Man - Machine - Gene - OSF

Man - Machine - Gene1

Predicting cognitive ability, non-cognitive traits and educational attainment from teacher2

assessments, short essays and the genome3

Tobias Wolfram∗ Felix C. Tropf†4

Abstract5

To what extent can nonstandard types of data predict psychological and social outcomes over the life6

course? We leverage a unique British dataset to study the predictive utility of short essays written at7

age 11 and genetic polymorphisms. Using state-of-the-art methods from natural language processing and8

genomics, we find that both approaches predict cognitive ability, non-cognitive traits and educational9

attainment with in part impressive precision: Performance based on the text samples (up to 61, 9 and10

25%) mirrors that of teacher evaluations (up to 66, 19, 29%) obtained at the same age. Prediction from11

genetic data is overall substantial, but measurably smaller (up to 17, 5, 19%). Combining all three sources12

of data explains 38% of variation in educational attainment and 70% in cognitive ability, approaching13

test-retest reliability of benchmark intelligence tests. We conclude that in order to improve predictive14

performance in the social and behavioral sciences, more attention should be paid to nonstandard data15

sources.16

Prediction, the ability to assert that certain changes will be accompanied by or lead to other changes, lies17

at the very heart of the scientific endeavour (p 339, Popper, 1962). This is no less true of the social and18

behavioral sciences: The task of forecasting individual and collective behavior from the circumstances of the19

present and past has been argued for by philosophers of science since the earlier days of the field, beginning20

with the concept of social prediction (Kaplan, 1940) and the conclusion that the causal explanation of a21

social phenomenon must also serve as the basis of its future prediction (Hempel and Oppenheim, 1948).22

However, such appeals had not found lasting resonance in disciplinary practice, until recently: Over the23

past few years numerous behavioral and social scientists began to argue for the importance of and potential24

for prediction in their respective fields (Watts, 2014; Kleinberg et al., 2015; Cranmer and Desmarais, 2017;25

Yarkoni and Westfall, 2017; Hofman et al., 2021), a development in part driven by the increased availability26

of data, computational power and sophistication of machine learning algorithms (Rahal et al., 2021).27

Such approaches can be especially beneficial when focusing on predicting life outcomes, allowing for the28

∗Bielefeld University, Faculty of Sociology, Universitätsstraße 25, 33615 Bielefeld, Germany†École Nationale de la Statistique et de L’administration Économique, Center for Research in Economics and Statistics,

91764 Palaiseu, France

1

identification of individuals of elevated risk with respect to various socially relevant variables (Chandler et al.,29

2011; Kleinberg et al., 2015; Berk et al., 2019) in order to intervene. They furthermore might lead to a30

better understanding of the rigidity of life trajectories and outcomes (Salganik et al., 2020), including family31

planning, professional careers (Geyer et al., 2006), longevity and health (Lleras-Muney, 2005) as well as values32

and beliefs (Weakliem, 2002) and the risk of receiving a criminal conviction (Lochner and Moretti, 2004).33

For this, sociologists and economists stress the importance of educational attainment, as it has been shown to34

be a central stratum of life course trajectories and careers. Raising the level of educational attainment has for35

a long time been a major goal of policymakers and the extent to which education works as a “great equalizer”36

is often viewed as a central measure in determining the degree of meritocracy within a society (Bernardi and37

Ballarino, 2016) leading to sizable proportions of GDP being invested into educational institutions (p 244,38

OECD, 2021b).39

Nevertheless, in developed societies, educational attainment is completed relatively late in life: On average40

across OECD countries, more than half of 18-24 year-olds and even 16% in the age group of 25-29 year olds41

have not yet finished their education (p 54, OECD, 2021b), limiting its potential for forecasting purposes as42

an early warning sign of detrimental development and making it necessary to find measures predictive of life43

outcomes that are available earlier and therefore allow for intervention and educational streamlining (Hart,44

2016). Furthermore, it is questionable, to what extent educational attainment might in itself be just a proxy45

for cognitive ability and non-cognitive traits of the individual, with little value in itself (Caplan, 2019).46

A commonly used alternative, i.e. in the case of academic tracking (Baeriswyl et al., 2011), are teacher47

assessments, as it can be assumed that the judgment of a pedagogically trained professional who spends48

significant amounts of time with a student on a daily basis is capable of correctly evaluating his potential from49

an early age on. Indeed, meta-analyses indicate that teacher assessments show moderate to high correlations50

with important traits like academic achievement, cognitive ability, creativity and social skills (Urhahne and51

Wijnia, 2021). However, they might also exhibit potential biases related to factors like sex, class or race52

(Campbell, 2015).53

Due to such drawbacks, debate on a new measure of children’s abilities, educational attainment and subsequent54

2

life course outcomes has recently gained traction: Genetic data (Hart, 2016; Plomin, 2018; Morris et al.,55

2020). While birth weight as an early biological proxy has been introduced to economic research a while ago,56

twin studies pointed towards strong genetic influences on educational achievement (Krapohl et al., 2014),57

attainment (Silventoinen et al., 2020) and other socioeconomic factors (Marks, 2017). Today, more and more58

studies can attribute these effects directly to actual variants in the genome i.e. Lee et al. (2018), allowing for59

the direct prediction of education and other outcomes from genotyped data (Selzam et al., 2017; von Stumm60

et al., 2020), in theory starting from the point of conception.61

Furthermore, there is growing evidence for the existence of a completely different, early available measure62

that contains valuable information about the individual: Textual data. Alvero et al. (2021) show that content63

and style of an essay are related to household income and SAT scores. Writing samples allow for modest64

prediction of personality facets Fast and Funder (2008); Cutler et al. (2021), mental health (Rodriguez et al.,65

2010), cognitive ability (Abramov and Yampolskiy, 2019) and educational achievement (Cöltekin, 2020).66

Rapid recent advances in deep learning based natural language processing (i.e. Brown et al., 2020) that have67

not been utilized in any of the aforementioned studies imply that the ceiling for text-based prediction of life68

outcomes might not have been reached yet.69

In this paper, we for the first time contrast the predictive power of teacher assessments (man), deep-70

learning-based prediction from textual data (machine), and genomic data (gene) in respect to cognitive,71

non-cognitive traits and educational attainment. In addition, we analyze the role of cognitive, non-cognitive72

and discriminatory factors contributing to the different prediction techniques.73

In order to achieve this goal, we rely on a unique data source: The National Child Development Study (NCDS,74

