1Jorge Luis Borges once described a band of monkeys locked up in a room solely furnished with typewriters [1]. Given a few eternities, he argued, the primate’s random strokes on the keyboards would produce not only meaningful sentences, but also worldclass literature and scientific discoveries. “Everything,” the Argentine novelist wrote, “would be in its blind volumes. Everything: the detailed history of the future, Aeschylus’ The Egyptians, the exact number of times that the waters of the Ganges have reflected the flight of a falcon, the secret and true name of Rome.” And, of course, the complete collected writings of Shakespeare. Facts and fiction alike would be present in this utopian “total library” (Borges [1939] 2000).
2This image of a battalion of immortal typist monkeys haphazardly trying out numerous combinations in an attempt to solve a given problem is probably how many consider machine learning today. In the view of its critics, a statistical method that tries to replace sound mathematical reasoning by sheer brute computation is about as likely to yield interesting results as an army of apes randomly stroking keyboards. [2] Machine learning proponents would certainly disagree with this characterization. They would for instance rightfully argue that their algorithms do not work at random. These algorithms certainly exploit randomness, but they do so while following very specific rules that allow them, contrary to Borges’ monkeys, to learn from their mistakes. The fact remains, however, that machine learners do tend to embrace some form of radical inductivism. As their argument goes, a learning machine can, through clever use of intense computation and with virtually no preestablished conceptions about the data at hand, efficiently uncover its hidden structure, and indeed detect complex correlations that would take mere human statisticians an eternity to find.
3These debates are nothing new, echoing classic oppositions about the primacy of data vs. models, or about induction vs. deduction. Almost twenty years ago, a provocative paper on the “two cultures” of statistics sparked an intense conversation within the statistical community. Written by machine learning figurehead Leo Breiman, the text made a powerful case for the “algorithmic” culture while trying to respond to the main criticisms leveled against it (see Breiman 2001 and its rejoinders). While this debate is still raging, the context has changed: over time, the machine learning approach has led to such important successes that it can no longer be dismissed on purely theoretical grounds. In the recent years, learning (or artificial intelligence) algorithms have written important pages in the history of science and technology. Disease detection, face recognition, realtime translation in multiple languages: all rely on some machine learning algorithm, and all perform their task better, or at least faster, than any human. Today, the apes are indeed writing Shakespearelike verses, and driving themselves around the streets of various cities.
4 These recent successes owe much to the rise in computational power and the growing availability of data, which have done much to increase the standing of machine learning. Moreover, they occur at a time of crisis for statistics, on two separate fronts. The more conventional statistical methods, with probabilistic hypotheses designed in the early twentieth century to deal with small random samples, are drowning in the rising tide of large, nonrandom and dirty datasets. Additionally, statisticians themselves are increasingly vocal in their fundamental critique of the many ill uses of the tools that have come to dominate large swaths of quantitative science – linear regressions and hypothesis testing (see Wasserstein and Lazar 2016). The time may thus seem ripe for a rapid takeover of science by machine learning algorithms. And indeed, the idea that they will deeply transform numerous disciplines is now regularly evoked. Famed computer scientist Pedro Domingos recently wrote that machine learning “follows the same procedure [as classic statistics] of generating, testing, and discarding or refining hypotheses. But while a scientist may spend her whole life [doing so], machine learning can do it in a fraction of second” (Domingos 2015: 13). Machine learning, he concluded, is “the scientific method on steroids”. To him, it is thus “no surprise that it is revolutionizing science” (ibid.).
5For machine learning to offer meaningful results in science, however, it will take more than just automation or computer power. While its achievements in certain areas are beyond dispute, and while scientists are increasingly discussing its relevance for their endeavors, certain central problems remain. The question of the respective roles of data and theory is one. Another is the lack of guarantees that comes with machine learning results: the very flexibility that makes the models so powerful makes it virtually impossible to mathematically demonstrate that their results are optimal. Yet another issue is even more problematic for scientific investigation: the merit of these algorithms is mainly judged in terms of predictive capacities, with little attention given to interpretation. Flexible models may be very accurate, but they are also much harder to interpret than a standard regression. In other words, they may be very good at telling whether a person will do something (buy a product, develop a disease, get married), but they have trouble telling us why. That marketing companies, internet giants or applied disciplines can make profitable use of techniques designed for prediction is understandable. What scientists who are more interested in the explanation of their chosen object can gain from them remains unclear.
6 A contribution to the ongoing debate on the role of machine learning for (social) science, this article offers an introduction to this class of statistical techniques. It starts by detailing its premises, its logic, and the challenges it currently faces. Because the field is much too diverse to warrant a uniform treatment, the rest of our argument is restricted to one key subclass of algorithms, called supervised learning. These algorithms are of particular interest for quantitative social science because they are aimed at generalizing one of its most dominant and ubiquitous tools, parametric regression. [3] The article systematically compares the two approaches, from a theoretical standpoint first, then on a largescale dataset. Beyond sheer methodological clarification, it argues that the statistical revolution announced by many is not likely to happen anytime soon, at least not in the terms spelled out by its proponents. Rather than ushering in a radically new scientific era in quantitative methods, the dissemination of machine learning is fostering an increased competition between the two approaches, one that brings about more uncertainty about quantified results. Surprisingly enough, we argue, this uncertainty may be good news for science overall.
What is machine learning?
7“Machine learning,” “artificial intelligence,” “statistical learning,” “data mining,” “pattern recognition”... the diversity of overlapping denominations reflects both the diversity of approaches and the diverging research goals of the various scientific communities that have contributed to the development of the field. As an intellectual endeavor, machine learning is probably best described as a crossover between disciplines, statistics and computer science in the first place. From the former, it retains the creation of mathematical models meant to capture the underlying structure of otherwise unmanageable datasets. From the latter, it borrows the skills to produce efficient algorithms for solving complex optimization problems.
8Machine learning is currently the most popular form of artificial intelligence (“AI”), a term it is sometimes equated with nowadays. In the early 1950s, the progressive dissemination of digital computers reignited a centuryold quest: producing a machine that could emulate human reasoning (Buchanan 2005), or at least acquire and improve knowledge. A wealth of methods and principles developed over time, most of which can be said to fall into one of the two great families of AI, namely the topdown and the bottomup approach (Copeland 2000). The topdown approach is also known as “symbolic AI”, because it consists in making computers manipulate symbolic representations of the world using logical rules, in an effort to mimic the way a human would juggle concepts and knowledge using her reason. [4] The bottomup approach takes the opposite stance: artificial intelligence can be generated from interconnected networks of simple calculating units. This approach, which comprises much of what is called machine learning today, was largely influenced by a theory called “connectionism,” in which animal intelligence is thought to emerge from the connections between its biological neurons.
9The history is partly technical, and both traditions experienced cyclical episodes of great expectations, followed by years of relative disinterest. Starting in the 1950s, researchers garnered sufficient results to attract interest in AI, both from the scientific community and from a wider audience. This initial golden era came to an end in the late 1960s, when several unsolved issues, such as the impossibility to carry out certain simple tasks or prohibitive computational requirements, started to be considered as unsurpassable. As often exaggerated promises had not been kept, both funding and interest plummeted abruptly, leading to what is known today as the first “AI winter.” For over a decade, AI and machine learning were regarded by many as past science fiction (Crevier 1993: 203). But in the 1980s, a new generation of artificial neural networks called “multilayer perceptrons” overcame these technical difficulties and, although they were quickly overshadowed by more efficient techniques, initiated a revival of the machine learning approach. Neural networks came back again in the 2000s, when socalled deep learning models suddenly gave much improved results when faced with much larger databases; they are currently one of the most fashionable methods in the whole field. Over the years, the progressive refinements of techniques, the rise in computational power and the increase in available data all concurred to help removing bottlenecks that had previously halted the development of AI.
10The history of the field is also partly commercial. In the 1950s, artificial intelligence received large subsidies from private companies and public administrations, which allowed bluesky research. Likewise, the current phase of intense development owes much to the massive investments made by numerous companies, including technology giants like Facebook, Google, Baidu and Microsoft, that dedicate large chunks of their Research & Development budgets to continually improving their standing in this highly competitive domain. Because they have significantly higher resources than universities, and can offer compensations that far exceed any academic salary, they attract numerous qualified researchers, PhD students and seasoned academics alike. For over a decade now, they have been a transforming force in the field, sometimes steering research toward their own goals, and blurring the line between academia and industry.
What does machine learning do?
11Machine learning is classically defined as “the field of study that gives computers the ability to learn without being explicitly programmed,” the art of programming computers to “optimize a performance criterion using example or past data” (Alpaydin 2010). Most definitions thus stress the role of empirics in the elaboration of the model, and intensive computation. Beyond this point however, there is little agreement about what machine learning is, and even less about what it should be.
