Explaining heterogeneous effects of education:
A study of the role of psychological resources in the relation between
education and important life outcomes.
Rosalie Joosten
10001637
Master thesis
Dr. T. Leopold
Dr. B. Völker
28 December 2015, Amsterdam
Research Master Social Sciences
Abstract
Taking the interplay between educational and psychological resources as its focus, the current study aims to understand heterogeneity in the effects of education on important life outcomes. Preliminary evidence on this interrelation suggests both mediating and moderating effects are at play. This study builds on this line of research by analysing new data1 for perceived control and self-control in relation to the life domains health (N=900) and personal finances (N=892) on the Dutch adult population (age >25). Results confirm the mediating mechanism for perceived control. Furthermore, although inconclusive, moderation results suggest a tentative pattern in the direction of a compensatory mechanism for both self-control and perceived control, mainly for the low educated. The findings are interpreted in relation to limitations of the current study, directions for future research and policy implications with regard to social disadvantage are suggested.
Key words: education, perceived control, self-control, moderation, mediation, self-rated health, personal finances
Introduction
The general, well-documented pattern is that higher educated individuals perform better with regard to all sorts of important life outcomes, such as health, the labour market, relations and criminality (Cutler & Lleras-Muney, 2006; Psacharopoulos & Patrinos; McNulty & Bellair, 2003; Lochner, Lance & Moretti; 2004). Heterogeneity in the effects of education however suggests that there are exceptions to this rule – higher educated that despite their schooling do not seem to get by, and lower educated that despite their lack of educational resources appear to do (very) well. Apparently, not everyone fits the general pattern.
The current study departs from the Bourdieuan idea that individuals need resources in order to be successful in life, deal with problems or prevent problems from occurring. Education and social networks are extensively researched, almost irrefutable examples of such resources. Within this framework however less attention exists with regard to what we call psychological resources.
Extensive research from the fields of personality, developmental and social psychology has shown the relevance of psychological resources for important life outcomes (for reviews see Almlund et al., 2011; Roberts et al., 2007; Tangey et al., 2004; De Ridder et al., 2012). Furthermore, recently a research line started developing at the intersection of psychology and economics, illustrating the independent effect of so-called non-cognitive skills or traits for a variety of life outcomes (see for example Borghans et al., 2008; Heckman & Rubinstein, 2001; Heckman & Kautz, 2012). Moreover, the increasing scientific interest in the topic of relevant skills other than IQ goes hand in hand with increased policy attention for the topic, a recent example of which is the OECD-report on skills that devoted a substantial part to non-cognitive skills (Kautz et al., 2014). The argument that acts like a red thread through these literatures is that apart from IQ, non-cognitive traits are relevant for important life outcomes as well. Although there is scholarly consensus that this type of resources2 matters, relatively little is known about how they relate to other resources that determine important life outcomes, such as education.
The research aim of the current study follows from these observations. On the one hand, it aims to understand how education and psychological resources interrelate in their influence on
important life outcomes. On the other hand, more specifically, it aims to investigate the extent to which psychological resources explain the heterogeneous effects of education. Two specific
psychological are selected to this purpose: perceived control and self-control. Perceived control, also known as mastery (Pearlin & Schooler, 1987), sense of control (Ross & Mirowsky, 1999) or locus of control (Rotter, 1954) refers to the extent to which one believes that generally people can influence outcomes through their own actions rather than that outcomes are determined by external factors.3 Self-control refers to the capacity for altering one’s own responses, especially to bring them into line with standards such as ideals, values, morals and social expectations, and to support the pursuit of long-term goals (Baumeister, Vohs and Tice, 2007). Whereas perceived control refers to ones beliefs about the world, self-control refers to an individual ability to control impulses. As such, both illustrate different types of psychological resources that influence
important life outcomes. Previous research with regard to these specific resources has shown their influence on a variety of important life outcomes, as outlined in the theory section (see for example Moffitt et al., 2011; Ross & Mirowsky, 1999).
In the case of perceived control previous research suggests that both mediation and moderation mechanisms may be at play with regard to the relation between education and important life outcomes (Matthews et al, 2010; Lachman & Weaver, 1998; Turiano et al., 2014). Therefore, analysing new data on a sample of 931 Dutch adult respondents, the current study investigates mediating as well as moderating effects of perceived control. To our knowledge, no studies taking the interplay between education and self-control as their research problem are currently available. Previous research on the independent effects of both factors however points in the direction of a moderation mechanism, which is tested in the current study. Finally, in order to illustrate how these mechanisms play out similarly for different life domains, analyses are conducted with regard to health as well as personal finances.
Taking the interplay between education and psychological resources at its central focus, the current study aims to contribute to existing research in multiple ways. Firstly, although the interaction between these types of resources has not been studied widely, it forms an important societal topic. Especially in the case of the Netherlands, where since the introduction of the participation society individual differences have come to be put in the spotlight. Secondly, this study contributes to the emerging line of research by analysing new data especially collected for this purpose by Dutch Scientific Council of the Government. Thirdly, tentative new insights are provided to the extent that deviations from the general educational pattern are explained by
3 As convincingly argued by Ross and Mirowsky (1999), essentially all these terms refer to the same concept. This argument is
widely accepted as illustrated by the interchanged use of the terms in the literature. Therefore, different terms are used interchangeably in the current study as well.
differences in self-control and perceived control. Preliminary evidence is presented with regard to the enhancing or downplaying influence of psychological resources on education as an early determinant of social disadvantage, for good or for ill.
Theory
Previous research on the relation between education, psychological resources and important life outcomes has shown how both mediating and moderating mechanism are at play (Lachman & Weaver, 1998; Matthews & Taylor, 2010; Ross & Mirowsky, 1999; Taylor & Seeman, 1999; Turiano et al, 2014). Literature in the field of the educational gradient suggests that similar mechanisms are at work for multiple important life outcomes, such as physical health, self-rated health, impulse buying, debts (Ross & Wu, 1995; Wood, 1998). This study builds on this research by focussing on the interrelated influence of education and the psychological resources perceived control and self-control on health and personal finances.
