Integrating Personality Psychology into Economics

This paper reviews the problems and potential benefits of integrating personality psychology into economics. Economists have much to learn from and contribute to personality psychology.

What can economists learn from and contribute to personality psychology? What do we learn from personality psychology? Personality traits predict many behaviors-sometimes with the same or greater strength as conventional cognitive traits. Personality psychology considers a wider array of actions than are usually considered by economists and enlarges the economist's way to describe and model the world.
Personality traits are not set in stone. They change over the life cycle. They are a possible avenue for policy intervention.
Personality psychologists lack precise models. Economics provides a clear framework for recasting the field. Economics now plays an important role in clarifying the concepts and empirical content of psychology.
More precise models reveal basic identification problems that plague measurement in psychology. At an empirical level, "cognitive" and "noncognitive" traits are not easily separated.
Moreover, personality psychologists typically present correlations and not causal relationships. Many contemporaneously measured relationships suffer from the problem of reverse causality. Economists can apply their tools to define and estimate causal mechanisms. In addition, psychological measures have substantial measurement error. Econometric tools account for measurement error, and doing so makes a difference.
Economists formulate and estimate mechanisms of investment-how traits can be changed for the better.
There are major challenges in integrating personality psychology and economics. Economists need to link the traits of psychology with the preferences, constraints and expectation mechanisms of economics. We need to develop rigorous methods for analyzing causal relationships in both fields. We also need to develop a common language and a common framework to promote interdisciplinary exchange.
There is a danger in assuming that basic questions of content and identification have been answered by psychologists at the level required for rigorous economic analysis. In explaining outcomes, how important is the person? How important is the situation? How important is their interaction? I address these issues in this paper.

A Brief History of Personality Psychology
Alfred Binet, architect of the first modern intelligence test that became the Stanford-Binet IQ test, noted that performance in school "...admits of other things than intelligence; to succeed in his studies, one must have qualities which depend on attention, will, and character; for example a certain docility, a regularity of habits, and especially continuity of effort. A child, even if intelligent, will learn little in class if he never listens, if he spends his time in playing tricks, in giggling, is playing truant." -Binet (1916, p. 254) All later pioneers have made similar statements. Many feature the Big Five trait "Conscientiousness" as a main determinant of success. 1 Before considering the Big Five traits, it is useful to briefly examine the modern concept of cognition by way of contrast.
2.0. Cognition: "g"-a single factor that is claimed to represent intelligence Traditional "g" is a product of early Twentieth Century psychology. The concept of "g" has been broadened even beyond the traditional subcomponents of "fluid" and "crystallized" intelligence. Figure 1 summarizes current thinking where "g" or general intelligence is at the top of a large pyramid of cognitive traits.  Ackerman and Heggestad (1997), based on Carroll (1993).

Personality Traits
Early pioneers used a lexical approach to define personality. They classified words that are used to describe people. This practice culminated in the "Big Five" derived from factor analysis of measurements of personality extracted from a variety of measures-observer reports, tests and measured productivity on the job (Costa and McCrae, 1992;Goldberg, 1993). No single "g p " explains all traits. There are strong correlations within clusters but weak correlations across clusters. Notes: Facets specified by the NEO-PI-R personality inventory (Costa and McCrae [1992b]). Trait adjectives in parentheses from the Adjective Check List (Gough and Heilbrun [1983]). *These temperament traits may be related to two Big Five factors. Source: Table adapted from John and Srivastava [1999].
Notes: Facets specified by the NEO-PI-R personality inventory (Costa and McCrae, 1992). Trait adjectives in parentheses from the Adjective Check List (Gough and Heilbrun, 1983). * These temperament traits may be related to two Big Five factors. Source: Table adapted from John and Srivastava (1999).
The Big Five predict many outcomes. The Big Five are defined without reference to any context (i.e., situation). This practice gives rise to an identification problem that I discuss below. The accumulated evidence speaks strongly against the claims of Mischel and the behavioral economists. 2

Personality Psychology After the Person-Situation
Debate Correlational evidence shows that for many outcomes, measured personality traits are as predictive, and are sometimes more predictive, than standard measures of cognition. Traits are stable across situations.
Situations also matter. Behavioral genetics show that personality traits are as heritable as cognitive traits.
Alterations in brain structure and function through accidents, disease and by experiments affect measured personality.