12One possible response is to consider what the algorithms do. Work in the field of machine learning is divided into several overarching tasks: numerical optimization, function approximation, visualization of multidimensional data, problem solving... Goals and tasks are thus varied, and not all of them are relevant for social science. A few of them are, though, insofar as they pursue the goals of classic statistical techniques with renewed means. One particularly relevant distinction is that between supervised and unsupervised learning. In supervised learning, the goal is to predict the values of an outcome variable y, based on the values of a set of predictor variables x. To achieve a good result, the algorithm should find good approximations to the unknown relationship between the predictors and the outcome variable. This is, of course, a standard task in statistics, called regression in the case of a continuous output (e.g. when trying to predict an individual’s income based on age, gender, level of education, etc.), or classification in the case of a categorical output (e.g. when predicting success or failure at an exam, based on students’ social backgrounds). Expanding on the standard method of linear regression, which has been known since the nineteenth century, many different algorithms tackle this problem from very different angles: feedforward neural networks, support vector machines, decision trees and their extensions such as random forests, are but a few popular examples.
13In unsupervised learning, there are no y values to predict, and instead the focus is on the detection of regularities in a set of x variables. [5] This family of methods can be further divided into two subtasks, clustering and dimensionality reduction. Clustering is the grouping of observations into coherent classes, whereas dimensionality reduction consists in mapping a multidimensional dataset in more manageable form, typically twodimensional plots. Some classic unsupervised algorithms are already part of the standard toolset of quantitative social science: hierarchical clustering and kmeans on the clustering side, factorial analysis (correspondence analysis such as it was developed by Benzecri and popularized by Bourdieu) for dimensionality reduction. Over the years, many techniques have been developed to generalize them and overcome some of their limitations: selforganizing maps are one example, dating from the 1980s, blending the clustering and the dimensionality reduction approaches; tSNE and deep autoencoders are more recent developments, highly publicized in the machine learning literature.
14Comparing machine learning to more classic statistical techniques can be difficult, because the frontier between the two is not always clearcut. For one thing, the main tasks they are asked to perform are largely the same. To add to the confusion, linear regression and PCA are often the first methods taught in machine learning classes. Some strong differences do nonetheless set apart the common econometric use of parametric models from the machine learning approach to prediction. The following section focuses on the differences between these two strands of supervised learning.
Four salient differences
Mathematical foundations versus empirical efficacy
15The first major difference pertains to the theoretical foundations of the two families of methods. The parametric modeling approach has solid foundations in mathematical statistics and probability theory. This can be readily observed in its methodological articles, which are usually centered around mathematical theorems, and thick with algebraic or probabilistic formulae. Indeed, the overall quality of a parametric method is mainly evaluated on the basis of its theoretical properties. The simple linear regression model, for instance, owes much of its popularity to the fact that its estimator is known to be unbiased, consistent and efficient [6] if its hypotheses are fulfilled.
16This stands in sharp contrast with the bulk of the supervised machine learning literature. Surely mathematics are not altogether absent from its papers and manuals, but they are mostly mobilized to explain what the models do, rather than to prove their desirable properties. More generally, its ties to mathematical statistics, and probability theory in particular, are much looser. Indeed, successful machine learning models owe their popularity to their empirical efficacy rather than to their theoretical properties. To put it bluntly, a good supervised learning model is one that gives good predictions on a wide range of empirical problems. Significantly, some of the most widely used algorithms are poorly understood from a mathematical standpoint. Random forests offer a prime example of this surprising fact: while they have been shown to be a highly effective and easytouse predictive model in many different settings, the question of just why they work so well remains open, largely due to the difficulty of expressing the workings of the complete algorithm into tractable mathematical formulae (Biau and Scornet 2016).
17This major difference is further refracted into various practical aspects of both types of methods. One striking difference is that parametric models typically have exact, unique, optimal and tractable solutions, while most machine learning methods have to rely on approximate optimization. Given a specified model and a dataset, a parametric regression’s parameters are computed using an algebraic formula, which is bound to yield the best possible solution – given the initial hypotheses. Most supervised machine learning models, on the other hand, are much too complex to have a known solution that can be expressed in a mathematical formula. Instead, the search for good parameters – the “training” phase – is algorithmic: it follows a procedure that typically consists in a series of steps that are repeated many times, in order to progressively approach a satisficing solution. The solutions found by machine learning algorithms are thus approximate, and their optimality is rarely guaranteed: there is almost always a possibility that a better solution exists but wasn’t found by the training procedure. [7] In addition to this, the fact that many training procedures are stochastic (i.e. make use of random number generation in their search for a good model) has an unsettling consequence: two consecutive runs of the same algorithm on the same dataset may yield different solutions.
18Perhaps the aspect in which supervised machine learning suffers most from its comparative lack of mathematical foundations is in the interpretation of its results, and inference. [8] The known theoretical properties of a parametric regression ensure that, once the model is trained, we know not only the value of the estimated parameter (e.g. the marginal effect of education level on personal income), but also its theoretical distribution (e.g. a normal distribution centered around the true value of the parameter). This makes it possible to evaluate the quality of the estimation, and the extent of its uncertainty. These properties can then be harnessed, usually in the form of inferential tests, to answer the question “is there a statistically significant wage gap between genders, when taking into account the level of education and social background of the studied individuals?” Results from machine learning procedures, on the other hand, rarely grant such definite inference: not much is known of their quality (apart from measures of goodness of fit), so that they can usually not answer such precise questions, or give a measure of uncertainty to their answer.
Explicit modeling versus flexible forms
19In the supervised setting, both parametric modeling and machine learning share a common goal: they try to model an outcome variable as a function of predictor variables, x_{1}, x_{2}, x _{3}, so that is approximated by y = f (x _{1}, x _{2}, x_{3}). Different specifications of the transformation are the main distinction between different kinds of models. In the parametric setting, it is almost entirely specified ex ante: the model, all the transformations of the variables and their potential interactions must be explicitly provided by the researcher. Fitting these models to the data then consists in tuning a few free parameters so that the predictions are as close as possible to the observed true values of the outcome variable. The model can take a very simple form, such as the familiar “naive” linear model:
20y = a + bx _{1} + cx _{2} + dx _{3}
21Alternatively, it can take a much more sophisticated form, and include any number of interactions between variables, or nonlinear (e.g. polynomial) transformations of the predictors or parameters.
22The specific mathematical form of the model should be guided by a detailed theory about the phenomenon under study (or, in technical terms, the “datagenerating process”), and the predictor variables should be chosen on strong empirical or theoretical grounds. In economics for instance, models of individual or collective behavior are supposed to reflect particular specifications of the actors’ utility functions. Moreover, it should be entirely defined ex ante: the progressive addition or transformation of variables to optimize a criterion (such as pvalues) is strongly advised against, a rule that is as strict as it is frequently transgressed. In common statistical practice, decisions as to which variables, transformations or interactions to include or exclude from a regression model are often based on preliminary results (such as descriptive plots or bivariate significance tests), or made by fitting many regression models sequentially. However common it might be, this practice is strictly prohibited by statistical theory, at least if further statistical inference is to be carried out, such as statistical significance tests. The reason for this interdiction is that iterative model construction tends to artificially increase statistical significance. Had the data been even slightly different, the many modeling decisions taken prior to computing the final model might have led to different choices. This uncertainty, however, is not taken into account in the reported standard errors and pvalues of statistical software. In other words, a routine and seemingly harmless procedure such as the iterative adjustment of the model frequently leads to artificially improved pvalues (a practice sometimes called “pvalue hacking”).
23In contrast, most supervised machine learning procedures start from the opposite stance: they treat the relationship between variables as fully unknown, and try to train a complex, flexible model to fit the data. Their starting point is that the reality expressed in the data is too complex to be known in advance, and that the model should be built from the data rather than specified ex ante. While they do feature a model (a set of mathematical formulae that link the predictors to the outcome), the key aspect of their model is flexibility, the capacity to automatically adapt to the data at hand. This is not to say that there are no choices to be made by the researcher when running machine learning algorithms. They are usually governed by two sets of parameters: one that fixes a level of complexity for the model (i.e. how far from linearity the model is allowed to go), and one that governs the way in which the algorithm searches for an optimal solution. [9]
24The flexibility of their models is what allows machine learning procedures to bear the promise of universal approximation: if a relationship does exist in the data, then the flexible model is, in theory at least, capable of finding and reproducing it. This allows them to do away with many constraints inherent to parametric modeling, the most obvious being monotonicity. In the simplest form of linear regression (a linear combination of the variables without further transformations or interactions, what we have called a naive model), the effect of each predictor on the outcome is the same for the whole population. While mathematically appealing, this hypothesis is regularly contradicted in empirical analyses. For instance, young women from workingclass families and/or from an ethnic minority have long outperformed the males of their social groups at school in France, a situation that was reversed at the other end of the social spectrum (Baudelot and Establet 1992). Or again, using GSS data from the 1990s, an article showed that sexual orientation used to play a different role respective to earnings, as lesbian women earned 2030% more than similar heterosexual women (Carpenter 2006: 258) – a fact the author attributed to the more limited (sexual) division of labor in the latter couples. In both cases, the situation may have changed over time, but the idea remains true: that the effect of predictor variables is contextdependent and overall nonlinear is the norm rather than the exception in the social sciences.
