Personality and positionality-evidence from survey experiments with alternative goods

Abstract

This paper employs survey experiments to examine the relationship between personality characteristics and positional concerns across a wide range of “goods”, e.g., income and market value of a car, and “bads”, e.g., infant mortality and poverty rates. Personality traits are measured using the five-factor model (Big-5), the locus of control, and reciprocity. We demonstrate that there are significant relationships between personality types and positional concerns, which differ both by the type of personality and by the nature of a good. The results are highly consistent with the predictions presented in the field of personality psychology. That is, while agreeableness is negatively associated, conscientiousness, neuroticism, and external locus of control are positively associated with positional concerns for most goods. Importantly, there is also a substantial heterogeneity in the mean degree of positional concerns across the low and high values of most personality characteristics and goods.

1 Introduction

Research in social psychology and more recently in mainstream economics has yielded a bulk of evidence that personality characteristics are significantly correlated with a wide range of individual preferences and outcomes (e.g., Barrick and Mount 1991; McCrae and Costa 1999; Borghans et al. 2008). Economists seem to agree that the dimensions of personality can be used as a proxy for the innate abilities, or “non-cognitive” skills, e.g., memory, empathy, attention, imagination, and social skills (Heckman et al. 2006; Almlund et al. 2011). Non-cognitive skills have been shown to be correlated with outcomes including productivity, earnings, and wages (e.g., Osborne 2005; Nyhus and Pons 2005; Heckman et al. 2006; Mueller and Plug 2006; Heineck and Anger 2010; Uysal and Pohlmeier 2011), job performance and job satisfaction (e.g., Barrick and Mount 1991; Tett et al. 1991; Van den Berg and Feij 1993; Judge et al. 2002), workplace deviance (e.g., Salgado 2002), leadership and academic/career success (e.g., Judge et al. 1999, 2002; Frolova and Mahmood 2019), unemployment dynamics (e.g., Cuesta and Budria 2012), well-being (e.g., DeNeve and Cooper 1998; Shuleska and Sterjadovska 2019) as well as several preference parameters, i.e., time preferences, risk aversion, altruism, and social preferences, which play crucial roles in economics (Borghans et al. 2008; Becker et al. 2012).

One important issue is how personality characteristics are related to positional concerns (“status” or “relative” concerns), i.e., that people’s utility is affected not only by their own level of consumption but also by their consumption level relative to that of comparable others (e.g., Solnick and Hemenway 2005; Ferrer-i-Carbonell 2005; Clark et al. 2008; Carlsson et al. 2007; Akay et al. 2013, 2019). The literature on positional concerns has been growing and several studies demonstrate that these concerns are strongly related to people’s utility in a varying degree across types of goods (Solnick and Hemenway 2005), levels of economic development and growth (Easterlin 1995; Senik 2004), and people’s observed characteristics (Akay and Martinsson 2018).¹ The literature has already identified important implications of these concerns in relation to a wide range of economic issues spanning from labor supply (Neumark and Postlewaite 1998) to migration (Akay et al. 2017) and how to optimize taxation to alleviate the negative impact of positional concerns on people’s utility (Aronsson and Johansson-Stenmann 2014).

However, how positional concerns are related to people’s “hard-wired” characteristics such as personality traits remains largely unknown (Cuesta and Budria 2015; Budria and Ferrer-i-Carbonell 2018; Friehe et al. 2018). Recently, using panels of subjectively reported well-being datasets and fixed-effects model specifications, Budria and Ferrer-i-Carbonell (2018) and Cuesta and Budria (2015) report that the influence of relative income position on life-satisfaction varies greatly depending on the respective person’s specific personality traits. Friehe et al. (2018) report a substantial heterogeneity of personality on income comparisons which also varies with respect to reference groups. The current paper contributes to this limited literature in two aspects: First, we investigate the relationship between personality characteristics and positional concerns using tailored stated choice experiments (Solnick and Hemenway 2005; Hillesheim and Mechtel 2013). One important advantage of this approach is that it identifies the direct relationship between personality and positional behavior, which might not be easy to capture with the subjective well-being data used in previous literature (Budria and Ferrer-i-Carbonell 2018). Second, we investigate the relationship between personality and positionality not only for “income” but also for comparisons with respect to a wide range of goods.

To ensure comparability with the existing literature, first we choose income/month after tax as the baseline good and use the Big-5 personality traits (extraversion, agreeableness, conscientiousness, neuroticism, openness-to-experience), internal/external locus of control (ILOC and ELOC hereafter), and positive/negative reciprocity (PR and NR hereafter) as personality characteristics (e.g., Budria and Ferrer-i-Carbonell 2018).² Second, our stated choice experiment is then extended to allow an alternative set of goods that are also expected to be consumed in different levels by personality types. We experiment on the market value of a luxury car, which might be considered as an “inherently” positional good. We then experiment with a public good using overall health expenditures in a country (billions TRY). In the paper, we refer to these first three items as “goods” as they involve positive and desirable elements. The experiment is also extended to include “bads” involving disutility and undesirable aspects including an item relating to leisure-work balance, working hours/week, and two public “bads” as infant mortality rates (per 1000), and poverty rates (%).³

We present highly robust findings that are consistent with the existing literature using both life-satisfaction (e.g., Budria and Ferrer-i-Carbonell 2018) and experiment data (e.g., Van der Zee et al. 1996, 1998). First of all, most personality characteristics are significantly related to positional concerns. Among Big-5 personality characteristics, we find strong evidence that agreeableness, conscientiousness, and neuroticism are significantly related to positionality. More specifically, agreeableness is negatively related to positionality while conscientiousness and neuroticism are positively related. There is a significant negative relationship between extraversion and positional concerns only for “bads”. There is no particularly strong relationship between openness-to-experience and positional concerns. While ILOC is weakly related to positionality, a significant positive relationship is found for ELOC, which is also highly consistent with the literature. We find a negative and significant relationship between PN and positional concerns only for the “bads”. We also conduct an extensive heterogeneity analysis that involves both parametric and non-parametric methods to unveil how positional concerns vary across levels of personality characteristics. One important result is that positional concerns are statistically significantly different from the overall mean level of positionality mostly among people who score lower on each personality trait. For people with medium-to-high levels of most personality characteristics, the degree of positionality is about the same as the overall mean level of positionality.

