Personality Traits and Household Consumption Choices

In this paper, we test whether consumption choices are affected by personality traits and whether this impact is different for singles and individuals living in couples. To fulfill this aim, we test the impact of personality on preferences for different commodities using the German Socio-Economic Panel (SOEP) and estimating a system of Engel curves that includes personality traits as demographic shifters. The analysis is conducted on four different samples: single men, single women, childless couples and couples with children.The inclusion of personality traits among demographic shifters helps to reduce unobserved heterogeneity and improves the goodness of fit of the Engel curves specification by an average of 15.7%. In comparing the results for singles and couples, we find evidence of a consumption-based marital surplus for Mental Openness and Conscientiousness. These traits are characterized by positive assortative mating, and they have a significant and consistent impact on the expenditure for several commodities for both singles and couples. For instance, similarly open-minded partners are likely to spend household resources on culture, and their joint consumption of such goods may be a potential reason for their marital surplus.


Introduction
In this paper, we examine the role played by personality traits in consumption decisions for both individuals and households. In the past two decades, a flourishing number of contributions in the economics literature have investigated the role played by personality traits in predicting both individual decision making and life outcomes across a wide variety of domains. First, in examining the interplay between education and individual outcomes in the labor market, economists have recognized that personality traits are an important component of human capital (Heckman et al., 2013;Rustichini et al., 2016;Almlund et al., 2011;Borghans et al., 2008;Bowles et al., 2001b).
The domains in which personality traits have been shown to be relevant go well beyond academic and labor market outcomes, mainly including credit scores, financial decisions, portfolio choices and business attitudes (Bucciol and Zarri, 2017;Caliendo et al., 2016;Conlin et al., 2015;Caliendo et al., 2014;Brown and Taylor, 2014).
With an even wider perspective, present-day economics contributions to personality research include analyses in which the economic facet is one among several: for example, personality has been proved to be an important moderator in determining how income and life events impact subjective well-being (e.g. Proto and Rustichini, 2015;Boyce and Wood, 2011) and social cohesion (Rapallini and Rustichini, 2016).
Similarly, the well-known evidence -among psychologists-that personality predicts marital satisfaction (e.g. Gaunt, 2006) is now common to economists who examine the role of personality in the marriage market (Dupuy and Galichon, 2014) as well as in intrahousehold bargaining on time use and partners' labor supply (Flinn et al., 2018). Indeed, the potential role of personality in couple formation was mentioned as early as Becker (1973) in his seminal model of the marriage market, where couples were characterized mainly by production complementarities and specialization between partners (Becker, 1981). In the Beckerian framework, marital surplus was generated by negative assortative mating on the personality traits that predict positive outcomes in the labor market and household production, with the husband being high in traits associated with better labor market performances and the wife in traits that imply higher productivity in domestic tasks. However, as women's labor force participation has increased, and the relative significance of household (rather than market) production has declined, complementarities in consumption have become more important sources of gains to marriage (Lundberg and Pollak, 2007;Stevenson and Wolfers, 2007). Accordingly, positive assortative mating on personality has been interpreted as evidence that consumption complementarities, such as those due to the joint consumption of public goods, are of greatest benefit when individuals with similar preferences for consumption and leisure are matched (Lundberg, 2012).
In this work, we link consumption preferences with personality for individuals living both in a one-person household and in a couple, being aware of the possible existence of assortative mating on personality traits. Stemming from the idea that personality traits shape consumption preferences (Lundberg, 2012), we test their impact on preferences for different commodities in samples of singles, couples and couples with children. We first estimate a system of Engel curves 1 by gender using a sample of one-person households. The study of the choices of male and female singles, for whom there is no couple's decision and an absence of children, allows us to draw conclusions regarding the influence of personality on consumption choices. Second, we estimate a system of Engel curves for people living in childless couples, including both the husband's and wife's personality traits. Then, we estimate a system of Engel curves for a sample of couples with children, again considering the personality traits of both partners. If the effects of a particular personality trait on a few consumption categories are significantly similar for singles and individuals living in couples, together with positive assortative mating on this trait, we interpret these findings as evidence of the presence of a consumption-based marital surplus. In other words, the persistence of the impact of a specific personality trait on different commodities from one-person households to couples is here interpreted as a direct test of the presence of a marital surplus in consumption-based couples. This finding is true when the sorting on a specific personality trait is positive, assuming that couples who are similar in this trait benefit more from joint consumption of household public goods. Furthermore, in such a case, the marital surplus could be due to the fact that partners agree quite easily on consumption decisions. Lundberg (2012) already provided evidence of similarity in a few traits of personality in the sorting of men and women into marriage and on the probability of divorce when personalities are too far divided. She interprets her findings as proof of the increasing spread of couples based on consumption complementarities among the younger cohorts living in Germany. We instead directly estimate the effect of personality on consumption choices for people living in households with different compositions; however, our interpretation is in line with the approach that she suggested.