The Predictive Power of Personality Traits
A growing body of evidence suggests that personality measures-especially those related to Conscientiousness, and, to a lesser extent, Neuroticism-predict a wide range of outcomes. The predictive power of any particular personality measure tends to be less than the predictive power of IQ but in some cases rivals or exceeds it.

Main Findings from Predictive Analyses
The predictive power of "g" decreases with the level of job complexity. Personality traits are predictive at all levels of job complexity. Conscientiousness is the most predictive Big Five trait across many outcomes such as educational attainment, grades, job performance across a range of occupational categories, longevity and criminality. Neuroticism (and related Locus of Control) predicts schooling outcomes and labor market search. Other traits play roles at finer levels. I now present examples of the power of personality traits.

Educational Attainment and Achievement
In explaining educational attainment, Conscientiousness plays a powerful role. See    GEDs earn at the rate of dropouts. Their lower levels of noncognitive skill leads to lower wages than ordinary high school graduates even though they have the same level of cognitive skills.
Cognitive   Notes: All correlations are significant at the 1% level. The correlations are corrected for scale reliability and come from a meta analysis representing a collection of studies representing samples of between N=31,955 to N=70,926, depending on the trait. The meta-analysis did not clearly specify when personality was measured relative to course grades. Source: Poropat (2009).

Labor Market Outcomes
Intelligence is the greatest single predictor of job performance, especially in complex tasks, but noncognitive skills are also important predictors. See  Notes: The values for personality are correlations that were corrected for sampling error, censoring, and measurement error. Job performance was based on performance ratings, productivity data and training proficiency. The authors do report the timing of the measurements of personality relative to job performance. Of the Big Five, the coefficient on Conscientiousness is the only one that is statistically significant with a lower bound on the 90credibility value of 0.10. The value for IQ is a raw correlation. Sources: The correlations reported for personality traits come from a meta-analysis conducted by Barrick and Mount (1991). The correlation reported for IQ and job performance come from Schmidt and Hunter (2004).

Longevity
Personality traits also predict longevity. In particular, Conscientiousness is a better predictor than IQ. See  It is important to distinguish personality traits from measured personality. One definition of personality by a leading psychologist is: "Personality traits are the relatively enduring patterns of thoughts, feelings, and behaviors that reflect the tendency to respond in certain ways under certain circumstances." - Roberts (2009, p. 140) His conceptual framework for personality is presented in Figure 10. Personality is a property of a system. This type of analysis is typical of the models used in personality psychology.

An Economic Framework for Conceptualizing and Measuring Personality and Personality Traits
How can we interpret personality within economic models? Through preferences (the standard approach), constraints  or through expectations? Or does it operate through all three?
10.1. Personality Affects Productivity Almlund et al. (2011) develop models in which productivity in task j depends on the traits of agents represented by trait vector θ, and the "effort" they expend on the task, e j : (1) Traits θ are endowments, like a public good.
J j=1 e j =ē.ē is endowment.
φ j (θ, e j ) is concave and increasing in e j ; with respect to {e j } J j=1 subject to the constraint J j=1 e j =ē. In general, as R j ↑ e j ↑. Effort in one task might diminish effort in another. If tasks are mutually exclusive, we obtain the Roy model (Heckman and Honoré, 1990;Heckman and Sedlacek, 1985).

Identifying Personality Traits From Measured Performance on Tasks
I next consider a basic identification problem. Some tasks may require only a single trait or only a subset of all of the traits. Divide θ into "mental" (µ) and "personality" (π) traits, θ µ and θ π . To use performance on a task (or on multiple measures of the task) to identify a trait requires that performance on certain tasks (performance on a test, performance in an interpersonal situation, etc.) depends exclusively on one component of θ, say θ 1,j , as well as on the effort used in the task. Thus measurement assumes task j output is generated by the following relationship: We need to standardize for effort at a benchmark level, say e * , to use P j to identify a measure of the trait θ 1,j .
The activity of picking a task (or a collection of tasks) that measure a particular trait (θ 1,j in our example) is called operationalization in psychology. Demonstrating that a measure successfully operationalizes a trait is called construct validity. Note, however, that we need to standardize for effort to measure the trait. Otherwise variation in effort produces variation in the measured trait across situations with different incentives.