The remainder of the paper is organized as follows. The next section presents the experiment setup, personality measures and hypotheses, and gives a summary of the experiment. Section 3 explains the econometric approaches and estimators used. Section 4 presents the main results of the experiment, a robustness analysis, and an extensive heterogeneity analysis based on parametric and non-parametric models. Finally, Sect. 5 concludes the paper.

2 The survey experiment

The survey experiment was conducted in a large lecture hall at Istanbul University in 2014 using 307 undergraduate students recruited from economics, psychology, and law departments.⁴ The experiment was explained by experimental assistants at the beginning of the session. The experiment consisted of two parts. First, the respondents answered a series of hypothetical stated choice questions. We used these responses to identify their degree of positional concerns. In the second part, the respondents were given a follow-up questionnaire, the responses to which we used to elicit several individual socio-demographic and socio-economic characteristics, personality measures as well as other data concerning the respondents’ attitudes, emotions, and feelings. At the end of the experiment, the participants received a pocket book, the value of which equaled the average labor market value of the time spent completing the experiment.

2.1 Identification of positionality

2.1.1 Preferences

The main objective of the experiment is to identify how degree of positional concern is related to personality characteristics for alternative sets of goods. To be able to measure the degree of positional concern, we specify a utility function $U^{g} (Y^{g} ,Y^{g} - Y^{gR} )$ that is additive in relative $Y^{g} - Y^{gR}$ and absolute $Y^{g}$ consumption levels of each good $g$ (e.g., Akerlof 1997; Knell 1999):

$$U^{g} (Y^{g} ,Y^{g} - Y^{gR} ) = (1 - \theta^{g} )Y^{g} - \theta^{g} (Y^{g} - Y^{gR} ),$$

(1)

where $\theta^{g}$ is the parameter measuring the marginal degree of positional concern, i.e., the fraction of the marginal utility that is due to the increase in relative consumption of good $g$. Equation (1) assumes that people compare their consumption level of a good with a reference (or comparison) consumption level $Y^{gR}$ of their reference group, e.g., others in their society. In order to investigate the relationship between personality and positional concern, an individual-specific measure for the marginal degree of positional concerns $\theta_{i}^{g}$ should be identified. To this end, we tailored a survey experiment involving binary choices between two societies that differed in terms of absolute and relative consumption levels. The levels of consumption of each good were specifically selected from the utility function (1) for a set of degree positionality $\theta^{g}$ for each good $g$.

2.1.2 Setup

In the first step of the experiment, the respondents were asked to make a series of decisions regarding where their imaginary grandchild should live given in a certain hypothetical scenario (Carlsson and Martinsson 2001; Johansson-Stenman et al. 2002; Akay et al. 2013).⁵ More specifically, the respondents were asked to choose one of two societies for their future grandchild, i.e., Society (A) or Society (B), which only differ in terms of their per capita absolute and relative amount of consumption of each good. The scenario and an example choice situation (in Fig. 1) were carefully described to the respondents just before the experiment started. The respondents were specifically informed that the prices and consumption levels of all other goods were identical in the two hypothetical societies (see Fig. 1 for details).

Fig. 1

The scenario and example choice situation

Having the scenario presented, an example choice situation, which is very similar to choices that respondents are going to make, is carefully described to respondents by experimental assistants. Figure 1 presents the choice situation used as an example which reads as follows: In Society (A), the respondent’s hypothetical future grandchild’s monthly after-tax income is 2000 TRY, while the average monthly after-tax income amounts to 2500 TRY.⁶ In Society (B), the future grandchild’s monthly after-tax income is 1800 TRY, while the average monthly after-tax income in this society is 1500 TRY. Thus, the future grandchild would be 200 TRY/month richer in Society (A) than in Society (B) in absolute terms, yet would earn $2500 - 2000 = 500$ TRY less per month than the average income earned by other people in society. That is, despite having less money, the future grandchild would be better off in relative terms in Society (B), as the grandchild would enjoy $1800 - 1500 = 300$ TRY more in monthly income than the average person in that society. In effect, by choosing Society (B), respondents state a particular degree of positional concern which is implicit in the assigned income levels.

In our survey experiment, respondents are asked to make repeated binary choices that vary in implicit degree of positional concerns. What we mean by “implicit” is that we select specific consumption levels of each good so that the distance between absolute and relative amounts in two societies corresponds to a particular degree of positionality $\theta$ in each choice situation.⁷ The logic of the design can be described as follows: If a respondent is indifferent between living in Society (A) and Society (B), then the respondent implies that $U^{g} (Y_{A}^{g} ,Y_{A}^{g} - Y_{A}^{gR} ) = U^{g} (Y_{B}^{g} ,Y_{B}^{g} - Y_{B}^{gR} )$ for a good $g$ (after-tax income/month in this case). In the case of additive utility function (1), we can write $Y_{A}^{g} - \theta^{g} Y_{A}^{gR} = Y_{B}^{g} - \theta^{g} Y_{B}^{gR}$ and the marginal degree of positional concerns is $\theta^{g} = Y_{A}^{g} - Y_{B}^{g} /Y_{A}^{gR} - Y_{B}^{gR}$. In the example choice situation, if the respondent chooses to live in Society (A), then the marginal degree of positional concern is $\theta^{g} = (2000 - 1800)/(2500 - 1500) = 0.20$ implying that the respondent’s degree of positionality is at least 0.20 ($\theta^{g} > 0.20)$ for after-tax income/month. The design cannot identify the exact degree of positional concerns of respondents. We can only observe the level of positionality in intervals by asking repeated questions with varying implicit degree of positional concerns. The respondents were asked to answer three sequential questions for each good and their answers provide information for the upper and lower bound of positionality in intervals $\theta^{g} \le 0.25$, $0.25 \le \theta^{g} \le 0.50$, $0.50 \le \theta^{g} \le 0.75$, and $\theta^{g} > 0.75$, which are implicit in the consumption levels assigned for each good (see “Appendix A” for three choice situations for after-tax income/month). Using the stated preferences of respondents for each good, we then aim to estimate the mean level of $\theta^{g}$ for each good $g$ conditional on the personality characteristics of respondents.