How education keeps away doctors and creditors
The literature on the educational gradient of important life outcomes roughly distinguishes between three mechanisms through which education operates. Firstly, the classical human capital argument is that education generates productive skills, which in turn yield economic resources (Becker, 2009; Grossman 1972). The highly educated generally earn substantially more than low educated (Cutler-Lleras-Muney, 2010), which enables them to buy healthy food, health insurances or other products that may be less accessible for low educated with less income. Moreover, with a higher budget one is more likely to be able to save, which in turn forms a buffer against adverse life events and reduces the chances of ending up in debts (Cole, Paulson & Shastry, 2012). The second mechanism describes how highly educated perform better with regard to important life outcomes because of their knowledge and information advantages. Not only do the highly
educated spend more time in school, the knowledge they gain is also more advanced compared to the lower educated. Moreover highly educated may have access to more information, and even if information is as accessible for everyone, better educated individuals are more likely to internalize the message (Cutler & Lleras-Muney, 2010). Finally, education is associated with important life outcomes through the behavioral pathways of time discounting, risk aversion or value of the future, although in their own study Cutler and Lleras-Muney (2010) find little evidence for these mechanisms. The psychological resources perceived control and self-control relate to these mechanisms in multiple ways, as will become clear in the next section.
How perceived control mediates the effect of education on important life outcomes
Reviewing the current state of scientific knowledge on a wide range of psychological resources, Matthews et al (2010) concluded that the findings thus far are mixed. The four studies that
included a measure of perceived control however yielded promising results. Schnittker et al (2004) showed how including mastery in the model significantly reduced the effect of education on multiple chronic conditions. For the Netherlands, Bosma, Schrijvers and Mackenbach (1999) found that the effect of SES on mortality was reduced by 37-57% after the inclusion of the perceived control measure. Moreover, the excess risk of CHD for low-SES individuals was reduced by 30% when controlled for locus of control (Bosma, Van Jaarsveld & Tuinstra, 2005). Finally, van Oort et al. (2005) showed how accounting for locus of control reduced the excess mortality risk of lower educated respondents by 21-48%.
The body of evidence on the mediating effect of psychological resources on the relation between education and important life outcomes is thin; however promising and worthy of further examination. The current study contributes to this by examining the mechanism on recent data for the Netherlands. First however the mechanism through which perceived control mediates the association between education and important life outcomes is elaborated.
Previous research has shown that less educated individuals, on average, demonstrate lower levels of perceived control than more educated individuals (Lachman & Weaver, 1998). The first mechanism through which education yields a sense of control originates in the human capital perspective. Mirowsky and Ross (1998, 2007) argue how individuals develop a sense of control through education, which in turn affects all sorts of health outcomes. Interpreting this
mechanism in a more general sense, Ross & Wu (1995) argue that education, through the
provision of skills and resources, produces effective and confident problem solving agents. More specifically, in education people learn communication skills such as reading, writing and debating, as well as analytical skills, such as interpreting, classifying, observing and building logical
arguments. Moreover, acquiring problem solving skills takes a central position in education. Not only do individuals learn and practice problem-solving skills in school, it also provides them with experiences of success in solving problems. This way, in education individuals develop skills as well as confidence that together provide them with a sense of control over their outcomes. Individuals with a high educational degree have spent more time in school developing skills. Moreover, the skills learnt and the problems solved in a higher level of education tend to be more advanced. As such, a higher level of education induces a higher sense of control compared to a low level of education, which in turn, leads them to deal more effectively with life tasks.
Literature from a more sociological perspective suggests a second mechanism through which education affects sense of control: ranking. In general, more educated end up higher in the social distribution and have therefore a beneficial rank compared to the less educated (Cutler & Lleras-Muney, 2006; Pearlin, 1989). A lower rank position is overall associated with more challenges rather than opportunities. Moreover, being in a lower rank position may lead
individuals to perceived outcomes as arbitrarily happening to them, as opposed to the feeling of control that is associated with being in a higher ranking (Ross & Wu, 1995).
Individuals with a high level of perceived control have been shown to deal more effectively with unemployment, debts, financial decisions, the losses associated with chronic illness and health behavior (Caliendo, Cobb-Clark & Uhlendorff, 2010; Davies & Lea, 1995; Cole, Paulson & Shastry, 2012; Ross & Wu, 1995; Thompson & Scglehofer, 2008). Moreover, high levels of perceived control have been associated with better adaptive functioning in confrontation with stress (Diehl & Hay, 2010). In general, people with a greater sense of control tend to put effort in dealing with problems, whereas people that perceive less control tend to adapt to a new situation and avoid efforts to change it.
In conclusion, previous research has suggested a mediating influence of perceived control on the relation between education and important life outcomes through two mechanisms, namely effective agency and ranking of positions in the social distribution. Therefore, our first hypothesis is phrased as follows:
H1: The favorable effect of education on the probability of having a) good self-rated health and b) financial problems is mediated by the extent to which one perceives control over their outcomes.
Exceptions to the general pattern: low educated doing well, high educated doing not so well
As described earlier, high level of education is generally associated with a variety of favourable important life outcomes. However, as the term ‘general’ implies: this pattern does not apply equally to everyone. That is to say, although highly educated are less likely to be in bad physical health (Ross & Mirowsky, 2007) or have debts (Cole, Paulson & Shastry, 2012), some end up in trouble nevertheless. The same occurs the other way around as well: although less educated individuals are more likely to experience disease and disability (Berkman & Kawachi, 2001), some fare (very) well nevertheless. Rather than examining this general pattern further, the second
research aim of the current study is to understand the extent to which psychological resources can account for the deviations from the general pattern.
The moderating effect of self-control
Previous research on the relation between education and dispositional self-control in their influence on important life outcomes has mainly focused on their independent main effects (Heckman & Lochner 2000; Heckman & Rubinstein 2001; Moffitt et al., 2012; Roberts et al., 2007; Tangey et al., 2004). The current section outlines why it merits the examination of the extent to which both factors act in a multiplicative manner as well.
As a fundamental dimension of the Big Five personality trait conscientiousness self-control has a large heritability factor. As shown by an extensive review by Roberts, Walton and Viechtbauer (2006), self-control is developed in early childhood and remains relatively stable thereafter. In general, high levels of self-control are associated with a higher ability of task persistence, temptations resistance, impulse overcoming, emotion regulation and choice making (De Ridder et al, 2012). Moreover, people with high levels of self-control are less likely to experience eating disorders, report aggression and substance abuse (Tangey et al., 2004).
Furthermore, low levels of self-control are associated with deviant behavior in other life domains than health as well, such as impulsive buying (Wood, 1998). De Ridder et al. even conclude, after reviewing around hundred studies of self-control that ‘self-control is thus one of the most beneficial traits in personality’ (2012: 92).