A Fundamental Identification Problem
Operationalization and construct validation require heroic assumptions. Even if one adjusts for effort in a task, measured productivity may depend on multiple traits. Thus two components of θ (say θ 1,µ , θ 1,π ) may determine productivity in j. Without further information, one cannot infer which of the two traits produces the productivity in j. In general, even having two (or more) measures of productivity that depend on (θ 1,µ , θ 1,π ) is not enough to identify the separate components.
Standardize measurements at a common level of effort e j = e j = e * . Note that if the supports of e j and e j are disjoint, no (θ 1,µ , θ 1,π ) exists. Assume that the φ k () are known. If the system of equations satisfies a local rank condition, then one can solve for the pair (θ 1,µ , θ 1,π ) at e * . Only the pair is identified. One cannot (without further information) determine which component of the pair is θ 1,µ or θ 1,π .
In the absence of dedicated constructs (constructs that are generated by only one component of θ), there is an intrinsic identification problem that arises in using measures of productivity in tasks to infer traits. Analysts have to make one normalization in order to identify the traits. However, we need only one such construct joined with patterned structures on how θ enters other task to identify the vector θ (e.g. one example is a recursive, triangular structure). See the discussion in Almlund et al. (2011).

Examples of Nonidentification
IQ and achievement test scores reflect incentives and efforts, and capture both cognitive and personality traits. Table 2 summarizes the evidence that paying disadvantaged students for correct answers on IQ tests substantially raises measured IQ. Almlund et al. (2011) summarize many other studies. At baseline (in the fall), there was a full standard deviation difference (10.6 points and SD was about 9.5 in this sample) between scores of children in the optimized vs standardconditions The nursery group improved their scores, but only in the standard condition.
"…performance on an intelligence test is best conceptualized as reflecting three distinct factors: (a) formal cognitive processes; (b) informational achievements which reflect the content rather than the formal properties of cognition, and (c) motivational factors which involve a wide range of personality variables. (p. 2) "…the significant difference in improvement in standard IQ performance found between the nursery and nonnursery groups was attributable solely to motivational factors…" (p. 10) Breuning and Zella [1978] Within and between subjects study of 485 special education high school students all took IQ tests, then were randomly assigned to control or incentive groups to retake tests. Subjects were below-average in IQ.

Incentives such as record albums, radios (<$25) given for improvement in test performance
Scores increased by about 17 points. Results were consistent across the Otis-Lennon, WISC-R, and Lorge-Thorndike tests.
"In summary, the promise of individualized incentives contingent on an increase in IQ test performance (as compared with pretest performance) resulted in an approximate 17-point increase in IQ test scores. These increases were equally spread across subtests… The incentive condition effects were much less pronounced for students having pretest IQs between 98 and 120 and did not occur for students having pretest IQs between 121 and 140." (p. 225) Holt and Hobbs [1979] Between and within subjects study of 80 delinquent boys randomly assigned to three experimental groups and one control group. Each exp group received a standard and modified administration of the WISC-verbal section.
Exp 1-Token reinforcement for correct responses; Exp 2 -Tokens forfeited for incorrect responses (punishment), Exp 3-feedback on correct/incorrect responses 1.06 standard deviation difference between the token reinforcement and control groups (inferred from t= 3.31 for 39 degrees of freedom) "Knowledge of results does not appear to be a sufficient incentive to significantly improve test performance among below-average I.Q. subjects…Immediate rewards or response cost may be more effective with below-average I.Q. subjects while other conditions may be more effective with average or above-average subjects." (p.

83)
A considerable fraction of the variance in achievement tests is explained by personality traits. See

Measures of Personality in Psychology Based on Linear Factor Analysis
Such measures account for measurement error, and identify factors that can be interpreted as traits. Cunha et al. (2010) develop nonlinear factor models (nonlinear and nonparameteric). Using these models they establish that measurement error is quantitatively important. The share of error variance for proxies of cognition, personality and investment ranges from 1%-90%. Not accounting for measurement error produces downward-biased estimates of self-productivity effects and perverse estimates of investment effects.

A Definition of Personality
I now add preferences and goals to the analysis. Preferences and goals also shape effort. They are personality traits broadly defined. Income is the return to productivity: Preferences are defined over final consumption goods X, productivity P and effort e: U (X, P, e | ψ) , ψ ∈ Ψ. (3) Agents have preferences over goods, agents may value the output of tasks in their own right and agents may value the effort devoted to tasks. The agents maximize (3)

Personality Traits
Personality traits are the components of e, θ and ψ that affect behavior. We observe measured personality-behaviors generated by incentives, goals, and traits.