2.1.3 Summary of the experiment

The experiment is summarized in Table 1, where we present own and others’ consumption levels for six goods and the share of respondents who chose the positional alternative for each choice situation. The first and second columns show the imaginary grandchild’s level of consumption and the consumption level of other people, respectively. For each good, we determined a Society (A) and three (B) societies, i.e., B (1), B(2), and B(3), for which the consumption levels were chosen to correspond to an increasing implicit marginal degree of positionality (0.25, 0.50, and 0.75) for each good (third column).⁸

Table 1 Summary of the experiment

The raw shares of respondents who chose the positional alternative in the respective choice situations are given in the fourth column. The overall raw mean shares of respondents who chose Society (B) across three choice situations for each good are given in bold fonts. The raw mean shares of positional choice are heterogeneous across goods. The share is 0.52 for income, and the respondents show the lowest positional concerns for working hours/week, at 0.39, and the highest for the market value of a car, at 0.56. The shares of respondents who chose the positional alternative fall when the implicit degree of positionality is higher, as expected. The mean share of positional choice tends to be lower (0.42) for the “bads” involving disutility (working hours/week) and suffering [infant mortality rates (per 1000) and poverty rates (%)] than for the “goods” (0.55) (after tax income/month, market value of a car, and overall health expenditures). Mann–Whitney U test suggests that the difference is highly statistically significant (p value < 0.001). The overall raw mean share of positional choice across all choice situations and goods is 0.48. This value is highly similar to the findings in previous studies using similar samples and methods (Johansson-Stenman et al. 2002; Alpizar et al. 2005; Carlsson et al. 2007; Akay et al. 2013).⁹

One important remark is that we assume our respondents are utility-maximizing agents and as such should comply with the standard assumptions of utility-maximizing behavior. However, in reality, some respondents may act inconsistently by switching back and forth between societies and violates the transitivity assumption, e.g., choosing Society (A) at the first-choice situation and then choosing Society (B) in the second- or third-choice situation. We identify all inconsistent respondents for each good and then calculate their share out of all respondents. The results presented in the second to last column of Table 1 suggest that the shares of respondents who responded inconsistently are similar across the goods and vary from 7.5% (infant mortality rate) to 12.6% (market value of a car). We exclude these respondents from the sample in our empirical analysis. The final sample size is given in the last column of Table 1 for each good.

2.2 Measures of personality characteristics and hypotheses

2.2.1 The Big-5 measure of personality

The model is one of the widely used taxonomies in the psychology and recently in economics, as it offers an integrative framework to understand individual differences in broader categories (for a comprehensive review see McCrae and Costa 1997). The measure is based on 15 subjective questions, which are presented in Table 5 (in “Appendix B”). The Big-5 personality measure identifies personality in five categories labeled as extraversion, agreeableness, conscientiousness, openness-to-experience, and neuroticism (or emotional stability when reversed). Respondents are asked to provide a response to each statement on a 7-point scale where 1 means “it does not apply at all” and 7 “it applies fully”. In a next step, Big-5 personality measures are calculated by adding the responses for the three questions for each personality category and thus, each of the five measures obtained ranges from 3 to 21 (please see Table 5 for the calculation of Big-5 personality characteristics). Each individual has a specific combination of the five mutually non-exhaustive characteristics, and each Big-5 personality trait is expected to be associated with specific attitudes and types of behavior.

A priori predictions about how Big-5 personality types relate to positional behavior might be difficult and the degree of correlation might also differ across the levels of each personality type and also on the type of good under consideration. Using information from the field of personality psychology, we suggest a series of hypotheses regarding the relationship between personality and positional concerns as follows. A high score on extraversion is associated with a higher frequency of social interactions and ambition, which in turn may be correlated with a higher positional concern. Indeed, evidence shows that extraverts value their status at work, for instance (Barrick et al. 2002; Landis and Gladstone 2017). Thus, the relationship between extraversion and positional behavior is expected to be positive. However, we note that these individuals are also friendly, seeking friendship, and sympathetic with others’ gains and losses, which might correlate negatively with positional concerns. That is, while extraverts might enjoy a status reward of having a higher position, they might also be sympathetic to the circumstances of comparable others, e.g., friends. Thus, the relationship between extraversion and positional concerns might be masked depending on the weights of these counteracting dimensions of extraversion.

A high score on agreeableness is associated with cooperative behavior and altruism, which might be associated with weaker positional concerns (McCrae and Costa 1999). These individuals are highly motivated for work and tend to avoid conflict with others and display less workplace deviation, which might also be related to lower levels of positional concerns, especially with respect to working hours (Salgado 2002; Graziano and Tobin 2002). Yet at the lower end of agreeableness, the positional behavior might look substantially different. Individuals scoring low on agreeableness might exhibit “spiteful” behavior and may tend to punish others if they perceive their status as a threat, which in turn may go hand in hand with stronger positional concerns (Marcus et al. 2014). Conscientiousness relates to hard work, work ethnics, self-discipline, competition, and a higher degree of goal-oriented behavior. A higher score on this personality relates to welfare positively and these individuals are more reactive to changes in their income and the behavior of others (Boyce and Wood 2011). Thus, a high level of conscientiousness might be associated with a higher level of positional concerns.