As mentioned earlier, to our knowledge no studies that focus on the interrelated influence of education and self-control on any important life outcome is currently available. On the basis of research thus far we know that self-control is developed in early childhood and remains relatively stable over the life course. Moreover, we know that it exerts a strong, independent influence on important life outcomes. Combining these literatures our tentative moderation analysis for self-control is phrased as follows:
H2: The favorable effect of being highly educated compared to low educated on the probability of having a) good rated health and b) financial problems is weaker for people with less self-control.
The moderating effect of perceived control
On the contrary to the interaction between education and self-control, the moderating of perceived control has been previously studied. Two studies investigating this effect suggested
significant moderation effects in relation to the health domain (Lachman & Weaver, 1998; Turiano et al., 2010). Moreover, studying the moderating effect of sense of control on the effect of social class on self-rated health and psychological well being, Lachman & Weaver (1998) even found that individuals from a lower social class with a high sense of control ‘resembled their higher social class counterparts than others in their own income group’ (p. 771). Investigating the moderating effect of perceived control on the relation between level of education and mortality risk Turiano et al. (2014) found similar results. However, their conclusion was even more restrictive in the sense that the moderating effect of perceived control only applied to the lower educated individuals in the sample, and not to the highly educated. Although the empirical basis of this moderating effect is thin, it is promising and merits further examination. Therefore, our third hypothesis is phrased as follows:
H3: The unfavorable effect of being low educated on the probability of having a) good self-rated health and b) financial problems is weaker for individuals with higher levels of perceived control.
Strucutralist versus self-deterministic
Coming from the literature on health, two perspectives on the interaction between education and psychological resources prevail: the structuralist hypothesis and the self-deterministic hypothesis (Cutler & Lleras-Muney, 2010; Turiano et al., 2014).
From the structuralist perspective, psychological resources are more likely to have an enhancing effect on health outcomes for the high educated, because the structural, unfavourable context of the low educated is likely to overpower individual attitudes or behaviors in their influence (Macintyre, 1997). This conceptualization of the interaction of educational and psychological resources builds on the idea that because education predisposes for a certain pathway in terms of career and earnings, one is likely to be surrounded by people with similar (social as well as economic) resources. This way a ‘life niche’ is created which is typically
associated with its own opportunities and challenges (Galobardes, Shaw, Lawlor, Lynch & Davey Smith, 2006a; 2006b). For example, being able to withstand unhealthy food because of high level of self-control may not make a difference if one’s environment in terms of social norms or material resources does not allow for buying healthy food or health insurances. Since the typical environment of the more educated, compared to the less educated, is generally characterized by opportunities rather than challenges, the effect of individual resources is expected to be stronger. Within this view, the societal structure is the main determinant for one’s outcomes here, rather than individual resources, explaining why it is known as the structuralist hypothesis.
The self-deterministic hypothesis works out the exact opposite thesis from the
structuralistic perspective. In the self-deterministic view, the typical environment of the highly educated may trump the potential influence of individual differences. According to this
conceptualization, should a highly educated individual due to a lack of self-control engage in ill-advised spending, material as well as social resources provide a large enough buffer in order to prevent the outcomes from becoming problematic. In case of the low educated, this view hypothesizes that individual psychological may indeed make a differences. Called trait-like tendencies by Donnellan et al (2009), these factors have been shown to have a stronger effect lower down the SES-gradiënt. The emphasis in this perspective is that individual resources may overpower social disadvantage, making it a self-deterministic view. Preliminary evidence on the moderating effects of individual psychological resources on the effect of education on important life outcomes points in the direction of the self-deterministic hypothesis (Lachman & Weaver, 1998; Turiano et al, 2014). Therefore, our fourth hypothesis is phrased in favour of the self-deterministic perspective.
H4: the difference in the effects of self-control and perceived control on the probability of having a) good self-rated health and b) financial problems is larger among the low educated than it is among the higher educated.
Methods
Operationalization
In order to study the relationship between education, psychological resources and important life outcomes in contemporary Dutch society new data was collected, commissioned by the Dutch Scientific Council of the Government (Wetenschappelijke Raad voor het Regeringsbeleid). A description of the full sampling and data collection procedure is included in the appendix. In total, 1014 respondents filled out the CASI (Computer Assisted Self Interviewing) survey. Because of the selective response and the increase of the two subsamples (people from the lowest social classes and people receiving welfare benefits), the composition of the sample differs from the population. Therefore, weights were applied so that it does represent the Dutch adult population.4 In order to be able to study people who were most likely to have completed their
education, respondents below the age of 25 were removed from the sample, making a total of 931 respondents.
Dependent variable: ‘Having problems’
Rather than testing different explanations with regard to a specific phenomenon captured by the dependent variable, the main interest of the current study is on the independent side. Moreover, the mechanisms studied here apply to multiple life domains in a similar way, as illustrated by a variety of previous research. Therefore, two rather than one outcome variables are selected that directly affect daily quality of life, namely health and personal finances. Moreover, in light of recent decentralizing policy trends in the Netherlands, these domains present especially interesting cases.
Important life outcomes in the health domain is measured by the item self-rated health: ‘Hoe is uw gezondheid over het algemeen?’ answered on a scale ranging from 1 (very good) to 5 (very bad). Answers were recoded into a binary variable so that a value of 1 indicates having a (very) good self-rated health and a value of 0 all other answers.5
The important life outcome in the domain of personal finances was measured using the item ‘having financial problems’: ‘Kunt u aangeven of u problemen ervaart op de volgende gebieden? Met uw financiën (voorbeelden; te weinig (zak)geld, kan schulden niet of nauwelijks betalen, kan touwtjes niet aan elkaar knopen, heb geen vast inkomen)’. Respondents reported their experiences in four categories: ‘Ja’, ‘Ja, enigszins’, ‘Nee’ or ‘Weet niet’. Answers were recoded into a binary variable so that (1) ‘Ja’ and ‘Ja, een beetje’ indicated having problems and (0) indicated no problems at all. Answers in the category ‘Weet niet’ were treated as missings.
Independent variables: education
Education is measured as the highest level of education completed, with seven answer categories. Answers were recoded into three categories: ‘LBO, VBO, VMBO (kader- en beroepsgerichte leerweg), MBO 1’, ‘MAVO, eerste 3 jaar HAVO en VWO, VMBO’ and HAVO en VWO
bovenbouw’ indicates a low level of education (value 1). ‘MBO 2, 3, 4 of MBO oude structuur’ is regarded ‘medium’ educated (value 2) and ‘HBO-/WO-propedeuse, HBO-/WO-bachelor or ‘kandidaats’ and ‘HBO-\WO-master of doctoraal’ indicates a high level of education (value 3). In total, 29% of the respondents were low educated. 37% of the respondents had a medium level of education and 34 % were highly educated.