Power and Elliott, 2006), an ongoing British birth cohort study started in 1958. At age 11, participants were75

requested to write an essay under the theme “Imagine you are 25.” of roughly 250 word length. At the same76

time, teachers were asked to give an assessment of the respondents abilities and behaviors. Eventually, in77

2002, blood samples of participants were collected and later genotyped (see Materials & Methods).78

Both essays and genotyping results confront us with the challenges of unstructured big data and require79

extensive feature engineering. We therefore leverage state-of-the-art deep learning language models (Vaswani80

3

et al., 2017) and over 500 lexicographic metrics to create more than 1000-dimensional numerical representations81

of each essay. Likewise, we reduce the complexity of our genetic data, which encompasses more than 35 million82

single nucletoide polymorphisms (SNPs): By utilizing publicly available summary statistics of genome-wide83

association studies (GWAS) we use different subsets of all available SNPs to construct polygenic scores (PGS)84

for a set of 33 curated traits, likely to be associated with our outcomes of interest in a multi-polygenic score85

approach to trait prediction (Krapohl et al., 2018). In contrast, only 22 items are used for teacher evaluations86

(see SI for a detailed list). Thus constructed covariates are used as cross-validated input to an ensembled87

array of established machine learning algorithms in the SuperLearner-framework (Van der Laan et al., 2007;88

Polley and Van Der Laan, 2010) to maximize prediction while guarding against overfitting.189

Figure 1: Overview of the research design: Genes, essays and teacher assessments are used to constructcovariates as input to an ensembling algorithm (SuperLearner) that is used to predict cognitive ability,noncognitive traits and educational attainment.

Following this design (also schematically displayed in Figure 1), our research is highly relevant to various90

current debates in the social and behavioral sciences and the broader societal discourse: First, the ability of91

computational social scientists to predict individual life outcomes in general has been questioned (Turkheimer92

and Waldron, 2000; Salganik et al., 2020). By looking at a broad range of variables measured at different93

points in time and multiple sets of innovative predictors, as well as by using best practices from the field of94

1All these aspects are described in more detail in the Materials & Methods section as well as in the SI.

4

machine learning, our work presents an important additional test of this gloomy prospect.95

Second, behavioral geneticists suggest that genetic factors are predictive of various life outcomes (Polderman96

et al., 2015), including psychological ones as well as educational attainment. Twin studies suggest that genes97

explain up to 50% of individual differences, while out of sample predictions based on molecular data remain98

smaller at around 17 percent (Okbay et al., 2022). Given the apparently consistent ability of genetic variants99

to predict education across Western populations (Rietveld et al., 2013), and the interpretation of non-genetic100

variation as unsystematic (Plomin, 2011) or luck (Jencks et al., 1972) and therefore unpredictive, it is even101

debated to use genes as incremental information for college admission (Plomin, 2018; Harden, 2021). Our102

direct comparison of genomic prediction to textual data and teacher assessments provides an important103

benchmark in debates on the applicability of genetic predictors in an educational context (Hart, 2016; Plomin,104

2018; Morris et al., 2020).105

Finally, the juxtaposition of “man” against “machine” has a long tradition (Jones, 2013) and progress in the106

fields of robotics and artificial intelligence (Silver et al., 2018; Brown et al., 2020) not only in the context of107

manual but also cognitive tasks and health care is causing a surge of “automation anxiety” (Feigenbaum and108

Gross, 2020) in developed countries. In this sense, AI-based automation of teaching is a key-issue in current109

debates on technology and education (Selwyn, 2021), as already a growing industry is dedicated to the task110

of automating essay scoring (Foltz et al., 2020). At the same time, automated essay evaluation might be111

considered as an opportunity, not only to support teachers but also for neutral assessments in contrast to112

human evaluation (Alvero et al., 2021).113

In the following, we investigate first the general predictive power of man, machine and gene on cognitive114

ability as well as non-cognitive traits which have been highlighted complementary for educational success115

of students during childhood, at the same or close to the time both the essays have been written and the116

teachers provided their assessments of the students’ abilities (age 11). Next, we quantify the predictability of117

educational attainment at age 33 by all three predictors separately, incrementally, and jointly and compare118

those results with alternative measures from the scientific literature. Finally, in particular in the context of119

the man and machine comparison, we investigate both predictive measures of the assessments and mediating120

5

and confounding factors respectively of their association with educational attainment.121

Results122

Essays, Teacher Assessments and Genes predict Cognitive and non-cognitive Traits123

First, we find quite substantial predictive power of both teacher evaluation and our essay-based deep learning124

(DL) algorithm: For cognitive ability at age 11 66% (64%-68% over all cross-validation folds, see Figure 2125

A) of individual differences in a general factor of cognitive ability can be predicted by teacher assessments,126

and 61% (58%-64% over all cross-validation folds) by essay-based DL. More specifically, for reading (Man =127

57%; Machine = 56%), verbal (Man = 57%; Machine = 54%), mathematical (62%; 53%) and non-verbal128

ability (45%; 36%), this is lower. Genetic predictions are in most cases measurable, but substantially smaller129

with 17% variance explanation in the prediction sample for general factor of cognitive ability and 14%, 13%,130

17% and 11% for reading, verbal, mathematical and non-verbal skills. In contrast, prediction is overall much131

weaker for non-cognitive traits such as job aspiration (Man = 12%; Machine = 9%; Gene = 4%), scholastic132

motivation (9%; 8%; 5%), externalizing (19%; 8%; 4%) and internalizing behavior (8%; 3%; 0%).133

Comparable prediction of Man, Machine, Genes and cognitive, non-cognitive skills.134

All three approaches predict educational attainment (see Fig. 2 B). Again, teacher assessment shows the135

highest prediction (29%), followed by DL (25%) and PGS predictions (19%). Combining teachers’ evaluation136

and the essay-based prediction only marginally improves the human prediction by 7% (2 percentage points,137

pp), but additional information from the teacher improves the DL prediction one quarter (6 pp). Polygenic138

prediction adds substantially to man (24%; 7 pp) and DL prediction (28%; 7 pp). Incrementally to the two139

other approaches, each man (19%; 6 pp), machine (6%; 2 pp) and gene (23%; 7 pp) add information to the140

prediction.141

The joint overall prediction of 38% is indeed comparable to the prediction based on cognitive abilities measured142

at age 11 as well as non-cognitive abilities (Fig. 2 C). The Figure also includes birth weight as a predictor143

since this has long been a biological proxy also in the social sciences associated with positive life course144

6

Figure 2: A: 10-Fold Cross-Validated Predictive R2 from SuperLearner-models based on essays, teacherassessments and genomic data for cognitive ability and non-cognitive traits. Whiskers mark lowest and highestR2 over all CV-Folds. B: Improvement of predictive R2 for combining Essays, Teacher Assessments andPolygenic Scores compared to each of the three Baseline-Predictions. C: 10-Fold Cross-Validated predictiveR2 from SuperLearner-Models based on Cognitive Ability, non-cognitive Traits, Birthweight and Height forhighest attained education at age 33. Whiskers mark lowest and highest R2 over all CV-Folds.