Similar to the bulk of the literature investigating the role of personality in economic decisions, we adopt the Big Five taxonomy. There is substantial agreement among personality psychologists on a five-factor structure to account for substantive covariations in personality descriptions (Costa and McCrae, 1989) and broad agreement about the labeling of these five factors, i.e., Extraversion (attitude toward being active, being forthcoming and desiring social relationships), Agreeableness (being friendly, warm and sensitive towards others), Conscientiousness (being systematic, goal-oriented and self-disciplined), Neuroticism (worrying, being nervous and being emotionally unstable) and Mental Openness (or Intelligence -being imaginative, creative, curious and unconventional).
We use the German SOEP, which records the Big Five personality traits assessment in several waves, including 2009, and comprehensive household consumption information in 2010 while referring to 2009 expenditures. By pooling the two waves and retaining those households (and individuals) for which both consumption expenditure and personality traits are observed, we are able to estimate a complete system of Engel curves and to address our research questions.
This paper aims to contribute to the literature along two main lines. First -as far as we know-consumption is a crucial sphere of economic decisions for which the effect of personality traits has not yet been investigated either at the individual or at the household level. We fill this literature gap by analysing the impact of personality on consumption decisions and how the inclusion of personality traits improves the estimation of Engel curves for each consumption category. Second, by focusing on household consumption choices, we shed light on the marital surplus in societies where women's participation in the labor market is increasingly widespread and the Beckerian model of marriage -based on the specialization of the partners-is probably decreasingly common.
In contrast, complementarities on consumption may become crucial to explain the formation, and success, of domestic partnerships. The remainder of the work is organized as follows. Section 2 introduces the background literature. Section 3 describes the data and the empirical strategy.
Section 4 discusses the main results. Section 5 concludes and suggests avenues for further research.
2 Background 2.1 Personality traits and economic outcomes Bowles et al. (2001b) surveyed for the first time the broad empirical evidence showing that apparently similar individuals, in terms of age, years of schooling, years of labor market experience, parents' level of schooling, occupation, and income, receive quite different earnings. In these seminal contributions, Bowles et al. (2001b,a) suggested that personality -or behavioral-traits be included in a theoretical human capital model. Personality traits are termed "incentive-enhancing preferences", meaning that they allow the employer to induce effort at lower cost, and they are clearly distinguished from cognitive skills. The individual's rate of time preference as well as the locus of control -measured by the Rotter scale-are the traits that were taken into account in these first papers. From then on, the idea that the labor market may remunerate not only cognitive skills but also the individual's noncognitive abilities achieved an increasing consensus among economists. In fact, evidence that productivity and the earnings of workers can be predicted by personality, mainly by the traits of Conscientiousness and Mental Openness, is emerging from more recent survey data analysis (Fletcher, 2013;Hanes and Norlin, 2011;Nyhus and Pons, 2005) and from studies based on experimental settings using real effort tasks (Carpenter, 2016;Cubel et al., 2016). Apart from the crucial role played by the two abovementioned traits, there is scholarly agreement that different traits predict earnings differently according to the worker's gender and that the magnitude of this effect is comparable to that of cognitive skills (e.g. Roberts et al., 2007;Mueller and Plug, 2006).
The comparison between the predictive power of cognitive and noncognitive skills and the research on the interplay between education and personality in the labor market is currently a crucial topic in the literature. This interplay is characterized by both direct and indirect effects of personality on the two domains. In the seminal model of Bowles et al. (2001b), personality has a direct effect on individual productivity, but personality may also have an indirect effect by affecting preferences for schooling and/or occupational choices. In addition, there is much evidence -well known among psychologists-that personality directly predicts educational outcomes (Poropat, 2009;Duckworth et al., 2007). Considering both the direct and indirect effects and examining a broader sample of behaviors, Heckman and colleagues showed that individuals who received a preschool intervention aimed at improving noncognitive skills during childhood, namely, the Perry preschool program, as adults have higher levels of educational attainment, employment and marriage and lower levels of crime than individuals who did not receive the intervention (Heckman et al., 2013). Moreover, they found evidence that although the Perry program did not produce long-term gains in IQ, it did create persistent improvements in personality skills, and the latter are crucial for long-term goals in all the aforementioned domains (Heckman et al., 2013;Almlund et al., 2011;Borghans et al., 2008).
Combined with the comparison between the role played by cognitive and noncognitive skills, the predictive power of personality is currently being investigated for the spheres of economic decisions in which risk attitude and time preference are the individual characteristics that are traditionally taken into account . The high correlation between these types of preferences and certain personality traits suggested new research venues. There is evidence, albeit not claimed to be complete, that personality -especially the Mental Openness and Extraversion traits-affects the transition into self-employment and that survival in self-employment is reduced for people who score high in the Agreeableness trait (Caliendo et al., 2014). The same study showed that risk tolerance has an influence on the decision to enter self-employment and on survival in self-employment, even when the authors control for the Big Five, and that the explanatory power of all the observed personality constructs amounts to 30% of all the observable variables. More specifically, in finance, Bucciol and Zarri (2017) show that the Agreeableness and Neuroticism traits have a significant negative correlation with financial risk taking, as measured by the holding and the amount of stock assets. Investigating both individuals and couples, Brown and Taylor (2014) analyzed the relationship between personality traits and financial decision making, focusing on unsecured debt and financial assets. They showed that Conscientiousness, Extraversion, and Agreeableness correlate with the amount of unsecured debt and savings. They also found significantly negative correlations with stock holding and the Extraversion trait in the sample of couples and the Agreeableness trait in the sample of singles. Within this framework, and closer to the topic of this paper, a promising field of research has analyzed how personality affects both couple formation and the economic decisions made within a household.