Actions
Actions are styles of behavior that affect how tasks are accomplished. They are aspects of behavior that go beyond effort. Smiling, cajoling, etc. are examples. Tasks are accomplished by taking actions. The i th possible action to perform task j is denoted a i,j , i ∈ {1, . . . , K j }. Array actions in a vector a j = a 1,j , . . . , a Kj ,j ∈ A.
Actions may be the same or different across the tasks. The productivity of the agent in task j depends on the actions taken in that task: P j = τ j a 1,j , a 2,j , . . . , a Kj ,j .
The actions themselves depend on traits θ and "effort" e i,j : where Kj i=1 e i,j = e j and J j=1 e j =ē.
Actions generalize the notion of effort to a broader class of behaviors.
Let M be the set of actions, including actions that do not directly contribute to productivity. Let M be the index set of items in M The agent solves max E [U (a, X, P, e | ψ) | I] with respect to X and e given the stated constraints.
We can introduce situations indexed by h ∈ H. For a person with traits θ and effort vector e j with action a i,j , using the specification (7), the action function can be expanded to be dependent on situation h:

A Definition of Personality
Let T ∈ T be a vector of traits (θ, ψ,ē). Personality is a response function.
The behavior that constitutes personality is defined as a pattern of actions in response to the constraints, endowments, and incentives facing agents given their goals and preferences.
Actions-not traits-constitute the data used to identify the traits. Personality psychologists use actions (e.g., "dispositions") to infer traits. Identification issues similar to those previously discussed apply to this broader set of measurements of behaviors.

Personality as Enduring Actions
Many personality psychologists define personality as "enduring patterns of thoughts, feelings and behaviors"

Average Actions
Consider task j and trait vector T = (θ, ψ,ē). Define the average action for information set I: where S T,I (h, e i,j ) is the support of (h, e i,j ) given T and I. g (h, e i,j | T = (θ, ψ,ē), I) is the density of (h, e i,j ) given T = (θ, ψ,ē) and information set I.ā T,j,I is the "enduring action" of agents across situations in task j with information I, i.e., the average personality. Only if ν i,j is separable in T , the marginal effect of personality trait vector θ is the same in all situations.
One can define the "enduring traits" in a variety of ways, say by averaging over tasks, j, situations, h, or both. Only under separability in T will one obtain the same marginal effect of θ. Epstein (1979) and a subsequent literature present evidence against nonseparability but in favor of an "enduring trait" that is common across situations. He argues strongly against the extreme form of situational specificity assumed in modern behavioral economics.    Roberts et al. (2006) and Roberts and Mroczek (2008). Reprinted with permission of the authors.  Roberts et al. (2006) and Roberts and Mroczek (2008). Reprinted with permission of the authors.  Roberts et al. (2006) and Roberts and Mroczek (2008). Reprinted with permission of the authors.

Processes of Development Discussed in the Literature
There are many hypothesized mechanisms of change. Two common processes discussed in the literature are ontogeny (programmed developmental processes common to all persons) and sociogeny (shared socialization processes). Personality also changes through external forces above and beyond common ontogenic and sociogenic processes. Such changes operate through alterations in normal biology, such as brain lesions and chemical interventions. A channel that receives a lot of attention in economics is investment: educational interventions and parental investment that affect personality throughout the lifecycle.

Life Cycle Dynamics
Let T v be traits at age v, v ∈ {1, . . . , V } ∈ V. Information I v may be updated through various channels of learning. The technology of skill formation Heckman, 2007, 2009) postulates the following equation of motion: Functions can be nonautonomous (v-dependent). Situations may change over time as a function of past actions, past situations, investment, information, and the like: Information I v may also change over the life cycle through experimentation and learning: Figure 17 summarizes the dynamics of skill formation as formulated in Heckman (2007, 2009).

Figure 17: A Life Cycle Framework for Organizing Studies and Integrating
Evidence: Period Life Cycle Cunha et al. (2010) estimate technology (10) using longitudinal data on the development of children with rich measures of parental investment and of child traits. Self-productivity becomes stronger as children become older, for both cognitive and noncognitive capability formation. The elasticity of substitution for cognitive inputs is smaller in the adolescent years, so that it is more difficult to compensate for the effects of adverse environments on cognitive endowments at later ages than it is at earlier ages.
This finding explains the evidence on ineffective cognitive remediation strategies for disadvantaged adolescents. Personality traits foster the development of cognition but not vice versa. Cunha et al. (2010) show that it is equally easy to substitute for deficits in personality traits at both early and late stages for socioemotional skills over the life cycle.
Overall, 16% of the variation in educational attainment is explained by factors extracted from adolescent cognitive traits, 12% is due to factors extracted from adolescent personality (socioemotional traits), and 15% is due to factors extracted from measured parental investments.