25Many of these limitations can of course be overcome in the parametric setting. One can introduce nonmonotonous effects by making the model more complex, e.g. by including a squared age variable, or interaction terms between gender and social background. However, all such refinements have to be specified (by hand) by the researcher, as parametric models are not designed to detect anything by themselves. Hence the temptation of trying to build a parametric model step by step. [10] But as stated above, this poses serious problems for further interpretation, especially in terms of pvalues. Machine learning solves this problem in a somewhat radical way: pvalues are virtually absent from the whole literature.
Validation criteria
26These remarks beg the question of the assessment of model quality. Here, too, machine learning and parametric modeling differ. In the latter, quality is mainly assessed through pvalues. While goodnessoffit measures, such as the R^{2}, AIC or BIC, are sometimes reported, they are mainly used for model comparison, and rarely interpreted for themselves. Pvalues, by difference, are central, as they are used to test specific hypotheses. Conventionally, a pvalue smaller than 0.05 for some parameter of interest in a regression model is interpreted as proof of the existence of a significant effect. Pvalues and statistical significance testing have a long history, going back to Neyman and Pearson in the early twentieth century, and have been central to the development of many scientific disciplines as we know them today. Their importance in current research cannot be overstated: in most quantitative research, pvalues decide whether a study makes it to print or not.
27One of the most disorienting aspects of supervised machine learning for the newcomer is thus that it is devoid of any such metric. Pvalues and statistical tests are almost altogether absent. Instead, its universal criterion of model quality is prediction. The logic is the following: if the model is able to predict well, then it has probably succeeded in capturing the patterns that link the predictors to the outcome variable. In other words, the ultimate measure of quality is the ability of the model to accurately guess the outcome value. While reminiscent of the common R^{2} of linear regression, the measures in machine learning differ in one key aspect: they are computed by crossvalidation, i.e., on observations that were not used to build the model. This is done in order to prevent overfitting: if the same data are used to train the model and evaluate its quality, there is a high risk that the model will stick too close to that specific dataset, and not generalize well to other data. [11]
28The simplest way to implement crossvalidation is to divide the dataset into two subgroups: the “train set” (typically comprising two thirds of the population’s observations) is fed to the algorithm during the training phase, in order to fit the model. The “test set” (the remaining third of the observations) is held out during training, and only used to evaluate the model’s quality: after training, the predictors of the test set are fed to the trained model, which yields predicted values that are then compared to the actual values of the outcome variable in the test set. This train/test approach is just one possibility among many, but the principle is clear: the quality of a model is assessed by its power of generalization, its ability to correctly predict observations from other samples of the same population.
Analysis of the results
29Finally, the type of results and interpretations also vary between the two approaches. With parametric regressions, the results mostly consist in interpreting the estimated parameters and their standard errors. Four main types of interpretation can be drawn from such models: the statistical significance of an effect (“controlling for other predictors, gender has a significant effect on wages, at the 5% level”), the sign of this effect (“women earn less than similar men”), its magnitude (“women earn a% less than similar men”), or a confidence interval for this magnitude (“the gender gap is between b% and c% with probability 95%”). Parametric models are generally sufficiently simple for their parameters to have an interpretable meaning, one that is shaped by the prior modeling choices. Being parsimonious in the way they link the predictors to the outcome, they provide simple answers to specific research questions – either a parameter for a given variable, or a binary answer to the question of significance. All these interpretations have direct implications in terms of understanding of the phenomenon under study, and can be immediately used outside the regression model they were based on. Furthermore, the very questions they answer are worked into the modeling choices, so that one typically tailors a parametric model to extract such knowledge from the data at hand.
30The situation is quite different in supervised learning models, which usually do not allow direct interpretation of their parameters. This is a notorious downside to their greater flexibility. Rather than parameters, the main result of most learning algorithms is, once again, their prediction, a type of information social scientists are not used to working with. [12] Once the model has been trained, it can output a predicted value of the outcome variable for any given observation. This type of information can certainly be relevant in engineering problems (such as assigning a category to an image), or for commercial uses. Likewise, it can reap – and indeed already has reaped – significant results in applied domains such as medicine, where some algorithms routinely detect certain pathologies better and faster than professionals. But what can more fundamental disciplines do with prediction, when their main questions have less to do with diagnosis and more with explanation? Since most academic disciplines are more concerned with answering “why” rather than “whether” someone will “do” something (be it success at an exam, consumption of cultural goods, or the spread of a disease), this feature of machine learning has long been described as a major impediment to its acclimatization in science.
31While they are certainly problematic, these traits are also well known by proponents of machine learning. Some even argue that prediction may not be such a bad tool overall, and that correct interpretation and good prediction actually go hand in hand (Breiman 2001). Their reasoning is the following: a model that achieves better prediction is bound to extract more information from the data, and to capture the phenomenon under study more precisely. Surely, the argument goes, there is more to be learned from an income model that accounts for 90% of income variance than from one that accounts for 60%. Furthermore, several tools have been developed over the years to “open the black box” and extract meaningful interpretations from the otherwise obscure models. Most are modelspecific, but a few are generalizable to all predictive models (including parametric models). Some offer insights into the global interpretability of the model and demonstrate which variables play the most important role; others explore the role of one variable across the dataset. These measures are detailed and illustrated in the following section, which compares parametric regression and supervised machine learning on an empirical case.
Differences in practice: determinants of salary in Sweden
32How do supervised learning methods and parametric regression compare in practice? The following pages present a comparative analysis of the two approaches on a classic question of social sciences, the determinants of individual wages, on a single dataset. To do so, we focus on moments central to quantitative analysis: model building, assessment of model quality, and interpretation of the results.
33The analysis is carried out on the Swedish labor force for 2012. The rationale for choosing this country is that Sweden practices population registration (folkbokföring). As such, it keeps detailed information about all of the persons living on its soil, and has done so since the seventeenth century at least. [13] The continued collection of varied information over decades, along with the merging of various administrative registers (census, taxes, land register, educational attainments), makes the Swedish register a wellknown trove of reliable, finegrained mass data, which can be accessed through Statistika centralbyrån (SCB), the country’s statistical service. For each individual living in Sweden, one can thus access data on a wide range of items including civil status, composition of the household, identity of spouse and children if applicable, as well as place and company of employment, eligibility to taxes, days of sick leave taken in the previous 12 months, etc. All these are updated and stored yearly. Other information, such as results from tests taken during conscription or realestate transactions, can also be retrieved. The register thus offers both largescale and quality data, a combination that is quite rare nowadays in the social sciences.
34With these sources, we analyzed income in Sweden in 2012. A classic question in the quantitative social scientific literature since Jacob Mincer’s famous equation (1974), its determinants are overall wellknown. For instance, certain regularities like the gender wage gap and the role of age have consistently been established in the literature, both in Sweden and abroad. This approach may lack the excitement of novelty, but it also offers precious landmarks when investigating a new method. For the sake of the analysis, not all items were used – only nine predictor variables were retained (see insert). The rationale for selecting only a few variables is that the use of many predictor variables (whose role is at best unclear) is likely to unduly favor the machine learning algorithms. The output variable for all models is the logarithm of the yearly gross wage. [14] We furthermore restricted the study to the employed population aged 18 to 65, with a gross salary exceeding 50,000 crowns a year (about 5,000 euros, less than a third of the fulltime observed minimum wage), to remove parttime workers. Again, this was done in order not to unduly unfavor the linear model, which is notoriously sensitive to outliers and heterogeneous samples. This left us with 4,109,447 observations.
35We fit three models to these data. The first one is the simplest form of linear regression, which takes as predictors the nine aforementioned variables, without any further transformation or interaction effect. We call this model “naive” because it does not require any domainspecific knowledge. The second model builds on this model by adding interaction effects and variable transformations. It does so using a “lasso,” a popular variant of the linear model in the machine learning context. Finally, the third model uses a fully flexible machine learning algorithm: random forests.
Box 1. – Description of variables
Age: A numerical variable, indicating the person’s age in 2012.
Gender: A categorical variable with two levels (man/woman).
Occupation: A categorical variable with seven ordered levels (Managers, Skilled employees, Intermediary occupations, Office and service activities [low qualifications]), Lowskilled workers, Military, and Other occupations).
Children between 0 to 10: A numerical variable indicating the number of children under the age of 10 in 2012.
Region: A categorical variable with 21 unique levels indicating the region of residence.
Family Type: A categorical variable with 4 levels indicating the type of relationship (single, couple) along with the presence of children in the household.
Educational Level: A categorical variable with 5 levels indicating the highest educational achievement (including an “Unknown education” level).
Unemployment: A continuous variable indicating the number of days of paid unemployment in the previous year.
Citizenship: A categorical variable with 4 levels indicating the individual’s citizenship (Swedish, Nordic countries, European Union, Others).
36All three models are presented in sequence. In order to be able to apply machine learning validation standards, the studied population was randomly divided into three samples prior to any treatment: the training sample (70% of the studied population, 2,876,613 observations) was used to fit the models, [15] the validation sample (15%) was used to finetune the models’ metaparameters, and the remaining 15% was used only once at the very end, as a test sample to estimate the generalization error of each final model.
Three models, one dataset and one question