Assortative mating and personality traits
The potential role of personality in couple formation was originally identified by Becker (1973) in his seminal model of the marriage market. Although most of the successive studies on the marriage market examined marital sorting based on a single-dimensional trait -mainly education or earnings, psychological traits were recognized by Becker (1973) as one of the possible dimensions of marital sorting. Thanks to the availability of personality inventories in large samples, a number of recent studies have examined the role of personality traits in the marriage market (Lundberg, 5 2012), assortative mating (Dupuy and Galichon, 2014) and couples labor supply (Flinn et al., 2018). Dupuy and Galichon (2014) proposed a theoretical model to test empirically for the dimensionality of sorting in the marriage market and to evaluate the importance of personality traits in couples' joint utility. Using data of Dutch households, they provided evidence that sorting occurs on multiple indices, including education, personality traits, BMI and the health status of the spouses, rather than on just a single dimension, as assumed in most of the current literature. This finding implies that individuals face important trade-offs between the attributes of their potential spouse. They observe homogamy only over the Conscientiousness trait and complementarity among the other traits. In detail, women face a trade-off between being attractive to more conscientious men and being attractive to more autonomous men. Similarly, among the Dutch couples, there was evidence that more conscientious women prefer more agreeable men, but more extraverted women prefer less agreeable men. Men therefore face a trade-off between being attractive to more conscientious women and being attractive to more extraverted women. In a nutshell, and excluding the Conscientiousness trait, the other traits matter differently for men and women. As a second finding, this study showed that personality traits explain a percentage of couples' joint utility in marriage similar to that of education: 17% vs. 28%.
The role of personality traits in couples' labor supply and resources allocation was structurally modeled by Flinn et al. (2018) using a sample of childless Australian couples. They found marital sorting on the Mental Openness and Neuroticism traits as well as a relation between positive sorting on these traits and higher levels of cooperation within a couple. They also found that personality is an important determinant of household bargaining weights and has an impact on wages comparable in magnitude to that of education. In particular, a percentage increase in the score of the Agreeableness trait reduces the Pareto weight by 0.8%. In addition, the Conscientiousness trait has an important positive impact on the labor market outcomes of both partners, and its difference within a couple also explains the gender wage gap.
Closer to our focus, Lundberg (2012) studied the role of production and consumption complementarities in couple formation and dissolution in Germany across different cohorts, examining the effects of personality traits. She argued that although production complementarities within the household have become less important than in the past because of the increase in women's labor force participation, there is not enough direct evidence that individual gains from marriage have become more consumption-based. To fill this gap, she used personality traits as proxies of preferences, and capabilities, to directly examine the marital surplus due to the joint consumption of household public goods. Returns to marriage due to production complementarities led to the standard prediction of gender specialization within the household and negative assortative mating (Becker, 1973), while complementarities in consumption were of greatest benefit among individuals with similar preferences (Lam, 1988) and implied positive assortative mating on traits related to preferences for household consumption among consumption-based marriages. Thus, Lundberg (2012) inferred that homogamy in a personality trait is related to the consumption benefits of 6 marriage. In particular, she provided evidence that the Mental Openness and Conscientiousness traits are positive predictors of marriage, and marital stability, for German couples of the younger cohorts, being -probably-associated with a high demand for marital public goods. For German couples of the older cohorts of the population, she instead found that psychological traits have more gender-specific effects, thus being consistent with the theoretical hypothesis of specialization in marriage. 2 3 Data and empirical strategy

Data and sample selection
This study uses data from the German SOEP, 3 a representative ongoing longitudinal survey of the German population. It suits the needs of our study because it records Big Five personality trait assessments in several waves, including 2009, and in the 2010 wave, it records comprehensive household consumption information from 2009. The two waves are pooled, and only those households (and individuals) for which both consumption expenditure and personality traits are observed are retained in the sample . 4 It is worth noting that although the methodology of collecting consumption expenditure data differs, and the aggregation of consumption items is broader with respect to typical household budget surveys, there is sufficient detail to estimate a complete system of Engel curves, which allows us to address our research questions. Finally, the sample of couples with children is composed of 1,982 families, which is reduced to 2 The different roles of marital sorting for older and younger cohorts in Germany found in Lundberg (2012) are consistent with quite persistent traditional gender roles in Germany, where labor market participation rates among married women are still low, especially in the former western part of the country, although constantly increasing. Also using the German SOEP, Pestel (2017) examined the relationship between marital sorting and female labor supply in East and West Germany. He observed a high level of assortative mating on education for couples where both partners have a medium or low level of education; for highly educated couples, there were marginal levels of sorting, especially in West Germany. He also found more attachment to the labour market among East German wives independent of the husband's earning quintile.