Neuroticism is associated with negative emotions, depression, and lower levels of life satisfaction. A higher score is expected to be positively correlated with positionality as these individuals seek (mostly downward) comparisons to enhance their mood (Van der Zee et al. 1996, 1998; Buunk et al. 2002). Finally, openness-to-experience is associated with diverse social interactions, valuing arts, and a tendency to socialize with people who represent different values and cultures. While a higher score on the openness-to-experience might be positively related to positional concern, as it would in the case of extraversion, this personality trait might lead to less positional behavior if these individuals use the information gained from the experiences and interaction with others to improve their status in the future (as in the case of “tunnel effect” of Hirschman and Rothschild 1973). Thus, the direction of the relationship is a priori unknown.

2.2.2 Internal and external locus of control

This model of personality aims to measure internal and external aspects of the degree of autonomy in the life decisions and actions of individuals (Rotter 1966). That is, it measures the degree to which individuals believe that they can control their own life decisions and the outcomes they lead to. There are two dimensions called internal and external locus of control, i.e., ILOC and ELOC. ILOC is a measure of the belief that an individual has control over one’s decisions and their outcomes. ELOC is a measure of how much individuals believe that external forces, e.g., other people, society, luck, or fate, determine their life outcomes. To measure ILOC and ELOC, we use Rotter’s (1966) inventory, which is based on the ten questions, or statements, presented in Table 6 (in “Appendix B”). Respondents are expected to respond to each question on a 7-point scale (1 = “it does not apply at all” and 7 = “it applies fully”). While ILOC ranges from 3 to 21, ELOC takes values ranged from 7 to 49.

There is substantial evidence that a higher ILOC score is positively correlated with career success, job performance, and income (e.g., Caliendo et al. 2014; Buddelmeyer and Powdthavee 2016; Andrisani 1981; Strauser et al. 2002; McGee and McGee 2016; Heineck and Anger 2010; Piatek and Pinger 2015; and see Cobb-Clark 2015 for a comprehensive discussion).¹⁰ While these individuals are expected to display weaker positional concerns, as they are more successful in general and do not depend on others’ behavior, they might also experience higher positional concerns as they regularly set higher life goals to challenge them (Cobb-Clark 2015). For instance, it is possible that higher consumption levels among others might be a pushing factor for people with higher ILOC scores. Thus, no a priori prediction can be made regarding the direction of the relationship between ILOC scores and degree of positionality. However, though not symmetrically, a higher ELOC score might be related to a dependency on others’ behavior (Cobb-Clark and Schurer 2013). Thus, individuals with high ELOC scores might engage in higher degree of comparisons with other people’s consumption levels.

2.2.3 Positive and negative reciprocity

The next trait-like construct that we investigate is the reciprocity, which is also considered to be one of the proxies for social preferences (Falk and Fischbacher 2006; Dohmen et al. 2009; Caliendo et al. 2014; Budria and Ferrer-i-Carbonell 2018). The reciprocity has to do with the degree to which a person tends to respond with kindness to kind behavior and unkindly to unkind behavior. Literature suggests that there are differences in the degree of responses to kind and mean behavior, and these differences are measured on two subscales for positive and negative reciprocity (PR and NR), respectively. The reciprocity inventory is presented in Table 7 (in “Appendix B”). It is based on six statements and measured using a 7-point scale as before, and a person’s total PR and NR scores can range from 3 to 21.

The reactions of individuals to other people’s behavior and attitudes are expected to differ based on how they perceive their and others’ position. One important finding in the earlier literature is that a higher PR scores are associated with higher levels of cooperation (Falk and Fischbacher 2006; Dohmen et al. 2009), which may lead to a less concern about one’s relative consumption level. In contrast, people with a high NR score might display lower levels of cooperative behavior, trust and also well-being (Dohmen et al. 2009; Budria and Ferrer-i-Carbonell 2018). These individuals are expected to be more positional especially if they perceive the positional behavior of others, e.g., conspicuous consumption, as unkind and thus they might reciprocate with a higher positional behavior.

3 Econometric specifications

3.1 The model

The repeated nature of our experimental setup allows us to identify the degree of positionality $\theta^{g}$ in intervals $( - \infty < \mu_{1} ],(\mu_{1} ,\mu_{2} ], \ldots ,[ - \infty > \mu_{V} )$ for each repeated choice situation or version $v = 1, \ldots ,V$ and for each good $g$. In our case, $V = 3$ and the interval boundaries are a priori known, $\mu_{1} = 0.25$, $\mu_{2} = 0.50$, $\mu_{3} = 0.75$, while the global lower and upper boundaries of positionality intervals are set to be censored. To be able to estimate the mean degree of positional concerns conditional on personality characteristics, we specify the following model:

$$\tilde{\theta }_{i}^{g} = X^{'} \beta + \alpha IA_{i} + \varPi_{p}^{'} \lambda^{p} + \varepsilon_{i}^{g} ,$$

(2)

where $\tilde{\theta }_{i}^{g}$ is the latent marginal degree of positional concern with respect to good $g$ for each individual $i$. $X$ is a matrix containing a set of control variables consisting of age, gender, household income (seven category dummies), household size, number of siblings, health-status (four dummies), university department enrolled at (dummies for economics, psychology, and law), and dummies for each good. $\beta$ is the corresponding vector of parameters to be estimated. The model also includes a subjective measure of attitude to inequality $IA_{i}$, i.e., subjective inequality aversion, to alleviate the potential bias due to omitted variables and $\alpha$ is the parameter to be estimated.¹¹ The model specification (2) controls for personality characteristics in $\varPi$, in separate regressions for Big-5, locus of control, and reciprocity. To allow for some flexibility on the relationship between personality and positional concerns, we use log-transformed personality characteristics. $\lambda^{p}$ is the corresponding vector for each personality measure $p$. Finally, $\varepsilon_{i}^{g}$ is the usual error term, which is assumed to be normally distributed.