37We start with a simple linear regression model, in which each predictor variable has a linear association with the output variable (logarithmic gross wage). This “naive” model is fitted through ordinary least squares on the training sample. The R^{2} is 0.36, meaning that the model bridges 36% of the error gap between a null model, which would predict the average wage for all observations, and a perfect prediction. This corresponds to a (root mean squared) prediction error of 0.44, which means that, on average, the predicted gross wage lies somewhere between 64% and 155% of the true observed gross wage. [16]
38The estimated coefficients and associated statistics are reported in Table 1. As could be expected given the large number of observations, nearly all coefficients have pvalues very close to zero, making this standard interpretation tool virtually useless. This is a wellknown limitation of significance testing: large numbers of observations push standard deviations (and thus pvalues) toward 0. Conceived at a time when data was scarce, statistical tests do not fare well in the era of big data. [17] On the other hand, the estimated parameters are readily interpretable: the coefficient for GenderWoman (0.26) means that, according to this model, women earn around 23% less [18] than comparable men, all other things being equal. Similarly, each additional year of age is found to be associated with a 1% increase in gross wage, ceteris paribus.
39The simplicity of these association effects reflects the simplicity of the chosen model: a 1% increase in wage for each additional year of age is an interesting and remarkably parsimonious result, but it is very probably an oversimplification, as wages are known to rise faster in the early stages of a professional career, and the temporal dynamics of wages are most probably different for men and women. It is thus tempting to try and specify a better model, by adding, removing and combining terms that might take into account such nonlinear and interaction effects. This is what we do for the second model.
Table 1. – Estimation results for the “naive” linear regression on Swedish register data
Table 1. – Estimation results for the “naive” linear regression on Swedish register data
Note: Dependent variable: logarithmic gross wage. Reference case: man, in a couple with no children, below high school education level, intermediary occupation, EU (except Nordic) citizenship. For better readability, 21 coefficients for county of residence were omitted from the table (of which 15 showed pvalues that were virtually 0).40The second model, which we shall call sophisticated linear, builds on the naive one by adding a squared age term and an interaction effect between age and gender. This allows a nonmonotonic association between age and wages, and specific age effects for men and women. Because we arrived at this specific model formulation by a process of trial and error, trying multiple specifications and choosing the best we could find, we used the lasso procedure instead of standard linear regression. [19] This lasso is still a linear model, but its optimization follows a machine learning approach (see insert), circumventing the problems that trialanderror poses to standard regression and hypothesis testing. We thus chose the model’s specification based on its prediction quality on the validation set. Once this was done, [20] the final generalization error was evaluated on the test set, yielding a R^{2} of 0.40.
Box 2. – The lasso: parametric models without pvalues
Instead of just finding the parameters that minimize the sum of squared errors, the lasso also seeks to limit the complexity of the fitted model, by forcing the parameters toward small absolute values. This is done to overcome some wellknown limitations of linear regression (possible overlearning, overinterpretation of pvalues...) while at the same time retaining its ease of interpretation. The intuition is that, for a given model specification, higher parameters mean more complex models: smaller parameters imply that a change in some predictor is associated with a small change in the outcome, making the overall model more careful.
The complexity of the model is measured by the sum of its absolute parameters, and is controlled by an additional parameter λ. For zero λ the lasso has the same solution as an unconstrained linear model, while for a very large λ all coefficients are forced toward zero. Between these two extremes lies an optimal value, the λ that yields the best generalization error. This optimal value is found, as typical for a machine learning procedure, by crossvalidation on the training sample.
Just like a standard parametric model, a lasso yields estimated parameter values that can be immediately interpreted. However, these parameters are not associated with standard errors and pvalues; instead, the constrained cost function ensures that “useless” variables have a parameter exactly equal to 0, so that they are left out of the model.
The lasso is thus a useful crossover between standard parametric modeling and machine learning. It has been extensively used in many empirical settings over the last two decades, and been given many extensions. Additionally, and quite uniquely for a machine learning model, it can be fully justified in terms of Bayesian statistics.
41The estimated coefficients are reported in Table 2, and while their value may be interpreted in roughly the same way as for the simple linear model, interpretation is made harder by the fact that all predictors were standardized prior to fit, and by the higher complexity of the model (the effect of age, for instance, can no longer be summarized by a single number, as it enters the predictive model through a linear term, a quadratic term, and an interaction term with one of the genders). Note the absence of standard errors and pvalues next to the coefficients, as such statistics cannot be computed using the lasso procedure. Instead, the lasso has forced some coefficients to zero (not reproduced in the table): according to the lasso procedure, and given the specific model we chose, these variables are better left out in order to achieve good generalization. The nonzero coefficients, on the other hand, are deemed useful for prediction; this is notably the case for the squared age term, the genderage interaction, and the number of children. Finally, because all variables were standardized, the absolute value of their parameters can serve as a measure of the relative contribution of each variable to the fit. The model can thus be said to be dominated by the age, gender and occupation variables.
Table 2. – Estimation results for the “sophisticated” linear regression, estimated by lasso on Swedish register data
Variable  Coefficient  Variable  Coefficient  
(Intercept)  7.9555  Occupation: Intermediary occupations  ref  
Age  0.8198  Occupation: Managers  0.0807  
Age (squared)  0.7107  Occupation: Military  0.0072  
Occupation: Office & Service (LowQ)  0.1094  
Gender: Man  ref  Occupation: Others  0.0792  
Gender: Woman  0.1655  Occupation: Skilled employees  0.0469  
Age x Gender: Man  0.0415  Occupation: Workers  0.0851  
Family: Couple with no children  ref  Days of unemployment  0.0884  
Family: Couple with child(ren)  0.0081  
Family: Single parent  0.0089  Education: Below high school  ref  
Family: Single & others  0.0030  Education: High school degree  0.0376  
Education: Higher ed – 2 years or less  0.0111  
Children 010  0.0245  Education: Higher ed – over 2 years  0.0409  
Education: Unknown  0.0006  
Citizenship: EU, Except Nordic  ref  
Citizenship: Nordic countries  0.0073  
Citizenship: Other countries  0.0157  
Citizenship: Sweden  0.0159 
Table 2. – Estimation results for the “sophisticated” linear regression, estimated by lasso on Swedish register data
Note: Dependent variable: logarithmic gross wage. All variables (dummy transformations and interactions included) were scaled to unit variance prior to estimation. Reference case: man, in a couple with no children, below high school education, intermediary occupation, and EU (except Nordic) citizenship. [21]42We thus achieved a slightly better generalization result by using a more sophisticated linear model than the first one. This is, of course, encouraging, but still begs the question: is this really the best possible model? Despite our best efforts to build a good and meaningful model, and our (limited) knowledge of the subject, it might very well be that we missed some important effects or interactions. We therefore turn to a fully flexible model to try and extract the most information from the data.
43The random forest algorithm was introduced in 2001, and has been acclaimed ever since for its adaptiveness and ease of use (see Hastie et al. 2009: 587603). Without going into the technicalities of the algorithm, it can be said that it uses multiple instances of an earlier supervised learning procedure called classification and regression trees (CART) [22] in what is called an ensemble. The idea of an ensemble is, in a sense, democratic: combine the opinions of many voters, and you get a single powerful decision. The random forest applies this principle by averaging the predictions of many lowbias, highvariance and decorrelated trees. [23] A typical forest is a collection of hundreds of trees, each of which is slightly perturbed in a process involving random numbers, hence the name. Its individual trees are relatively poor predictors, but as an ensemble it is a powerful lowbias, lowvariance model. The resulting complexity of the random forest also makes it a typical black box.
44Simply giving the nine predictor variables and the outcome variable to a random forest implementation [24] yields an R^{2} of 0.45 on the test sample. This is a notably better prediction score than our best effort with linear models (see Figure 1 for diagnostics plots of the three trained models). This was, furthermore, achieved effortlessly since the model construction effort was carried out by the computer, without the researcher doing a thing. The algorithm has thus captured a more accurate picture of how wages vary along with social characteristics in Sweden. However, immediate interpretation was lost in the process. Indeed, there is no reading of coefficients in random forests, as the combination of variables it produces is an intricate entanglement of tree branches. In fact, should the full mathematical equation of the trained model be written down, it would fill many pages (this is thus never done). Interpretation, instead, has to rely on prediction.
Model interpretation
45The first key of interpretation of supervised models, especially in the machine learning context, is their predictive power. In terms of generalization error, the random forest is clearly the best of the three: its R^{2} of 0.45 means that it bridges 45% of the gap between the null model and perfect prediction (compared to 36% and 40% for the naive and sophisticated linear models). Its root mean squared error on the test set is 0.4063. In terms of nominal gross wages, this means that most random forest predictions lie between 67% and 150% of the true observed wages (65%153% for the sophisticated linear, 64%155% for the naive linear model). This is certainly a step forward, although admittedly still far from perfect. [25] In exactly what way does the random forest surpass the linear models? The diagnostics from Figure 1 give us a first indication, showing that the flexible model does a better job at capturing variations at the ends of the wage spectrum: its predictions are better for the nearly 40% of the population whose logwages are under 7.5 or above 9, and its predicted wages are less narrowly distributed than is the case for both linear models. All in all, it is thus a better prediction machine, in the sense that it gives a more accurate answer to the question “how much does someone with this set of social properties earn?”
Figure 1. – Diagnostics plots, observed vs. predicted wages
Figure 1. – Diagnostics plots, observed vs. predicted wages
46Prediction alone, of course, is not very useful to a social scientist, and the black box aspect of the random forest model, with individual parameters that are meaningless for direct interpretation, can be unsettling. Fortunately, a variety of tools have been designed to extract interpretable information from predictive models. The next pages present two of the most common, which share the benefit of being applicable regardless of the type of model used: permutation importance and partial dependence.
Box 3. – Permutation importance: which variables matter for prediction?
Figure 2. – Variable importance estimation for the three models (effect of random permutation of each variable on prediction accuracy)
Figure 2. – Variable importance estimation for the three models (effect of random permutation of each variable on prediction accuracy)