3 DOI: 10.5684/soep.v32.1, see Wagner et al. (2007). 4 The SOEP data sample has been prepared using the PanelWhiz stata add-on (Haisken-DeNew and Hahn, 2010). 5 Such a sample selection clearly impedes any claim of representativeness for the study. Nevertheless, we are interested in the microaspect of the interaction between personality traits and household consumption choices and consequently prefer to avoid the notable theoretical and empirical issues that would arise from including older children and also dependent elderly relatives or other forms of complex families.
1,035 once all missing values are accounted for. We consider only children younger than 16, as when older children are present, it would be implausible to assume that consumption decisions are made only by the two adults. Actually, when a couple's children are adolescents -or young adults-they should also be considered decision makers.

Consumption variables
In wave 2010, the SOEP collected information on household consumption in 2009, 6 asking for information on expenditures on 16 aggregated commodities and services: food at home, food out of the home, clothing and shoes, personal hygiene, health, telecommunications, education, culture, leisure time, vacations, life and pension insurance, other insurance, car repairs, transportation, furniture, and other expenditures. For each item, three questions were asked: whether there was an expenditure on the specific item, the amount of the monthly expenditure and the amount of the yearly expenditure (see Appendix A). 7 We use the monthly expenditure as our main consumption measure, substituting the yearly expenditure divided by twelve when the monthly value is missing and zero when both are missing and the household declared that it did not consume the item.
Because of the large number of missing records in certain consumption categories, we aggregated life and pension insurance and other insurance into a unique insurance category and car repairs and transportation into a broader transportation category. As a consequence, the final system includes 13 categories: food at home, food out of the home, clothing and shoes, personal hygiene, health, telecommunications, education, culture, leisure time, vacations, insurance, transportation, and other expenditures. 8 Although slightly aggregated, all expenditure categories were preserved, producing a complete consumption expenditure system for nondurable goods (as in Pollak and Wales, 1981, for instance), where total expenditure is the sum of expenditures in all categories. Descriptive statistics on budget shares and total expenditure are reported in the first parts of Tables 1 and 2 for singles and couples. The largest share of the household budget is spent on food at home, accounting for roughly 34-38% in all samples, followed by insurance, with 10-16%, and vacations, with about 7% for singles and 10-12% for couples. All other budget shares are below 10%, and the smallest ones are education and culture at approximately 1% for both singles and couples.
Finally, as detailed in Section 3.5, one possible concern when estimating demand systems is the potential endogeneity of total expenditure. The main cause is measurement error, due either to the infrequency of purchases or to recall errors. As instruments for total household 6 Appendix A reports the questions used by the SOEP to assess household consumption. 7 The pros and cons of collecting survey data on consumption by recall, vs. the diary method, have been reviewed by Crossley and Winter (2014), who showed that information collected by the former method is quite accurate. Marcus et al. (2013) compared consumption data collected in the SOEP (2010) with those of the Income and Expenditure Survey (EVS) that was conducted in 2008 and was based on diary records. After pointing out several reasons that a perfect overlap between the consumption distributions of the two data sources should not be expected, they concluded that the two likely reflect the relevance of certain consumption categories in similar ways.
8 The last category, other expenditures, accounting for about 3% of the total expenditure on average, is omitted from the demand system to avoid collinearity.  (Jackson, 2005), all of which have an eigenvalue significantly greater than 1 in all samples, a criterion often used to assess the number of relevant components (see Figure 1).  Given the scope of this study, an important feature of personality traits is their stability across ages and life events as well as external and cyclical shocks. The stability of personality traits is a research question that has been debated at length among psychologists (Specht et al., 2014;Boyce et al., 2013;Specht et al., 2013;Lucas and Donnellan, 2011;Specht et al., 2011;Roberts and DelVecchio, 2000;Roberts, 1997) and more recently among economists as well (Cobb-Clark and Schurer, 2012; Almlund et al., 2011;Borghans et al., 2008). To clarify different possible answers to this research question, the notion of change should be clearly defined. At the population level, at least two measures have been considered: mean-level changes and rank-order changes.
"Mean-level change reflects shifts of group of people to higher or lower values on a trait over time" (Specht et al., 2011, pag. 863), while "rank order consistency reflects whether groups of people maintain their relative placement to each other on trait dimensions over time" (Specht et al., 2011, pag. 863). By adopting these two notions of stability, the literature has found that the most evident changes occur during adolescence and old age, while the degree of consistency is much higher during middle age (Specht et al., 2013;Lucas and Donnellan, 2011;Borghans et al., 2008;Fraley and Roberts, 2005;Caspi and Roberts, 2001;Roberts and DelVecchio, 2000). effects on consumption choices can be generalized without the concern that they depend on 13 the year of the data available and, in particular, that the year investigated in this study was characterized by negative economic trends. Third, we are able to compare the effect of personality traits on consumption for individuals who are potentially in different phases of their life cycle, i.e., when they are singles, in couples and in couples with children, without the concern that the change in status may affect the individual personality. The sample of couples confirms both the empirical evidence of positive assortative mating for the Mental Openness and the Conscientiousness traits (see e.g. Flinn et al., 2018;Lundberg, 2012) and the power of these two traits of predicting individuals' education level and wage.