3.2 Estimators

The model specification (2) is an interval regression in which we use the marginal interval of positionality as the dependent variable. Using the normal distribution assumption, the probability that a respondent’s degree of positionality is in a particular positionality interval is

$$\Pr (\mu_{v} < \tilde{\theta }_{i} \le \mu_{v + 1} ) = \varPhi (\mu_{v + 1} |X,\beta ,\lambda ) - \varPhi (\mu_{v} |X,\beta ,\lambda ),$$

(3)

where $\varPhi (.)$ is the standard normal distribution function. The likelihood function for the uncensored observations is:

$$\ln (L(\beta |X) = \sum\limits_{i = 1}^{N} {\sum\limits_{v = 1}^{V = 3} {\delta_{i,v} \ln } } \left[ {\varPhi (\mu_{v + 1} |X,\beta ,\lambda ) - \varPhi (\mu_{v} |X,\beta ,\lambda )} \right],$$

(4)

where $\delta_{i,v}$ is a dummy variable indicating the positional choice of individual i for the experimental version v. The global lower and upper bounds can also be chosen arbitrarily or set to be censored below zero and above one. In our baseline model specification (2), we consider them as censored. Yet we also compare estimation results with the neutral choice as 0 for the lower boundary of the first and 1 for the upper boundary of the last interval. Then, by calibrating the likelihood function (4) for the censored observations, the estimates of $\beta$ and $\lambda$ can be obtained by maximizing (4).

The model specification (2) is also estimated with alternative model specifications to investigate the robustness of our results. First, an ordered probit model, which maps the marginal positionality intervals the positionality intervals, is estimated. Second, binary choice on an increasing ordinal sequence and allows unknown cut-off points for models are estimated by assuming that an individual makes binary decisions in each choice situation. Third, the repeated nature of the experimental data is exploited to allow for the unobserved individuals effect. A panel dataset is defined over the respondents and repeated choices (as time dimension of the panel data) and panel data (Mundlak type) correlated random-effects models are estimated based on alternative auxiliary functions of unobserved individual heterogeneity. The characteristics used in the auxiliary distributions are well-being (e.g., positive and negative moods, feelings and happiness), emotions (e.g., envy) and self-esteem. Detailed results from these alternative estimators will be presented in our robustness analysis. Finally, a non-parametric estimator, i.e., Spearman–Karber, which is robust to violation of parametric assumptions and sample size, will also be employed when we predict the mean degree of positionality in relation to low and high levels of each personality trait.

4 Results

We first look at the results obtained by combining all goods and then present the heterogeneity in the positional concerns for “goods” and “bads”, and also for each good separately. Second, we present the results from our robustness checks focusing on model specifications, confounding factors, and experimental biases. Finally, we conduct an extensive heterogeneity analysis to investigate how mean degree of positional concern varies across the levels of each personality trait.

4.1 How do personality characteristics relate to positional behavior?

4.1.1 Main results

We first focus on the sign and significance of personality characteristics on the marginal degree of positional concern. To be brief, the tables present results only for personality measures.¹² First, we merge the experimental data obtained from all experimented goods in one dataset and estimate model specification (2). In this model, the dependent variable is the lower and upper boundary value of the marginal positionality interval for each individual over all goods. Having deleted the inconsistent respondents and missing information, the total sample consists of 1544 respondent-good observations. The main estimation results of the relationship between personality and positionality for all goods combined are given in Column I of Table 2.

Table 2 Main results

First of all, among the Big-5 personality characteristics, agreeableness, conscientiousness, and neuroticism are significantly related to positionality conditional on the socio-demographic characteristics, inequality aversion, and indicators for goods and choice situations. Also, the signs of the estimated parameters are in line with the predictions discussed above and those in the literature (Budria and Ferrer-i-Carbonell 2018). Agreeableness is negatively associated with positionality, while conscientiousness and neuroticism are positively related. The sign of extraversion and openness-to-experience is negative but they are not statistically significant on positional concerns. The main results for ILOC and ELOC are presented in the second block of Table 2, Column I. Both variables are positively related to positional concerns, yet only ELOC is statistically significant, in line with our predictions and the literature. In the final block, we present the results for PR and NR. The negative sign of the estimated parameter of PR and the positive sign of the estimated parameter of NR are also consistent with the predictions. However, there is no significant relationship between reciprocity and positional behavior when we combine all goods.

Do the results differ between “goods” and “bads”? We will now investigate the heterogeneity on the relationship between personality and positional behavior across goods. To this end, we combine the data from after-tax income/month, market value of a car, and overall health expenditures experiments as “goods”, and the remaining three items, working hours/week, poverty and infant mortality rates, as “bads”. Columns II and III present results for the “goods” and “bads” using the baseline model specification (2). These results are highly consistent with those in Column I, with some exceptions. The estimated coefficients of conscientiousness, neuroticism, and ELOC have the same sign and significance levels as in the baseline coefficients in Column I. Other than that, there are some important differences between the columns that should be highlighted. First, positional concerns of extraverts react differently to “goods” and “bads”. These individuals exhibit significantly lower positional concerns regarding “bads” than “goods”, possibly because of their higher levels of friendliness and sympathy. A higher score on agreeableness is significantly associated with positional concerns only for “bads”. Conscientious individuals are also slightly more positional regarding the “bads” (0.11 vs. 0.21), yet this difference is not statistically significant (p value = 0.135). People with a high conscientiousness score might desire and also believe they deserve to live in a better society given their high level of work effort and discipline. In contrast, people with a high neuroticism score are more positional regarding “goods” than “bads” (0.14 vs. 0.07) with a partially statistically significant difference (p value = 0.077). Finally, people with high PR scores show significantly lower positional concern (p value = 0.051) regarding “bads”, possibly because of the high degree of empathy and cooperative behavior demonstrated by people with this personality.