7

outcomes as well as body height, which is an example from social psychology explaining socio-economic145

success (Stulp, 2013). However, those measures only predict education by 1% and 3%. Even one of the146

key sociological predictors of educational attainment, parental education, is only able to explain 12% of the147

variance.148

Cognitive Ability, Noncognitive Traits and Parental SES mediate the Prediction of Essays,149

Teacher Assessments and Genes on Educational Attainment.150

What pathways drive the association from Man, Machine and Gene to educational attainment? Next to151

race, parental socio-economic status (SES) and sex are potential discriminatory factors in school assessment152

(Campbell, 2015), beyond cognitive and non-cognitive skills which teachers are supposed to evaluate. They153

might also be caught by our DL algorithm from textual cues in the essays. Furthermore, recent research154

has shown, that polygenic scores for educational attainment capture both direct as well as indirect effects,155

the latter potentially being confounded by parental SES (Wang et al., 2021). To better understand the156

signals picked up by our nonstandard predictors, we fit a multiple mediation model including cognitive ability,157

non-cognitive traits and potential non-psychological confounders, namely height, birthweight, sex, parental158

SES (measured by education).159

Our model shows that measured factors explain most (85%) of the DL prediction, with cognitive ability160

emerging as the strongest factor (53%), followed by in total 17% for scholastic motivation and occupational161

aspiration, parental SES 9% and externalizing behavior 4%. The overall pattern for genetic prediction and162

teachers evaluation is very similar. However, for teachers’ evaluation, cognitive abilities explain only 40%163

of the association and overall only 70% can be explained. For the genetic prediction, cognitive ability only164

explains 26% and overall only half of the total effect can be explained.165

Discussion166

In this study, we have demonstrated that social and psychological traits can be predicted from nonstandard167

data, available at early stages in life. While previous studies have questioned the predictability of social168

and psychological variables, our best model is substantially above the range of predictive performance set in169

8

Table 1: Multiple Mediation Model from Essays, Teacher Assessments and Genes to Educational Attainment

Essay Teacher Assessments GenesUnexplained Effect 0.148** (0.056) 0.296*** (0.05) 0.512*** (0.032)CAMSIS-Prestige of first aspired Job (Age 11) 0.074*** (0.014) 0.072*** (0.013) 0.046*** (0.01)Factor of Scholastic Motivation (Age 16) 0.107*** (0.016) 0.097*** (0.014) 0.083*** (0.013)General Factor of Cognitive Ability (Age 11) 0.532*** (0.049) 0.4*** (0.045) 0.263*** (0.024)Externalizing Behavior (Age 16) 0.037*** (0.01) 0.039*** (0.011) 0.027*** (0.008)Internalizing Behavior (Age 16) 0.01. (0.006) 0.01. (0.005) 0.005 (0.003)Parental SES (Age 16) 0.09*** (0.016) 0.073*** (0.013) 0.067*** (0.014)Height (Age 16) 0.008 (0.005) 0.008 (0.005) 0.006 (0.005)Sex -0.009 (0.011) 0.004 (0.008) -0.009 (0.01)Birthweight 0.002 (0.004) 0.002 (0.003) 0.001 (0.002)

the most recent Fragile Families Challenge (FFC, Salganik et al., 2020). Especially measures of cognitive170

ability and educational attainment achieve much higher predictive values, ranging from 38% for highest171

obtained degree at age 33 to 70% for a general factor of cognitive ability, as shown in the SI, a value that is172

approaching test-retest reliability of benchmark child intelligence tests, (Canivez and Watkins, 1998). This173

stands in contrast to the best predictions in the FFC (23% for material hardship and 19% for GPA).174

Compared to cognitive ability, in our setting noncognitive traits are less predictable. However, predictions175

based on teacher assessments for each analyzed traits substantially exceeds the benchmark prediction of176

the only noncognitive outcome in FFC - grit (6%). This even holds for some dimensions of the Big 5 even177

though they were measured almost 40 years later in life (see SI). Predictive performance is therefore highly178

dependent on the traits in question and the available covariates.179

Compared to essays and teacher assessments, prediction based on genetic polymorphisms was less precise.180

However, these results are overall in line with published benchmarks on polygenic prediction (i.e. Krapohl181

et al., 2018). The utility of gene-based prediction for personalized education is therefore currently limited182

(Morris et al., 2020). Having said that, the incremental contribution, the information added to man and183

machine, is largest amongst the three. In general, the overall performance of polygenic prediction for many184

outcomes was comparable to those of hundreds of social variables in the aforementioned Fragile Families185

Challenge. The expectation is that larger sample sizes, more precise phenotypical measures and whole genome186

sequencing (Wainschtein et al., 2022) increase performance in the future.187

9

Judging the individual performance of our three approaches, teacher assessments achieved overall the best188

prediction. In light of their limited scope, this finding is quite remarkable. Provided they are more focused on189

a characteristic of interest, significantly better results might be obtained, stressing the importance of expert190

opinion in a time that emphasizes the potential for automation. Nevertheless, in most cases, textual samples191

were on par with the teacher-evaluation-based prediction. It is therefore possible to predict various traits to a192

surprising degree from a ~250 word essay using state-of-the-art natural language processing models. This in193

itself is a noteworthy achievement that warrants further research: Longer texts combined with more advanced194

models might lead to even better results.195

Potential automation anxiety due to work force replacement is not necessary given the unique contribution of196

teachers to the education prediction. However, since the DL is highly redundant to the human prediction,197

second essay corrections or other supportive tasks are potentially to be implemented given the high costs of198

the educational system in the UK (4.1% of GDP; in US it is 3.9% OECD, 2021a) annually.199