47Figure 2 shows the results of permutation importance for the three trained models. For the first linear model, the importance measure yields results that are in stark contrast with the common interpretation of its coefficients and pvalues. The first difference is formal: while significance tests give a binary (yesno) answer, permutation importance yields a score and an ordering. More importantly, whereas nearly all regression parameters were deemed significantly different from 0 (at the 0.1% level) and showed sizeable marginal effects, from a permutation point of view only Occupation, Age, Gender (and to some extent the number of days of unemployment in the past year) play an important part in prediction. [27] In comparing both sets of results, however, one should keep in mind that the two importance measures do not answer quite the same question. Consider, for instance, a dataset in which a predictor has a definite but very small effect on outcome y, independent from the other regressors. While a (welldefined) parametric model should find the effect to be significantly different from zero, it would score low on permutation importance. Depending on what one means by “importance,” one of several possible measures can thus be favored.
48For the sophisticated linear (lasso) model, the permutation importance measure can be confronted to the estimated coefficients, again with contrasting results: while many variables have nonzero estimated parameters (such as the education and county of residence dummies), again only three or four variables can be deemed important for prediction, based on the permutation measure.
49Comparing these models on permutation results alone, we find that all three agree on the main lines: Occupation, Age and Gender are always the most important variables, followed by numbers of days of unemployment. Additionally, the random forest draws attention to Education by giving it a moderate importance level. The order of the variables, however, varies from one model to another: the importance of Age increases as the complexity of the model increases (around 20% increased error for the naive linear model, 35% for the sophisticated linear, over 60% for the random forest). We may infer from this that wages and age are associated in a complex way, or at least one that is hard to accurately formulate into an explicit equation.
50Such diverging results raise the question of which of the three importance measures is the most accurate, the answer to which is debatable. From a machine learning perspective, the fact that the random forest yields better (crossvalidated) prediction makes it more trustworthy. However, the lack of inference on the permutation importance values (in the sense of a measure of uncertainty of the computed values) makes it somewhat dubious from a statistical perspective. This flaw applies more generally to most interpretation results of supervised machine learning, although in recent years some effort has been made in that direction: see Efron and Hastie (2016: chap. 20).
51The partial dependence measure is another method that harnesses prediction power to gain insight into a model’s representation of the data. In the naive linear model, the marginal effect of a given variable can be directly computed from its estimated parameter, and interpreted in the common ceteris paribus way. The sophisticated linear model has slightly more complex marginal effects due to the interaction terms, and these effects are allowed to vary between different subpopulations, as prespecified in the model’s construction. When using flexible models such as the random forest, the effects of a given predictor are allowed to vary freely according to the predictor’s value, in a nonmonotonic and nonprespecified way. Contextual interpretations can then be made about the local effects of variables (age, for instance, is not specified ex ante to play a monotonic or quadratic role, or to be independent from the other regressors). Partial dependence is a popular method to extract such information from black box models. As it cannot be summarized by a single digit, it is typically represented on bivariate “partial dependence plots.”
Box 4. – Partial dependence: isolating the effect of a single variable in a complex model
52Figure 3 shows the partial dependence plots for the age regressor, for all three fitted models. The shape of the curves is a telling visual indicator of the way the regressor is used in the model. The naive model’s curve is by design straight, while the sophisticated linear model’s curve describes a section of a parabola: predicted wages rise steadily from 18 years on, peaking just before 50, and slowly decreasing afterwards. In these models, the shape of partial dependence is determined by the initial modeling choices: the naive linear specification can only yield a linear curve, the sophisticated linear is allowed to take a quadratic shape. The random forest, on the other hand, is free to let each regressor take an arbitrary form in the predictive model. As noted before, it predicts wages on a broader spectrum than the two other models; on Figure 3 this is visible from the fact that it predicts much lower average wages for the youngest in the sample (1825 years old). Its partial dependence curve is reminiscent of the quadratic shape of the sophisticated linear, but its irregular form is actually closer to piecewise linear: fast rising wages for the young, then a slower increase from 25 to 40, followed by a plateau between 40 and 60 years old, and a steady decrease afterwards. Such an irregular form would be nearimpossible to attain
Figure 3. – Partial dependence plot for age, computed from three predictive models
Figure 3. – Partial dependence plot for age, computed from three predictive models
53with parametric models, [29] the plateau shape being especially difficult to build using polynomial transformations. It is also probably a more accurate description of the association between age and wage in Sweden, judging by the lower generalization error of the random forest.
54It must be noted that the same partial dependence method can be extended to the estimation of combined effects of two or more predictors, although the required amount of computation grows exponentially with the number of variables. The search for interesting combined effects of predictors, the modeling of which is theoretically one of the strong points of machine learning, thus largely falls onto the researcher, who must try out different small subsets of predictor variables and visualize them on 2D or 3D plots. Another weakness of the partial dependence measure arises from the way it tries to force counterfactual prediction: when estimating the partial dependence on occupation, for instance, the model will be asked to predict synthetic observations that have no close equivalent in the training dataset (for instance women with a high education level doing manual construction labor), which can yield misleading results. Finally, as in the case of permutation importance, faith in the interpretation of partial dependence rests entirely on the credit given to the predictive power of the model, rather than on statistical theory.
The advent of statistical learning?
55In the light of the previous developments, it should be clear that machine learning offers an alternative approach to quantification. Will it replace more classic methods altogether, starting with parametric regression? The time may seem ripe for the computational approach to take over quantitative social science. Vast amounts of funding are being poured into the field. Just like big data in the last decade, “machine learning” has become the fashionable keyword to include in an abstract when applying for a grant or a prestigious conference. Similarly, the hype surrounding the field, whose results regularly make the headlines of international media, attracts masses of scholars and students. More importantly perhaps, these successes come at a time of heightened defiance towards the dominant method of parametric regression and its longtime companion, statistical significance testing. Pvalues, in particular, are suspected to lie at the heart of the crisis of replicability that has been documented in many areas of science, and their use and abuse has lately been condemned by the largest professional association in mathematical statistics (Wasserstein and Lazar 2016). However, the current results offered by statistical learning, along with lessons from the history of science, should warn against unfettered optimism: the oftevoked machine learning revolution, advocated by some and feared by others, may have to wait.
Three challenges for machine learning in science
56One reason has to do with the type of results produced by these techniques. As evoked before, prediction is not a tool many scientists are used to working with. Surely, various tools have been developed to extract information from blackbox models and to offer interpretation possibilities akin to those of parametric regression. However, the large variety of both machine learning models and interpretation tools, along with the lack of solid statistical inference, means that no golden standard for interpretation has appeared as of yet. This, combined with the fact that learning methods are not commonly taught in social science departments, makes publishing machine learning results much costlier than sticking to standard parametric modeling. Furthermore, the lack of guarantees as to the optimality of any given trained machine learning model, and the fact that their high flexibility could make them easier to manipulate toward a preconceived result, might make their results even more suspicious than those of parametric models.
57In addition to the types of information offered by learning algorithms, the results themselves are often not as spectacular as could be hoped for. There are, of course, undeniable successes in certain areas. Automatic translation, selfdriving cars, or, closer to science, the detection of rare diseases all bear witness to the efficiency of the approach – and the list is regularly growing longer. But even on its own criterion of choice, (crossvalidated) predictive power, machine learning is regularly outperformed by more classic methods on social science datasets.
58Here lies another challenge for machine learning: the size of the datasets. While parametric regression does not usually improve significantly beyond a few hundreds of observations, machine learning algorithms can require a very large number of training cases (observations) to live up to their promise of universal approximation. There is no rule as to how many observations are enough, as this is entirely dependent on the intrinsic complexity of the relationships to be modeled. And of course, this complexity remains largely unknown in advance. [30]
59Figure 4 illustrates how the three different models presented above were affected by size: all three were trained multiple times on random samples on the Swedish wages dataset, with sample sizes varying from 300 to 300,000 observations. [31] The linear and random forest models are the same as in section 3. XGBoost is another flexible learning model; it refers to “Extreme Gradient Boosting,” a recent variation on the popular gradient boosting algorithm (Hastie et al. 2009: chap. 10). The linear model’s prediction quality increases at first, when data is scarce, but reaches a plateau after 3,000 observations: its rigid specification prevents it from extracting more information from additional data. The random forest’s prediction power, on the other hand, shows a steady increase as the dataset expands, with a curve hinting at the fact that it might do better still on a larger dataset (as indeed it does, on the full nearthreemillion training set in section 3). The XGBoost curve is even more interesting: while this powerful model performs slightly better than the random forest after the 30,000 observations mark, it does consistently worse than even the linear model on datasets with under 10,000 observations. Unfortunately, 10,000 observations are already a large dataset for social sciences.
Figure 4. – R^{2} for three different predictive models, on random subsamples of varying sizes taken from the Swedish wages dataset
Figure 4. – R^{2} for three different predictive models, on random subsamples of varying sizes taken from the Swedish wages dataset