Independent variables: perceived control and self-control
In order to measure perceived control, a Dutch translation of the widely used and validated Pearlin and Schooler Mastery scale (1987) was included in the survey. This scale consists of seven items such as ‘sommige van mijn problemen kan ik met geen mogelijkheid oplossen’ and ‘er is weinig dat ik kan doen om belangrijke dingen in mijn leven te veranderen’.6 Items were answered on a scale ranging from ‘heel erg op mij van toepassing’ (value 1), to ‘helemaal niet op mij van toepassing’ (value 5). In total, a scale was constructed on the basis of 5 items7, with an internal consistency value of α =.81.
Self-control was operationalized using the Brief Self-Control Scale was used (Tangey et al., 2004). This scale consists of 13 items, such as ‘ik kan verleidingen goed weerstaan’ and ‘mensen zeggen dat ik een ijzeren zelfdiscipline heb’. Like for the mastery index, items were answered on a scale ranging from ‘heel erg op mij van toepassing’ (value 1), to ‘helemaal niet op mij van
toepassing’ (value 5). These answers were recoded such that a higher score indicates more self-control. One index was created out of these 13 items, with an internal consistency value of α =.80.
Controls
Age and gender are known confounders of the relation between education and having problems as well as the relations between both perceived control and self-control and having problems (Mirowsky, 2013; Ross & Mirowsky, 2002). They are accounted for in the analyses as control variables. Gender is recoded such that 1 indicates female and 0 indicates male. Age is controlled for in a linear sense. 8,9
Data analysis
Both outcome variables under study are recoded in a binary way. In order to be able to compare the regression coefficients across the models and for different outcomes, we estimated Linear Probability Models. Coefficients can thus be interpreted as percentage points change in the probability of having (very) good health or financial problems.
6 The full survey is included in the appendix.
7 Factor analysis showed that a scale based on five items had a better fit to our data.
8 Ideally, we would control for all other variables that determine education but cannot be affected by it, such as parental
characteristics. No data on these variables was available however.
9 In order to account for the nonlinear effect analyses were employed including age squared as well. The models did not improve
Results
Descriptives
Table one describes the weighted samples of the self-rated health analyses (N=900) and financial problem (N=892) analyses, which are very similar. The mean age is 52 (SD=15.42) years old in both samples. With regard to gender, 50 per cent of the self-rated health sample is female, for the financial problems sample this percentage is 49. The average self-control score for both samples is 3.43 (SD=.51) on a scale ranging from 1-5. For mastery, the average lies around 3.53 (SD=.68) for both samples. Around 56 per cent of the in the health sample reports a (very) good self-rated health (SD=.50). In case of the financial problems sample, 30 per cent reports having (some) financial problems (SD=.46).
Table 1. Descriptive statistics
aGender: male=0, female =1
Self-rated health sample Financial problems sample
Mean SD Min Max Mean SD Min Max
Education 2.058 .792 1 3 2.063 .793 1 3 Age 51.52 15.42 25 92 51.55 15.42 25 92 Gendera .494 .500 0 1 .496 .500 0 1 Self-control 3.430 .506 1 5 3.432 .508 1 5 Mastery 3.526 .683 1 5 3.527 .683 1 5 Self-rated health .564 .496 0 1 - - - - Financial problems - - - - .295 .456 0 1 N 900 892
The general pattern: how education keeps away doctors and creditors
In order to test our hypotheses, a stepwise model building procedure was followed. Models 1a and 1b in table 2 present the main effects of education on the probability of having good self-rated health and having financial problems respectively. In order to investigate the extent to which the effect of education is mediated by perceived control, this variable is added to the analyis in models 2a and 2b. To understand the mediating influence of perceived control on a more detailed level, model 3a and 3b include the effects in a differentiated way.
Table 2. Stepwise LPM regression analyses predicting self-rated health and financial problems
Self-rated health Financial problems
Model 1a Model 2a Model 3a Model 1b Model 2b Model 3b Medium education 0.062 0.040 0.042 -0.040 -0.021 -0.015 (Ref.: low) (0.042) (0.040) (0.040) (0.039) (0.037) (0.037) High education 0.100** 0.052 0.054 -0.187*** -0.143*** -0.145*** (Ref.: low) (0.043) (0.041) (0.041) (0.040) (0.038) (0.038) Age -0.006*** -0.006*** -0.006*** -0.003*** -0.002** -0.003** (0.001) (0.001) (0.001) (0.001) (0.001) (0.001) Gender -0.073** -0.064** -0.061* -0.010 -0.019 -0.022 (Ref.: male) (0.033) (0.031) (0.032) (0.031) (0.029) (0.029) Mastery 0.202*** -0.192*** (0.023) (0.021) 2nd group mastery 0.145*** -0.279*** (Ref.: lowest) (0.048) (0.044) 3d group mastery 0.276*** -0.375*** (Ref.: lowest) (0.050) (0.046) 4th group mastery 0.382*** -0.424*** (Ref.: lowest) (0.048) (0.044) Constant 0.843*** 0.883*** 0.656*** 0.529*** 0.495*** 0.794*** (0.075) (0.072) (0.079) (0.071) (0.068) (0.073) N 900 900 900 892 892 892 R2 0.051 0.127 0.124 0.033 0.114 0.130
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
As shown in table 2, being highly educated as opposed to being low educated is associated with a significant increase in the probability to report good self-rating health with .10 percentage points, controlled for gender and age. Having a medium level education compared to low education is associated with an increase of .06 percentage points. This effect however falls outside the regularly held significant range. A similar pattern is found for the probability of having financial problems outcome. Whereas the effect of being medium educated as opposed to low educated is associated with a small decrease in the probability of having financial problems of .03 percentage points that is not significant, being high educated is associated with a relatively strong, significant decrease of .17 percentage points. These results confirm the well-known favourable aspects of education on important life outcomes: holding everything else constant, highly educated
individuals have 10% more chance to report good self-rated health and have 17% less chance to be in financial problems than low educated.
In line with our the first research aim of understanding how educational resources and
psychological resources interrelate in their influence on health and finances, models 2a and 2b estimate the mediating effect of perceived control on health and personal finances respectively. In order to aid the interpretation of the main effect of perceived control the beta coefficient is calculated.10 One standard deviation increase in perceived control is associated with an increase of .28 percentage points in the probability of reporting good self-rated health, holding all other variables constant. Moreover, one standard deviation increase in perceived control is associated with a 29% smaller chance to experience financial problems. Both effects are significant and relatively strong. Of main interest here however is the extent to which the main effect of education decreases after perceived control enters the model. Comparing the coefficients of model 1a and 2a, and 1b and 2b respectively, we find that perceived has a small mediating effect. That is to say, entering perceived control to the model is associated with a reduction in the effect of education of .05 percentage points for both self-rated health and having financial problems, holding all other variables constant.11 Moreover, in the case of health, the effect of being highly educated is no longer significant after inclusion of perceived control, which suggests that indeed a mediating mechanism is at play.
