Heterogeneous returns to personality: the role of occupational choice

While the measures of the Big Five inventory provide a concept to capture all superior facets of personality that are intrinsic to a person, reciprocity, and locus of control serve as measures of behavior or attitude related to outcomes. The Big Five traits can be characterized as follows: the first facet, conscientiousness, relates to whether a person is reliable, organized, and responsible. The second, extraversion, corresponds to an enthusiastic, outgoing attitude, while the third, agreeableness, relates to a kind and compassionate attitude. Neuroticism, being the fourth, instead is defined with respect to being unstable, worrying, and anxious, and finally, the fifth, openness to experience, refers to imaginative, original individuals with wide interests. Reciprocity aims at measuring the propensity to symmetrically react to friendly or hostile behavior, whereas locus of control captures the attitude of how self-determined (internal) or heteronomous (external) a person regards her own life.

Although cognitive and non-cognitive skills partly are substitutes, they are seen here as complements, since we can assume that cognitive ability is a very important requirement for job entry, which works as a hurdle before non-cognitive skills are considered.

Judge et al. (1999) relate these occupational personality measures with the concept of the Big Five and reveal significant correlations of e.g., openness to experience and the artistic type.

The Guilford-Zimmerman Temperament Survey (see Guilford et al. 1976) distinguishes the ten facets activity level, restraint, sociability, domination, emotional stability, objectivity, friendliness, thoughtfulness, personal relation skills, and masculinity. Sociability, friendliness, and emotional stability correspond to extraversion, agreeableness, and neuroticism (reversely defined) of the Big Five inventory.

Stevens et al. (1993) summarize evidence on this indirect link. They report results of several studies emphasizing different strategies and outcomes of salary negotiation for men and women. As a possible link, Stevens et al. (1993) name gender differences in self-efficacy, a concept that directly relates to other measures like locus of control or self-esteem.

NLSYW stands for National Longitudinal Survey of Young Women, whereas, NCDS stands for National Child Development Study.

BIBB denotes the Bundesinstitut für Berufsbildung, IAB the Institut für Arbeitsmarkt- und Berufsforschung der Bundesagentur für Arbeit.

See Appendix 8 for a more detailed discussion on the construction of our empirical measures for the personality traits. Moreover, we provide a commented STATA do-file, enabling reproducibility of the data, as an online resource.

Since individuals are not required to be at least 30 years when choosing their occupation, we rely on rank-order stability only when measuring the impact of personality on occupational choice.

The group of agricultural employees is not considered separately. Due to the small number they are counted in the group of laborers.

Gross labor income includes overtime premiums, but no special payments like e.g., leave pay.

P values of testing the hypothesis of equal means are displayed in detail in Table 5 in Appendix 9.

Note that for overlapping domains for all measures of non-cognitive skills, an individual can sort into more than one possible occupation.

There is a longstanding tradition to use a multinomial logit model to estimate occupational choice. Early general applications of using a multinomial logit to estimate occupational attainment were provided by Boskin (1974) and Schmidt and Strauss (1975). The method has also been applied to issues of occupational mobility, for example, of different ethnic minorities like in Kossoudji (1988) using the 1976 Survey of Income and Education or Chiswick and Miller (2009) using US Census data from 2000. Cobb-Clark and Tan (2011) likewise employ a multinomial logit model to the Australian HILDA data in order to estimate gender specific occupational attainment. Besides, Constant and Zimmermann (2003) make use of a multinomial logit model for German SOEP data to estimate the influence of parents occupation on occupational choice of children.

See Ham et al. (2009b), who apply the same test to justify validity if IIA.