60There is an important lesson to be gained from this: when faced with datasets that are too small (even with tens of thousands of observations in the case of complex patterns), machine learning procedures may not fulfil their promise of universal approximation. They may not even do better than more classic approaches. In fact, it is worth noting that while doing research for this article, the authors have tried out supervised learning algorithms on a good dozen datasets with varied properties and size. More often than not, standard parametric regression performed at least as well as advanced learning algorithms. This issue could be progressively solved by the everincreasing mass of available data. The growing use of computers, the dissemination of sensors capturing varied aspects of everyday life, or the ease with which one can now format and merge databases, have led to a sharp increase in the size and number of datasets at hand. This might however not be sufficient. The majority of the newly and massively available data is, in fact, quite limited in information. As the former head of the U.S. Census bureau once put it, the current rise in data is not primarily due to an increase in what he called “design data” – data collected with a research question in mind. Instead, it consists in a new abundance of what he termed “organic data,” information collected for other purposes (the functioning of an administration, of a service, or for business matters). And while the latter may be converted into relevant information, this is in no way guaranteed (Grove 2011). In the social sciences, big data all too often remain poor data.
61There is yet another reason why machine learning may not replace standard methods altogether. The history of science reminds us that this is not the first time that (social) sciences have discussed the relative merits of a more inductivist approach over a theorydriven one. The cyclical debates about the primacy of empirics bear witness to this. This is of course the case in sociology – one need only think of the periodic resurgences of such claims, for instance under the guise of “serendipity” or of “grounded theory.” This is also the case in economics, where empiricism is sometimes denounced as a form of “measurement without theory” (Koopmans 1947), but sometimes vindicated too. Likewise, the stern defense of flexible models coming from proponents of machine learning echoes the oppositions between proponents of modeling and supporters of rich descriptions in many disciplines. The lessons of the past must be heeded here: the lack of resolution of these centuryold disputes has less to do with the absence of an appropriate method, and more with the fact that by choosing to focus on one aspect, one necessarily loses sight of the other. There is no easy way out of established antinomies.
62History also suggests that, in science at least, revolutions are less frequent than selective appropriation. In his book Chaos of Discipline (Abbott 2001), Abbott defines this as a process of “fractalization,” whereby key insights from one approach are integrated within the other, while the two traditions keep on existing on their own. Such a process is surely ongoing between machine learning and parametric regression. In a recent example of what the data avalanche could do to econometrics, Hal Varian (2014) mentioned a few elements that could be fruitfully imported into this discipline from statistical learning, most notably tools for variable selection (the choice of which predictors to include in a regression model). Coming from econometrics too (Charpentier et al. 2017) make a case for a broader use of machine learning, and in particular for ascertaining the correct specification of the model. The article also evokes the use of prediction and the question of causal inference, two aspects that were largely tackled by Susan Athey (2017), who in recent years has consistently looked for ways to improve the classic econometric approach in the light of the practices of statistical learning. Conversely, causal modeling, an endeavor that used to be the hallmark of econometrics, is gaining currency in the machine learning literature, following the influential work of Judea Pearl (2009).
Knowledge and uncertainty
63Hybridization does not mean that no change is under way. In the words of Abbott, “a fractal distinction produces both change and stability” (2001: 21). To capture this change, one must switch focal lenses and consider not what social sciences can do with machine learning, but rather what the rise of this set of techniques can do to social sciences. When it comes to their methods, many disciplines have been through deep introspection. Initiated in the first decade of the twentyfirst century with a series of proven frauds and blatant invention of data, a movement for more transparency in scientific research emerged. It was reinforced by several failed attempts at replicating various studies. One of the most famous may be the Reproducibility project, spearheaded by psychologist Brian Nosek. Looking at 100 recently published studies in the field, he and his colleagues were able to reproduce only a third of the results proclaimed by the papers (Open Science Collaboration 2015). Although the authors themselves nuance this result by stating that several reasons may account for such a low figure, the overall picture is grim. One reason is well known, and has to do with the organization of scientific production. The reduction of research budgets and the increased conditioning of grants on publications have led to more publication bias: publishing only conclusive results – and being in constant dire need to publish some – tends to generate false conclusions.
64This is nonetheless not the only, and maybe not the even the main cause for this crisis that looms over most contemporary applied quantitative research. In a remarkable discussion of the issues plaguing contemporary quantitative research, John Ioannidis explained that the reason why “Most Published Research Findings Are False” had to do with the ubiquity of statistical significance testing (2005). And in fact, since its introduction in social sciences after the 1950s, hypothesis testing, usually by way of parametric regression, has become omnipresent. With variations between disciplines and countries, they have largely reshaped the way researchers deal with numbers in the social sciences. This omnipresence, though, came at a cost. Speaking about U.S. sociology, Abbott famously contended that the adoption of the “generalized linear model” gave way to a cognitive change: increasingly, researchers started to think about the world as if it had the properties of the model, thus giving way to a generalized linear vision of reality (1988). The quasi universal adoption of this one tool certainly forced many to format their research so as to be able to implement a regression – even when the method was not warranted. Having become the main token of scientificity, parametric regression narrowed the scope of possible quantitative research.
65Here may lie the most lasting impact of machine learning for social science, once the considerable hype surrounding it settles down. The rapid dissemination of statistical learning in many disciplines is indeed favoring the emergence of an alternative approach to quantification. By offering results that are sometimes, but certainly not always, in line with standard regression; by advancing different validation criteria; and most importantly by promoting another way to work with quantitative data. As the fundamental critique of parametric models and statistical testing gains currency, and as a plausible alternative emerges, parametric regression is being rooted out of its quasi monopolistic position. The aforementioned limitations of the machine learning approach, however, make it somewhat incommensurable, in the terms famously proposed by Kuhn. Its results are generally of a different nature than those of regression, and this prevents it from fully filling the opening space. [32]
66Paradoxically, the greater diversity of statistical techniques and criteria is thus making scientists more ignorant about their object of study: on the one hand, the statistical certainties of the yester year are now known to be illfounded, on the other hand the newer methods do not even pretend to give any definite answers regarding scientific hypotheses. All in all, this forces scientists to adopt a humbler position towards their data. In this sense, greater uncertainty may not be such a problem. Controlled ignorance might indeed be a scientific blessing in this time of patented replicability crisis. In a more recent endeavor, Nosek assembled three dozen teams of researchers to assess the magnitude of implicit bias in quantitative research. Each starting with the same question and the same database, the teams reached very different conclusions (Silberzahn et al. 2017). The article drawn from this experience subsequently made a powerful case for collective scrutiny. According to the authors, crowdsourcing of quantitative methods is necessary to debunk potential biases. Though obviously commendable, this option is not always feasible, as it would require too many people’s time and energy – especially if it is to be done at every stage of research. But the multiplication of quantitative standards can offer such a critical eye. The presence of an alternative approach to quantification could become a way to ascertain our results. By forcing us to multiply standpoints, it would trigger an unsettling but eventually productive confrontation. After all, it has long been known that monoculture yields decreasing returns after a while. [33]
67Clifford Geertz once wrote that the immense success experienced by certain ideas is as justified as it is problematic. Speaking about concepts such as that of “culture” in the early twentieth century, he wrote that these ideas “resolve so many fundamental problems that they seem also to promise that they will resolve all fundamental problems, clarify all obscure issues.” However, the anthropologist quickly noted, this achievement often becomes an issue in itself, as it can quickly transform into solipsism. “The sudden vogue of a grande idée,” he added, is always at risk of “crowding out almost everything else for a while” (Geertz 1973: 310). The same goes for methods. In quantitative research, few techniques have been as central as parametric regression (in its various forms of linear regression, logistic, Poisson, nonlinear least squares, etc.). Born at the turn of the twentieth century, it has produced momentous changes in numerous scientific areas. It also became so ubiquitous that its use has not only become rigidified – it may have become unproductive.
68Judging from its resounding successes of the early twentyfirst century, supervised machine learning may seem slated to be the next statistical grande idée. Emerging at a time of crisis for conventional statistics in general, and the common econometric use of parametric models in particular, this renewed approach to quantification may have sparked a new hope. By freeing itself from constraining and easilyviolated probabilistic hypotheses, by being able to harness the evergrowing masses of data and computation power, and to automatically detect intricate patterns the researcher could not even have formulated, it bears the promise of making regression great again. While all this is true, the present article argues that it is very unlikely that machine learning will become the next methodological hegemon, at least in social science. The lack of solid understanding or the difficult interpretation of their results, need for large datasets to perform well are or again the serious obstacles to its prophesied rise to prominence. What machine learning does seem likely to do, however, is to force scholars to denaturalize their methods. By widening the scope of possible uses of quantification, and exposing the weak points of yesterday’s certainties, it is favoring a humbler attitude toward statistical results.
69 There are, no doubt, some unnerving aspects to machine learning. Not the least is the arrogance of some of its users, who buy into (or try to sell) the rhetoric of an always upcoming revolution. Against such naiveté, Geertz again had some cautionary words: “after we have become familiar with the new idea [...] it no longer has the grandiose, allpromising scope, the infinite volatility of apparent application it once had.” Past that point, “our attention shifts to isolating [what the idea is good for], to disentangling ourselves from a lot of pseudoscience to which, in the first flush of celebrity, it has also given rise” (Geertz 1973: 311).
Notes