Our findings further open up an exciting avenue for many large social- and behavioral scientific surveys which200

so far to our knowledge do not contain anything comparable to the essays available in the dataset used in this201

study. So far, such data would have been of limited use to quantitative researchers. Our approach presents202

a promising example of how it can be made useful. In general, more research on the relationship between203

classic constructs from social- and behavioral sciences and textual data is necessary, especially in light of the204

rapidly growing use of such models in commercial contexts (Dale, 2021).205

Overall, given our results, simply accepting that successful prediction in the social and behavioral sciences is206

not possible, or only possible to a very limited extent, seems premature.207

10

Materials and Methods208

A more detailed report of the study design and variable construction is available in SI Appendix.209

Data210

Outcomes211

The dataset used for our analyses is the National Child Development Study (NCDS), an ongoing british birth212

cohort study started in 1958. We seek to evaluate the extent to which genomic data, teacher assessments,213

and the information embedded in the essays are able to predict 18 outcome variables that can be broadly214

categorized under three different themes: Cognitive abilities, non-cognitive traits and socioeconomic outcomes.215

Information on the generation of the outcome variables is provided in the appendix.216

Teacher Evaluations217

We use nine variables overall: At age 11, teachers of the respondents were asked to assess their students218

general knowledge, number work, use of books and oral ability on a scale from one to five, as well as certain219

behaviors and motor-abilities on a scale from one to three. In addition, a series of behavior descriptions220

was given to the teacher, who is asked to underline the descriptions which best fit the child. By summing221

the number of selected items, a quantitative assessment of the child’s adjustment to school is obtained for222

multiple categories of behavior.223

Essays224

Also at age 11, study participants were asked to write an essay under the theme “Imagine you are 25.” In total225

10,511 essays of varying length (ranging from 1 to 1239 words) were transcribed. We apply various approaches226

to extract the maximum amount of information from them, by using a pre-trained and publicly available,227

state-of-the-art deep-learning NLP model, RoBERTa (Liu et al., 2019) to extract representations of the first228

250 words of all essays in lower dimensional space. These embeddings capture the semantic importance of229

each word in context and allow to express the texts by numeric vectors along 768 dimensions. In addition,230

we measure 535 linguistic metrics of lexical diversity (TAALED, Kyle et al., 2021), lexical sophistication231

11

(TAALES, Kyle et al., 2018) and sentiment (SEANCE, Crossley et al., 2017), 31 metrics of readability and232

grammatical and typographical error/word-ratios.233

Genomic Data234

For multiple subsets of NCDS participants genotyped data was available. We combined all available genomic235

data into a single file and restricted the available SNPs to those in common with the 1000 genomes reference236

panel. The final sample contains 37 772 588 variants on 6437 individuals. Using PRSice2 (Choi and O’Reilly,237

2019), we then applied publicly available summary statistics of genome-wide association studies to construct238

polygenic scores for a set of 33 curated traits (see appendix) likely to show associations with the outcomes in239

question, which span the realms of cognition, mental health, personality, physical composure, social behavior240

and substance abuse.241

Analytical Strategy242

SuperLearner243

To guarantee that we extract the maximum predictive validity from the three data sources of interest (teacher244

assessments, essays, genomic data), we use a so-called SuperLearner (Van der Laan et al., 2007; Polley and245

Van Der Laan, 2010) approach an ensembling algorithm that estimates the performance of a selection of246

machine learning models for predicting the analyzed outcomes based on cross-validation. From these, the247

SuperLearner determines a weighted average based on their performance. We run the SuperLearner with248

nested cross validation, using both 10 folds in the inner and the outer loop with the MSE as the cost function249

using the SuperLearner package in R.250

Input Algorithms251

We use a diverse range of state-of-the-art machine-learning models as input to the SuperLearner: Extreme252

Gradient Boosting, as implemented in the package xgboost (Chen et al., 2015), RandomForest, as implemented253

in the package ranger (Wright and Ziegler, 2015), a shallow neural network, as implemented in the package254

nnet, SVM with a Radial Basis Kernel, as implemented in the package ksvm, a simple linear regression model255

12

and the mean of the outcome in the training set. As an extremely large number of variables are extracted256

from the essays in the form of word embeddings, etc. the LASSO (implemented in glmnet) is used as a257

pre-screening algorithm to limit computation time.258

Evaluation Metric259

We focus on R2Holdout (or predictive R2) as our main evaluation criterion (Salganik et al., 2020), which can260

be viewed as the predictive equivalent of the well-known coefficient of variation R2. It rescales the squared261

error of a model’s prediction by the squared error when the prediction is only based on the mean value in the262

training set. For each cross-validation fold we compute263

R2Holdout = 1 −

∑i∈Holdout(yi − yi)2∑

i∈Holdout(yi − yT raining)2 .

Analogous to R2, a value of R2Holdout = 1 implies a perfect prediction, while R2

Holdout ≤ 0 indicates that the264

model performs worse than using just the sample mean for prediction.2265

2Five extreme outliers (> 10 SD) were identified in a single fold of a single trait (neuroticism) and set to the mean prediction.This does not influence results in any qualitative way but changes a misleading and not interpretable minimum fold R2

Holdout of-216 to 0.0024.

13

References266

Polina Shafran Abramov and Roman V Yampolskiy. Automatic iq estimation using stylometric methods. In267

Handbook of Research on Learning in the Age of Transhumanism, pages 32–45. IGI Global, 2019.268

AJ Alvero, Sonia Giebel, Ben Gebre-Medhin, Anthony Lising Antonio, Mitchell L Stevens, and Benjamin W269

Domingue. Essay content and style are strongly related to household income and SAT scores: Evidence270

from 60,000 undergraduate applications. Science advances, 7(42):eabi9031, 2021.271

Franz Baeriswyl, Christian Wandeler, and Ulrich Trautwein. «Auf einer anderen schule oder bei einer anderen272

lehrkraft hätte es für’s gymnasium gereicht»: Eine untersuchung zur bedeutung von schulen und lehrkräften273

für die übertrittsempfehlung. Zeitschrift für pädagogische Psychologie, 25(1):39–47, 2011.274

Richard Berk, Drougas Berk, and Drougas. Machine Learning Risk Assessments in Criminal Justice Settings.275

Springer, 2019.276

Fabrizio Bernardi and Gabriele Ballarino. Education, Occupation and Social Origin: A Comparative Analysis277

of the Transmission of Socio-Economic Inequalities. Edward Elgar Publishing, 2016.278

Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind279

Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, et al. Language models are few-shot learners.280

Advances in neural information processing systems, 33:1877–1901, 2020.281

Tammy Campbell. Stereotyped at seven? Biases in teacher judgement of pupils’ ability and attainment.282

Journal of Social Policy, 44(3):517–547, 2015.283

Gary L Canivez and Marley W Watkins. Long-term stability of the wechsler intelligence scale for Chil-284

dren—Third edition. Psychological Assessment, 10(3):285, 1998.285

Bryan Caplan. The Case against Education. Princeton University Press, 2019.286

Dana Chandler, Steven D Levitt, and John A List. Predicting and preventing shootings among at-risk youth.287

American Economic Review, 101(3):288–92, 2011.288

14

Tianqi Chen, Tong He, Michael Benesty, Vadim Khotilovich, Yuan Tang, Hyunsu Cho, et al. Xgboost:289

Extreme gradient boosting. R package version 0.4-2, 1(4):1–4, 2015.290

Shing Wan Choi and Paul F O’Reilly. PRSice-2: Polygenic Risk Score software for biobank-scale data.291

Gigascience, 8(7):giz082, 2019.292

Cagrı Cöltekin. Predicting educational achievement using linear models. Proceedings of the GermEval 2020293

Task, 1:23–29, 2020.294

Skyler J Cranmer and Bruce A Desmarais. What can we learn from predictive modeling? Political Analysis,295

25(2):145–166, 2017.296

Scott A Crossley, Kristopher Kyle, and Danielle S McNamara. Sentiment Analysis and Social Cognition297

Engine (SEANCE): An automatic tool for sentiment, social cognition, and social-order analysis. Behavior298

research methods, 49(3):803–821, 2017.299

Andrew D Cutler, Stephen W Carden, Hannah L Dorough, and Nicholas S Holtzman. Inferring grandiose300

narcissism from text: LIWC versus machine learning. Journal of Language and Social Psychology, 40(2):301

260–276, 2021.302

Robert Dale. GPT-3: What’s it good for? Natural Language Engineering, 27(1):113–118, 2021.303

Lisa A Fast and David C Funder. Personality as manifest in word use: Correlations with self-report,304

acquaintance report, and behavior. Journal of personality and social psychology, 94(2):334, 2008.305

James Feigenbaum and Daniel P Gross. Automation and the fate of young workers: Evidence from telephone306

operation in the early 20th century. Technical report, National Bureau of Economic Research, 2020.307

Peter W Foltz, Duanli Yan, and André A Rupp. The past, present, and future of automated scoring. In308

Handbook of Automated Scoring, pages 1–10. Chapman and Hall/CRC, 2020.309

Siegfried Geyer, Örjan Hemström, Richard Peter, and Denny Vågerö. Education, income, and occupational310

class cannot be used interchangeably in social epidemiology. Empirical evidence against a common practice.311

Journal of Epidemiology & Community Health, 60(9):804–810, 2006.312

15

Kathryn Paige Harden. The Genetic Lottery: Why DNA Matters for Social Equality. Princeton University313

Press, 2021.314

Sara A Hart. Precision education initiative: Moving toward personalized education. Mind, Brain, and315

Education, 10(4):209–211, 2016.316

Carl G Hempel and Paul Oppenheim. Studies in the Logic of Explanation. Philosophy of science, 15(2):317

135–175, 1948.318

Jake M. Hofman, Duncan J. Watts, Susan Athey, Filiz Garip, Thomas L. Griffiths, Jon Kleinberg, Helen319

Margetts, Sendhil Mullainathan, Matthew J. Salganik, Simine Vazire, Alessandro Vespignani, and Tal320

Yarkoni. Integrating explanation and prediction in computational social science. Nature, 595(7866):181–188,321

July 2021. ISSN 1476-4687. doi: 10.1038/s41586-021-03659-0.322

Christopher Jencks et al. Inequality: A reassessment of the effect of family and schooling in america. 1972.323

Steven E Jones. Against Technology: From the Luddites to Neo-Luddism. Routledge, 2013.324

Oscar Kaplan. Prediction in the social sciences. Philosophy of science, 7(4):492–498, 1940.325

Jon Kleinberg, Jens Ludwig, Sendhil Mullainathan, and Ziad Obermeyer. Prediction policy problems.326

American Economic Review, 105(5):491–95, 2015.327

Eva Krapohl, Kaili Rimfeld, Nicholas G. Shakeshaft, Maciej Trzaskowski, Andrew McMillan, Jean-Baptiste328

Pingault, Kathryn Asbury, Nicole Harlaar, Yulia Kovas, and Philip S. Dale. The high heritability of329

educational achievement reflects many genetically influenced traits, not just intelligence. Proceedings of the330

national academy of sciences, 111(42):15273–15278, 2014. doi: 10.1073/pnas.1408777111.331

Eva Krapohl, Hamel Patel, Stephen Newhouse, Charles J Curtis, Sophie von Stumm, Philip S Dale, Delilah332

Zabaneh, Gerome Breen, Paul F O’Reilly, and Robert Plomin. Multi-polygenic score approach to trait333

prediction. Molecular psychiatry, 23(5):1368–1374, 2018.334

Kristopher Kyle, Scott Crossley, and Cynthia Berger. The tool for the automatic analysis of lexical335

sophistication (TAALES): Version 2.0. Behavior research methods, 50(3):1030–1046, 2018.336

16

Kristopher Kyle, Scott A Crossley, and Scott Jarvis. Assessing the validity of lexical diversity indices using337

direct judgements. Language Assessment Quarterly, 18(2):154–170, 2021.338

James J. Lee, Robbee Wedow, Aysu Okbay, Edward Kong, Omeed Maghzian, Meghan Zacher, Tuan Anh339

Nguyen-Viet, Peter Bowers, Julia Sidorenko, Richard Karlsson Linnér, Mark Alan Fontana, Tushar340

Kundu, Chanwook Lee, Hui Li, Ruoxi Li, Rebecca Royer, Pascal N. Timshel, Raymond K. Walters,341

Emily A. Willoughby, Loïc Yengo, Maris Alver, Yanchun Bao, David W. Clark, Felix R. Day, Nicholas A.342

Furlotte, Peter K. Joshi, Kathryn E. Kemper, Aaron Kleinman, Claudia Langenberg, Reedik Mägi,343