Socio-economic variables are dummy variables for German citizenship, the presence of children younger than 16 years, being married, and age coded into three dummy variables for being 40–49 and 50–55 years with 30–39 years as the reference group. Regarding education, dummy variables for basic, lower secondary, and higher secondary education, possessing a vocational degree and having a university degree are included. The reference group for education is no educational degree. Level of education of parents is also included in the analysis, but here a coarser classification is applied: dummy variables for possessing a lower secondary, higher secondary degree, or a university degree are coded for mother and father, respectively. Parents’ occupation is regarded as well using the same classification as for the individuals included in the estimation. Hence, there are eight dummy variables for group of occupation for mother and father, respectively. Regional information contains local unemployment rates and GDP measured at the level of federal states and dummy variables for regions East, West, North, South, and city state. East comprises federal states Mecklenburg-Western Pomerania, Thuringia, Saxony, Saxony-Anhalt and Brandenburg. Region West corresponds to federal states North Rhine-Westphalia, Rhineland-Palatinate, and Hesse. The northern region stands for federal states Schleswig-Holstein and Lower Saxony. South (the reference region) comprises federal states Bavaria and Baden-Wuerttemberg. City states correspond to federal city states Berlin, Hamburg, and Bremen.

Labor market experience is measured in quadratic polynomials of years spent in full-time and part-time employment as well as of years in unemployment. Job characteristics included are dummy variables for working in the public sector and for being employed in a company with at least 200 employees. In addition, tenure and the required training of the position held are considered. Required training is a dummy variable equal to one if the employment requires having a diploma from a university or a university of applied sciences.

Laborers are the group with the lowest qualification requirements.

Table 6 in the Appendix displays all estimated coefficients of the personality variables.

All scores are standardized to have mean zero and standard deviation one to enable comparison across groups.

See Card (1999) for a summary of empirical findings (Table 6, p. 1849–1850).

We apply a back-of-the-envelope calculation, where marginal effects are multiplied with the average wage rate in the sample times 4.29 (average number of weeks per month) times 40 (assuming full-time employment of 40 h per week). Earnings are measured in prices of the year 2000.

We thank an anonymous referee, who suggested this evaluation of the results.

Referring to the notation in Sect. 4, we use \(\mathbf {Z}\) to denote background variables in the occupation equation and \(\mathbf {X}\) for background variables in the wage equation.

Swope et al. (2008) undertake a similar analysis to classify interaction of certain traits: Using the Meyer-Briggs-typology they describe relative frequencies of types built as joint appearance of four attitudes.

Using factor analysis to analyze questionnaire items aiming at measuring personality shows that factors load on more than one factor. This in turn leads to correlated factors. Correlation analysis for the Big Five personality traits within our sample reveals a negative relationship between neuroticism and the other four traits. Conscientiousness and agreeableness as well as extraversion are substantially positively correlated. The same applies for extraversion and openness to experience. These relationships can be found in other samples, too, see Biesanz and West (2004).

This personality profile accounts for around 11.5 % of the sample. Assuming a uniform distribution, each type would occur with a share of 3.125 %. A share of 11.5 % thus represents a more than threefold more probable occurrence of that specific type.

Marginal effects for all personality types are displayed in Appendix 9 in Table 8.

We apply a back-of-the-envelope calculation, where marginal effects are multiplied with the average wage rate in the sample times 4.29 (average number of weeks per month) times 40 (assuming full-time employment of 40 h per week). Earnings are measured in prices of the year 2000.

The theoretical model as well as the estimation equation is presented in Appendix 10.

Estimating the relationship of personality and earnings assumes that there is no correlation with other factors influencing returns, as for example cost of effort. Estimates therefore can only provide gross effects, since these factors are unmeasurable and are likely correlated with personality traits. Besides, returns to non-cognitive skills may interfere with compensating wage differentials that measure wage premiums for specific (adverse) characteristics of the employment.

Piatek and Pinger (2010) proceed in a similar way when extracting locus of control from the SOEP questionnaire.

\(m(P_j)\) corresponds to \(E(u_j), j=1,\ldots ,8\) given that outcome \(j\) has been chosen and \(m(P_k) \frac{P_k}{(P_k-1)}\) corresponds to \(E(u_k), k=1,\ldots ,8\) and \(k\ne j\) given that outcome \(j\) has been chosen.