[1]
The authors are grateful to JeanYves Bart, Marie Bergström, Arthur Charpentier, Sébastien Chauvin, Illias Garnier, Satu Helske, Estiatorio Margaritas, JeanPhilippe Touffut, the participants of the IAS lectures series (Linköping University), of the CREST (ENSAE) workshop, of the Berkeley sociology colloquium series, and of the University of Chicago Mixed Methods workshop. We are also grateful to the editorial board of the Revue Française de Sociologie for insightful comments. The data was accessed via Linköping University and paid for by several grants, including ERC grant agreement no 324233, Riksbankens Jubileumsfond (DNR M120301:1), and the Swedish Research Council (DNR 44520137681 and DNR 34020135460). The research was supported by the Excellence Initiative of the University of Strasbourg.

[2]
In fact, when the experiment was effectively carried out in the 2000s, the results turned out to be quite disappointing. In thirty days, the six macaques locked up with a computer only produced five pages of text, mostly made of “s.” After a short while, the lead animal started bashing the keyboard with a rock, while others used it as a lavatory. See http://news.bbc.co.uk/2/hi/3013959.stm, retrieved on February 25th, 2018.

[3]
In all the following, we will refer to parametric regression for the wide family of models comprising linear regression, its logistic and Poisson cousins, nonlinear least squares and other methods that both rely on an explicit formulation of the model and are interpreted through statistical significance tests. They are parametric in the sense that their main objective is to identify the value of certain parameters, which are hypothesized to play a key role in the phenomenon under study.