Is the evidence consistent with predictions? The results so far indicate highly significant relationships between various personality characteristics and positional concerns. In Columns IV (a) and (b) of Table 2, we summarize our predictions about the relationship between specific personality characteristics and positional concern (a), and whether our results are consistent with the predictions (b). Column IV (a) uses the symbols (−) (+), and (~) to characterize expected relationships as negative, positive, and a priori unknown, respectively. Column (b) uses symbols to indicate whether the predicted and estimated signs are consistent (o), inconsistent (x) or partially consistent o (p). The results suggest that the evidence is highly consistent with our predictions, except for extraversion (inconsistent) and NR and NP (partially consistent). As discussed above, extraversion is expected to be positively associated with positionality. Yet, extraverts might also have lower positional concerns as they might be more empathetic to other people’s situations. Thus, the negative relationship found for “bads” might be due to the nature of the good considered. NR is expected to be positively related to positional concerns, which is in line with our estimates. However, our estimates are statistically imprecise.

4.1.2 Personality traits and positional behavior across goods

The baseline model specification is estimated for each good in separate regressions. The heterogeneity results presented in Table 3 unveil important patterns. First, there is a substantial consistency in the signs and magnitudes of parameter estimates across goods compared with the ones presented previously. However, the significance levels show important variations for some personality characteristics and goods as well. Extraverts are significantly less positional when it comes to infant mortality and poverty rates, as reported above. The relationship is slightly stronger for the former. Agreeableness is negatively and statistically significantly related to positionality only for working hours/week and health expenditures. Conscientiousness and neuroticism are the two personality characteristics with the most consistent associations on positional concerns across the evaluated goods.

Table 3 Results across goods

Conscientiousness is positively associated with positionality, and the relationship is statistically significant for almost all goods. Neuroticism is also positive across all goods, but it is statistically significant only for income, market value of a car, and health expenditures. ILOC is significantly and positively related to positionality only for working hours/week, while ELOC is positively associated with positionality for the market value of a car, health expenditures, and poverty rates. NR is also positive and significant on positionality for market value of a car and health status. As previously reported, the parameter estimate of PR is negative and large for “bads”. Overall, most results are highly consistent with the predictions discussed above.

4.2 Robustness

4.2.1 Alternative model specifications

The model specification chosen for our baseline is an interval regression as we measure positionality in intervals with predetermined borders. One possibility is to interpret the positionality intervals as an ordinal data by assigning increasing ordered numbers for each marginal positionality interval. In this case, the specification (2) is a standard ordered probit model where the borders of marginal positionality intervals are assumed to be unknown cut-off points. Using the baseline sample, which combined the experimental results for all goods in one dataset, an ordered probit model is estimated and the results are presented in Column I of Table 4. Note that the magnitudes of parameter estimates cannot be directly compared with those of the baseline. We compare only signs and significance of estimates. The ordered probit model specification produces the same signs and significance levels for the Big-5 characteristics, ILOC, and ELOC compared to those obtained from the baseline. Yet, this model specification suggests no statistically significant relationship between PR and positional concerns, albeit the sign is the same as in baseline. Among the unreported results, several alternative model specifications have also been estimated. First, we estimated the model with OLS and the results hardly changed. Second, we experimented with the vector of control variables used in the model specification. The results presented in Table 2 turned out to be robust with respect to combinations of control variables used and show hardly any noticeable difference compared with that of the baseline.

Table 4 Robustness checks

4.2.2 Unobserved heterogeneity and endogeneity of personality traits

As in the bulk of the previous literature, this paper assumes that personality characteristics are relatively stable constructs that do not change significantly over time (Borghans et al. 2008; Heineck and Anger 2010; Cobb-Clark and Schurer 2013). People’s personalities have been found to develop before or during adolescence (age 10–19) and then remain stable for a long time.¹³ Thus, it is fair to assume that the respondents in our experiment are old enough (ages 18–27) to have completed the development of their personalities (Becker et al. 2012). The assumption that a person’s personality is fixed, at least for quite a few years, then allows us to consider personality characteristics as exogenous variables (Cobb-Clark and Schurer 2013). However, some omitted variables, e.g., self-esteem, emotions, and mood characteristics, may be correlated with personality characteristics and therefore lead to bias in our estimators. To deal with this potential problem, we have already allowed our baseline model specification for the subjective inequality aversion, which is expected to be one of the key potential omitted variables. Yet, allowing the inequality measure in the model does not affect the results substantially.

Ideally, to deal with the endogeneity of personality, we would need a panel dataset with a fixed-effect model specification or a proper instrument for personality to identify the causal effects. Unfortunately, though, the survey experiment does not allow us to use either of these approaches. That is, the results presented in this paper are still simple correlations and should be interpreted with caution. However, we can suggest an alternative strategy. First, since our respondents make repeated choices characterized by a gradually increasing marginal degree of positional concern, we can interpret the experimental data as a panel dataset defined over the respondents and repeated choices for each good. Second, we specify a panel data random-effects model with alternative sets of potential “time-invariant” characteristics that could capture unobserved individual characteristics in a Mundlak-type formulation, i.e., a correlated effects model.

We have elicited several proxy measures that are relatively stable and might be correlated with personality characteristics. The first one is PANAS (Positive and Negative Affects Scale),¹⁴ which aims at measuring positive and negative dimensions of moods and feelings (Watson et al. 1988). Individuals tend to experience positive and negative affects in a relatively stable pattern (Diener and Larsen 1984) and these affects are found to be correlated with personality characteristics as well (Steel et al. 2008). Second, certain emotions such as envy of other people’s success or consumption levels may be correlated with both positional behavior and some personality characteristics, e.g., neuroticism. The approach suggested to measure individual differences in tendency to envy others is based on the Dispositional Envy Scale¹⁵ developed by Smith et al. (1999). Finally, a measure of self-esteem is generated based on the Rosenberg (1985) inventory.¹⁶ The literature identifies important relationships between people’s self-esteem, personality, and the degree to which they compare their consumption, appearance, and success with others (e.g., Aspinwall and Taylor 1993; Suls et al. 2002; Vrabel et al. 2018).

