How Behavioral Habits Mediate the Relationship between Personality Traits and Savings

Evidence from the UK


Bachelor Thesis, 2016
62 Pages, Grade: 4.00

Free online reading

Table of Contents

1. Introduction

2. Conceptual Framework: Determinants of Household Saving Behavior
2.1. Standard Economic Model and its Limitations
2.2. Evidence from Behavioral Economics Theory
2.2.1. Empirical Link between Personality Traits and Accumulation of Savings
2.2.2. Empirical Link between Personality Traits and Particular Saving Behaviors
2.2.3. Empirical Link between Particular Saving Behaviors and Accumulation of Savings

3. Methodology
3.1. Preconditions for Mediation Analysis
3.2. Significance and Strength of Mediation
3.3. Multiple Mediators, Control Variables and Adjustments

4. Data and Variables
4.1. Independent Variables (Personality Traits)
4.2. Mediating Variables (Saving Habits)
4.3. Dependent Variables (Saving Outcomes)

5. Results
5.1. Results from Preconditions for Mediation Analysis
5.2. Result from Significance and Strength of Mediation
5.3. Results from Multiple Mediators, Control Variables and Adjustments

6. Discussion
6.1. Discussion of Separate Relationships
6.2. Discussion of Mediation Effects and Combined Model

7. Conclusion

8. Limitations

9. Reference List

10. Appendices

1. Introduction

The household saving ratio in the UK has fallen from 11.5% in 2010 to a historically low level of 3.8% in the last quarter of 2015 (Office of National Statistics, 2016). This rapid decline is part of a downward trend starting in 2000 which was only interrupted by a sharp increase in response to the financial crisis (Watkins, 2015). In fact, a recent study by Which? (2014) has found that 41% of the households do not hold the savings buffer recommended by the Money Advice Service and the UK government. This development is worrying as household savings, while they are also important for the national economy (Benton, Meier, & Sprenger, 2007), have various direct implications on individuals and families such as their overall level of living, retirement security, ability to cope with emergency situations (Fisher & Anong, 2012), vulnerability to rising interest rates, economic downturns and high levels of unemployment (Hirsh, 2015).

In order for policy makers to successfully take action, however, it is necessary to understand the determinants of saving behavior first (Crossley, Emmerson, & Leicester, 2012; Wärneryd, 1989). In doing so, two complementary approaches in the existing literature can be identified. Traditional theory takes a purely economic perspective and describes saving behavior as an expected utility maximizing, optimal decision making process (Benton et al., 2007; Keynes, 1936; Modigliani, 1988). However, despite its substantial predictive power (Crossley et al., 2012), it fails to provide guidance for policy design (Wärneryd, 1989), to account for the complexity of saving behavior (Veldhoven & Groenland, 1993) and to fully explain the decline in the saving ratio (Guidolin & La Jeunesse, 2008). Further, it is in direct conflict with the strong field evidence that underlines the importance that behavioral concepts play for the successful design of saving stimulating policies (Ashraf, Gons, Karlan, & Yin, 2003; Beshears, Choi, Laibson, & Madrian, 2006; Madrian & Shea, 2000; Saez, 2009; Thaler & Benartzi, 2004). To fill the shortcomings of traditional theory, researchers and policy makers increasingly turn to behavioral economics theory (Crossley et al., 2012). In this field, three related bodies of research help to understand saving behavior by establishing direct links between individual personality traits and saving outcomes (Brown & Taylor, 2014; Nyhus & Webley, 2001), by relating individual personality traits to specific saving habits (Canova, Rattazzi, & Webley, 2005; Otero- López & Villardefrancos Pol, 2013) and by analyzing the effects of these saving habits on saving outcomes (Fisher & Anong, 2012; Lunt & Livingstone, 1991; Which?, 2014).

Thus, a considerable amount of evidence suggests that there is a significant influence of personality traits on individuals’ and households’ savings. However, the findings of these three bodies of literature exist in isolation and no significant efforts are made in order to combine them into a single framework. Therefore, the existing research primarily focuses on the question if personality traits have an effect on individual or household savings, thereby assessing the direct relationship between the variables. Little effort, however, has been made to understand in what way certain personality traits affect saving outcomes. At the same time, the huge successes of the aforementioned policy initiatives that appeal directly to behavioral and psychological incentives rather than economic ones (Ashraf et al., 2003; Crossley et al., 2012; Duckworth, Gendler, & Gross, 2016; Madrian & Shea, 2000; Saez, 2009) demand that the underlying dynamics of these policy initiatives are understood. In fact, only the conceptualization of why these initiatives are so successful allows designing them in a more structured way and on a larger scale. This in turn calls for a detailed assessment of the way in which personality characteristics affect saving outcomes.

Taking these two arguments together, it can be concluded that there is currently both a lack of and a need for understanding in what way personality traits affect saving outcomes. To address this gap, the three related bodies of literature are combined into a single framework. In doing so, it can be assessed in what way personality traits, saving habits and saving outcomes are related. This assessment contributes to the field of behavioral economics in two ways. Firstly, it contributes to the academic research as it can help to better understand the link between personality traits and household saving by illuminating the driving forces of this relationship. This connects the links provided in the existing literature and allows assessing in what way personality traits and saving outcomes are related. Secondly, discerning underlying behavioral habits that lead to overall household saving success and their relation to individuals’ personality traits can support the design of new governmental initiatives aimed at influencing household saving behavior. More specifically, if the personality traits of people who are likely to commit to a certain behavioral habit can be identified, new policy initiatives can be set up which appeal directly to the personality trait underlying this saving habit. This conceptualization is especially needed in order to design these policies at a larger scale and in a more systematic way than it has been done for far.

Thus, the aim of this paper is to analyze whether and to what extent particular saving habits mediate the relationship between personality traits and saving outcomes. For this analysis, firstly a literature review will present both standard economic models and evidence from behavioral economics theory related to saving behavior. Afterwards, the methodology as well as the data and variables used in this paper are presented. Next, the results of the statistical analysis are described before being discussed and interpreted. Lastly, limitations of the research are outlined.

2. Conceptual Framework: Determinants of Household Saving Behavior

2.1. Standard Economic Model and its Limitations

In the past, economists described saving behavior as an optimal decision making process in which each individual acts in line with his or her long term best interest (Benton et al., 2007). The most widely applied model in the context of saving behavior is the life cycle-model (Hubbard, Skinner, & Zeldes, 1994) which was first introduced by Modigliani and Brumberg (1954). Despite multiple generalizations made to this initial model, its central element remains the notion that individuals keep the marginal utility of their consumption constant (Browning & Lusardi, 1996). This in turn is predicted to cause a consumption smoothing behavior that leads to similar levels of consumption across periods (Wärneryd, 1989) as the marginal utility function of individuals is assumed to be decreasing with wealth (Crossley et al., 2012). Further, the model assumes that projected future utilities are discounted at a regular fashion (Samuelson, 1937).

One reason why the basic model was subject to various modifications is its inability to account for the whole set of motives that incentivize individuals to save. A list of these motives can be found in Keynes’ (1936) initial work in this field. There, a total of eight motives are listed of which two are precaution and foresight. While precaution is defined as building up “a reserve against unforeseen contingencies”, foresight is defined as providing “for an anticipated future relation between income and needs of the individual or his family different from what exists in the present” (Keynes, 1936, p. 73). These two are of particular interest as - opposed to the other motives - they particularly motivate the smoothing of individual consumption over periods of time as predicted by the life-cycle model (Modigliani, 1988).

As emphasized by Crossley et al. (2012) the basic life-cycle model has made valuable contributions for the conceptualization of saving behavior. For example, it highlighted the difference between active (saving today) and passive (capital gains from existing wealth) savings and that savings are expected to increase when income is high; current needs are low and the expected returns of saving are high etc. In practice, however, individuals often do not follow the optimal behavior that these models suggest (Benton, 2007). Three reasons for this are outlined by Thaler (1994). While they are not exhaustive, they do serve to develop a basic understanding of the shortcomings of the life-cycle model. Firstly, the model assumes that households are able to solve multi-period dynamic maximization problems which can be unrealistic in the light of bounded rationality and costly information (Crossley et al., 2012). Secondly, the model does not account for a lack of self-control that might prevent individuals from pursuing a set of actions that individuals themselves believe to be optimal. Thirdly, alternative behavioral models might lead individuals to follow another utility maximizing pattern which does not assume the same form of rationality as suggested by the life-cycle model. As an example, a body of literature proposes that individuals or households have so-called mental accounts which do not all entail the same marginal propensity to consume (Thaler, 1985).

Next to this theoretical assessment of the traditional models, also empirical evidence can be found that illustrates the shortcomings of these models. Madrian and Shea (2000) provide such evidence by analyzing the effect of automatic enrollment into 401 (k) retirement accounts on saving behavior in the US. They find that automatic enrollment substantially increases initial participation in the 401 (k) saving scheme and that employees who are automatically enrolled retain their default contribution rate of the saving scheme even though the employees who are not automatically enrolled in the saving scheme rarely chose these default settings. These findings suggest that individuals do not behave consistent when making saving decisions and that forcing people to save via a mandatory system might more effective than relying on individual rationality as suggested by the traditional models. Further evidence is found by Huffman and Barenstein (2004) who explain that consumption expenditures are steadily declining between paydays and that the magnitude of these declines is stronger than the constant, exponential discounting as assumed in the life-cycle model would suggests. Rather the decline in consumption expenditure is found to follow a reference-dependent preferences model.

2.2. Evidence from Behavioral Economics Theory

As emphasized by Wärneryd (1989) there is no use of psychology in the life-cycle model since all elements of the model are either based or translated into economic and demographic factors. With this background, he emphasizes that even though the life-cycle model is able to explain saving behavior to a significant degree, it has limited use as a basis to design policies aimed at influencing saving behavior. Similarly, Xu, Beller, Roberts, and Brown (2015) claim that in nearly every study of wealth accumulation, there is a substantial amount of variation which cannot be explained by economic or demographic variables. In order to solve these shortcomings, a psychological perspective that goes beyond optimization and stable preference assumptions should receive increased attention (Crossley et al., 2012; Wärneryd, 1989). In line with this argumentation, Veldhoven and Groenland (1993) suggest that the complexity of saving behavior is derived from the fact that the act of saving is embedded in a larger behavioral layer of financial management and thus cannot be captured with purely economic models. In addition to this, a more practical perspective on this issue suggests that even if economic variables are found to have a larger predictive relevance for saving behavior (Lunt & Livingstone, 1991), it is much more sensible to target psychological factors with saving policies than to influence economic variables (i.e. income or interest rates) to this end.

In fact, much field evidence can already be found that emphasizes the effectiveness of policy initiatives which appeal directly to behavioral and psychological rather financial incentives (Crossley et al., 2012). How these policies can affect saving behavior will be exemplified in the following. As already mentioned, Madrian and Shea (2000) found striking evidence from the US that automatic enrollment in the 401 (k) saving scheme has substantially increased participation and thus individuals’ saving rates. Beshears et al. (2006) builds on this conclusion and reviews further evidence for the “tremendous influence that defaults exert on realized savings” (Beshears et al., 2006, p. 27). The authors highlight that the behavioral concepts underlying these findings stand in direct contradiction to the predictions of standard economic theory. Building on this evidence, the Pension Act of 2008 has introduced a similar policy in the UK by prescribing employers to automatically enroll their employees in a pension scheme (UK Government, 2008). This new policy came into effect in October 2012 and marks a radical reform of the pension system (Crossley et al., 2012). In addition to these automatic enrollment policies, Ashraf et al.