[4]
Expert systems were a very successful example of symbolic AI, whose many commercial applications in the 1980s included medical diagnosis, chemical analysis and credit authorization.

[5]
Whereas in prediction tasks the outcome variable plays the role of a supervising variable, where only the changes in x that are relevant to those in y must be accounted for, in unsupervised learning all variables are treated equally, as the observations are not “labelled” by an y value; hence the name unsupervised.

[6]
Following the GaussMarkov theorem.

[7]
The fact that most machine learning results cannot be guaranteed to be optimal has one important consequence in practice, in the case of negative results. If a given learning procedure fails to find a good relationship between the predictors and the outcome variable, the failure may either be due to the fact that no meaningful relationship exists in the data, or to a failure of the training algorithm to find a good solution to its intricate optimization problem. Negative results (such as the absence of a relationship) are thus very hard to establish using supervised machine learning tools.

[8]
Here the word inference is understood in the general meaning of what can be said about the results, their stability and how they confirm or contradict a given assertion about the datagenerating process, see Efron and Hastie (2016: 38).

[9]
The sensitivity to these two types of parameters varies widely among different types of machine learning models. Some, such as random forests, need practically no tuning, and can be successfully run with default parameters in most situations. Others, such as feedforward neural networks, allow (and often require) intensive finetuning of both sets of parameters.

[10]
Some approaches, such as the popular stepwise regression, even automate this trial and error process. This has sometimes been called datamining, often disparaging term coming from statisticians (Chatfield 1995 notes that the practice, especially when combined with interpretation of pvalues, has been called “logically unsound and practically misleading,” or even “a quiet scandal”). When used carefully and without hypothesis testing, it may be regarded as a primitive and crude form of machine learning (it is typically addressed at the very start of the discipline’s manuals).

[11]
A model that overfits is akin to a student who, in order to prepare for a math exam, would learn all of the numerical answers given in class, without understanding the reasoning. This student would succeed in the unlikely case the test perfectly replicates the examples seen in class, but fail should the teacher change any number in the exercise. Learning the answers is not the same as being able to generalize them to other contexts.

[12]
It should be stressed that the term “prediction,” in this context, does not mean predicting the future. A supervised model acts like a function: when given a set of input values, it produces an output value, which is called a prediction because it represents what the model “thinks” the outcome variable should be. It can thus apply to any observation.

[13]
A prerogative of the Swedish church until the official separation of church and state, population registration is now conducted by the tax office.

[14]
The logarithm is the most common transformation for the econometric study of wage equations. It helps linear models cope with the usually nonGaussian distribution of wages, so that the partial effect of each variable is relative rather than absolute (e.g. an extra year of schooling might be associated with, say, a 5% increase in wages, rather than an increase of 3,000 Swedish crowns).

[15]
While it is unusual to split the data into training and validation samples in the context of parametric regression, it is worth noting that in this specific case where the data describe the complete population, fitting a linear regression on a random sample makes it conform better to the probabilistic hypotheses of inference (pvalues, for instance, lose their theoretical justification when computed on a complete population).

[16]
Because our models predict logarithmic wages, the models’ errors have to be turned into percent of change to be interpretable in terms of nominal wages. A root mean squared error of A thus becomes a (100 x exp(+/ A))% error in nominal wages.

[17]
Saporta (2006: xxxii) gives a telling example of this fact. When measuring the linear correlation between two continuous variables with one million observations, the absolute correlation need only be 0.002 to be significant at the 5% level. Such a small correlation is, of course, of no interest for interpretation.

[18]
Because gross wage enters our regression function in logarithmic form, an estimated coefficient b for predictor X means that a unit increase in X is, on average, and all other variables held equal, associated with a 100 x (exp(b) – 1)% variation of gross wage.

[19]
The attentive reader might note that, moving from our first to our second model, we make two changes at once: the model’s parametric specification and the optimization algorithm, thus making comparison more difficult. We would argue, however, that if anything the lasso is supposed to perform better than ordinary least squares in terms of generalization error, so that our second model is actually giving parametric specifications their best chance.

[20]
All lasso computations were executed using the glmnet package for R (Friedman et al. 2010), using the default parameters and 10fold crossvalidation.

[21]
For improved readability, 21 coefficients for county of residence were omitted from the table (all of which except Stockholm showed very small absolute coefficients, three of which were 0).

[22]
Classification and regression trees are simple algorithms for supervised learning (see Hastie et al. 2009: 30512). They work with recursive binary partitioning: first find the explanatory variable that best separates the output variable into wellseparated groups, split the population in two, then repeat the operation on each of these resulting subpopulations, then on the four subsequent subpopulations, etc. The result has a tree form that makes it easy to interpret.

[23]
The individual trees of a random forest are said to have low bias in the sense that, on average, they give a prediction that is close to the desired target value; they have high variance because a small change in the data can result in a large change in the fitted tree; they are decorrelated in the sense that they are forced, by various random perturbations, to fit the data in many different ways.

[24]
For all random forest computations we used the very efficient ranger package (version 0.8.0) for R (Wright and Ziegler 2017), with 500 trees and 4 variables per split.

[25]
A word on perfect prediction might be in order here. While it is, in theory, the ultimate goal of any predictive model, it can of course never be attained in practice. Even complex models trained on millions of cases cannot compensate for lowquality predictor variables: measurement errors, dubious proxies and omitted variables are very common in social scientific research, and are most probably the real bottleneck that prevents good prediction.

[26]
Prediction importance could also be measured by training multiple models, and measuring the degradation in prediction quality that results from removing each variable; this, however, is not practical, as it demands training a large number of different models, a number that grows exponentially with the number of regressors.

[27]
Of course, pvalues by themselves are not sufficient to assess whether a variable does really play an important role in the model, as reminded by the American Statistical Association’s statement. Further interpretation should rely on the size of the effects, as measured by the estimated parameters (and always keeping in mind that these strongly depend on the chosen model form). For instance, the marginal difference between men and women on gross wage is equivalent to a 20year difference in terms of age, so that the effect of gender can be said to be more important.

[28]
As in the case of permutation importance, it is interesting to note that, while the partial dependence measure is used to answer a question similar to that of the ceteris paribus effects of variables in the linear model, it does not measure the exact same thing. While the estimated parameters of the linear regression measure the marginal effect of each variable, independently from the other predictors, the partial dependence measures the average effect of a given predictor, for all observed values of the other predictors. In the former case, the other predictors are “controlled” by the specific form of the model, while in the latter they are “averaged out” (see Hastie et al. 2009: 36970 for a treatment of this difference).

[29]
This is at least true for models were age is kept as a continuous variable. Piecewise linear functions are easy to construct if the turning points are already known (in this case, ages 25, 40 and 60), but the whole point of the random forest (and similar flexible models) is that it was able to detect these thresholds by itself.

[30]
It should be noted that machine learning procedures are in fact capable of surpassing parametric models on small datasets, as witnessed by the many examples of only a hundred observations that can be found in machine learning textbooks. In the authors’ experience, however, this is not the rule on social science datasets.

[31]
In order to assess this effect of size, we used 7 sample sizes, from 300 to 300,000 observations. Each model was trained on 20 random samples of each size, for increased robustness. As we used three different models, this adds up to a total of 7 x 20 x 3 = 420 trained models.

[32]
If anything, Bayesian statistics might be the best candidate for this office, as it provides a rigorous theoretical framework on which to build more traditional and interpretable models – albeit again at the cost of increased complexity and loss of familiarity.

[33]
In other places or disciplines, the statistical hegemon may be different. The problem of methodological monoculture extends well beyond parametric regression.