Joey W. Trampush, Shefali Setia Verma, Yang Wu, Max Lam, Jing Hua Zhao, Zhili Zheng, Jason D.344

Boardman, Harry Campbell, Jeremy Freese, Kathleen Mullan Harris, Caroline Hayward, Pamela Herd,345

Meena Kumari, Todd Lencz, Jian’an Luan, Anil K. Malhotra, Andres Metspalu, Lili Milani, Ken K. Ong,346

John R. B. Perry, David J. Porteous, Marylyn D. Ritchie, Melissa C. Smart, Blair H. Smith, Joyce Y.347

Tung, Nicholas J. Wareham, James F. Wilson, Jonathan P. Beauchamp, Dalton C. Conley, Tõnu Esko,348

Steven F. Lehrer, Patrik K. E. Magnusson, Sven Oskarsson, Tune H. Pers, Matthew R. Robinson, Kevin349

Thom, Chelsea Watson, Christopher F. Chabris, Michelle N. Meyer, David I. Laibson, Jian Yang, Magnus350

Johannesson, Philipp D. Koellinger, Patrick Turley, Peter M. Visscher, Daniel J. Benjamin, and David351

Cesarini. Gene discovery and polygenic prediction from a genome-wide association study of educational352

attainment in 1.1 million individuals. Nature Genetics, 50(8):1112–1121, August 2018. ISSN 1546-1718.353

doi: 10.1038/s41588-018-0147-3.354

Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke355

Zettlemoyer, and Veselin Stoyanov. Roberta: A robustly optimized bert pretraining approach. arXiv356

preprint arXiv:1907.11692, 2019.357

Adriana Lleras-Muney. The relationship between education and adult mortality in the United States. The358

Review of Economic Studies, 72(1):189–221, 2005.359

Lance Lochner and Enrico Moretti. The effect of education on crime: Evidence from prison inmates, arrests,360

and self-reports. American economic review, 94(1):155–189, 2004.361

Gary N Marks. The contribution of genes and the environment to educational and socioeconomic attainments362

17

in Australia. Twin Research and Human Genetics, 20(4):281–289, 2017.363

Tim T Morris, Neil M Davies, and George Davey Smith. Can education be personalised using pupils’ genetic364

data? Elife, 9:e49962, 2020.365

OECD. Education at a Glance 2021: OECD Indicators. Organisation for Economic Co-operation and366

Development, Paris, 2021a.367

OECD. Education at a Glance 2021. 2021b. doi: https://doi.org/10.1787/b35a14e5-en.368

Aysu Okbay, Yeda Wu, Nancy Wang, Hariharan Jayashankar, Michael Bennett, Seyed Moeen Nehzati, Julia369

Sidorenko, Hyeokmoon Kweon, Grant Goldman, Tamara Gjorgjieva, Yunxuan Jiang, Barry Hicks, Chao370

Tian, David A. Hinds, Rafael Ahlskog, Patrik K. E. Magnusson, Sven Oskarsson, Caroline Hayward, Archie371

Campbell, David J. Porteous, Jeremy Freese, Pamela Herd, Chelsea Watson, Jonathan Jala, Dalton Conley,372

Philipp D. Koellinger, Magnus Johannesson, David Laibson, Michelle N. Meyer, James J. Lee, Augustine373

Kong, Loic Yengo, David Cesarini, Patrick Turley, Peter M. Visscher, Jonathan P. Beauchamp, Daniel J.374

Benjamin, and Alexander I. Young. Polygenic prediction of educational attainment within and between375

families from genome-wide association analyses in 3 million individuals. Nature Genetics, 54(4):437–449,376

April 2022. ISSN 1546-1718. doi: 10.1038/s41588-022-01016-z.377

Robert Plomin. Commentary: Why are children in the same family so different? Non-shared environment378

three decades later. International Journal of Epidemiology, 40(3):582–592, June 2011. ISSN 0300-5771.379

doi: 10.1093/ije/dyq144.380

Robert Plomin. Blueprint: How DNA Makes Us Who We Are. Mit Press, 2018.381

Tinca J. C. Polderman, Beben Benyamin, Christiaan A. de Leeuw, Patrick F. Sullivan, Arjen van Bochoven,382

Peter M. Visscher, and Danielle Posthuma. Meta-analysis of the heritability of human traits based on fifty383

years of twin studies. Nature Genetics, 47(7):702–709, July 2015. ISSN 1546-1718. doi: 10.1038/ng.3285.384

Eric C Polley and Mark J Van Der Laan. Super learner in prediction. 2010.385

Karl Raimund Popper. Conjectures and Refutations: The Growth of Scientific Knowledge, volume 15.386

Routledge, 1962.387

18

Chris Power and Jane Elliott. Cohort profile: 1958 British birth cohort (national child development study).388

International journal of epidemiology, 35(1):34–41, 2006.389

Charles Rahal, Mark Verhagen, David Kirk, et al. The rise of machine learning in the academic social sciences.390

Technical report, Center for Open Science, 2021.391

Cornelius A. Rietveld, Sarah E. Medland, Jaime Derringer, Jian Yang, Tõnu Esko, Nicolas W. Martin,392

Harm-Jan Westra, Konstantin Shakhbazov, Abdel Abdellaoui, Arpana Agrawal, Eva Albrecht, Behrooz Z.393

Alizadeh, Najaf Amin, John Barnard, Sebastian E. Baumeister, Kelly S. Benke, Lawrence F. Bielak,394

Jeffrey A. Boatman, Patricia A. Boyle, Gail Davies, Christiaan de Leeuw, Niina Eklund, Daniel S. Evans,395

Rudolf Ferhmann, Krista Fischer, Christian Gieger, Håkon K. Gjessing, Sara Hägg, Jennifer R. Harris,396

Caroline Hayward, Christina Holzapfel, Carla A. Ibrahim-Verbaas, Erik Ingelsson, Bo Jacobsson, Peter K.397

Joshi, Astanand Jugessur, Marika Kaakinen, Stavroula Kanoni, Juha Karjalainen, Ivana Kolcic, Kati398

Kristiansson, Zoltán Kutalik, Jari Lahti, Sang H. Lee, Peng Lin, Penelope A. Lind, Yongmei Liu, Kurt399

Lohman, Marisa Loitfelder, George McMahon, Pedro Marques Vidal, Osorio Meirelles, Lili Milani, Ronny400