(2003) use data from developing countries to emphasize that commitment saving products can contribute greatly to encouraging saving and thus increase saving rates. This finding can be explained with a model derived by Duckworth et al. (2016) who outline how i.e. situational and intrapsychic strategies can lead to greater self-control. One policy example in this field is the highly successful “SMartT” program (Crossley et al., 2012). This program is specifically aimed to employees who lack the necessary self-control to commit to a certain saving goal. To nudge these people to save more, it was tailored to overcome a variety of underlying behavioral biases. For example, one of its main features is that the employees enrolled in the program pre-commit to an automatic increase in their individual contribution rate that occurs with their pay increase expected in the distant future. This feature utilizes behavioral principles of future utility discounting (Thaler & Benartzi, 2004). In addition to the commitment strategy, Saez (2009) shows how even the way in which information of a policy is presented can influence the behavioral response of individuals. More specifically, he compares the take-up and contribution rates for three alternative retirement accounts which all provide the same economic incentive, while only the presentation of the alternative policies is different. Similar to the other research papers mentioned in this section, Saez (2009) concludes that it is not only the implied economic incentive that matters for policies’ effect on saving choices but that also subtle psychological factors play a role. Similarly, a report for the European Commission (Chater, Huck, & Inderst, 2010) provides further evidence for this phenomenon by noting that the very way in which information is presented has a significant effect on individuals’ financial decision making

From the concrete, practical example in the previous paragraph, we can conclude that public policies that appeal to psychological and behavioral factors of individuals can be highly successful. However, in order to design these policies at a larger scale and in a more systematic way than it has been done so far, there is a need to conceptualize the underlying dynamics between psychological factors and saving behavior. In order to achieve this and account for the complexities of the underlying relationships, this paper will combine evidence from three related bodies of literature and subsequently combine these into a single framework. These three bodies of literature will be presented separately in the following subsections as they largely exist in isolation without any significant attempts being done so far to combine them into a single model.

2.2.1. Empirical Link between Personality Traits and Accumulation of Savings

The first body of literature upon which the framework will be based assesses the direct link between psychological factors and saving outcomes. As mentioned by Crossley et al. (2012), there are already multiple existing behavioral concepts which are related i.e. to bounded rationality, mental accounting, loss aversion and reference points, as well as time inconsistency and self-control. However, Nyhus and Webley (2001) emphasize that not much of this existing research has been concerned specifically with the importance of personality traits even though highly significant, albeit contradictory relationships are found in the existing papers.

More specifically, Nyhus and Webley (2001) find that both emotional stability and introversion are positively related to household savings and negatively related to household debt. Empirical evidence provided by Brandstätter (1996), Hirsh (2015) and Wärneryd (1996) is in line with these findings. Nyhus and Webley (2001) further suggest that autonomy and agreeableness, in contrast, have inverse effects, meaning that they are positively correlated with household debt and negatively correlated with household savings. Interestingly, the authors find no significant effect of conscientiousness or neuroticism on household savings or debt. While Brown and Taylor (2014) draw similar conclusion with respect to the personality traits neuroticism and extraversion, they find contrasting evidence with respect to the personality trait agreeableness. More specifically, in contrast to Nyhus and Webley (2001) the paper by Brown and Taylor (2014) shows that agreeableness does not have a significant association with financial assets of a household.

Also in contrast to Nyhus and Webley (2001), the paper by Brown and Taylor (2014) provides evidence that higher conscientiousness scores have a significantly negative effect on the amount of debt holdings in a household and a positive effect on the likelihood that a household has never been in debt. Similarly, Brandstätter (1996) concludes that conscientiousness is linked to increased savings . Further, he finds that emotional stability and introversion significantly affect household savings. More specifically, Brandstätter (1996) finds that the positive effect of conscientiousness on savings is even increased when an individual is also introverted. In agreement with Brandstätter (1996), also Wärneryd (1996) concludes that conscientiousness is positively linked to increased savings. He further argues that it is the most important personality trait associated with self-control and thus saving behavior. Donnelly, Iyer, and Howell (2012) in turn find evidence that higher scores of conscientiousness are linked to improved money management which is in turn associated with higher savings and lower debt holdings. Also, Letkiewicz and Fox (2014) provides evidence that emphasizes the role of conscientiousness on increased amounts of savings. Again contradictions in the research findings surface, since Nyhus and Webley (2001) do not find any effect of conscientiousness on household savings or debt, while among others Brown and Taylor (2014) find that high conscientiousness scores decrease the likelihood of debt holdings. Lastly, Xu et al. (2015) find evidence that young adults with high conscientiousness or extraversion scores are less likely to experience financial distress, while higher agreeableness and openness scores increase the likelihood of financial distress.

2.2.2. Empirical Link between Personality Traits and Particular Saving Behaviors

The findings provided by the first body of literature give clear evidence that there is indeed a relationship between personality traits and household saving. This is a necessary foundation for this paper as the underlying dynamics of saving policies that are built on psychological consideration are analyzed. However, the evidence is not sufficient to derive meaningful interpretations to this end since the research findings are mainly concerned with the relationship of personality traits and the amount of assets and debt holdings in a household as such. By itself, this evidence does not outline in what way the different personality traits affect saving outcomes. Other authors, in contrast, try to fill this gap by deriving more specific behavioral implications that are associated with certain personality traits. This approach helps to extract the particular behaviors that are related to personality characteristics and thus improves the understanding of the dynamics underlying the link between personality traits and saving outcomes. Therefore, the second body of research upon which this paper is based is concerned with particular behavioral habits that are associated with personality traits.

As an example, Otero-López and Villardefrancos Pol (2013) find that the phenomenon of compulsive buying behavior is positively linked to neuroticism and negatively linked to conscientiousness, while the effects of agreeableness, extraversion and openness remain unclear. Mowen and Spears (1999) arrive at similar findings as they also find that high conscientiousness scores are associated with lower compulsive buying behavior. These findings, however, are partially in disagreement with previous research in this field as emphasized in the paper of Otero-López and Villardefrancos Pol (2013) . Related to this, Gathergood and Weber (2014) find that impulsive spending behavior together with high levels of financial literacy can explain the co-holding phenomenon which occurs when individuals or households simultaneously hold credit products on which high interest rates are paid and liquid assets on which low interest rates are earned. Taylor (2015) derives even more concrete behavioral outcomes of personality traits by focusing on charitable behavior. More specifically, their findings suggest that conscientiousness and neuroticism are inversely related to charitable giving while scoring high on openness has a positive association. In addition, Brandstätter and Güth (2000) set up a saving game to find the relationship between personality traits of individuals to their intertemporal consumption patterns. They find that introverts respond most and extroverts least to changes in life expectancies which imply that previously made consumptions decisions were too high or too low given the new information.

In addition, both Brown and Taylor (2014) as well as Nyhus and Webley (2001) find evidence that there are significant relationships between personality traits and the types of saving and debt holdings of households. More specifically, Brown and Taylor (2014) find that personality traits have different effects on different types of debt holdings, while they do not find significant relationship between personality traits and different types of financial assets with the exception of stock holdings. Extraversion, for example, is found to be positively associated with holding credit card debt while conscientiousness decreases the likelihood of holding credit card debt. Nyhus and Webley (2001), in contrast, are able to identify relationships between individual financial assets of households and particular personality traits. For example, emotional stability is found to be positively related to liquid assets while agreeableness is negatively associated with liquid assets. Similar relationships were identified with respect to risky asset holdings and insurance savings.

Moreover, Canova et al. (2005) reveal in their study about the superordinate goals of saving that the saving goals which are highest in the hierarchy are of psychological nature. From this finding it can be derived that personality traits also influence the motives and goals which make people save. Further evidence for this conclusion is provided by Wärneryd (1996) who analyses how personality traits are correlated with different saving attitudes, saving motives and saving behaviors. According to an interpretation by Lunt and Livingstone (1991), even Keynes (1936) acknowledges that while individuals’ consumption pattern may be rational, saving behaviors are based on irrational and psychological motivations such as precaution. Further, Lunt and Livingstone (1991) find evidence that different psychological factors affects how regular individuals save money. Similarly, Fisher and Anong (2012) conclude that psychological factors such as risk tolerance and the tendency to consider a long time horizon when making decisions positively affects the regularity with which individuals save money.

2.2.3. Empirical Link between Particular Saving Behaviors and Accumulation of Savings

From the second body of literature, it can be concluded that there are various relationship between personality traits and specific saving behaviors. This insight helps to better understand the links between personality traits and saving outcomes as analysed in the first body of literature. However, in order to assess whether these specific saving behaviors indeed have explanatory value for the relationships found in the first body of research, it is necessary to assess which specific saving behaviors have a positive and which ones have a negative association with an increased accumulation of household savings. This analysis will characterize the third body of literature upon which this paper is based.

In particular, a report issued by the UK based consumer body Which? (2014) suggests three behavioral habits to be strongly linked to increased household savings. Namely, these are the tendencies to put savings aside with monthly regularity, to keep savings separate from money used for other purposes, as well as to save for a rainy day. Especially the latter of these three factors is widely researched in the academic literature. While the standard economic models mostly treat saving motives as assumptions and focus on a single saving motive, behavioral economics theory acknowledges that multiple saving motives work together (Xiao & Noring, 1994). However, there is strong evidence that the precautionary saving motive is the most significant in predicting the accumulation of household savings (Guariglia, 1998). Dardanoni (1991), for example, has found that 60% of savings in the United Kingdom arise from precautionary motives. Hubbard et al. (1994) further emphasize the importance of considering precautionary saving motives when studying household saving decisions and policy tools.

Besides the literature emphasizing the importance of precautionary saving, also evidence for the effectiveness of the behavioral habit to save regularly can be found in the existing literature. More specifically, Moore et al. (2001) argue that saving regularity is an effective saving strategy as it is associated with greater savings in individual development accounts. The research leaves unanswered, however, which individuals make conscious commitments to save regularly. In fact, the second body of research outlined in this paper is concerned with questions of this kind. More specifically, Fisher and Anong (2012) as well as Lunt and Livingstone (1991) provide evidence that certain psychological factors are associated with the tendency to save regularly. The significance of saving regularity on the accumulation of savings is further emphasize by the Money Advice Service (2016b) which is an independent body created by the Financial Services Act 2010 (UK Government, 2010).

In addition to the effectiveness of the third behavioral habit to keep savings separate from other money for which Which? (2014) provides empirical evidence, a fourth effective behavioral habit can be derived from a publication by the Fairbanking Foundation which is a charitable organisation registered in the United Kingdom. More specifically, an important criterion for their assessment of the fairness of banks is whether or not a goal-based saving account is offered. This is based on research conducted by the foundation itself that identifies setting saving goals to be a significant factor that encourages people to save (Fairbanking Foundation, 2013).

From the review of the third body of literature, it can be concluded that certain behavioral habits have a positive association with the total amount of accumulated savings. These insights provide the last link needed in order to construct a model that helps to understand the way in which personality traits influence household saving outcomes. Following the literature review this model is based on the following realizations. Firstly, personality traits are found to have significant effects on saving outcomes as suggested by the first body of literature. At the same time, personality traits are found to have significant effects on specific saving behaviors as suggested by the second body of literature. Lastly, the third body of literature suggest that these specific saving behaviors in turn have a significant relationship with saving outcomes. By combining these insights, this paper will assess whether or not the specific behavioral habits mediate the relationship between personality traits and household savings.

3. Methodology

The methodology of this paper will be presented in three stages. In the first stage, the three preconditions for the mediation analysis are going to be presented. Only those mediation triangles for which all three preconditions are fulfilled will be further analyzed in the next stage. Here, mediation triangles refer to sets of three variables in which one is viewed as the independent variable, one as the mediating variable and one as the dependent variable. In the second stage, it will be assessed whether the mediation effects in these mediation triangles are significantly different from zero and what percentage of the total effects that the independent variables have on the dependent variables are mediated. Those mediating relationships which are found to be significant will be further analyzed in stage three. There, it will be analyzed how the strength of the mediation effect will be affected when the other mediators, as well as independent variables are controlled for. In addition, various adjustments are made to the model in this stage.