Education and wage are the two individual characteristics traditionally considered for positive assortative mating. To obtain a closer idea of the correlations between the personality traits, as well as education, of husband and wife, Table 2 reports the statistics of an association index, , where x is a specific trait, and its standard deviation is computed with the household as a reference. When husband and wife have exactly the same trait, the standard deviation is 0 and the association index is 1; when the trait takes the two extreme values, 1 and 7, the association index takes the value 0.19. The highest association index, on average, is observed for the Conscientiousness trait, at approximately 0.7, followed by the Mental Openness and the Agreeableness traits, while the lowest is observed for the Neuroticism trait. A similar picture emerges in examining the correlation coefficients of husband and wife personality traits, summarized by Table 3, which also highlights the association between different traits, such as the Conscientiousness trait and the Agreeableness trait. It also shows that overall, there is a closer association in education than in personality traits, while the income association is quite low. 9 Figure 4 visually confirms that a certain level of assortative mating is observable for two personality traits, Mental Openness (as in Flinn et al., 2018, andLundberg, 2012) and Conscientiousness (as in Lundberg, 2012).  income from work and pension rather than wage rates, as we do not limit our sample to working-age individuals.

Other individual and family characteristics
The control variables, used in all specifications, are age and its square, being a part-time worker, being a full-time worker, being retired, and being an immigrant. For the samples of couples with and without children, all of these variables are provided for both the husband and the wife when relevant. Other household characteristics include the property status of the dwelling (1 if owner), whether the household is in a urban settlement, whether it is in East Germany, and the number of children.
An important aspect is the age selections in each sample. In particular, the samples of single and childless couples are composed mainly of relatively elderly people, although both contain a cluster of young people. For this reason, in each analysis, it is important to control for the age of the respondents and additionally for the retirement status. The sample of couples with children aged 16 or below is less problematic from this point of view, but we still control for age.
In more detail, the sample of singles has a different composition when separated by gender: the majority of female singles are retired (57%), and only 37% of them work, either full-time (22%) or part-time (15%). The figures are different for single males, of whom 47% work full-time and 12% part-time, whereas 32% are retired. The difference in average age, about 10 years, is also quite relevant, leading to a significant difference in the education level, which is higher for males by 0.6 years. The proportion of immigrants is similar for males and females at about 5%. Given such a heterogeneous composition of the sample of singles, we decided not only to analyze consumption using the two samples of male singles and female singles but also to perform a specific robustness check by separating in the estimations those who are less than 60 years old from those who are more than 60 years old.
The sample of couples is also quite heterogeneous when those with and without children are considered separately: in the sample of childless couples, the average age of husbands and wives is higher by almost 22 years than that of the couples with children; the husband works full time in 82% of families with children, versus 34% of childless families, and the wife is less likely to work full-time when there are children, 16% vs 25%. In contrast, in families with children, the wife is much more likely to work part-time, 57% vs 21%, and as expected, the proportion of retired individuals is negligible, below 2% for both husbands and wives, versus 58 and 46%, respectively.
In the sample of couples with children, being younger, both husband and wife have a higher level of education, with little difference between them. In contrast, in the sample of childless couples, a relevant difference exists between the education level of the spouses (about 0.6 years). Among couples with children, both parents are much more likely to be immigrants (about 12-13% vs about 7%). The sample of couples with children aged 16 or below has on average 1.8 children, 49% of whom are girls.

Engel curves specification
The use of Engel curves to study consumption when no price information is available dates back to Engel (1857), and they have been applied countless times in economics since then. In this study, we apply the quadratic budget share specification of Engel curves proposed by Banks et al. (1997), incorporating observed heterogeneity as linear demographic translating functions (Lewbel, 1985;Pollak and Wales, 1981;Gorman, 1976). 10 With this specification, we estimate both the probability of consuming different commodities and their share of total expenditure on consumption. We thus specify two different consumption models: where o ij and w ij are the probability of non zero consumption and the budget share of the j th commodity category for individual i, respectively; coefficients β Given the possible concern about endogeneity of total household expenditure, instead of the standard instrumental variables method, which in nonlinear models is biased and inconsistent (Terza et al., 2008), the control function approach is used for both equation (1) and (2). The control function approach is a two-step procedure: in the first stage, total expenditure is regressed on the same set of covariates as in the main model plus the exclusion restrictions. As established in the literature, as exclusion restrictions, we use individual income from work and pension and the first three components of a principal component analysis on ownership of a number of items (detailed in the previous section), which can be interpreted as composite indices for household wealth. The first-stage equation can be formalized as follows: In the second stage, the prediction of the idiosyncratic component of the first stageˆ ij is included as an additional regressor in equations (1) and (2) .
Equations (3) are estimated using a logit model with indepedent equations for each commodity; however, the Engel curves system specified in equation (4) is estimated by means of seemingly unrelated regressions, allowing for correlation of the error terms.