In order to understand this effect in more detail, scores on perceived control are divided
10 Ideally one would directly report beta coefficients to interpret the influence of the indexes. However, since our interest here is
in four groups, more or less equal in size.12 As shown in model 3a, the effect of being in the second lowest perceived control group compared to the lowest perceived control group does not fall within commonly held significance boundaries. Being in the second highest or highest group however, is associated with an increased chance of 18% and 22% of having good self-rated health respectively. Moreover, the probability of having financial problems is already significantly
smaller (11%) for individuals in the second lowest group of perceived control compared to the lowest group. Being in the second highest group of perceived control is also associated with a decreased probability of having financial problems of 11%. Individuals in the highest perceived control group have even a 24% smaller probability to experience financial problems compared to individuals in the lowest perceived control group.
The above-presented results suggest that both education and perceived control independently strongly affect both the outcomes of health and personal finances. Adding perceived control to the model substantially increases the explained variance for both outcomes. The mediating influence of perceived control is small, in terms of the reduction in the effect of education observed after perceived control enters the model. It seems that in the case of personal finances, educational resources and psychological resources operate rather independently. In the case of health however, the main effect of education loses significance after inclusion of perceived control, which does suggest a relatively strong mediating effect. Together they still explain only 12% and 13% of the variance in health and personal finances outcomes respectively. The extent to which their interaction can explain additional variance in important outcomes is investigated in the next section.
Exceptions to the general pattern: low educated doing well, high educated doing not so well
The usual interpretation of the characterization of the educational groups presented in table 3 would probably compare the percentages across educational groups. Indeed, the percentage of individuals experiencing financial problems is higher among the low educated (36%) than it is among the highly educated (20%). Our interest however lies in the opposite interpretation of these numbers, namely deviations to the general pattern. That is to say, 64% percent of the low educated experiences no problems with regard to their personal finances at all. Furthermore,
12 Ideally one would construct equal quartile groups. However, to the extent that specific scores may be overrepresented, it is not
possible to create perfectly equal groups. The aim of the current analysis is to understand the linear effect of perceived control on a more detailed level, so the relative effect of being in each group is important. A precise description of the groups can be found in the appendix.
although the large majority of the highly educated does not expierence any financial problems, a substantial part of 20% does, despite the favorable prospects generated by their level of
education. With regard to health, the differences are less explicit, however still interesting. A small minority of the low educated (46%) reports (very) good self-rated health. Among the highly educated, a substantial part of 37% reports not being in good health. Although the face
interpretation of numbers is obviously biased, because age and gender differences are not taken into account, they provide an illustrative introduction to the second part of analyses.
Table 3. Mean level of self-control and perceived control by educational group
The moderating effect of self-control
In order to investigate the extent to which psychological resources can explain the heterogeneous effects of education on important life outcomes, the current section presents the moderation analyses results for the self-control variable. Should it be the case that psychological resources and education indeed interrelate in their influence on important life outcomes, then this would be confirmed by sizeable, significant interaction effects.
First, the interaction terms of education and the linear effects of self-control13 are
included in models 6a and 6b (figure 1). As is shown in table 4, the interaction terms for medium
Low Medium High
Overall PCT 28.57 36.99 34.45 Self-control M 3.404 3.434 3.445 SD .479 .498 .536 Perceived control M 3.416 3.507 3.635 SD .734 .648 .661 Good self-rated health PCTa 13.24 21.47 21.75 COLPCTb (46.23) (58.27) (62.99) Financial problems PCTc 10.10 12.54 6.920 COLPCTd (35.53) (34.10) (19.88)
a Total percentage in good self-rated health by level of education
b Percentage in good self-rated health within each educational group
c Total percentage having financial problems by level of education
.45 .5 .55 .6 .65 L in e a r Pre d ict io n
low medium high
Level of education Predictive Margins of education with 95% CIs
.15 .2 .25 .3 .35 .4 L in e a r Pre d ict io n
low medium high
Level of education Predictive Margins of education with 95% CIs
and high education do not fall within the regularly held boundaries of significance for either of the two outcome variables. For size and significance are different aspects, beta coefficients are produced in order to interpret the interaction term of the linear effect of self-control nonetheless.
Figure 1. Margins plot of the linear interaction with self-control on health (left) and personal finances (right).
For both outcomes it applies that the interaction term is small. That is to say, the positive effect of being highly educated on the probability of reporting good health decreases with .03
percentage points for every standard deviation increase in self-control, holding all other variables constant.14 The negative effect of being highly educated on the probability of having financial problems however becomes .03 percentage points smaller for every unit increase in self-control as well. In other words, highly educated individuals that report one standard deviation more self-control have 15% chance to experience financial problems, whereas their highly educated counterparts with a lower level of self-control have an 18% chance.15 One should be cautious however in drawing too firm conclusions on the basis of these results since they do not fall within the regular significance range.
Although the interaction between education and the linear self-control variable was not significant, further analyses are employed in order to see whether more differentiated models add to our understanding.16 Therefore, models 7a and 7b include the interaction effects between the two levels of education and three groups of self-control.17
14 In this model this means that the probability refers to men of age 52.
15 These numbers are presented to illustrate the difference in the effect of being highly educated across levels of self-‐
control, rather than to interpret the absolute probability for an individual with specific values on each variable, because this latter interpretation is problematic in case of a Linear Probability Model as applied here.
16 Groups are divided in the same way as was done for perceived control in the previous section. A full description of the groups
is to be found in the appendix.