Merging the data for the respondents and their answers for each good and in each choice-situation generates 4650 respondent-good-choice observations. In Column II of Table 4, we first present results from a linear random effects model using the variables controlled for in the baseline. The panel results are highly similar to those for the baseline except when it comes to agreeableness and PR. The parameter estimates of agreeableness are similar to those for the baseline, yet they are not precisely estimated. The estimated parameter of PR is very small compared with that of the baseline, and it is also statistically insignificant. The results of our correlated random-effects panel data model are given in Columns III and IV for two alternative auxiliary functional forms of the unobserved individual heterogeneity. The correlated effects model given in Column III includes PANAS and the Dispositional Envy Scale. The results differ very little from those for the baseline (Column I of Table 2). Next, we add the self-esteem measure into the auxiliary distribution of the heterogeneity in Column IV. The results are very similar, except for extraversion. In this specification, we find a negative and partially significant relationship between extraversion and positionality. Among the unreported results, we also combined alternative sets of proxy measures in separate correlated-effects models and found highly similar measurements to those in Columns III and IV. In another specification check, we estimated a non-linear correlated effects model using PANAS, the Dispositional Envy Scale, and the self-esteem measure in a random-effects probit model. The parameter estimates (not the marginal effects) are presented in Column V of Table 4. In this model specification, the dependent variable is defined as one if a respondent chooses the positional alternative in any choice situation, and zero otherwise. The results are highly consistent with the baseline and the linear random-effects models presented in the previous columns. Finally, we restrict the dependent variable by using only the answers from the first-choice situation for each good. The dependent variable is a dummy if the respondent chooses the positional alternative at the first choice and otherwise zero. The results are highly consistent (Column VI).

4.2.3 Order effect

An important concern in the repeated stated choice experiments is the order effect, which might generate bias due to trend in the repeated answers to the binary choice questions. Respondents might get bored or tired, which can lead to bias in the observed preferences depending on the order of the questions asked. To alleviate this potential bias, we a priori design six alternative versions of the same experimental questionnaire where the goods are presented in different orders. Column VII of Table 4 presents results from baseline model specification (2), which includes six order dummies. The results are hardly affected.

4.3 Heterogeneity of positional concerns by levels of personality

The results presented above suggest that several personality measures are significantly related to positional concerns. We now turn our attention to the heterogeneity of predicted mean degree of positionality across the levels of each personality trait. We investigate how the degree of positional concerns differs for the low and high values of personality characteristics using our baseline parametric model. Then, a robust non-parametric estimator is exploited to investigate the heterogeneity for each good separately.

4.3.1 Parametric results

The mean degree of positionality is predicted using the estimated baseline specification (2), i.e., interval regression, for the levels of each personality trait while keeping other observed characteristics of each respondent at their mean levels. To be brief, mean predicted degree of positionality for the levels of each personality trait are presented in Fig. 2. We use our pooled experimental dataset that includes all “goods” and “bads” and the baseline specification given in Eq. (2). The three graphs show the predicted mean degree of positionality on the vertical axis and the observed levels of the respective personality traits on the horizontal axis. To add some flexibility, we include the quadratic function of each personality trait. We also tried alternative model specifications that included only a first- or third-degree polynomial, and the results turned out to be highly similar.

Fig. 2

Patterns of positionality by personality trait. The figures are obtained from the baseline interval regression. Separate models are estimated for Big-5 (a), locus of control (b), and reciprocity (c). We calculate the mean predictions and the standard errors for each value of personality and draw the 90% confidence intervals (vertical lines with caps). The horizontal line represents the overall mean degree of positional concerns, 0.492

Panel (A) of Fig. 2 shows the heterogeneity by the levels of each Big-5 personality measure. The vertical lines at each observed personality value are the 90% confidence intervals, which are used to investigate whether the mean predicted degree of positionality for a particular value of a personality trait is statistically different from both the overall mean positionality (the horizontal line at 0.492) and positionality for any other value of personality trait. The mean predicted degree of positionality is heterogeneous around the overall mean for the low and also partially high levels of each personality trait for some particular values. For instance, lower values of agreeableness (filled circles) are related with a higher level of positionality compared to overall mean and the positionality for the higher values of agreeableness. These findings are consistent with previous findings that low agreeableness is correlated with less prosocial, i.e., lower levels of cooperative behavior, and a higher degree of spiteful behavior, which might trigger higher positional behavior. However, the mean degree of positional behavior is gradually lower as the level of agreeableness gets higher.

Low levels of conscientiousness (diamonds) are associated with very low levels of positional behavior, while the mean predicted level of positionality is about the same as the overall mean degree of positionality for the higher values of conscientiousness (around median level, 12). Indeed, this result is highly in line with the predictions of personality psychology. Low conscientiousness is characterized by a tendency to be laid back and a lower degree of success-driven behavior. Also, comparing the confidence intervals across the conscientiousness values suggests that the degree of positional concerns differ across the low and higher values.

A partially similar pattern is observed for neuroticism (triangles). That is, lower values of neuroticism, i.e. higher emotional stability, are associated with lower mean positionality values (lower than the overall mean), and vice versa. We also find that the mean level of positionality is higher than the overall mean degree of positionality for very low values of extraversion (empty circles) and of openness-to-experience (empty squares). Note that the mean degree of positionality is almost the same for the median levels (12–14 on the horizontal axis) of each Big-5 personality trait, and for higher values of these traits, the mean degrees of positionality converge toward the overall mean degree of positionality, i.e., the horizontal line at 0.492.

The results for ILOC and ELOC are given in Panel (B). The mean degree of positionality varies marginally across the ILOC values; there is only a slight positive slope. As for ELOC, the mean degree of positionality is significantly lower than the overall mean only at lower values. Comparing confidence intervals across the ELOC levels suggests that positional concerns are heterogenous across these levels. The final panel of Fig. 2 (Panel C) presents the patterns for NR and PR. On average, positionality does not vary with reciprocity to any degree of significance.