While the parameters λ j of the translating functions can be straightforwardly interpreted as consumption shifters or preference parameters (Lewbel, 1985), 11 the information contained in the total expenditure parameters of the Engel curves can be effectively synthesized by computing income elasticities, which return the percentage increase in expenditure on commodity j if total expenditure y increases by 1%. Formally, the income elasticities are defined as follows: where e j is expenditure on commodity j, and the structure of equation (2) permits translating the coefficients into income elasticities as follows:

Results and discussion
In the Engel curves estimation, standardized personality traits (and education level) are included as shifting parameters, allowing us to capture how these individuals' characteristics may change the level of the budget share devoted to each category of expenditure (see Equation 4). To make the comparison of the effects of the same trait across different household compositions easier, the coefficients of each trait are shown in a single table with household types by column. In particular, Table 4 reports the R 2 goodness of fit statistics, and tables from 5 to 10 show the coefficients of the Big Five personality traits and education level obtained separately from the estimations of the system of Engel curves in equation 4 for the samples of single men, single women, women and men in childless couples, and women and men in couples with children. 12 In  For childless couples, personality traits are particularly relevant to explain clothing and leisure expenditures (with an increase of more than 25%), while for couples with children, they improve the R 2 by far more than 15% for all consumption categories except food at home, telecommunications and vacations. Particularly large R 2 increases are found for clothing (96%), culture (almost 49%) and transportation (almost 62%). These results suggest that accounting for personality in the consumption analysis captures a larger portion of consumers' preferences than more standard approaches do.
Regarding the coefficients of the Big Five personality traits and education level, a first interesting result pertains to the individuals who score high in the Mental Openness trait, i.e., those  20 who are more curious, intelligent, and unconventional, who show a higher share of budget devoted to expenditure on culture (see Table 5). This finding is verified for single men and single women, for both the male and the female component of childless couples, and for the female partners of couples with children. In the case of male partners in childless couples, the coefficient is also positive and statistically significant for expenditure on education. An examination of the impact of the Mental Openness trait on the probability of non-zero consumption, as expressed by equation (3), reported in Appendix C, Table C.3, indicates that this effect is even stronger because it includes a positive and statistically significant coefficient not only for culture but also for a companion type of expenditure, i.e., on education, which is verified for all types of household in which the individual can potentially live.
In terms of magnitude of effect, expenditure for culture has a relatively small share of the budget, constituting between 1 and 1.5% of the total expenditure of the households analyzed here (see Tables 1 and 2 Regarding the interpretation, we should first consider that the Mental Openness trait has one of the higher correlation coefficients in the couple, i.e., is one of the traits with higher positive assortative mating (see Table 3). Then, we can argue that agreement between the partners on using household resources for consumption related to culture is highly probable, or we may expect that the joint consumption of these goods is a potential candidate for explaining the marital surplus of couples in which both partners share this trait of personality. This interpretation is reinforced by examining the Mental Openness trait score average in the couple instead of the personality score of each partner (see Table D.2). In fact, when the average score is considered, the coefficients are statistically significant for both culture and education items and for both couples with children and childless couples. The effect is, instead, not statistically significant when the association index is considered. In other words, the effect is not verified when both partners have the same trait, low or high, in comparison with partners who score differently in the trait; in addition, the effect is statistically significant when both are high in the trait in comparison to partners who both score low in it.
In examining the Mental Openness trait, a final remark emerges from our analysis regarding the very similar effects on consumption of education level (see Table 6). The coefficients of the education level variable for both the culture and the education shares of the budget are positive and statistically significant for both male and female singles, and they persist when childless couples are considered. This finding is confirmed in examining the average of the trait in childless couples (see Table D.1). Similar evidence emerges if we examine the probability of nonzero consumption (see Table C.1), with marginal effects similar to those of the Mental Openness trait (see Tables

C.2 and C.2).