Table 4. LPM regressions including the interaction of education and self-controla
Self-rated health Financial problems
Model 6a Model 7a Model 6b Model 7b
Medium education 0.046 -0.073 -0.027 -0.133 (Ref.: low) (0.041) (0.087) (0.038) (0.082) High education 0.085** 0.162* -0.173*** -0.275*** (Ref.: low) (0.042) (0.086) (0.039) (0.080) Age -0.007*** -0.007*** -0.002* -0.002* (0.001) (0.001) (0.001) (0.001) Gender -0.074** -0.074** -0.008 -0.010 (Ref.: male) (0.032) (0.032) (0.030) (0.030) Self-control 0.166*** -0.180*** (0.062) (0.058) 2nd group selfcontrol 0.022 -0.235*** (Ref.: lowest) (0.084) (0.079) 3d group self-control 0.116 -0.308*** (Ref.: lowest) (0.088) (0.082) 4th group self-control 0.239*** -0.269*** (Ref.: lowest) (0.086) (0.081) Interactions Medium education*Self-control 0.088 0.018 (Ref.: low) (0.081) (0.076) High education*Self-control -0.041 0.044 (Ref.: low) (0.080) (0.075) Medium education*2nd self-control group 0.106 0.219** (0.114) (0.107) Medium education*3d self-control group 0.254** 0.152 (0.118) (0.111) Medium education*4th self-control group 0.110 0.026 (0.117) (0.109) High education*2nd self-control group -0.091 0.139 (0.118) (0.110)
High education*3d
self-control group -0.072 0.189* (0.116) (0.109) High education*4th self-control group -0.145 0.068 (0.115) (0.107) Constant 0.903*** 0.796*** 0.475*** 0.683*** (0.075) (0.088) (0.070) (0.083) N 900 900 892 892 R2 0.088 0.098 0.063 0.076
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
An easy way to interpret the interaction effect between two discrete variables is usually to calculate the probability of a certain outcome for each subgroup. The downside of a Linear Probability Model however is that it allows for probability estimates outside the range [0, 1], which are problematic to interpret. In order to circumvent this problem and be able to interpret the interaction term nevertheless, we focus on the relative size of the probabilities. As such, we compare the four subgroups resulting from the interaction between high education and highest self-control group dummies, since they provide the most extreme cases. However, this effect is not significant, which should be kept in mind while interpreting the size and direction of the effect.
If we look at the health domain, table 4 (model 7a) shows that the probability of
reporting good health is .24 percentage points higher for low educated individuals in the highest self-control group than it is for low educated individuals in the lowest self-control group. That is to say, compared to their counterparts in the lowest self-control group, individuals with a low level of education but a high level of control have a 24% higher chance to report good self-rated health. However, having a high level of education but being in the highest self-control group yields only a slightly higher probability of having good self-rated health (1.05) than being high educated and being in the lowest group of self-control (.96). Furthermore, low educated individuals in the highest self-control group, have a probability of reporting good health almost equal to that of highly educated individuals in the highest self-control group (1.04 and 1.05 respectively). However, again, this effect is not significant, so one should be careful in substantiating claims on the basis of these findings.
With regard to finances, a similar pattern is observed. For coherence reasons we focus again on the interaction between being highly educated and being in the highest self-control group, although two other interaction terms are actually significant.18 Ranking the probabilities on having financial problems according to the interaction between education and self-control, the following sequence is observed: holding all other variables constant, highly educated individuals in the highest group of self-control have the smallest probability to experience financial
problems, namely .20 percentage points. Second, being highly educated but being in the lowest self-control group yields a probability of having financial problems of .40. Third, with only one percentage point more, the low educated individuals in the highest group of self-control follow. Finally, low educated individuals with a low level of self-control have, within this comparison, the highest probability of experiencing financial problems, namely 68%.
18 Namely, between being medium educated and in the second lowest self-control group and between being highly educated and
.45 .5 .55 .6 .65 L in e a r Pre d ict io n
low medium high
Level of education Predictive Margins of education with 95% CIs
.15 .2 .25 .3 .35 .4 L in e a r Pre d ict io n
low medium high
Level of education Predictive Margins of education with 95% CIs
The above presented analyses lead to multiple sub conclusions. Firstly, although some of the differentiated interaction effects are significant, this does not apply to the full pattern. Therefore, on the basis of significance, we interpret these findings to be inconclusive. Secondly, however, a tentative pattern can be observed interpreting on the size and direction of the effects. The
favourable effect of being highly educated is lower for individuals with a low level of self-control. Moreover, the unfavourable effect of being low educated can be downplayed by a high level of self-control, such that it reaches the probability of being highly educated and having either the highest level of self-control (in the case of health) or having the lowest level of self-control (in the case of personal finances). In conclusion, it seems that the determining effect of education on important life outcomes is at least to some extent conditional on the level of self-control.
The moderating effect of perceived control
Table 5 presents the moderation analyses for the linear effect of perceived control19 on the relation between education and good self-rated health (model 8a) and personal finances (model 8b). In the case of health, the results show a significant and positive, however small, interaction term between being highly educated and the linear effect of perceived control (Figure 2, left). With regard to the size of this effect, it is shown that for every standard deviation increase in perceived control, the enhancing effect of being highly educated compared to being low educated on the probability of reporting good self-rated health increases with .08 percentage points,
holding all other variables constant. So, a highly educated individual that has a level of perceived control one standard deviation higher than average, has a 8% higher chance to report good self-rated health than their highly educated individual counterpart with an average level of perceived control. These findings suggest that apart from a mediating effect, perceived control may also exert a moderating influence on the relation between education and important life outcomes.
Figure 2. Margins plot of the linear interaction with mastery on health (left) and personal finances (right).
In the case of personal finances, the effect is positive as well, although smaller and it does not fall within the regular range of significance. The decreasing effect of being highly educated rather than low educated on the probability of having financial problems decreases with .05 percentage points for every unit increase in perceived control. This suggests that among the highly educated, the probability of having financial problems is actually larger for people with more perceived control than highly educated who perceive less control. This would suggest that the favourable effect of perceived control might in fact be nonlinear. These results invite for more differentiated analysis, in order to get more insight in how education interrelates with psychological resources as operationalized in terms of perceived control.
Therefore, model 9a and 9b present the results of the differentiated moderation analyses for health and personal finances respectively.20 As for the self-control analyses, the main interest lies in the comparison of the four groups differing in terms of high versus low education and being in the highest perceived control group versus the lowest. Calculating the probabilities for the different subgroups the same approach as for self-control is applied, with attention for their relative size rather than absolute probability. This yielded the following observations.