Patterns by goods: non-parametric results. The results from the baseline parametric interval regression model in Fig. 2 suggest that mean degree of positionality is heterogeneous for the low and high levels of most personality traits. We will now present more detailed results for each good and will in particular focus on how the mean degree of positionality differs between the marginally low and high levels of each personality trait. In order to define the low and high levels, we use the first and third quartiles of each personality trait. Then, we compare the mean degrees of positional concerns of these two groups for each good. One important remark is that the cell size in the first and third quartiles is small to obtain meaningful heterogeneity results with our baseline parametric model specification. To deal with potential bias due to small sample size, we use the Spearman–Karber estimator, which is a non-parametric and robust with respect to potential bias due to low sample size (Carlsson and Martinsson 2001; Johansson-Stenman et al. 2002).

The idea behind our choice of estimator is as follows. If a respondent chooses the positional alternative in a choice situation (which signals that the respondent has a higher degree of positional concern than the one implied in the choice situation), then the respondent “survives” until the next choice situation and so on. There are three choice situations where respondents can terminate in any choice situation or survive all three. Each choice situation is assigned an implicit degree of positionality, $\tilde{\theta }_{1}$, $\tilde{\theta }_{2}$ and $\tilde{\theta }_{3}$, and for each case a proportion of respondents survive, $P_{1}$, $P_{2}$, and $P_{3}$. The Spearman–Karber estimator is then formed as:

$$SK_{g} = \frac{{\sum\nolimits_{j = 0}^{4} {(P_{j + 1} - P_{j} )(\tilde{\theta }_{j} + \tilde{\theta }_{j + 1} )} }}{2},$$

(5)

where we assume $P_{0} = 0$, $P_{4} = 1$, and $\tilde{\theta }_{0} = 0$, $\tilde{\theta }_{4} = 1$ for the lower and upper boundaries, and $\tilde{\theta }_{1} = 0.25$, $\tilde{\theta }_{2} = 0.50$, and $\tilde{\theta }_{3} = 0.75$. The variance of the estimator is:

$$SK_{g} = \frac{{\sum\nolimits_{j = 0}^{4} {(\tilde{\theta }_{j} + \tilde{\theta }_{j + 1} )^{2} P_{j} (1 - P_{j} )/} n_{j} }}{4},$$

(6)

where is the number of respondents who choose the positional alternative in each choice situation. Equation (6) is then used to calculate the confidence intervals of estimates.

Figure 3 illustrates the results for each good. The vertical (dashed) lines represent good-specific overall mean level of positionality estimated with the non-parametric estimator. The mean degree of positionality for the low values of the personality traits is given by a triangle symbol and for high values of personality by circles. The lines around the circles and triangles represent 90% confidence intervals. Therefore, the figure shows not only whether the low and high values of each personality trait statistically differ from the good-specific overall mean but also whether the mean degrees of positionality of the low and high values of each personality trait statistically differ from each other for each good. We draw a box around the cases where there is a statistically significant difference between the estimated mean degrees of positional concern associated with the low and high levels of the respective personality trait. Overall, the mean degrees of positionality associated with the low and high levels of the personality traits are highly heterogenous for several personality types and goods.

Fig. 3

Heterogeneity of positionality by goods: non-parametric estimator

5 Concluding discussion

There is a growing interest in the relationship between personality characteristics and the economic outcomes of individuals. Drawing on this literature, this paper examines the relationship between a wide range of personality characteristics and the degree of positional concerns with respect to a number of goods using tailor-made survey experiments. Our results suggest that there are substantial relationships between personality characteristics and positional concerns. In most cases, the signs and significance of the relationships are found to be consistent with recent studies (Budria and Ferrer-i-Carbonell 2018; Cuesta and Budria 2015) based on the life-satisfaction surveys. The results further demonstrate that the relationships between personality traits and positional concerns differ not only across different goods but also across the values of each personality trait. The results are robust with respect to estimators, potential confounding factors, and order effects.

Our paper can be concluded as follows. First, as previously demonstrated by many studies in the literature, respondents show a substantial level of positional concerns, although we note considerable variation across goods in this regard. The overall mean level of positional concern is found to be about 0.50, which is highly in line with the previous literature (e.g., Alpizar et al. 2005; Akay et al. 2013). Second, most personality characteristics are significantly related to the degree of positional concern. The association between agreeableness and positional concerns is negative, while conscientiousness, neuroticism, and external locus of control are positively related. Third, the relationship between personality and positionality differs depending on the good considered. While conscientiousness and neuroticism are significantly related to positional concerns regarding most goods, extraversion and positive reciprocity play a significant role only for the “bads”. The signs and significance of these relationships are highly consistent with the suggestions provided in the field of personality psychology. Finally, we show that the mean degree of positionality across the levels of personality characteristics is heterogeneous across the lower and higher values of most personality characteristics. In most cases, the relationship between personality and positionality is more pronounced for the lower values of each personality characteristics. A low score on agreeableness, for instance, relates to a very strong positional behavior, while low conscientiousness and neuroticism scores relate to less positional behavior. The analysis also unveils that while the degrees of positionality differ significantly between the lower and higher values of some personality characteristics, the mean level of positionality converges toward the overall mean level of positionality for the medium-to-high values of most personality characteristics.

Overall, this paper brings a new set of evidence regarding the relationship between personality characteristics and positional concerns using a survey-based experiment. Our results offer important conceptual implications. We find that the relationship between personality and positional concerns are heterogenous not only across personality characteristics but also within each personality characteristics. Our results call for further investigation of the relationship both in methodological and conceptual terms and suggest that the relationship between personality and economic outcomes should be interpreted with caution in economic models. Most importantly, future research should investigate why personality characteristics and positional concerns are strongly associated mostly at the lower levels and why most non-cognitive skills tend to generate similar positionality outcomes at their higher levels.

Appendices

Appendix A. The survey experiment (after-tax income/month)

Appendix B. Personality inventories and measures

2.1 B.1: Big-5 inventory

See Table 5.

Table 5 Big-5 inventory

2.2 B.2. Internal and external locus of control

See Table 6.

Table 6 Locus of control inventory

2.3 B.3. Positive and negative reciprocity

See Table 7.

Table 7 Reciprocity inventory