Our interpretation begins again by considering the correlation coefficient of male and female education levels in the couple, which is the highest among those considered, with the Mental Openness trait being second in magnitude (see Table 3). Furthermore, as we have underlined in Section 2.1, the Mental Openness trait and the Conscientiousness trait are the two individual features that best predict an individual's educational outcomes. From these premises, two contrasting hypotheses emerge: the first is that the positive assortative mating on the Mental Openness trait, and then its predictive power for expenditure on education and culture, can be considered a type of prior effect with respect to education level. In other words, two individuals who score high in the Mental Openness trait are predicted to have a high education level, or to choose a demanding school career; then, they will have a high probability of becoming partners in the marriage market and devoting relatively higher shares of their budget to consuming items classified as culture or education. The contrasting hypothesis is that the Mental Openness trait generates positive assortative mating and also predicts preferences for expenditure on education and culture, but the latter effect is not necessarily mediated by the level of education. To test these two hypotheses, as a robustness check, we separate singles with a secondary education from those who do not have this level and also separate households in which least one spouse has a secondary education from those in which neither spouse has a secondary education. As illustrated in the next section, this check supports the second hypothesis, i.e., that the effect of the psychological trait is independent from that of the education level (see Table F.1 and F.2). Table 7 shows the coefficients of the Conscientiousness trait estimated for the four systems of Engels curves for single men, single women, and couples with and without children. In comparison with the Mental Openness trait, the Conscientiousness trait affects consumption in a more genderspecific way. For instance, there is a negative effect on expenditure on culture, but only for single men. Regarding the probability of consuming, highly conscientious single men and women are predicted to have a lower probability of consuming culture and vacations, but the effect tends to disappear in couples except for women in childless couples (see Table C.5). In addition, in contrast to our observations for the Mental Openness trait, in this case, the effect on consumption is imputable to the association of partners instead of to the average trait in the couple (see the results for childless couples in Table D.3). Considering that the Conscientiousness trait is the third most relevant in terms of positive assortative mating, after the education level and the Mental Openness trait (see Table 3), we may interpret this finding as another example of similarities in consumption that can potentially generate a marital surplus. Of course, in this case, the marital surplus is generated by the agreement between the partners in reducing the consumption of these items rather than by increased consumption or a public good joint consumption.
Another gender-specific effect of the Conscientiousness trait is detectable in examining the shares of the budget devoted to food at home in comparison with the shares of the budget for the consumption of food out of the home. The share of the budget for food at home is higher both for the most conscientious single men and for the male partners of couples with children (see   Table 7). In contrast, we verified that for women -and in particular for those who are singlethe Conscientiousness trait has a negative and statistically significant coefficient for food out of the home. In the latter case, the sign is confirmed by the probit estimation of the nonzero probability of consuming (see Table C.5). In terms of magnitude, consider that the share of the budget of a single women devoted to this item is 5.7%; thus, an increase of one standard deviation in her Conscientiousness trait reduces consumption of this item by 1.3% (see Table   C.6). When individuals must exert self-discipline in reducing consumption, women moderate their expenditures for food out of the home, while men increase their propensity to consume food at home when they are either single or in a family with children. Regarding the latter outcome, in couples with children, the male preference seems to be so strong as to remain statistically significant when the couple average score is taken into account (see Table D.3).
Expenditure on food at home is also affected by the Extraversion trait. Men and women who are high in this trait, and thus are forthcoming and desire social relationships, are relatively more prone to devoting high shares of their budget to food at home, probably because they invite their friends to their home (see Table 8). For women, this interpretation is reinforced by the effect of the Agreeableness trait, which positively predicts a higher budget share of this item if they are single or in households with children (see Table 9). The aforementioned interpretation is also applicable to explain the positive and statistically significant effect in terms of the probability of the nonzero consumption of food at home for the sample of childless couples when both score high in the Extraversion trait (see Table C.7). The interplay between the Extraversion and Agreeableness traits in shaping the consumption of food -at home vs. out of the home-is also detectable in the predictive power of the average of the two traits in childless couples: the former predicts a higher share of the budget for dining out, while the latter predicts a lower share for this kind of consumption choice (see Table D.5 and Table D.4).
Finally, the Neuroticism trait plays a role in increasing both the shares of the expenditure and the probability of nonzero consumption of health quite consistently across gender and household composition (see Table 10 and Table C.11); this finding is confirmed in examining the average trait in childless couples (see Table D .6). As this trait is a measure of an individual's tendency to be emotionally unstable and inclined to worry, it is easily understandable that this feature leads to increased expenditures on health. show large differences in elasticities for men and women or according to household type. For example, food out of the home has an elasticity greater than one for two samples: single women and couples with children. In these cases, an increase of 1% in income returns an increase in expenditure on food out of the home of more than 1%. The same is true for the expenditure on clothing and shoes: the elasticities are greater than one for women and couples with children.
Single men show an elasticity greater than one for health and telecommunications, with the former also being greater than one for couples with children. Furthermore, there is a group of items, we find a plausible effect of the transition from one-person households to couples, as the decrease in the magnitude of the elasticity is an expected result owing to the couple formation being a first way to be insured.

Robustness and heterogeneity analysis
In order to consider the heterogeneous composition in terms of age, whether in the samples of singles or of childless couples, as a first robustness check, we separate in the estimations those who are less than 60 years old from those who are more than 60 years old. The positive effect of the Mental Openness trait on the share of budget devoted to culture still holds, whether for singles or childless couples (see Table E.1 and Table E.4). In contrast, the negative effect of the Conscientiousness trait on expenditure on culture is confirmed only for young singles and is not true for young childless couples or elderly singles and childless couples (see Table E.2 and Table   E.5). A confirmed outcome, independent of the selection by age of the estimation sample, is the positive effect of the Neuroticism trait on expenditures on health (see Table E expenditure on culture is driven mainly by individuals with a relatively high level of education or whether it is also detectable for individuals with a lower level of education. As Tables F.1 and F.2 show, the predictive power of this personality trait is confirmed for all typologies of households independent of the level of education, as it is also true for the sample of individuals with a lower level of education. 14 These results confirm the importance of accounting for personality traits when analyzing consumption choices: their contribution to explaining consumption preference does not stem uniquely from their association with educational achievements. Rather, the similar impact of personality on the consumption of individuals with higher and lower levels of education suggests that the education level is a poor proxy of personality, and more specifically of the Mental Openness trait, in consumption studies.