3.1. Preconditions for Mediation Analysis

Before outlining the specific methodology used in this paper, it is necessary to review the basic concepts of mediation analysis. This will serve as a framework for the methodic decisions made in this paper and as the basis for the interpretation of the results. As noted by Jose (2013), the term "mediate” can be defined in multiple ways. The definition that comes closest to the subject of statistical mediation, however, is “to serve as the medium for affecting a result or conveying an object or information” (Jose, 2013, p. 6). More specifically, a mediating variable is conveying information from the independent variable to the dependent variable and can, therefore, be regarded as the medium of information between the independent and the dependent variable.

The customary nomenclature for mediation analysis as described Jose (2013) and MacKinnon (2008) is to call the independent variable or causal variable “X”, the mediating variable “M” and the dependent variable “Y”. As mentioned by Baron and Kenny (1986), James and Brett (1984) as well Judd and Kenny (1981b), the steps that have to be taken for testing mediation are always the same regardless of which method of regression analysis is being used. It is worth noting here that a correlation analysis could be used at this point instead of the regression analyses. However, as will be seen later, the fact that not all variables used in the model are measured as interval or ratio variables makes regression analyses necessary.

The first step prescribed is to run a regression analysis with X as the predictor variable and Y as the dependent variable. Jose (2013) calls this the basic relationship as it tests whether the causal variable is correlated with the dependent variable and thus, whether there is an effect at all that can be mediated. Judd and Kenny (1981b) emphasize that it makes little sense to continue the analysis if no effects are found in this step. In fact, this first test coincides with the relationships analyzed in the first body of behavioral economics literature from section 2.2.1. The basic relationship is formalized in Model 1, in which c is the coefficient of the basic relationship, i1 denotes the intercept and e1 is the error term which summarizes the variability in Y which is not explained by the independent variable X.

Model 1: Y = i1 + c X +e1

Next, as noted by Judd and Kenny (1981b), it must be tested whether each variable in the causal chain affects the variable that follows it in the chain. From this, the second and third steps can be derived. Specifically, in the second step, it is analyzed whether the independent variable X has a significant relationship with the mediating variable M. In other words, the mediating variable is treated as the outcome variable in the second model. This test coincides with the empirical evidence reviewed in section 2.2.2. It can be formulized in the following way, in which a denotes the coefficient of the effect of the independent variable X on the mediating variable M:

Model 2: M = i2 + a X + e2

After it is found that the independent variable has a significant relationship with the dependent variable Y (Model 1) as well as the mediating variable M (Model 2), the last precondition for the execution of a formal mediation analysis is tested. In particular, it must be shown that the mediating variable M significantly affects the dependent variable Y. However, also the independent variable X has to be included in the regression analysis, since the mediating variable M and the outcome variable Y may have a significant relationship only because both the mediating variable M and the dependent variable Y are associated with the independent variable X. This test coincided with the empirical evidence reviewed in section 2.2.3 and can be formalized as follows:

Model 3: Y = is + c’ X + b M + es

The coefficient for the relationship between the independent variable X and the dependent variable Y is now c’ rather than c as in model 1. The reason for this is that Model 3 adjusts the coefficient for the inclusion of the mediating variable M in the model. Similarly, the ”b” is the coefficient for the effect of the mediating variable M on the dependent variable Y when the independent variable X is controlled for.

It is important to note here that the analysis in this paper will deviate slightly from the aforementioned basic models. The reason for this is that all mediating variables and dependent variables in this paper are measured on a dichotomous scale. Thus, all three steps have to be applied using logistic regressions rather than traditional linear regression analyses. As first described by MacKinnon and Dwyer (1993) and further interpreted by Kenny (2016b), a variable has a different scale depending on whether it is used as the independent or the dependent variable in a logistic regression. This does not change the basic logic of the 3 models as outlined above. However, the differences in scale have statistical consequences and therefore have to be accounted for. This leads to the following adjusted models, in which the apostrophes reflect the differences in scale:

Adjusted Model 1: Y’ = i1 + c X +e1

Adjusted Model 2: M’ = i2 + a X + e2

Adjusted Model 3: Y’’ = i3 + c’ X + b M + e3

3.2. Significance and Strength of Mediation

As explained by Kohler, Karlson, and Holm (2011), it is difficult to interpret the magnitude of logit coefficients as they are measured on arbitrary scales. In fact, this claim holds for all coefficients of total, direct and indirect effects in this paper. Thus, instead of measuring the total magnitude of the mediation effects, this paper focuses on their significance and their strength. This is in line with the aim of this paper to analyze the direction and the extent to which behavioral habits function as a mediator for the relationships between personality traits and savings rather than to analyze the total magnitude of these relationships.

After conducting the analyses described in the first three steps, all triplets (mediation triangles) for which the mediation preconditions were met are passed on to the fourth step. At this point, even though the fulfillment of all preconditions suggests that a mediation effect is to be found, the statistical significance and the strength of the mediation effect is still to be determined. In order to do so, it is necessary to first ensure comparability of the coefficients. This is needed as the scales of coefficients are distorted by the use of logistic regression in this paper. For the rescaling of the coefficients, Herr (2016) derives equations from MacKinnon and Dwyer (1993) that suggest the multiplication of each of the coefficients a, b, c and c’ with the standard deviation of the predictor variable as well as division with the standard deviation of the outcome variable. In doing so, the variances of the outcome variables are defined as:

1. Variance (Y’): c2 V(X) + π2/3
2. Variance (M’): a2V(X) + π2/3
3. Variance (Y’’): c’2V(X) + b2 V(M) + 2 b c’ Cov (X,M) + π2/3

The rescaled coefficients can then be tested for statistical significance. Specifically, it is tested whether the coefficients of the mediation effects are significantly different from zero. For this, a bootstrapping approach as suggested by Bollen and Stine (1990) as well as Shrout and Bolger (2002) will be applied. According to Bollen and Stine (1990), bootstrap “is an approach to estimating properties of estimators based on samples drawn from the original observations” (Bollen & Stine, 1990, p. 3). More specifically, the indirect effects are computed from each of these samples in order to empirically generate a sampling distribution (Kenny, 2016a). This sampling distribution will be used in order to extract 90% confidence intervals. A variable is found to be significantly different from zero whenever this confidence interval does not include the value 0, indicating that the p-value of the coefficient is below 10%. As suggested by Jose (2013), the bootstrap function should be performed multiple times as the combination of all generated outcomes gives a more reliable estimate of the outputs generated. Thus, in order to ensure the accuracy of the bootstrapping approach, 300 replications will be used for each significance test. Even with 300 replications, however, bootstrapping cannot give definite results and provides slightly varying outcomes. Nevertheless, only those mediation triangles in which the indirect effect proves to be significant are considered in the next step.

While the last step has used the rescaled coefficients to test the statistical significance of the mediation triangles, the next step evaluates the strength of the indirect effects. This is different from a statistical significance test in that a significance test only analyses whether a coefficient is different from zero while a test of strength indicates how large the mediation effect is (Jose, 2013). In order to conduct a test of strength, both Baron and Kenny (1986) and Judd and Kenny (1981b) prescribe to identify whether the mediating variable M completely mediates the relationship between the independent variable X and the dependent variable Y. This would suggest a 100% mediation effect. However, for the purpose of this paper it is necessary to not only identify complete mediation but also partial mediation as it is unlikely that the behavioral habits fully mediate the relationship at hand. Therefore, a ratio measure suggested by MacKinnon (2008) is used in order to estimate the strength of the indirect effect. This is in line with the suggestion by Kohler et al. (2011) to use a confounding percentage measure. In particular, it will be measured how much of the total effect of the independent variable Y on the dependent variable X is mediated by the mediating variable M. For this, the mediation effect of M is divided by the total effect of the independent variable X on the variable Y. The same measure of strength will be in stage 3 of the analysis.

3.3. Multiple Mediators, Control Variables and Adjustments

At his point in the analysis, multiple mediation triangles are detected and the strength of each indirect effect is assessed. To gain further insights from the data, however, two additional analyses will be executed. Both of these analyses will help to merge all individually derived mediation triangles into distinct models. Firstly, the possibility that multiple mediating variables affect the dependent variable will be analyzed. Again, the fact that logistic regression models are used complicates the analysis. In particular, Karlson and Holm (2011) explain that the scale identification issue prohibits the simple decomposition of effects. The authors, thus, refer to the KHB model which is found to be at least as good as the other models which could be used to overcome this issue (Kohler et al., 2011).

In line with their argumentation, this paper will use the KHB model in the following way. All mediation triangles whose indirect effects were proven to be significant in stage 2 will be combined for each of the independent variables. In other words, the KHB model will be used to set up a mediation analysis in which one independent variable X, one dependent variable Y and all the mediating variables M which significantly mediate the relationship between independent variable M and dependent variable Y will be included. This analysis will reveal whether the individual mediating variables have overlapping mediation effects for the relationship between independent variable X and dependent variable Y. Also, it will be revealed which of the mediation effects has the strongest unique mediation effect when all other mediating variables are controlled for.

After this analysis has been conducted, depending on the statistical interpretations, further adjustments are made to the model. These adjustments are based on additional insights into statistical theory. For example, Judd and Kenny (1981a) and Jose (2013) explain that precondition 1 from stage 1 of the statistical analysis in this paper is irrelevant in the case when inconsistent mediation effects (suppression effects) are found. Therefore, the model derived in the previous step of the analysis will be tested for this phenomenon. Also, up to this point the statistical analysis was designed in a highly strict and conservative way. In fact, Kenny (2016a) acknowledges that the three precondition analyses can be seen in terms of zero and nonzero coefficients rather than in terms of statistical significance. With this claim, he corrects the instructions made in his initial paper on mediation analysis (Baron & Kenny, 1986). Also, adjustments are going to be made in order to ensure greater comparability between the models. Whether and to what way these adjustments will take place is dependent on the statistical results and cannot be outlined in advance. Thus, all preceding steps are executed without considering these potential adjustments. Only then can it be decided whether or not certain phenomena have occurred and whether or not adjustments are required.

As soon as all adjustments are made, in a last step the models will be further improved by not only controlling for all mediating variables the models, but by also explicitly controlling for all other independent variables in the analysis. This in turn will show the mediation effect of each mediating variable on the relationship between one independent variable and one dependent variable when all other independent variables and mediating variables are accounted for. It will help to discern the unique mediation effect that a combination of one of the independent variables together with one of the mediating variables suggests.

4. Data and Variables

In order to answer the research question, a representative survey of 1000 respondents in the United Kingdom will be used. It includes information about the respondents’ personality traits, saving methods, saving goals, saving regularity and saving motives. Importantly, the survey data at hand provides information at the household level and includes data only from those household members who are solely or jointly responsible for their households’ finances. The survey was conducted by Which? (2014) which is a consumer body in the United Kingdom. The legal rights of consumer bodies in the United Kingdom are outlined in the Enterprise Act 2002 (UK Government, 2002) and determine the title “consumer body” to be an official designation made by the Secretary of State. The role of consumer bodies is to complaint to the Office of Fair Trading whenever certain features of the market in the UK appear to be significantly harming the interests of consumers. “Which?” is the largest of these consumer bodies in the UK and claims to operate completely independent from any interest groups. Given the legal character, significance and independence of “Which?” it can be regarded as a reliable source for this research. In the following, the independent, mediating and dependent variables that are derived from the literature review are presented, together with a reference to the survey questions from which the variable data is obtained. All relevant questions can be found in Figure 1 of the Appendix.

4.1. Independent Variables (Personality Traits)

The independent variables for the research concern the measure of the responds’ personality traits. As has been done in previous research in this field (Donnelly et al., 2012; Otero-López & Villardefrancos Pol, 2013; Taylor, 2015), this paper will use the Big Five personality dimensions to this end which were developed by (Costa & McCrae, 1992). Namely, these are openness, neuroticism, extraversion, conscientiousness and agreeableness. The Big Five personality dimensions framework suggests that most differences in the personality of humans can be categorized into these five broad, empirically derived domains (Gosling, Rentfrow, & Swann, 2003). While not being without criticism, there is consensus in the field of personality psychology, suggesting the Big Five personality dimensions as being the general taxonomy of personality traits (John & Srivastava, 1999). Importantly, some evidence is found that the Big Five personality traits are stable over time (Cobb-Clark & Schurer, 2012).