First, the probability of having good self-rated health for the high educated being in the highest group of perceived control (1.13) is much larger than the probability of the high educated in the lowest group of perceived control (.70), holding all other variables constant.21 The same pattern applies for the low educated: being low educated and in the lowest group of perceived control yields a probability of reporting good self-rated health of .69 percentage points, whereas being low educated and in the highest perceived control group is associated with a probability of 1.04. Furthermore, what these numbers tell us is that being low educated but perceiving a lot of control yields a similar probability to being high educated and being in the highest perceived control group. Moreover, being low educated and being in the highest perceived control group trumps being highly educated but having the lowest level of perceived control in the ranking with regard to the probability of reporting good self-rated health. However, since this effect does not fall within significance boundaries, we cannot draw firm conclusions on the basis of these results. With regard to the personal finances domain, a similar pattern occurs. Whereas being low educated and being in the lowest group of perceived control is associated with a probability of having financial problems of .78, being low educated and reporting the highest level of perceived control yields a probability of .32. Furthermore, the probability of this latter group is similar to its highly educated counterpart, which has a probability of 27%. Having completed higher education and being in the lowest group of perceived control yields a probability coefficient of .61,
20 The same differentiated groups as in the mediation analyses were included.
Table 5. LPM regressions including the interaction of education and perceived controla
Self-rated health Financial problems
Model 8a Model 9a Model 8b Model 9b
Medium education 0.044 0.009 -0.020 0.059 (Ref.: low) (0.040) (0.094) (0.037) (0.086) High education 0.050 0.007 -0.151*** -0.173* (Ref.: low) (0.041) (0.096) (0.039) (0.089) Age -0.006*** -0.006*** -0.003** -0.002** (0.001) (0.001) (0.001) (0.001) Gender -0.065** -0.065** -0.019 -0.021 (Ref.: male) (0.031) (0.032) (0.029) (0.029) Mastery 0.144*** -0.199*** (0.040) (0.037)
Second group mastery 0.177** -0.245***
(0.082) (0.076)
Third group mastery 0.157* -0.311***
(0.089) (0.083)
Fourth group mastery 0.348*** -0.459***
(0.086) (0.079) Medium education*Mastery 0.074 -0.039 (Ref.: low) (0.056) (0.052) High education*Mastery 0.102* 0.060 (Ref.: low) (0.056) (0.052) Medium education*2n mastery -0.020 -0.073 (0.115) (0.106) Medium education*3d mastery 0.145 -0.166 (0.124) (0.114) Medium education*4th mastery 0.023 -0.040 (0.120) (0.110) High education*2nd mastery -0.081 -0.042 (0.121) (0.112) High education*3d mastery 0.187 -0.012 (0.125) (0.116) High education*4th mastery 0.077 0.120 (0.119) (0.110) Constant 0.877*** 0.687*** 0.499*** 0.776*** (0.072) (0.092) (0.068) (0.085) N 900 900 892 892 R2 0.131 0.130 0.117 0.137
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
aModels 6a and 6b include the linear effects of mastery centered around the mean. Models 8a and 9a
compared to an effect of .27 for highly educated with a level of perceived control that falls in the highest group. In conclusion, being low educated but having a high sense of control yields a probability of having financial problems more favourable than that for highly educated
individuals that report a low sense of control. However, as for the health domain, the interaction effect of being high educated and being in the highest group of perceived control was not significant.
The above presented analyses lead to multiple sub conclusions. Firstly, with regard to the health domain, the linear interaction term is significant, whereas the more differentiated analyses are not. With regard to the personal finances domain, neither the linear interaction term nor the differentiated analyses yield significant results. Therefore, on the basis of significance, we
interpret these findings to be inconclusive. With regard to size and direction however, the results of the differentiated analyses were in the expected direction. Likewise to the effect of
self-control, it appears that the determinant influence of education on important life outcomes is moderated by the extent to which one perceives control over these outcomes. However, compared to the self-control findings, in drawing conclusion on the basis of the results of these perceived control moderation analyses even more caution should be exercised.
Self-deterministic hypothesis
Finally, we aimed to examine the extent to which the compensatory or enhancing effect of psychological resources applies equally across levels of education. More specifically, we phrased our fourth hypothesis in favour of the self-deterministic perspective, predicting that the
difference in the effects of both self-control and perceived control on health and personal finances among the lower educated would be larger than the difference in the effects of these resources among the higher educated. In order to test this, an equivalent but seemingly different interpretation of the interaction effects presented in tables 4 and 5 is required.
The added value among the low educated of belonging to the highest self-control group is 24% (see table4). Among the higher educated, this difference is only 9%. Dividing these differences by each other, we conclude that in the case of health, the added value of being in the highest instead of the lowest self-control group is more than twice as large for low educated as it is for the higher educated. Although the difference is substantially smaller (factor 1,4), this applies to the case of personal finances as well. Looking at perceived control, we find the pattern to be similar to that of self-control in the case of personal finances. That is to say, the difference in being in the highest rather than the lowest perceived control group is 1,3 (health) or 1,4 (finances) times the size of this difference for the low educated as it is for the higher educated.
To the extent that the difference is larger for the lower educated, these results suggest support for our fourth hypothesis. However, as it calculated using the same, insignificant interaction coefficients as earlier, taking caution in drawing conclusions on the basis of these results is advised.
Conclusion
The overall aim of the current study was to gain understanding in how education and
psychological resources interplay in their influence on important life outcomes. More specifically, it was investigated to what extent psychological resources explain the heterogeneous effects of education. To these aims, both mediation and moderation analyses in the case of perceived control, and moderation analyses in the case of self-control, were conducted with regard to the influence of education on self-rated health and experiencing financial problems.
In line with previous research, our findings support the hypothesis that perceived control
mediates the relationship between level of education and important life outcomes in the domains of health and personal finances. With regard to our moderation hypotheses, the findings were less conclusive. In case of the psychological resource self-control, the differentiated interaction terms suggested some support for our second hypothesis. The majority of the interaction terms were not significant for both outcomes. However, regardless of the significance interpretation, the overall pattern in terms of size and direction pointed in the direction of the expected pattern. On the basis of these inconclusive findings, we do not accept nor refute our second hypothesis. In the case of the psychological resource perceived control, the interaction term of being highly educated with the linear effect of perceived control in the health domain was the only significant effect. However, as was the case for self-control, interpreting the size and direction of the differentiated interaction term, the pattern appeared to be in the expected direction. Therefore, on the basis of these findings, we do not accept nor refute our third hypothesis either. Finally, in line with preliminary earlier findings, our results suggested underpinning of the fourth hypothesis that the advantageous effects of self-control and perceived control on important life outcomes are stronger for the low educated than the high educated. However, since this observation is based on the same, insignificant results, we cannot accept our fourth hypothesis on the basis of these findings.
cross-sectional design suffer from some serious limitations. Firstly, endogeneity is a problem to the extent that outcomes in the different domains influence the score on the psychological scales. This is especially a concern for perceived control, since this is influenced by factors other than education as well. It might well be that experiencing financial problems for a longer period eventually influences one’s sense of control. Furthermore, self-control may operate as
confounder, influencing both education and important life outcomes. Moreover, unexpected life events are a known confounder of the relations studied here. In response to these issues, a recommended suggestion for future research on this topic would be to employ a time-series design, so that these limitations can be accounted for. Moreover, it allows testing for reverse causality, which would mean an interesting contribution to the existing field.
