

  2. Fair enough. Yea, I’m being a little polemical with “total bust”. I had in mind the longstanding declarations of “single best predictor” and so on.

Leave a Reply

Your email address will not be published. Required fields are marked *