In the survey underlying this research, the personality traits were measured using 14 questions. More specifically, two questions were used to assess the responds’ conscientiousness score, while three questions were used to derive insights into the other four personality traits. As described by Gosling et al. (2003) there is a large consensus that the usage of more questions to assess a single personality trait tends to have better psychometric properties than i.e. single-item scales. This raises concerns about the reliability of the psychometric test at hand since only few questions were used for the assessment of each personality trait. Besides, the usage of reverse- coded items in the questionnaire imposes further concern. In particular, Swain, Weathers, and Niedrich (2008) identify three potential sources of misresponse that can arise from the presence of reversed items. These are respondent inattention, which occurs when respondents allocate insufficient cognitive resources when filling out the questionnaire, respondent acquiescence, which is defined as uncritical agreement with items and item verification difficulty, which describes that reversed items are more confusing or difficult to process than non-reversed items. Given the concern imposed due to the brief measure of the personality dimensions, as well as the use of reversed items, the variables for each personality dimension are analyzed and constructed individually. In doing so, the reliability measure Cronbach’s alpha will be used to analyze the internal consistency of the constructed variables. This is in line with the approach of Brown and Taylor (2014) and is suggested by Gliem (2003).

The questionnaire includes two items as a measure of the personality dimension “conscientiousness”. One of these items is measured on a reverse-scale. When combining both items in a construct a low Cronbach’s alpha of α=0.394 is found, suggesting that the construct is unreliable. Hence, in this paper only one of the items is used as a proxy for the personality dimension conscientiousness. More specifically, the reverse-scaled conscientiousness measure is used as it is better able to discriminate between respondents due to its higher standard deviation and its mean which is closer to the center than the other conscientiousness measure. For the personality dimension extraversion three items are included in the questionnaire. One of them is measured on a reverse-scale. When combining the three items into a construct, a Cronbach’s alpha of α=0.71 is found. A more detailed analysis, however, shows that the exclusion of the reverse-scaled item improves the Cronbach’s alpha to α=0.77, while at the same time leading to a larger standard deviation which further improves the discriminatory power of the construct. The normality of the distribution of the construct variable is not affected by this change. Hence, the personality dimension extraversion will be proxied by a construct of the two items which are not reverse-scaled. For neuroticism, it is insignificant whether the reverse-scaled item is included in the analysis or not as Cronbach’s alpha is α=0.77 and α=0.79 with and without this item, respectively. However, the standard deviation of the construct after excluding the reverse-scaled variable is significantly higher, allowing the variable to better discriminate among respondents. At the same time the exclusion of the reverse-scaled item mitigates the concerns expressed by Swain et al. (2008). Lastly, the personality dimension openness is also measured with three items in the questionnaire. However, in this case there is no reverse-scaled item. Therefore, the Cronbach’s alpha is α=0.72 and only gets lower when one of the variables is excluded from the construct.

In contrast, to the personality dimensions extraversion and neuroticism, only in the case of agreeableness it was decided not to exclude the reverse-scaled item: When combining all three items in a construct a Cronbach’s alpha of α=0.57 is found. The exclusion of the reverse-scaled variable would indeed lead to a small improvement of this measure to α=0.63. However, in contrast to the analysis of the extraversion construct, in this case the histogram is not as normal as with the inclusion of the reverse-scaled item. Also, in contrast to the neuroticism measure, there is no large increase in the standard deviation after the exclusion of the reverse-scaled item. Thus, the inclusion of the reverse-scaled variable has a positive effect on the internal consistency of the construct, while having a negative effect on the fulfillment of the assumptions for the regression analysis, without having a large impact on the standard deviation. Additionally, following Gosling et al. (2003) argument that the inclusion of more items is recommended if all else is equal, the reverse-scaled item will be included and thus, the personality dimension agreeableness will be proxied by all three items in this paper.

4.2. Mediating Variables (Saving Habits)

After the specification of how personality traits are going to be constructed, the research outline requires saving habits to be identified that potentially mediate the relationship between personality traits and household saving. The selection of these saving habits will be based on the findings of Which? (2014) which finds three distinct saving habits that are strongly linked to successful saving. These habits will be included as mediating variables in this research and the claims that these behavioral habits are strongly linked to successful saving (Which?, 2014) will be tested with a formal statistical analysis. In the following, the three behavioral habits are introduced and the ways in which the variables will be defined for the statistical analysis are outlined in depth.

Firstly, saving every month is named as a significant contributor to building up and maintaining the recommended savings buffer. Hence, this paper will use “Saving Regularity” as a saving habit and therefore as a potential mediator. In specific, the variable “Saving Regularity” will be coded as a dummy variable which takes on the value 1 when the respondent indicated in Q17 to have saved at least in 5 of the past 12 months. This way, the total sample is divided almost equally into the 484 (50.47%) respondents who save regularly and the 475 people (49.53%) who did not. A total of 41 respondents who indicated that they do not know how many months in the last 12 months they have saved money will be omitted from the analysis.

Secondly, Which? (2014) has identified that respondents who save for a rainy day are more likely to have the recommended savings buffer. Here, “rainy day savings” are defined as savings for unexpected expenditures or to help for making ends meet occasionally. This definition is in line with the precautionary saving motive as defined by Keynes (1936). In order to account for this, the variable “Saving Motive” is included as a saving habit into the analysis. However, there are two questions in the questionnaire with which we can find data for this variable. On the one hand side, respondents are asked whether they see all or parts of their savings as bearing earmarked for different things in Q11. On the other hand side, in Q24 respondents are asked directly why they are saving. In this paper, data collected from the latter question will be used to analyze the variable “Saving Motive” as the second question asks more directly about the saving motive while the first question assesses how the savings are viewed after they have already been accumulated. The variable is coded as a dummy variable and takes on the value 1 when respondents indicate that saving for a rainy day is one of their saving motives and 0 if it not. This approach divides the respondents into 349 (54.36%) who do save for a rainy day and 293 (45.64%) who do not. The 358 respondents who indicated in Q17 that they have not saved in the last 12 months and those who indicated that they do not know if they have are omitted when this variable is used.

Thirdly, Which? (2014) identifies that keeping savings separate from other money is affecting the likelihood of an individual to have the recommended savings buffer. This finding is included as the saving habit and potential mediating variable “Saving Method” into the analysis. As a proxy for this variable respondent data from two different survey questions can be used. Firstly, Q34 provides insights into the main savings product each respondent uses. It is possible to categorize these savings products into those which allow individuals to separate their savings from their other money and those which do not. However, this categorization is problematic as i.e. Instant Access Cash Individual Savings Accounts (ISAs) allow individuals to retrieve cash immediately and without notice from their savings accounts (Nationwide Building Society, 2016), thereby having similar characteristics to a regular current account. In order to overcome this classification problem the data gathered from Q20 will be used as a proxy for “Saving Method”. This question directly asks the respondents whether their main method to save money is to keep it in a savings product or in a current account and cash. More specifically, the variable is coded as a dummy variable which takes on the value 1 when the main saving method is to keep the money in a current account or to keep it in cash and 0 if the money is mainly kept in a savings product. It is found that a total of 584 (68.07%) of the respondents mainly keep their savings in a designated savings product while 274 (31.93%) mainly use current accounts and cash holdings. The remaining 142 respondents are omitted from the analysis as they indicated that they either do not know the answer, that they do not save at all or that they use another method. The latter option is not further specified and therefore does not allow for a distinction between those who keep the money in a separate account dedicated to savings and those who keep their savings with their other money.

Beyond the findings of Which? (2014), a fourth variable will be included into the analysis as a saving habit and thus a potential mediating variable. Namely, the variable “Saving Target” which takes on the value 1 for those respondents who indicated in Q23 to have a certain saving target (specific money amount, specific goal (i.e. saving for a holiday, car etc.) and proportion of income) and 0 for those who do not. In order to derive meaningful interpretation from the analysis, the 276 respondents who do not save and those who do not know whether they have saving targets are omitted from the statistical analysis. This leaves 502 (69.34%) who do not have specific saving targets and 222 (30.66%) who do.

4.3. Dependent Variables (Saving Outcomes)

In this paper, two different dependent variables will be used to describe saving outcomes. The dummy variable “Cover 3 Months” will refer to the achievement of a savings buffer that is sufficiently high to meet the recommendation of the Money Advice Service and the UK government. Specifically, this means having three months’ or more of essential expenditure put aside in liquid savings (Which?, 2014). The variable value is 1 when the respondent successfully meets this benchmark and 0 otherwise. The data for this variable is extracted from survey question Q16. In total, 399 (45.76%) of the respondents indicated that they meet this benchmark, while 473 (54.24%) indicated that they do not. The 128 respondents who indicated that they do not know whether they meet this benchmark will be omitted in the statistical analysis.

While the first dependent variable focuses on savings only, the second dependent variable will additionally include information about the debt holdings of a household. More specifically, the second dependent variable “More Debt” takes on the value 1 when it can be inferred from the respondents’ survey data that they have more debt than savings and 0 otherwise. This information will be derived from the data gathered in Q12 which asks the respondent to indicate a range for their total amount of savings and Q14 which asks the respondents to indicate a range for the amount of debt that they are holding. In 670 (71.73%) of the cases the respondent indicated a higher range for his debt holdings than for his savings. From this, it is inferred that the respondent has more debt than savings and the variable More Debt is coded 1. In contrast, 264 (28.27%) of the respondents indicated a higher range for their savings than for their debt holdings. It is inferred that the respondents have more savings than debt and thus, the variable More Debt is coded 0. There are, however, 97 (9.62%) respondents who indicated the same range of values for both their debt holdings and savings. In these cases, it cannot be inferred from the data whether the respondent has more debt holdings than savings or vice versa. In order to ensure that the “More Deb” variable takes on the value 1 only when it can be concluded with certainty that the respondent has more debt holdings than savings, these 97 cases are going to be coded 0. Further, 66 cases in which respondents did not know or preferred not to say the total amount of their debt holdings or savings will be omitted from the analysis. Lastly, it must be noted that this measure does account for any mortgage and student debt while it accounts for the total amounts of savings in the household.

5. Results

In the following, the results from the three stages of analysis will be presented. Given the statistical outcomes in the other stages, adjustments to the models are executed in stage 3. In all statistical tests an alpha level of 5% is applied. However, since the strength of the mediation is assessed following the analysis of the preconditions, mediation triangles which fulfill the preconditions at an alpha level of 10% are also passed to the next stage. It will be mentioned whenever this is the case.

5.1. Results from Preconditions for Mediation Analysis

In order to test the first precondition of the mediation analysis, a total of 10 logistic regressions are run. In specific, 5 of these logistic regressions use the saving outcome Cover 3 Months, while the other 5 logistic regressions use the saving outcome More Debt as the dependent variable. A separate logistic regression must be run for each of the individual personality traits which function as the independent variables. The regression coefficients, z-scores and p-values for the regressions can be found in tables 1a and 1b. It is shown that the personality traits agreeableness, neuroticism, extraversion and openness have a negative impact on the likelihood that a respondent has the recommended savings buffer as defined in section 4.3. The negative effect of the personality trait extraversion, however, is only significant at the 10% alpha level. In contrast, the personality trait conscientiousness has a positive effect on the saving outcome Cover 3 Months. When testing the first precondition for the saving outcome More Debt, it is found that the personality traits agreeableness and neuroticism have significantly positive effects on the likelihood that responds' debt holdings are higher than their savings. Besides, the personality trait conscientiousness has a negative effect on the saving outcome More Debt at an alpha level of 10%. The variables extraversion and openness do not have a significant effect on the variable More Debt.