Conclusions
In this study, we have analyzed the role played by personality traits in consumption decisions for both individuals and households. Our investigation is based on the estimation of a system of Engel curves and probit equations for single men, single women, men and women in childless couples and men and women in couples with children, with each group considered separately. We show that personality traits are relevant predictors of household expenditure and participation in consumption of several categories of goods and services. The inclusion of personality traits among demographic shifters helps to reduce unobserved heterogeneity and improves the goodness of fit of the Engel curves specification. These results suggest that accounting for personality in consumption analysis captures a larger portion of consumers' preferences than more standard approaches do.
A key contribution of the paper is the test of which traits are consistent between the consumption choices of people living alone and in couples. Given this kind of evidence and taking into account assortative mating on personality traits, we are able to characterize the marital surplus based on joint consumption or on the agreement to reduce/increase the share of the budget devoted to certain items. For example, we find that the expenditure on culture (and its probability of nonzero consumption) is higher for men and women who score high in the Mental Openness trait. This is true for men and women not only when living alone but also when living as a couple without children. Furthermore, this finding remains true for couples with children, even if only in terms of consumption probability and not of the budget share. Considering that the Mental Openness trait has a large correlation coefficient in the couple (it is the personality trait with the highest positive assortative mating), we conclude that the joint consumption of culture is a potential candidate to explain the marital surplus in couples where both partners score high in the Mental Openness trait. A robustness analysis allows us to show that this effect is not entirely due to the level of education of the partners: by separating the estimations for couples 14 Estimation outcomes for the other traits are available upon request.
with at least one person with a secondary education from those in which neither spouse has this level, we find that the predictive power of personality is independent of the education level of the partners. We therefore conclude that the Mental Openness trait not only is a sort of antecedent of the education level but also generates positive assortative mating and increases consumption expenditure on culture independently of the education level of the partners.
Furthermore, we show that the probability of spending money on leisure time, vacations and culture is lower for individuals who score high in the Conscientiousness trait. As this behavior is common among singles -men and women-and in both husband and wife of couples with children, and considering that the Conscientiousness trait is the second most relevant in terms of positive assortative mating, we interpret this finding as another example of joint consumption that can generate a marital surplus. In this case, the agreement between the partners does not imply an increase in the probability of consuming certain items; instead, it represents a decrease in the resources devoted to these items. All of the income elasticities are greater than 1, which is evidence that this group of items is particularly sensible to income fluctuations to which conscientious individuals are adverse.
Regarding the share of food, which is the most relevant in terms of size, we find statistically significant effects for two personality traits, Conscientiousness and Extraversion. Conscientious single men, as well as the male partners in couples with children, tend to spend more on food at home, as do men and women who are high in the Extraversion trait. The most conscientious women -in particular those who are singles-tend to spend less for food out of the home.
We conclude with a caveat: our analysis is to some extent limited by the simplified model of household decision making adopted for couples. In fact, we assume that households are maximizing a single utility function because we are unable to explicitly differentiate partners' preferences within a couple using individual utility functions or to model any kind of strategic interaction between the two. 15 Nonetheless, we assume that the individual personality traits, as well as other individual characteristics, of both partners can have different impacts on household utility. Consequently, we are able to test whether two partners who score high -or low-in a certain trait are predicted to allocate the household budget in a certain way, and whether this budget allocation is similar to that of women and men who score similarly and live alone. In the affirmative case, we infer that personality affects consumption similarly when individuals are single or live in a couple. We cannot infer much more because i) in our sample, people are not the same individuals changing their status from single to married, and ii) we do not account for partners' individual utility functions in modeling the household decision-making process. Given the relevance of the household decision-making process in the economics literature, as well as the usefulness of personality traits for understanding household labor supply (as shown e.g. by Flinn et al., 2018) and consumption decisions, further insights into this issue remain for future research. Below are different qualities that a person can have.
You will probably find that some apply to you perfectly and that some do not apply to you at all. With others, you may be somewhere in between.
Please answer according to the following scale: 1 means "does not apply to me at all", 7 means "applies to me perfectly".
With values between 1 and 7, you can express where you lie between these two extremes.. An example of a less clear effect is that of the Neuroticism trait on the probability of spending on insurance: the coefficient for single women is statistically significant and negative, and that for single men is not statistically significant, while among couples, we find a positive and statistically significant coefficient for husbands in childless couples and wives in couples with children. The expected result, i.e., that people who are inclined to worry are more prone to buy insurance, is not always verified.               Note: estimation also include the following control variables, not reported in the table: age, age-squared, being immigrant, working part-time, working full-time, being retired, being owner of the dwelling, living in an urban settlement and living in East-Germany. Full estimation tables are available upon request. *** p < 0.01, ** p < 0.05, * p < 0.1 G First-stage estimations for total expenditure endogeneity