Myhre, Marja-Liisa Nuotio, Christopher J. Oldmeadow, Katja E. Petrovic, Wouter J. Peyrot, Ozren401

Polašek, Lydia Quaye, Eva Reinmaa, John P. Rice, Thais S. Rizzi, Helena Schmidt, Reinhold Schmidt,402

Albert V. Smith, Jennifer A. Smith, Toshiko Tanaka, Antonio Terracciano, Matthijs J.H.M. van der403

Loos, Veronique Vitart, Henry Völzke, Jürgen Wellmann, Lei Yu, Wei Zhao, Jüri Allik, John R. Attia,404

Stefania Bandinelli, François Bastardot, Jonathan Beauchamp, David A. Bennett, Klaus Berger, Laura J.405

Bierut, Dorret I. Boomsma, Ute Bültmann, Harry Campbell, Christopher F. Chabris, Lynn Cherkas,406

Mina K. Chung, Francesco Cucca, Mariza de Andrade, Philip L. De Jager, Jan-Emmanuel De Neve, Ian J.407

Deary, George V. Dedoussis, Panos Deloukas, Maria Dimitriou, Gudny Eiriksdottir, Martin F. Elderson,408

Johan G. Eriksson, David M. Evans, Jessica D. Faul, Luigi Ferrucci, Melissa E. Garcia, Henrik Grönberg,409

Vilmundur Gudnason, Per Hall, Juliette M. Harris, Tamara B. Harris, Nicholas D. Hastie, Andrew C.410

Heath, Dena G. Hernandez, Wolfgang Hoffmann, Adriaan Hofman, Rolf Holle, Elizabeth G. Holliday,411

Jouke-Jan Hottenga, William G. Iacono, Thomas Illig, Marjo-Riitta Järvelin, Mika Kähönen, Jaakko412

Kaprio, Robert M. Kirkpatrick, Matthew Kowgier, Antti Latvala, Lenore J. Launer, Debbie A. Lawlor,413

Terho Lehtimäki, Jingmei Li, Paul Lichtenstein, Peter Lichtner, David C. Liewald, Pamela A. Madden,414

19

Patrik K. E. Magnusson, Tomi E. Mäkinen, Marco Masala, Matt McGue, Andres Metspalu, Andreas415

Mielck, Michael B. Miller, Grant W. Montgomery, Sutapa Mukherjee, Dale R. Nyholt, Ben A. Oostra,416

Lyle J. Palmer, Aarno Palotie, Brenda Penninx, Markus Perola, Patricia A. Peyser, Martin Preisig, Katri417

Räikkönen, Olli T. Raitakari, Anu Realo, Susan M. Ring, Samuli Ripatti, Fernando Rivadeneira, Igor418

Rudan, Aldo Rustichini, Veikko Salomaa, Antti-Pekka Sarin, David Schlessinger, Rodney J. Scott, Harold419

Snieder, Beate St Pourcain, John M. Starr, Jae Hoon Sul, Ida Surakka, Rauli Svento, Alexander Teumer,420

Henning Tiemeier, Frank JAan Rooij, David R. Van Wagoner, Erkki Vartiainen, Jorma Viikari, Peter421

Vollenweider, Judith M. Vonk, Gérard Waeber, David R. Weir, H.-Erich Wichmann, Elisabeth Widen,422

Gonneke Willemsen, James F. Wilson, Alan F. Wright, Dalton Conley, George Davey-Smith, Lude Franke,423

Patrick J. F. Groenen, Albert Hofman, Magnus Johannesson, Sharon L.R. Kardia, Robert F. Krueger,424

David Laibson, Nicholas G. Martin, Michelle N. Meyer, Danielle Posthuma, A. Roy Thurik, Nicholas J.425

Timpson, André G. Uitterlinden, Cornelia M. van Duijn, Peter M. Visscher, Daniel J. Benjamin, David426

Cesarini, and Philipp D. Koellinger. GWAS of 126,559 Individuals Identifies Genetic Variants Associated427

with Educational Attainment. Science (New York, N.Y.), 340(6139):1467–1471, June 2013. ISSN 0036-8075.428

doi: 10.1126/science.1235488.429

Aubrey J Rodriguez, Shannon E Holleran, and Matthias R Mehl. Reading between the lines: The lay430

assessment of subclinical depression from written self-descriptions. Journal of personality, 78(2):575–598,431

2010.432

Matthew J. Salganik, Ian Lundberg, Alexander T. Kindel, Caitlin E. Ahearn, Khaled Al-Ghoneim, Abdullah433

Almaatouq, Drew M. Altschul, Jennie E. Brand, Nicole Bohme Carnegie, Ryan James Compton, Debanjan434

Datta, Thomas Davidson, Anna Filippova, Connor Gilroy, Brian J. Goode, Eaman Jahani, Ridhi Kashyap,435

Antje Kirchner, Stephen McKay, Allison C. Morgan, Alex Pentland, Kivan Polimis, Louis Raes, Daniel E.436

Rigobon, Claudia V. Roberts, Diana M. Stanescu, Yoshihiko Suhara, Adaner Usmani, Erik H. Wang, Muna437

Adem, Abdulla Alhajri, Bedoor AlShebli, Redwane Amin, Ryan B. Amos, Lisa P. Argyle, Livia Baer-Bositis,438

Moritz Büchi, Bo-Ryehn Chung, William Eggert, Gregory Faletto, Zhilin Fan, Jeremy Freese, Tejomay439

Gadgil, Josh Gagné, Yue Gao, Andrew Halpern-Manners, Sonia P. Hashim, Sonia Hausen, Guanhua He,440

Kimberly Higuera, Bernie Hogan, Ilana M. Horwitz, Lisa M. Hummel, Naman Jain, Kun Jin, David441

20

Jurgens, Patrick Kaminski, Areg Karapetyan, E. H. Kim, Ben Leizman, Naijia Liu, Malte Möser, Andrew E.442

Mack, Mayank Mahajan, Noah Mandell, Helge Marahrens, Diana Mercado-Garcia, Viola Mocz, Katariina443

Mueller-Gastell, Ahmed Musse, Qiankun Niu, William Nowak, Hamidreza Omidvar, Andrew Or, Karen444

Ouyang, Katy M. Pinto, Ethan Porter, Kristin E. Porter, Crystal Qian, Tamkinat Rauf, Anahit Sargsyan,445

Thomas Schaffner, Landon Schnabel, Bryan Schonfeld, Ben Sender, Jonathan D. Tang, Emma Tsurkov,446