The second precondition for the mediation analysis is tested by executing 20 logistic regressions. These consist of 5 regressions for each of the four saving habits which are treated as dependent variables in this step. The results for the logistic regressions with Saving Method, Saving Target, Saving Regularity and Saving Motive as the dependent variable can be found in tables 2a, 2b, 2c and 2d, respectively. Again, each one of the Big Five personality traits separately functions as an independent variable in all logistic regressions. It is found that the personality traits extraversion, agreeableness, neuroticism and openness have a significant, positive effect on the likelihood that a respondent does not use a separate savings product for his savings. In contrast, the variable conscientiousness has a significant, negative effect on this likelihood. When Saving Target is used as the dependent variable the direction in which the Big Five personality traits affect the dependent variable are the same as in the case of Saving Method. Besides, it is found that the personality trait extraversion has a significant, positive effect on the variable Saving Regularity, while the personality trait neuroticism has a significant, negative relationship. The variables conscientiousness, agreeableness and openness do not have any significant relationship with Saving Regularity. Lastly, the logistic regressions provide evidence that the personality traits extraversion, agreeableness and neuroticism have a significant, negative relationship with the variable Saving Motive. The personality traits conscientiousness and openness, however, do not have a significant relationship.

In order to conduct the assessment of the third precondition in this paper, it is necessary to analyze the coefficients of 40 more logistic regressions. These consist of 10 regressions for each of the four saving habits, which are further subdivided into 5 regressions for each of the two saving outcome variables. More specifically, for each combination of one of the four saving habits and one of the two saving outcomes 5 separate regressions are run, in which the five personality traits are individually accounted for. Thus, both the 4 saving habits, as well as the Big Five personality traits are treated as independent variables here. The logistic regressions including the behavioral habits Saving Method, Saving Target, Saving Regularity and Saving Motive can be found in tables 3a-3h. It is found that Saving Method has a significantly negative effect on the saving outcome Cover 3 Months and a positive effect on the saving outcome More Debt, regardless of which personality trait is controlled for. On the contrary, the behavioral habits Saving Regularity and Saving Motive have a significantly positive effect on the saving outcome Cover 3 Months and a significantly negative effect on the dependent variable More Debt, regardless of which personality trait is controlled for. The behavioral habit Saving Target only has negative effects on Cover 3 Months at the 10% alpha level, when the personality traits conscientiousness, extraversion, neuroticism and openness are controlled for.

When taking together the results from the assessment of all three precondition assessments, a total of 20 mediation triangles can be identified which pass these conditions. More specifically, the statistical analyses suggest that significant mediation effects are expected to be found in the following mediation triangles: Firstly, it is suggested that the behavioral habits Saving Method and Saving Target mediate the relationship between the personality trait conscientiousness and the saving outcome Cover 3 Months. Further, it can be inferred that the habit Saving Method mediates the relationship between the personality trait conscientiousness and the saving indicator More Debt. For the personality trait agreeableness, the behavioral habits Saving Method and Saving Motive are expected to have mediation effects on both saving outcomes, namely Cover 3 Months and More Debt. Besides, it is found that all four behavioral habits, namely Saving Method, Saving Target, Saving Regularity and Saving Motive, mediate the relationship between the personality trait extraversion and the saving indicator Cover 3 Months. Similarly, all of the behavioral habits are suggested to mediate the relationship between neuroticism and the saving outcome Cover 3 Months. The relationship between neuroticism and More Debt, however, is only mediated by three of the behavioral habits, being Saving Method, Saving Regularity and Saving Motive. Lastly, the behavioral habits Saving Method and Saving Target are suggested to mediate the relationships between the personality trait openness and the saving indicator Cover 3 Months.

5.2. Result from Significance and Strength of Mediation

In this section, a formal statistical test will be applied to assess the significance of the 20 potential mediation triangles that were identified in the previous section. The results from the bootstrapping analysis needed for this assessment are listed in tables 4a-4d. It is found that the positive relationships between the personality trait conscientiousness and the saving outcomes Cover 3 Months and More Debt are mediated by the behavioral habit Saving Method. Further, the negative relationship between the personality trait extraversion and the saving outcome Cover 3 Months is significantly mediated by the saving habits Saving Method, Saving Regularity and Saving Motive. Besides, it can be seen that the habits Saving Method and Saving Motive mediate the negative effect of the personality trait agreeableness on the saving indicator Cover 3 Months, while only the behavioral habit Saving Method mediates its positive effect on the saving outcome More Debt. The negative relationship between the personality trait neuroticism and the saving outcome Cover 3 Months is found to be significantly mediated by the behavioral habits Saving Method, Saving Regularity and Saving Motive. In contrast, the positive relationship between the personality trait neuroticism and the saving outcome More Debt is only found to be significantly mediated by the behavioral habits Saving Method and Saving Regularity. Lastly, the negative relationship between the personality trait openness and the saving indicator Cover 3 Months is found to be mediated only by the behavioral habit Saving Method.

In sum, 14 mediation triangles were identified which show a significant mediation effect. It is striking that there is no mediation triangle in which the behavioral habit Saving Target has a significant mediation effect. Further, the bootstrapping analysis revealed that the behavioral habit Saving Motive shows relatively weak significant mediation effects or no significant indirect effect in all cases with one exception, namely in the relationship of the personality trait extraversion and Cover 3 Months.

Now, the 14 mediation triangles which showed a significant mediation effect in the bootstrapping analysis will be further analyzed by calculating the confounding percentages. The results can be found in tables 5a and 5b. For example, the percentage of the total effect of the personality traits conscientiousness, extraversion, agreeableness, neuroticism and openness on the saving outcome Cover 3 Months that is mediated by the behavioral habit Saving Method is 38.6%, 41.7%, 28.4%, 20.0% and 26.8%, respectively. Besides, the behavioral habit Saving Method mediates 26.1%, 17.6% and 9.45% of the total effects that the personality traits conscientiousness, agreeableness and neuroticism have on the saving outcome More Debt, respectively. Moreover, -64.5% and 20.4% of the relationships between the personality traits extraversion and neuroticism with the saving outcome Cover 3 Months is mediated by the behavioral habit Saving Regularity, respectively. The relationship of the personality trait neuroticism to the saving outcome More Debt in turn is mediated to 21.9% by the saving habit Saving Regularity. Lastly, the behavioral habit Saving Motive respectively mediates 36.4%, 17.6% and 8.43% of the effect that extraversion, agreeableness and neuroticism have on the saving outcome Cover 3 Months. Thus, it can be concluded that the mediation effects which were discerned in the previous steps are not only significant but also high in individual mediation strength. The meaning of negative confounding percentages will be addressed in the next section.

5.3. Results from Multiple Mediators, Control Variables and Adjustments

In this stage individual mediation effects which were analyzed in the previous steps will be disentangled. The first step to do so is to combine the mediation triangles which include the same personality trait and the same saving indicator into individual models. By doing so, the unique mediation effect of each behavioral habit is discerned, while all other behavioral habits are controlled for. The results from this approach can be found in table 6a. It is striking that for four of the 8 models, there is only one saving habit left in the analysis. In these models, the mediation effect is by default the same as found in the previous stage. The mediation effects in the other 4 models, however, differ from the effects found in stage 2 since the models now control for the inclusion of other mediating variables.

Before listing the confounding percentages that each of the mediation effects wields individually and in total, however, the final models will be adjusted for two reasons. The first reason can be derived when analyzing the outcome of model 2. In this model, it is tested how strong the mediation effects of the saving habits Saving Method, Saving Regularity and Saving Motive are for the relationship between the personality trait extraversion and the saving outcome Cover 3 Months. It is striking that the confounding percentage of the mediating variable Saving Regularity is found to be negative. This finding demands further analysis and subsequent adjustments to the final models. It is explained by MacKinnon, Krull, and Lockwood (2000) that it is generally assumed in mediation analysis that the inclusion of mediating variables (in this case the saving habits) reduces the magnitude of the relationships between the independent and the dependent variables. The concept of inconsistent mediation (suppression), however, distorts this basic assumption. A suppressor variable is defined as “a variable which increases the predictive validity of another variable by its inclusion in a regression equation” (Conger, 1974, pp. 36-37). Thus, in this paper a suppressor variable is found whenever the inclusion of a behavioral habit into the regression increases the magnitude of the relationship between a personality trait and a saving outcome. Judd and Kenny (1981a) as well as Jose (2013) acknowledge that in the case of a suppressor variable the first precondition of stage 1 is not relevant anymore. More specifically, it is argued that an independent variable can have i.e. a positive direct effect on a dependent variable while at the same time having a negative indirect effect or vice versa. Thus, the total effect, which is tested with precondition 1, might then be insignificant as the direct and indirect effect cancel each other out. This leads to a situation in which there is indeed a significant mediation effect even though precondition 1 is not fulfilled. In line with this argumentation and based on the statistical findings of this paper, it can be concluded that the mediating variable Saving Regularity which was discarded from some of the models in previous steps, should be reinserted into the eight final models. Also, a ninth model must be added which tests the relationship between the personality trait Extraversion and the saving indicator More Debt. The reason is that this model was discarded from the analysis in stage 1 as precondition 1 was not found to be fulfilled. As it turns out, however, the likely cause for this was the suppressing effect of the saving habit Saving Regularity. Evidence for the inconsistent mediation effect at hand can be found in table 6.2. A model which tests mediation for the relationship between the personality trait openness and the saving outcome More Debt, in contrast, is not reinserted into the analysis. The reason is that for this model no inconsistent mediation effect is found after reinserting Saving Regularity. Therefore, the fact that this mediation triangle has not fulfilled precondition 1 in stage 1 was a valid argument to discard the model from the analysis.

At this point, the final outcome of the statistical analysis was adjusted to 9 models, of which 5 include the saving outcome Cover 3 Months as the dependent variable, while the remaining 4 have the saving outcome More Debt as the dependent variable. Following from the statistical analysis in all previous steps, each of these models contain the saving habits Saving Method and Saving Regularity as mediating variables. The habit Saving Motive, however, is only contained by four of the nine models. In order to ensure the comparability of the final models and thus the validity of the interpretations made from these outcomes, the saving habit Saving Motive will be added to the five models which currently do not contain it. In fact, this step is not unreasonable for the following reasons. Firstly, the margins at which the relationships in stage 2 were assessed to be significant or not are based on a bootstrapping approach. A repetition of these bootstrapping analyses reveals that the error margin of the outcomes is large enough to argue that the habit Saving Motive was discarded using too conservative measures. The same argument is not made for the behavioral habit Saving Target as not a single case was found in which this variable has a statistically significant mediation effect. Thus, for this habit a reinsertion is not necessary and at the same time not justifiable. Secondly, Kenny (2016a) and Jose (2013) argue that the preconditions can also be seen in terms of zero and nonzero coefficients rather than in terms of statistical significance. Thus, Kenny (2016a) corrects the claims he made in his initial paper (Baron & Kenny, 1986). It can be concluded, therefore, that a small deviation from the approach of this paper to strictly test each single precondition and discard models for which only one precondition is not found to be significant can be justified. Thirdly, if in fact the habit Saving Motive has no significant indirect effect in the models in which it will be reinserted, then the impact that this reinsertion has on the model is expected to be marginal. Lastly, the mediation effect of Saving Motive was only tested in in the individual mediation triangles. Its mediation effect might change, however, if the variable is included while the other behavioral habits are controlled for. This is the case in the final model.