The Causal Effects of Schooling on Cognitive and Personality Traits
Using the methodology of Hansen et al. (2004), it is possible to estimate the causal effect of schooling on cognitive and noncognitive measurements. See Figures 18-21. Schooling has substantial effects on both types of traits.

Figure 18: Causal Effect of Schooling on ASVAB Measures of Cognition
Notes: Effect of schooling on components of the ASVAB. The first four components are averaged to create male's with average ability. We standardize the test scores to have within-sample mean zero, variance one. The model is estimated using the NLSY79 sample. Solid lines depict average test scores, and dashed lines, confidence intervals. Source: Heckman et al. (2006, Figure 4).

Figure 19: Causal Effect of Schooling on ASVAB Measures of Cognition
Notes: Effect of schooling on components of the ASVAB. The first four components are averaged to create male's with average ability. We standardize the test scores to have within-sample mean zero, variance one. The model is estimated using the NLSY79 sample. Solid lines depict average test scores, and dashed lines, confidence intervals. Source: Heckman et al. (2006, Figure 4).

Figure 20: Causal Effect of Schooling on ASVAB Measures of Cognition
Notes: Effect of schooling on components of the ASVAB. The first four components are averaged to create male's with average ability. We standardize the test scores to have within-sample mean zero, variance one. The model is estimated using the NLSY79 sample. Solid lines depict average test scores, and dashed lines, confidence intervals. Source: Heckman et al. (2006, Figure 4).

The Evidence from Interventions
The Perry Preschool program intervened early in the lives of disadvantaged children. It has a 7-10% rate of return per annum. (See Heckman et al., 2010.) The Perry Preschool Program did not have a lasting improvement on cognitive ability, but it did improve important later-life outcomes through changes in personality . Notes: IQ measured on the Stanford-Binet Intelligence Scale (Terman and Merrill, 1960). Test was administered at program entry and each of the ages indicated. Source: Cunha et al. (2006) and Heckman and Masterov (2007) based on data provided by the High Scope Foundation.
The Perry Preschool Program worked primarily through socioemotional channels. It raised scores on achievement tests but not IQ tests. As previously noted, socioemotional factors and cognitive factors both explain performance on achievement tests (Duckworth, 2007;Borghans et al., 2009).

Personality and Preference Parameters
Measures of personality predict a wide range of life outcomes that economists study. Personality psychologists define traits as relatively stable, person-specific determinants of behavior. Preferences are the natural counterpart of these traits in economics. However, the exact link between personality and preferences is unclear. Table 3 shows one possible correspondence between conventional economic preference parameters and personality measures.   The evidence relating personality to time preferences is mixed. Using data from an experiment involving college students, Daly, Delaney and Harmon [2009] find that a factor that loads heavily on self-control, consideration of future consequences, elaboration of consequences, affective mindfulness, and Conscientiousness, is negatively associated with the discount rate.

Summary and Conclusions
What can economists take from and contribute to personality psychology? What do we learn from personality psychology? Personality traits predict many behaviors sometimes with the same strength as conventional cognitive traits. Personality psychology considers a wider array of actions than are usually considered by economists. It enlarges the economist's way to describe and model the world. Cognition is one aspect of personality broadly defined.
Personality traits are not set in stone. They change over the life cycle. They are a possible avenue for intervention and policy.
Personality psychologists lack precise models. Economics provides a framework for recasting the field.
More precise models reveal basic identification problems that plague measurement in psychology. Such analyses show that, at an empirical level, "cognitive" and "noncognitive" traits are not easily separated.
Personality psychologists typically present correlations-not causal relationships. Many contemporaneously measured relationships suffer from the problem of reverse causality. Econometric tools can be used to define and estimate causal mechanisms and to understand the causes of effects. Psychological measures have substantial measurement error. Econometric tools account for measurement error, and doing so makes a difference. Economists can formulate and estimate mechanisms of investment-how traits can be changed for the better.
There are major challenges in linking the traits of psychology with the preferences, constraints and expectation mechanisms of economics. Developing rigorous methods for analyzing causal relationships in both fields remains to be done. Developing a common language and framework to promote interdisciplinary exchange is required. There is a danger in assuming that basic questions of content and identification have been answered by psychologists at the level required for rigorous economic analysis.