Austin van Loon, Onur Varol, Xiafei Wang, Zhi Wang, Julia Wang, Flora Wang, Samantha Weissman,447

Kirstie Whitaker, Maria K. Wolters, Wei Lee Woon, James Wu, Catherine Wu, Kengran Yang, Jingwen448

Yin, Bingyu Zhao, Chenyun Zhu, Jeanne Brooks-Gunn, Barbara E. Engelhardt, Moritz Hardt, Dean Knox,449

Karen Levy, Arvind Narayanan, Brandon M. Stewart, Duncan J. Watts, and Sara McLanahan. Measuring450

the predictability of life outcomes with a scientific mass collaboration. Proceedings of the National Academy451

of Sciences, 117(15):8398–8403, April 2020. ISSN 0027-8424, 1091-6490. doi: 10.1073/pnas.1915006117.452

Neil Selwyn. Education and Technology: Key Issues and Debates. Bloomsbury Academic, London ; New York,453

3rd edition edition, December 2021. ISBN 978-1-350-14554-2.454

Saskia Selzam, Eva Krapohl, Sophie Von Stumm, Paul Francis O’Reilly, Kaili Rimfeld, Yulia Kovas, PS Dale,455

JJ Lee, and Robert Plomin. Predicting educational achievement from DNA. Molecular psychiatry, 22(2):456

267–272, 2017.457

Karri Silventoinen, Aline Jelenkovic, Reijo Sund, Antti Latvala, Chika Honda, Fujio Inui, Rie Tomizawa,458

Mikio Watanabe, Norio Sakai, Esther Rebato, Andreas Busjahn, Jessica Tyler, John L. Hopper, Juan R.459

Ordoñana, Juan F. Sánchez-Romera, Lucia Colodro-Conde, Lucas Calais-Ferreira, Vinicius C. Oliveira,460

Paulo H. Ferreira, Emanuela Medda, Lorenza Nisticò, Virgilia Toccaceli, Catherine A. Derom, Robert F.461

Vlietinck, Ruth J. F. Loos, Sisira H. Siribaddana, Matthew Hotopf, Athula Sumathipala, Fruhling Rijsdijk,462

Glen E. Duncan, Dedra Buchwald, Per Tynelius, Finn Rasmussen, Qihua Tan, Dongfeng Zhang, Zengchang463

Pang, Patrik K. E. Magnusson, Nancy L. Pedersen, Anna K. Dahl Aslan, Amie E. Hwang, Thomas M.464

Mack, Robert F. Krueger, Matt McGue, Shandell Pahlen, Ingunn Brandt, Thomas S. Nilsen, Jennifer R.465

Harris, Nicholas G. Martin, Sarah E. Medland, Grant W. Montgomery, Gonneke Willemsen, Meike Bartels,466

Catharina E. M. van Beijsterveldt, Carol E. Franz, William S. Kremen, Michael J. Lyons, Judy L. Silberg,467

21

Hermine H. Maes, Christian Kandler, Tracy L. Nelson, Keith E. Whitfield, Robin P. Corley, Brooke M.468

Huibregtse, Margaret Gatz, David A. Butler, Adam D. Tarnoki, David L. Tarnoki, Hang A. Park, Jooyeon469

Lee, Soo Ji Lee, Joohon Sung, Yoshie Yokoyama, Thorkild I. A. Sørensen, Dorret I. Boomsma, and Jaakko470

Kaprio. Genetic and environmental variation in educational attainment: An individual-based analysis of 28471

twin cohorts. Scientific Reports, 10(1):12681, July 2020. ISSN 2045-2322. doi: 10.1038/s41598-020-69526-6.472

David Silver, Thomas Hubert, Julian Schrittwieser, Ioannis Antonoglou, Matthew Lai, Arthur Guez, Marc473

Lanctot, Laurent Sifre, Dharshan Kumaran, Thore Graepel, et al. A general reinforcement learning474

algorithm that masters chess, shogi, and Go through self-play. Science (New York, N.Y.), 362(6419):475

1140–1144, 2018.476

Gert Stulp. Sex, stature and status: Natural selection on height in contemporary human populations. 2013.477

Eric Turkheimer and Mary Waldron. Nonshared environment: a theoretical, methodological, and quantitative478

review. Psychological bulletin, 126(1):78, 2000.479

Detlef Urhahne and Lisette Wijnia. A review on the accuracy of teacher judgments. Educational Research480

Review, 32:100374, 2021.481

Mark J Van der Laan, Eric C Polley, and Alan E Hubbard. Super learner. Statistical applications in genetics482

and molecular biology, 6(1), 2007.483

Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser,484

and Illia Polosukhin. Attention is all you need. Advances in neural information processing systems, 30,485

2017.486

Sophie von Stumm, Emily Smith-Woolley, Ziada Ayorech, Andrew McMillan, Kaili Rimfeld, Philip S. Dale,487

and Robert Plomin. Predicting educational achievement from genomic measures and socioeconomic status.488

Developmental science, 23(3):e12925, 2020.489

Pierrick Wainschtein, Deepti Jain, Zhili Zheng, L Adrienne Cupples, Aladdin H Shadyab, Barbara McKnight,490

Benjamin M Shoemaker, Braxton D Mitchell, Bruce M Psaty, Charles Kooperberg, et al. Assessing the491

22

contribution of rare variants to complex trait heritability from whole-genome sequence data. Nature492

Genetics, pages 1–11, 2022.493

Biyao Wang, Jessie R Baldwin, Tabea Schoeler, Rosa Cheesman, Wikus Barkhuizen, Frank Dudbridge, David494

Bann, Tim T Morris, and Jean-Baptiste Pingault. Genetic nurture effects on education: A systematic495

review and meta-analysis. bioRxiv, 2021.496

Duncan J Watts. Common sense and sociological explanations. American Journal of Sociology, 120(2):497

313–351, 2014.498

David L Weakliem. The effects of education on political opinions: An international study. International499

Journal of Public Opinion Research, 14(2):141–157, 2002.500

Marvin N Wright and Andreas Ziegler. Ranger: A fast implementation of random forests for high dimensional501

data in C++ and R. arXiv preprint arXiv:1508.04409, 2015.502

Tal Yarkoni and Jacob Westfall. Choosing prediction over explanation in psychology: Lessons from machine503

learning. Perspectives on Psychological Science, 12(6):1100–1122, 2017.504

23