Thus, the statistical analysis has resulted in 9 models of which each one includes the three behavioral habits Saving Method, Saving Regularity and Saving Motive. The percentage of the models’ total effects that is mediated by each of the behavioral habits can be found in table 6b. In these models, however, currently only the other saving habits are controlled for. Thus, in the last step it will be ensured that not only the other saving habits are controlled for when assessing the confounding percentage but that at the same time the remaining personality traits are controlled for. The results of this final adjustment can be found in table 6c. For example, the statistical outcome for Model 1 reveals that the total effect of the personality trait conscientiousness on the saving outcome Cover 3 Months is mediated by the saving habits Saving Method, Saving Regularity and Saving Motive with a total strength of 62.71%, 25.11% and 8.32%, respectively. In sum, 96.14% of the total effect is mediated by the saving habits. Negative confounding percentages in the statistical outcome indicate that inconsistent mediation effects are at hand. This is the case for Model (2), Model (3), Model (7) and Model (8). As the measure summary statistic “Total Confounding Percentage” is meaningless when inconsistent mediation effects are found, the measure is slightly adjusted for these models. More specifically, for these models the absolute value of the individual confounding effects are summed up for the total confounding percentage measure, thereby avoiding the inclusion of negative measures in the summary statistic. It is important to note that this adjustment allows the total confounding percentages to be above 100% and that it changes the interpretation for the confounding measures slightly. More concrete interpretations can be found in section 6 of this paper.

6. Discussion

In line with the existing literature in this field, the findings will first be evaluated in isolation. This enables initial interpretations of the results and contributes to each individual body of literature separately. Also, this approach allows identifying whether the results follow consistent pattern and sets the frame for explanations where they do not. Subsequently, the findings of the mediation analysis and the combined model are interpreted. With this approach, both theoretical contributions are made for the research in this field as well as practical contributions for policy assessment and policy design.

6.1. Discussion of Separate Relationships

Firstly, it is striking that all five personality traits have a significant effect on the likelihood that a household has accumulated the savings buffer recommended by the UK government. This finding emphasizes the importance that personality traits have on saving outcomes. Since the existing empirical evidence concerned with this relationship is characterized by large contradictions, it is not possible to generalize whether the findings of this paper are in line with previous findings. Importantly, however, the personality trait conscientiousness has a positive effect on the likelihood that an individual holds the savings buffer recommended by the UK government and a negative effect on the likelihood that the household has more debt than savings. This is in direct disagreement with evidence provided by Nyhus and Webley (2001), while it aligns with the evidence provided by various other researchers (Brandstätter, 1996; Brown & Taylor, 2014; Donnelly et al., 2012; Letkiewicz & Fox, 2014; Wärneryd, 1996). This finding is not surprising as conscientious people tend to be organized, dependable and not impulsive (John & Srivastava, 1999). All these are characteristics which are likely to result in a saving behavior which is in line with recommendations made by the UK government.

Further, it is striking that scoring high in three of the five personality traits, namely conscientiousness, agreeableness or neuroticism, has a reverse effect on the likelihood that an individual has more debt than savings as compared to the individual’s likelihood to hold the savings buffer recommended by the UK government. For example, extraversion has a negative effect on the likelihood that an individual holds the recommended savings buffer, while it has a positive effect on the likelihood that a respondent has more debt than savings. This pattern is expected since a personality trait which is associated with more savings is likely also to be associated with having more savings than debt holdings. However, the personality traits extraversion and openness do not follow this pattern. For both of these personality traits a negative effect is found on both the likelihood that a household achieves the recommended savings buffer as well as the likelihood that a household has more debt than savings. The negative effect on the likelihood that an individual has more debt than savings, however, is not even significant at the 10% significance level. Thus, this unexpected pattern in the statistical output is not particularly worrying for the analysis. Nevertheless, a significant positive coefficient was expected, given that the coefficients relating extraversion and openness to the saving outcome Cover 3 Months are negative. Thus, potential reasons for the described phenomenon are provided in the following. Firstly, it is possible that these personality traits have different effects on saving behavior depending on other demographic or psychological factors, so that for some households high scores in extraversion or openness result in higher savings, while they lead to higher debt holdings in other households. It might also be the case that the co­holding phenomenon as it was described by Gathergood and Weber (2014) occurred. However, while the authors relate compulsive buying behavior to the co-holding phenomenon, it is not entirely clear in what way the personality traits extraversion and openness affect this relationship. A third potential explanation for the insignificant coefficients is that the dependent variable More Debt was measured imprecisely. The reason is that all respondents for whom the data did not reveal whether they have more debt or savings were assumed not to have more debt than savings. As mentioned in section 4.3, this was the case for 9.62% of the respondents. It might have given more expected results if these cases were excluded from the dataset.

Furthermore, the data analysis shows that all personality traits have a significant relationship with almost all saving habits. This is in line with various findings in the existing literature which suggest that personality traits indeed do not only affect saving outcomes as such but particular saving habits too. The evidence from the existing literature was presented in section 2.2.2. It is striking that the variable Saving Method was found to be significantly affected by all five personality traits. This is line with findings made by Nyhus and Webley (2001) who conclude that personality traits have an effect on the saving products an individual uses. Evidence found in this paper, however, is not directly comparable with findings made by Nyhus and Webley (2001), since they have focused on specific saving products while this paper has analyzed whether or not respondents primarily use saving accounts for their savings at all or whether they keep their savings with their other money. It is worth noting at this point that high conscientiousness scores have a positive effect on the likelihood that a household holds the savings buffer recommended by the UK government, while it also increases the likelihood that a designated savings account is used for the household’s savings. For the other four personality traits, the exact opposite is true, meaning that higher scores in these personality traits both lower the likelihood that a household holds the recommended savings buffer as well as the likelihood that a household primarily uses a designated savings account for its savings. Thus, at this point it can already be expected that Saving Method has a strong mediation effect on the relationship between the personality traits and Cover 3 Months. In fact, this prediction is confirmed in further stages of the analysis and thus underlines the consistency of the findings.

The opposite pattern is found when substituting Saving Target for Saving Method in the previous paragraph. In that case, the personality trait conscientiousness which was found to increase the likelihood that the recommended savings buffer is achieved decreases the likelihood that a household sets specific saving targets. For the other four personality traits which were found to decrease the likelihood that the recommended savings buffer is achieved it is concluded that they increase the likelihood that a respondent sets specific saving targets. Thus, with just the results from the first two preconditions it can already be expected that setting a specific saving target has a negative effect on the likelihood that a household achieves the recommended savings buffer. In fact, this expectation is confirmed in the assessment of precondition 3 during the statistical analysis. In line with this argument, the analysis for precondition 3 also shows that the reverse is true for the effect of Saving Target on More Debt. These two findings confirm that no matter which personality trait is controlled for, setting a specific saving target increases the likelihood that a household has more debt than savings while it decreases the likelihood that a household meets the recommended savings target. The fact that it was possible to predict this from the analysis of the first two preconditions alone emphasizes the consistency of the findings. In sum, the evidence provided disagrees with suggestions made by the Fairbanking Foundation (2013) in that setting specific saving targets has an inverse effect on households’ savings.

For the remaining saving habits Saving Regularity and Saving Motive it is not possible to derive equally general interpretations as for Saving Method and Saving Target since only the personality traits extraversion, agreeableness and neuroticism are found to have a significant effect on them. From this, evidence that policy initiatives have to be tailored to specific personality traits is derived since not all personality traits equally affect particular saving habits or affect them at all. As the existing literature was not specifically concerned with the relationship of the Big Five personality traits and the saving habits Saving Regularity and Saving Motive, it is not possible to assess whether the findings of this paper align with previous research findings. It can be confirmed, however, that these relationships exist, as was already emphasized by Canova et al. (2005), Lunt and Livingstone (1991) and Wärneryd (1996).

Further, for all of the saving habits a consistent effect on the saving outcomes Cover 3 Months and More Debt is found regardless of which personality trait is accounted for. Thus, neither the sign, nor the significance of the effects changes when using different personality traits as control variables. The only exception for his is Saving Target as here some coefficients are found to be insignificant. A potential reason for this can derived from explanations found in Fairbanking Foundation (2013) and Which? (2014). There, setting specific saving targets is viewed in conjunction with saving regularity. It is claimed that whenever a saving target is achieved a new saving target is likely to be set. Therefore, it is argued that Saving Target can be viewed as a proxy for saving regularity. This might be a reason why the variable Saving Target itself is found to be insignificant in many cases. Importantly, the variable Saving Regularity is part of the final model, and thus the habit Saving Target is also partly accounted for.

In addition, for all saving habits the effects on the saving outcome Cover 3 Months are always the reverse of their effects on More Debt. For example, the saving habit Saving Regularity has a positive effect on Cover 3 Months regardless of which personality trait is accounted for and a negative effect on More Debt regardless of which personality trait is accounted for. This consistent pattern is most likely based on the notion that a saving habit which increases the likelihood that a certain amount of savings is reached by definition also increases the likelihood that total savings are higher than total debt. Thus, the recognition that this pattern can be found for every saving habit further emphasizes the consistency of the findings.

6.2. Discussion of Mediation Effects and Combined Model

When assessing the mediation effects in the final models it can be seen that striking evidence is found suggesting that the relationship between personality traits and saving outcomes is mediated by the saving habits Saving Method, Saving Regularity and Saving Motive. In fact, up to 96.14% and at the least 18.34% of the relationships are mediated by these three habits. It is important to note that these figures are calculated when controlling for all other saving habits and personality traits simultaneously. Thus, the confounding percentages mentioned here are discerning the unique effect that the personality traits at hand have on saving outcomes. Further, it is striking that the behavioral habit Saving Method has the strongest mediation effect for most of the relationships between personality traits and saving outcomes. From this, it can be derived that the way in which personality traits affect saving outcomes is to a high extent explained by the decisions that people with different personality traits make regarding their saving products. More generally, it can be concluded that a substantial portion of the effects that personality traits have on saving outcomes arise due to the effect which these personality traits have on saving habits.

A closer look at the mediation effects shows, however, that saving regularity has an inconsistent mediation effect in the four models which include the independent variables extraversion and agreeableness. This is, for example, the case when the effect of extraversion on the saving outcome Cover 3 Months is assessed. When the mediating variable Saving Regularity is included in the model, this effect is strengthened rather than being reduced. In order to fully understand these dynamics, a detailed step by step assessment must be conducted. Firstly, it is known from the statistical analysis that extraversion has a negative total effect on Cover 3 Months. Apparently, however, extraversion is at the same time positively affecting the regularity in which households save money. A higher saving regularity in turn has a positive effect on Cover 3 Months. Thus, a certain part of the effect that extraversion has on Cover 3 Months is positive. More specifically, extraversion positively affects the regularity in which individuals save which in turn has a positive effect on Cover 3 Months. However, at the same time extraversion has negative effects on Cover 3 Months in various other ways which are not controlled for and thus not captured by the model. These additional negative effects outweigh the positive effect that extraversion has on Cover 3 Months. The total effect of extraversion on Cover 3 Months is negative since it captures both the smaller positive portion of the effect and at the same time the negative effects which dominate. In the mediation analysis, the positive effect that extraversion has on Cover 3 Months is extracted from the total effect since the habit Saving Regularity is controlled for. Specifically, this means that a positive portion of the overall effect is excluded from the total effect. This positive portion was reducing the negative total effect previously. As soon as it is controlled for in the model at hand, the partial positive effect of extraversion is excluded from the total effect measure which - as a result - gets more negative. This argumentation can be summarized in the following way: High extraversion scores have overall a negative effect on the likelihood of a household to meet the recommended savings buffer. This negative effect, however, is partly reduced since the personality trait extraversion has also a positive effect on Cover 3 Months due to the mediator Saving Regularity. Nevertheless, the overall effect is negative since the partial positive effect is not strong enough. Similar interpretations can also be formulated for the three remaining models in which saving regularity has an inconsistent mediation effect. It is, however, not in the scope of this study to assess why i.e. the personality traits extraversion and agreeableness have a positive effect on Saving Regularity. The evidence that higher regularities of saving are associated with these two personality traits, however, is striking. Hence, policy initiatives can be designed which appeal to the tendency of people who score high on extraversion and agreeableness to save regularly. This finding is especially valuable, as people who score high on these two personality traits are unlikely to hold the recommended savings buffer.