Secondly, the use of self-reported health outcomes, although widely used, is subject to critics, because of confounding of prevalence and awareness. However, it is a widely used measure in the field and has practical advantages in terms of the larger design of the current survey. Third, the current analysis was built using data collected for this purpose. Although this means a large advantage in terms of recent data, it carries some disadvantages as well. The most important of which is that since this was the first time this survey was used, we cannot be certain of its validity and reliability. Despite the poor fit between the research design and the data, however, interesting preliminary results were found, which merit further examination by future research. Moreover, should the tentative pattern as suggested here be confirmed by future analyses, this may have important policy implications for contemporary Dutch society.
The most important policy implication arguably follows from the evidence supporting the self-deterministic hypothesis. Should this thesis be confirmed, this suggests that the disadvantageous effect of having a low level education can be compensated on the individual level by
psychological resources. However, the policy implications are very different depending on the resource at hand. In the case of perceived control, for example, it has been suggested that lower educated individuals that maintain a high sense of control are less likely to end up in trouble than their counterparts with a lower sense of control. Interventions aimed at enhancing the position of low educated individuals may as such consider an approach aimed at increasing sense of control. With regard to self-control however, the results from interventions aimed at training this ability are at best mixed. This suggests that policy measures to this regard might more effectively be found in the field of nudging Moreover, to the extent that psychological resources cannot be trained nor nudged, this has important implications for a policy in which individual difference plays the leading part.
References
Almlund, M., Duckworth, A. L., Heckman, J. J., & Kautz, T. D. (2011).Personality psychology and
economics (No. w16822). National Bureau of Economic Research.
Becker, G. S. (2009). Human capital: A theoretical and empirical analysis, with special reference to education.
University of Chicago Press.
Kawachi, I., & Berkman, L. F. (2001). Social ties and mental health. Journal of Urban health, 78(3),
458-467.
Bosma, H., C.H. Van Jaarsveld & J. Tuinstra. 2005. Low control beliefs, classical coronary risk factors, and socio-economic differences in heart disease in older persons. Soc. Sci. Med. 60: 737–745.
Bosma, H., C. Schrijvers & J.P. Mackenbach. 1999. Socioeconomic inequalities in mortality and importance of perceived control: cohort study.
Borghans, L., Duckworth, A. L., Heckman, J. J., & Ter Weel, B. (2008). The economics and psychology of personality traits. Journal of Human Resources,43(4), 972-1059.
Caliendo, M., Cobb-Clark, D. A., & Uhlendorff, A. (2010). Locus of control and job search strategies, Discussion papers // German Institute for Economic Research, No. 979 Cole, S. A., Paulson, A. L., & Shastry, G. K. (2012). Smart money: The effect of education on financial behavior. Harvard Business School finance working paper, (09-071).
Cutler, D. M., & Lleras-Muney, A. (2008). Education and health: evaluating theories and evidence (No.
w12352). National Bureau of Economic Research.
Cutler, D. M., & Lleras-Muney, A. (2010). Understanding differences in health behavior by education. Journal of Health Economics, 29(1), 1-28.
Davies, E., Lea, S.E.G., 1995. Student attitudes to student debt. Journal of Economic Psychology, 16, 663-679.
Diehl, M., & Hay, E. L. (2010). Risk and resilience factors in coping with daily stress in
adulthood: the role of age, self-concept incoherence, and personal control. Developmental
psychology, 46(5), 1132.
Donnellan, M. B., Conger, K. J., McAdams, K. K., & Neppl, T. K. (2009). Personal
characteristics and resilience to economic hardship and its consequences: Conceptual issues and empirical illustrations. Journal of Personality, 77, 1645–1676.
socioeconomic position (Pt. 1). Journal of Epidemiology and Community Health, 60, 7–12. Galobardes, B., Shaw, M., Lawlor, D. A., Lynch, J. W., & Davey Smith, G. (2006b). Indicators of socioeconomic position (Pt. 2). Journal of Epidemiology and Community Health, 60, 95–101. Grossman, M. (1972). On the concept of health capital and the demand for health. The journal of
political economy, 223-255.
Heckman, J. J., & Kautz, T. (2012). Hard evidence on soft skills. Labour economics, 19(4), 451-464. Heckman, J. J., & Rubinstein, Y. (2001). The importance of noncognitive skills: Lessons from the
GED testing program. American Economic Review, 145-149.
Kautz, T., Heckman, J. J., Diris, R., ter Weel, B., & Borghans, L. (2014). Fostering and Measuring
Skills. OECD-library.org.
Lachman, M. E., & Weaver, S. L. (1998). The sense of control as a moderator of social class differences in health and well-being. Journal of personality and social psychology, 74(3), 763. Lochner, Lance, and Enrico Moretti, 2004, The Effect of Education on Crime: Evidence from Prison Inmates, Arrests, and Self-Reports. American Economic Review, 94, 155-189. Macintyre, S., & Hunt, K. (1997). Socio-economic position, gender and health how do they interact? Journal of Health Psychology, 2(3), 315-334.
Matthews, K. A., Gallo, L. C., & Taylor, S. E. (2010). Are psychosocial factors mediators of socioeconomic status and health connections? A progress report and blueprint for the future. Annals New York Academy of Sciences, 1186, 146–173.
McNulty, T. L., & Bellair, P. E. (2003). Explaining racial and ethnic differences in adolescent violence: Structural disadvantage, family well-being, and social capital. Justice
Quarterly, 20(1), 1-31.
Mirowsky, J. (2013). Depression and the Sense of Control Aging Vectors, Trajectories, and Trends. Journal of health and social behavior, 54(4), 407-425.
Mirowsky, J., & Ross, C. E. (1998). Education, personal control, lifestyle and health A human capital hypothesis. Research on Aging, 20(4), 415-449.
Mirowsky, J., & Ross, C. E. (2007). Life Course Trajectories of Perceived Control and Their Relationship to Education1. American Journal of Sociology, 112(5), 1339-1382.
Moffitt, T. E., Arseneault, L., Belsky, D., Dickson, N., Hancox, R. J., Harrington, H., ... & Sears, M. R. (2011). A gradient of childhood self-control predicts health, wealth, and public safety. Proceedings of the National Academy of Sciences,108(7), 2693-2698.