As was already hinted at in the last paragraph, the findings of this paper can be used in order to understand why previously executed policy initiatives have worked so well and also to design new policy initiatives in a more structured way. To illustrate this point, the automatic enrollment saving scheme can be used as an example. As Madrian and Shea (2000) outline, introducing an automatic enrollment into 401 (k) saving accounts in the US has increased the participation in this saving scheme dramatically. This policy initiative shows how very small changes (making enrollment the default options even though employees can still easily opt out) can have dramatic effects on the saving behavior of individuals. Equipped with the models derived in this paper, some dynamics underlying the success of this policy can be illuminated. In fact, it can be seen that the setup of the automatic enrollment retirement savings plan appeals to all three saving habits derived in this paper. Firstly, the 401 (k) is a savings account. Thus the funds allocated to the account cannot be accessed before the retirement age is reached unless significant penalties are paid. This is linked to the variable Saving Method in this paper as savings are clearly not kept together with other money. In addition, contributions to the retirement saving plan are made monthly so that a high regularity is also ensured. Lastly, the saving motive with which the saving account is set up is by definition for retirement. With a broader definition of the term precautionary saving, this can be counted as a precautionary motive as well since it characterizes the savings for times in which so-called rainy days are expected.

The findings at hand do not, however, only help to understand the success of the 401 (k) savings plan better but also enable policy makers to design it in an even more effective way. Now, not only the saving habits which increase the likelihood of meeting the recommended savings buffer are known but also how personality traits play into this relationship. More specifically, the 401 (k) savings scheme or also the automatic enrollment as introduced with the Pension Act 2008 in the UK (UK Government, 2008) can be set up to more specifically appeal to people which score high on certain personality. With this approach individuals can be targeted who are especially unlikely to have accumulated the recommended savings buffers. For example, the finding that people who score high on extraversion generally have difficulties with meeting the recommended savings buffer while they are at the same time counterbalancing this tendency with their habit to save regularly. Thus, the special status of the habit Saving Regularity can be utilized in the context of policy design. The first step is to derive from the final models of this paper which personality traits are associated with which of the three saving habits. In doing so, two complementary approaches can be followed. If people who score high on personality traits which are positively associated with certain saving habits are targeted - as for example the positive relationship conscientiousness has with Saving Method - the saving policy can build on this by including features which appeal to this specific saving habit. On the other hand, a relationship between a personality trait and a saving habit can be chosen which has a negative association as for example the relationship between agreeableness and having a precautionary saving motive. Then, a saving policy can be designed in a way to specifically target people with high agreeableness scores and subsequently nudge them into pursuing precautionary saving behavior. In this case, a saving habit which is currently lacking is overcome.

Another example which emphasizes how the findings of this paper can be used in practice is derived from a report to the European Commission (Chater et al., 2010). It underlines that standard economic models of rational and self-interested economic agents fail to describe human decision making. They provide evidence for this by noting that the very way in which information is presented has a significant effect on individuals’ financial decision making. This argument is in line with the evidence provided by (Saez, 2009) as mentioned in section 2.2. This evidence can be used in conjunction with the findings of this paper. More specifically, the way in which information about new saving policies is presented can be set up in a way that directly appeals to the saving habits outlined in this paper. If, for example, those people who are especially unlikely to meet the recommended savings buffer are targeted, then the information can appeal specifically to the saving habits which these people already tend to adhere to. For example, people who score high on extraversion are unlikely to hold the recommended savings buffer. Nevertheless, they tend to score high on Saving Regularity. Thus, information about a new saving policy can be presented in a way to appeal to this particular saving habit.

Also, saving policies which introduce commitment saving schemes in which individuals are incentivized to increase their contribution rates over time or in which they are committed not to withdraw money from their savings accounts (Ashraf et al., 2003; Crossley et al., 2012) can be improved with the findings of this paper. In fact, many examples can be found in which these commitment accounts are dedicated to a specific saving motive. One example are Christmas saving schemes in the UK (Money Advice Service, 2016a) in which the savings can only be dipped into by the account holders after a day close to Christmas. Given the findings of this paper, it can be suggested that similar saving accounts can be set up which are not dedicated to Christmas savings as the driving motivation but for example precautionary savings in some form. Such a policy could, for example, be directed towards people who score high on neuroticism in order to motivate them to accumulate more precautionary savings. At the same time, people who score high on conscientiousness and thus already largely save with a precautionary motive can be given an opportunity to do so with a savings account dedicated to precautionary saving motives. Thus, saving policies which are already appealing to personality factors can be designed in a more effective way by tailoring them to the relationships that personality traits and certain saving habits have on saving outcomes.

7. Conclusion

The aim of this paper is to analyze whether and to what extent particular saving habits mediate the relationship between personality traits and saving outcomes. This aim was derived from four realizations. Firstly, the sharp decline in the household savings ratio in the UK requires government action. Secondly, researchers and policy makers increasingly turn to behavioral economics theory to understand the decline and to derive saving policy initiatives, as standard economic theory fails to explain the recent developments. Thirdly, several existing saving policies which are specifically based on psychological and behavioral incentives are found to be highly successful, revealing that slight changes in saving schemes can have dramatic effects on their effectiveness. Fourthly, the existing literature in the field of behavioral economics related to saving is scattered into explaining the separate parts of the relationship between personality traits, saving habits and saving outcomes in isolation. Not only does this impede that holistic theoretical frameworks can be derived but also does it hinder the use of the findings for the assessment of existing saving policies and the design of new saving schemes in a more structured way, as well as at a larger scale.

In order to overcome this gap, three related bodies of literature which each explain a single link between personality traits, saving habits and saving outcomes are combined into a single framework. The variables used as proxies for these factors are derived from the existing literature. In order to understand the dynamics which relate these variables to each other a dichotomous mediation analysis is conducted. For this, a representative survey of 1.000 UK respondents is used. It is found that the links between personality traits and saving outcomes are highly significant and that they are substantially mediated by three behavioral habits, namely whether or not households keep their savings separate from their other money, whether or not they save regularly and whether or not they save out of precautionary motives. Thus, it can be concluded that a substantial portion of the effects that personality traits have on saving outcomes arise due to the effect which these personality traits have on specific saving behaviors. These findings can be used to design saving schemes which not only address the saving habits that were found to be highly effective but at the same time connect them with an appeal to selected personality traits of individuals.

8. Limitations

Firstly, limitations regarding the interpretations and practical implications of this paper will be addressed. Here, it is especially important to emphasize that the findings derived in this paper present a framework for policy makers which assists them in designing saving schemes that nudge selected individuals to save more. However, it is not the goal of this paper to answer whether governments should aim to influence the saving behavior of their citizens at all and with which motivations they should do so. These are questions which cannot be answered without taking considerations from economic and political thought into account which are beyond the scope of this paper. Besides, this paper does not write out nor prescribe concrete saving policies. It does, however, provide a framework which suggests which saving habits in combination with which personality traits saving schemes have to appeal to in order to achieve higher effectiveness. In other words, rather than designing saving schemes itself this paper provides a framework which supports the design of saving schemes.

Also, limitations with respect to the methodology and research design will be outlined. Most importantly, it must be noted that this paper focuses on the extent to which saving habits mediate the relationships between personality traits and saving outcomes. This analysis is different from questions regarding the magnitude of the effects that personality traits have on saving outcomes. While i.e. the effect of the personality trait conscientiousness on the likelihood that a household has the recommended savings buffer is found to be mediated almost entirely by the three saving habits mentioned above, it is only assessed whether the total effect itself is significantly different from zero. The question whether this total effect has a considerable magnitude, however, is left unanswered. The second limitation is concerned with the level of measurement used. Since most of the answers in the underlying survey were not measured on interval or ratio scales, a dichotomous mediation analysis was used in this paper. This does, for example, only discriminate respondents into those who do hold the recommended savings buffer and those who do not. It fails, however, to account for differences between respondents’ savings beyond this basic classification. Thus, two respondents who both hold the recommended savings buffer are entering the analysis in the same way, even though it is possible that one of the respondents surpasses the recommended savings buffer by a much larger extent.

9. Reference List

Ashraf, N., Gons, N., Karlan, D. S., & Yin, W. (2003). A Review of Commitment Savings Products in Development Countries. Retrieved from

Baron, R. M., & Kenny, D. A. (1986). The moderator-mediator variable distinction in social psychological research: conceptual, strategic, and statistical considerations. Journal of personality and social psychology, 51(6), 1173-1182.

Benton, M., Meier, S., & Sprenger, C. (2007). Overborrowing and undersaving: Lessons and policy implications from research in behavioral economics. Federal Reserve Bank of Boston Community Affairs Discussion Paper, 7(4).

Beshears, J., Choi, J. J., Laibson, D., & Madrian, B. C. (2006). The Importance of Default Options for Retirement Savings Outcomes: Evidence from the United States Working paper series. (12009).

Bollen, K. A., & Stine, R. (1990). Direct and indirect effects: Classical and bootstrap estimates of variability. Sociological methodology, 20(1), 15-140.

Brandstätter, H. (1996). Saving Income, and Emotional Climate of Households Related to Personality Structure. CentER for Economic Research, Tilburg University.

Brandstätter, H., & Güth, W. (2000). A psychological approach to individual differences in intertemporal consumption patterns. Journal of Economic Psychology, 21(5), 465-479.

Brown, S., & Taylor, K. (2014). Household finances and the ‘Big Five’ personality traits. Journal of Economic Psychology, 45, 197-212.

Browning, M., & Lusardi, A. (1996). Household Saving: Micro Theories and Micro Facts. Journal of economic literature, 34(4), 1797-1855.

Canova, L., Rattazzi, A. M. M., & Webley, P. (2005). The hierarchical structure of saving motives. Journal of Economic Psychology, 26(1), 21-34.

Chater, N., Huck, S., & Inderst, R. (2010). Consumer decision-making in retail investment services: A behavioural economics perspective. Report to the European Commission/SANCO.

Cobb-Clark, D. A., & Schurer, S. (2012). The stability of big-five personality traits. Economics Letters, 115(1), 11-15.

Conger, A. J. (1974). A revised definition for suppressor variables: A guide to their identification and interpretation. Educational and psychological measurement, 34(1), 35-46.

Costa, P. T., & McCrae, R. R. (1992). Normal personality assessment in clinical practice: The NEO Personality Inventory. Psychological assessment, 4(1), 5.

Crossley, T. F., Emmerson, C., & Leicester, A. (2012). Raising household saving. Institute For Fiscal Studies, London February.

Dardanoni, V. (1991). Precautionary savings under income uncertainty: A cross-sectional analysis. Applied Economics, 23(1), 153-160.

Donnelly, G., Iyer, R., & Howell, R. T. (2012). The Big Five personality traits, material values, and financial well-being of self-described money managers. Journal of Economic Psychology, 33(6), 1129-1142.

Duckworth, A. L., Gendler, T. S., & Gross, J. J. (2016). Situational Strategies for Self-Control. Perspectives on psychological science : a journal of the Association for Psychological Science, 11(1), 35-55.

Fairbanking Foundation. (2013). Fairbanking Ratings: Reaching for the Stars. Retrieved from http://www.fairbanking.org.uk/FairbankingRatingsReport2013.pdf

Fisher, P. J., & Anong, S. T. (2012). Relationship of Saving Motives to Saving Habits. Journal of Financial Counseling & Planning, 23(1).

Gathergood, J., & Weber, J. (2014). Self-control, financial literacy & the co-holding puzzle. Journal of Economic Behavior and Organization, 107(Part B), 455-469.

Gliem, J. A. (2003). Calculating, interpreting, and reporting Cronbach’s alpha reliability coefficient for Likert-type scales.

Gosling, S. D., Rentfrow, P. J., & Swann, W. B. (2003). A very brief measure of the Big-Five personality domains. Journal of Research in Personality, 37(6), 504-528.

Guariglia, A. (1998). Understanding Saving Behaviour in the UK: Evidence from the BHPS: Citeseer.

Guidolin, M., & La Jeunesse, E. A. (2008). The Decline in the U.S. Personal Saving Rate: Is It Real and Is It a Puzzle? CFA Digest, 38(2).

Herr, N. R. (2016). Mediation with Dichotomous Outcomes. Retrieved from http://www.nrhpsych.com/mediation/logmed.html

Hirsh, J. B. (2015). Extraverted populations have lower savings rates. Personality and individual differences., 81, 162-168.

Hubbard, R. G., Skinner, J., & Zeldes, S. P. (1994). Expanding the life-cycle model: Precautionary saving and public policy. The American Economic Review, 84(2), 174-179.

Huffman, D., & Barenstein, M. (2004). Riches to Rags Every Month? The Fall in Consumption Expenditures Between Paydays. Bonn :: IZA.

James, L. R., & Brett, J. M. (1984). Mediators, moderators, and tests for mediation. Journal of Applied Psychology, 69(2), 307-321.

John, O. P., & Srivastava, S. (1999). The Big Five trait taxonomy: History, measurement, and theoretical perspectives. Handbook of personality: Theory and research, 2(1999), 102­138.

Jose, P. E. (2013). Doing statistical mediation and moderation: Guilford Press.

Judd, C. M., & Kenny, D. A. (1981a). Estimating the effects of social intervention: CUP Archive.

Judd, C. M., & Kenny, D. A. (1981b). Process analysis estimating mediation in treatment evaluations. Evaluation review, 5(5), 602-619.

Karlson, K. B., & Holm, A. (2011). Decomposing primary and secondary effects: A new decomposition method. Research in Social Stratification and Mobility, 29(2), 221-237.

Kenny, D. A. (2016a). Mediation. Retrieved from http://davidakenny.net/cm/mediate.htm

Kenny, D. A. (2016b). Mediation - Extensions - Dichotomous Variables. Retrieved from http://davidakenny.net/cm/mediate.htm#EX

Keynes, J. M. (1936). The general theory of employment, interest and money. London :: Macmillan for the Royal Economic Society.

Kohler, U., Karlson, K. B., & Holm, A. (2011). Comparing coefficients of nested nonlinear probability models. Stata Journal, 11(3), 420-438.

Letkiewicz, J. C., & Fox, J. J. (2014). Conscientiousness, financial literacy, and asset accumulation of young adults. Journal of Consumer Affairs, 48(2), 274-300.

Lunt, P. K., & Livingstone, S. M. (1991). Psychological, social and economic determinants of saving: Comparing recurrent and total savings. Journal of Economic Psychology, 12(4).

MacKinnon, D. P. (2008). Introduction to statistical mediation analysis. New York [etc.] :: Erlbaum.

MacKinnon, D. P., & Dwyer, J. H. (1993). Estimating mediated effects in prevention studies. Evaluation review, 17(2), 144-158.

MacKinnon, D. P., Krull, J. L., & Lockwood, C. M. (2000). Equivalence of the Mediation, Confounding and Suppression Effect. PREVENTION SCIENCE, 1(4), 173-182.

Madrian, B. C., & Shea, D. F. (2000). The power of suggestion : inertia in 401(k) participation and savings behavior. Cambridge, MA :: National Bureau of Economic Research. Modigliani, F. (1988). The Role of Intergenerational Transfers and Life Cycle Saving in the Accumulation of Wealth. Journal of Economic Perspectives, 2(2), 15-40.

Modigliani, F., & Brumberg, R. (1954). Utility analysis and the consumption function: An interpretation of cross-section data. Franco Modigliani, 1.

Money Advice Service. (2016a). Saving For Christmans. Retrieved from https://www.moneyadviceservice.org.uk/en/articles/christmas-savings-clubs-and-schemes

Money Advice Service. (2016b). Why it pays to save regularly. Retrieved from https://www.moneyadviceservice.org.uk/en/articles/why-it-pays-to-save-regularly

Moore, A., Beverly, S., Sherraden, M., Sherraden, M., Johnson, L., & Schreiner, M. (2001). Saving and asset-accumulation strategies used by low-income individuals. St. Louis, MO: Center for Social Development, Washington University in St. Louis.

Mowen, J. C., & Spears, N. (1999). Understanding compulsive buying among college students: A hierarchical approach. Journal of Consumer Psychology, 8(4), 407-430.

Nationwide Building Society. (2016). What is an ISA? Retrieved from http://www.nationwide.co.uk/products/isas/isas-explained Nyhus, E. K., & Webley, P. (2001). The role of personality in household saving and borrowing behaviour. European Journal of Personality, 15(S1), S85-S103.

Office of National Statistics. (2016). Time Series: Households & NPISH saving ratio: CP: SA. Retrieved from

https://www.ons.gov.uk/economy/grossdomesticproductgdp/timeseries/nrjs Otero-López, J. M., & Villardefrancos Pol, E. b. (2013). Compulsive buying and the Five Factor Model of personality: A facet analysis. Personality and individual differences., 55(5), 585-590.

Saez, E. (2009). Details Matter: The Impact of Presentation and Information on the Take-up of Financial Incentives for Retirement Saving. American Economic Journal: Economic Policy, 1(1), 204-228.

Samuelson, P. A. (1937). A Note on Measurement of Utility. The Review of Economic Studies, 4(2), 155-161.

Shrout, P. E., & Bolger, N. (2002). Mediation in experimental and nonexperimental studies: New procedures and recommendations. Psychological Methods, 7(4), 422-445.

Swain, S. D., Weathers, D., & Niedrich, R. W. (2008). Assessing Three Sources of Misresponse to Reversed Likert Items. JMR. Journal of marketing research, 45(1), 116.

Taylor, K. (2015). Charitable behaviour and the big five personality traits evidence from UK panel data. Bonn: IZA.

Thaler, R. H. (1985). Mental Accounting and Consumer Choice. Marketing Science, 4(3), 199­214. Retrieved from

Thaler, R. H. (1994). Psychology and Savings Policies. The American Economic Review, 84(2), 186-192.

Thaler, R. H., & Benartzi, S. (2004). Save More Tomorrow: Using Behavioral Economics to Veldhoven, G. M. v., & Groenland, E. A. G. (1993). Exploring saving behaviour: A framework and a research agenda / G. M. van Veldhoven, E. A. G. Groenland. Journal of Economic Psychology, 14(3Special Issue), 507.

Wärneryd, K. (1989). On the psychology of saving: An essay on economic behavior. Journal of Economic Psychology, 10(4), 515-541.

Wärneryd, K. (1996). Personality and Saving. CentER for Economic Research, Tilburg University.

Watkins, C. (2015). The Saving Ratio: How is it affected by Households’ and Non-Profit Institutions Serving Households’ income and expenditure? National Accounts articles, Ι­ό. Retrieved from

Which? (2014). The three habits of successful savers: How learning from their behaviour could get the UK saving, 1-33. Retrieved from http://www.which.co.uk/ website: http://www.staticwhich.co.uk/documents/pdf/the-three-habits-of-successful-savers-how- learning-from-their-behaviour-could-get-the-uk-saving-377855.pdf

Xiao, J. J., & Noring, F. E. (1994). Perceived saving motives and hierarchical financial needs. Financial Counseling and Planning, 5(1), 25-44.

Xu, Y., Beller, A. H., Roberts, B. W., & Brown, J. R. (2015). Personality and young adult financial distress. Journal of Economic Psychology, 51, 90-100.

I would like to thank Patrick Gerhard who always took the time to support me with my research.

10. Appendices

Figure 1: All Questions from the Questionnaire Used in this Paper

illustration not visible in this excerpt

Tables 1a – 1b: Logistic Regressions for the Assessment of Precondition 1 in Stage 1

illustration not visible in this excerpt

Tables 2a– 2b: Logistic Regressions for the Assessment of Precondition 2 in Stage 1

illustration not visible in this excerpt

Tables 2c– 2d: Logistic Regressions for the Assessment of Precondition 2 in Stage 1

illustration not visible in this excerpt

Tables 3a– 3b: Logistic Regressions for the Assessment of Precondition 3 in Stage 1

illustration not visible in this excerpt

Tables 3c– 3d: Logistic Regressions for the Assessment of Precondition 3 in Stage 1

illustration not visible in this excerpt

Tables 3e– 3f: Logistic Regressions for the Assessment of Precondition 3 in Stage 1

illustration not visible in this excerpt

Tables 3g– 3h: Logistic Regressions for the Assessment of Precondition 3 in Stage 1

illustration not visible in this excerpt

Tables 4c– 4d: Results from Bootstrapping for the Analysis of Significance in Stage 2

illustration not visible in this excerpt

Note: The model names indicate which mediation triangle is tested. For this, the following abbreviations are uses: For the personality traits: Consc = Conscientiousness, Extr = Extraversion, Agre = Agreeableness, Neur = Neuroticism, Open = Openness. For the saving habits: SMe = Saving Method, ST = Saving Target, SR = Saving Regularity, SMo = Saving Motive. For the saving outcomes: C3M = Cover 3 Months, MD = More Debt.

illustration not visible in this excerpt

Note: The model names indicate which mediation triangle is tested. For this, the following abbreviations are uses: For the personality traits: Consc = Conscientiousness, Extr = Extraversion, Agre = Agreeableness, Neur = Neuroticism, Open = Openness. For the saving habits: SMe = Saving Method, ST = Saving Target, SR = Saving Regularity, SMo = Saving Motive. For the saving outcomes: C3M = Cover 3 Months, MD = More Debt.

Tables 5a- 5b: Confounding Percentages for the Analysis of Strength in Stage 2

illustration not visible in this excerpt

Note: The model names indicate which mediation triangle is tested. For this, the following abbreviations are uses: For the personality traits: Consc = Conscientiousness, Extr = Extraversion, Agre = Agreeableness, Neur = Neuroticism, Open = Openness. For the saving habits: SMe = Saving Method, ST = Saving Target, SR = Saving Regularity, SMo = Saving Motive. For the saving outcomes: C3M = Cover 3 Months, MD = More Debt.

Tables 5a– 5b: Confounding Percentages for the Analysis of Strength in Stage 2

Stage 2: Assessment of Mediation Strength through Confounding Percentage Calculation

illustration not visible in this excerpt

Note: The model names indicate which mediation triangle is tested. For this, the following abbreviations are uses: For the personality traits: Consc = Conscientiousness, Extr = Extraversion, Agre = Agreeableness, Neur = Neuroticism, Open = Openness. For the saving habits: SMe = Saving Method, ST = Saving Target, SR = Saving Regularity, SMo = Saving Motive. For the saving outcomes: C3M = Cover 3 Months, MD = More Debt.

Table 6a: Disentangled Mediation Strengths in Combined Model without Adjustments

illustration not visible in this excerpt

Table 6b: Disentangled Mediation Strengths in Combined Model after Adjustments

illustration not visible in this excerpt

Table 6c: Disentangled Mediation Strengths in Final Model

illustration not visible in this excerpt

62 of 62 pages

Details

Title
How Behavioral Habits Mediate the Relationship between Personality Traits and Savings
Subtitle
Evidence from the UK
College
Maastricht University
Grade
4.00
Author
Year
2016
Pages
62
Catalog Number
V339859
ISBN (Book)
9783668296794
File size
801 KB
Language
English
Tags
Behavioral economics, savings, household saving, personality traits, big five, nudging, saving policy
Quote paper
Ufuk Altunbüken (Author), 2016, How Behavioral Habits Mediate the Relationship between Personality Traits and Savings, Munich, GRIN Verlag, https://www.grin.com/document/339859

Comments

  • No comments yet.
Read the ebook
Title: How Behavioral Habits Mediate the Relationship between Personality Traits and Savings


Upload papers

Your term paper / thesis:

- Publication as eBook and book
- High royalties for the sales
- Completely free - with ISBN
- It only takes five minutes
- Every paper finds readers

Publish now - it's free