The impact of socioeconomic conditions on health is well documented: those living in poverty experience greater physical morbidity and live shorter lives than the wealthy (Adler et al., Reference Adler, Boyce, Chesney, Cohen, Folkman, Kahn and Syme1994; Lawlor et al., Reference Lawlor, Sterne, Tynelius, Davey Smith and Rasmussen2006; Galobardes et al., Reference Galobardes, Lynch and Davey-Smith2008). The adverse effects of socioeconomic inequalities are evident at each stage of the life course and extend beyond physical health. Across diverse populations and contexts, lower socioeconomic status (SES) predicts almost every measure of mental health including measures of self-reported distress (Power and Manor, Reference Power and Manor1992; Molarius et al., Reference Molarius, Berglund, Eriksson, Eriksson, Lindén-Boström, Nordström, Persson, Sahlqvist, Starrin and Ydreborg2009; Melchior et al., Reference Melchior, Chastang, Head, Goldberg, Zins, Nabi and Younes2013) and diagnoses of clinical disorders (Dohrenwend et al., Reference Dohrenwend, Levav, Shrout, Schwartz, Naveh, Link, Skodol and Stueve1992; Ritscher et al., Reference Ritscher, Warner, Johnson and Dohrenwend2001; Gilman et al., Reference Gilman, Kawachi, Fitzmaurice and Buka2002; Stansfield et al., Reference Stansfield, Clark, Rodgers, Caldwell and Power2011). Further, the contribution of SES is evident above the poverty threshold: mental health follows a gradient across the full socioeconomic hierarchy. For instance, a meta-analysis of studies examining the social gradient in depression found an almost twofold increase in risk of depression for individuals from low relative to high SES individuals, and this pattern followed a dose–response relation in income and education (Lorant et al., Reference Lorant, Deliege, Eaton, Robert, Philippot and Ansseau2003).
Whilst mental health disparities have been consistently demonstrated, the potential consequences of low SES are not necessarily fixed or universal. Many of those raised in disadvantaged circumstances experience few mental health problems subsequently (Masten, Reference Masten2001; Fergusson and Horwood, Reference Fergusson, Horwood and Luthar2003). Resilience research has sought to identify the factors that either attenuate or exacerbate the negative psychological consequences of early-life adversity and environmental stressors (Luthar, Reference Luthar, Cicchetti and Cohen2006). One candidate attribute highlighted by this work is high cognitive ability (frequently referred to as intelligence or intelligence quotient/IQ), which has been shown to predict positive academic and behavioural outcomes in the face of adverse life events (Fergusson and Lynskey, Reference Fergusson and Lynskey1996; Masten et al., Reference Masten, Hubbard, Gest, Tellegen, Garmezy and Ramirez1999; Pargas et al., Reference Pargas, Brennan, Hammen and Le Brocque2010; Flouri et al., Reference Flouri, Mavroveli and Panourgia2013). Crucially, the protective effect of high cognitive ability has been shown to extend to the psychological effects of stressful life events, including moderating the onset of symptoms of adolescent depression (Riglin et al., Reference Riglin, Collishaw, Shelton, McManus, NG-Knight, Sellers, Thapar, Frederickson and Rice2016) and the occurrence of emotional internalizing problems (Flouri et al., Reference Flouri, Midouhas and Joshi2014). Further, resilient youths, defined in part by the absence of recurrent depression, have been shown to have higher levels of cognitive ability (Pargas et al., Reference Pargas, Brennan, Hammen and Le Brocque2010).
In line with this idea, using two British cohorts of individuals born in 1970 and 1958, we recently showed that high levels of childhood cognitive ability were associated with a substantial reduction in the link between social disadvantage and psychological distress from young adulthood to midlife (measured using the ‘Malaise Inventory’; Rutter et al., Reference Rutter, Tizard and Whitmore1970; Bridger and Daly, Reference Bridger and Daly2017). In both cohorts, early-life disadvantage was unrelated to future risk of distress among those with high levels of cognitive ability (i.e. 1 s.d. above the mean). This buffering effect was not explained by the potential compensatory role of adult SES. As such, acquiring the financial resources, prestige and knowledge absent from their initial circumstances did not explain why the adverse consequences of disadvantage were no longer apparent in adulthood for those with high cognitive ability.
We suggest that the buffering effect of cognitive ability may reflect the adaptive nature of intelligence which, by definition, is tied to an individual's ability to successfully interact with their environment (Legg and Hutter, Reference Legg and Hutter2007). This ability might protect against the potentially distressing effects of adverse circumstances by enabling individuals to respond flexibly to stressors and challenges, profit from learned experiences and subsequently counter environmental threats more effectively. Greater cognitive resources may serve not only to reduce the number of threats encountered in this way but may also help regulate negative emotions in the face of unavoidable stressors (Schmeichel and Tang, Reference Schmeichel and Tang2015; Riglin et al., Reference Riglin, Collishaw, Shelton, McManus, NG-Knight, Sellers, Thapar, Frederickson and Rice2016) and directly protect individuals against the psychological sequelae of repeated exposure to stress.
Building directly on our prior work which examined two birth cohorts using a non-standard mental health measure, the objectives of the present study were to (i) establish whether the potential stress-buffering effect of cognitive ability generalises to a population-representative sample including individuals from across the full adult life span and (ii) assess the consistency of the buffering effect across a broad set of well-established clinically relevant mental health measures [e.g. Short Form-12 (SF-12) Mental Component, General Health Questionnaire (GHQ), clinical diagnosis of depression]. The Understanding Society [UK Household Longitudinal Study (UKHLS)] study records this information in a large population representative sample of UK households along with rich sociodemographic records. Using these data, we aimed to estimate the potential mental health effects of one's family background as gauged by parental education and occupational class levels. We examine family background disadvantage in this way to avoid the possibility of reverse causation associated with adult SES which partially reflects the influence of mental health on income and employment prospects throughout life (Chatterji et al., Reference Chatterji, Alegría, Lu and Takeuchi2007; Smith and Smith, Reference Smith and Smith2010; Egan et al., Reference Egan, Daly and Delaney2016). Critically, the UKHLS also includes measures of cognitive functioning which allows the potential buffering effect of cognitive ability to be estimated. We hypothesised that cognitive ability would moderate the association between early-life social status and mental health such that those with high levels of cognitive ability would tend to experience few adult mental health consequences of their disadvantaged backgrounds relative to other individuals.
Method
Participants
The UKHLS or Understanding Society is a survey of approximately 40 000 households from across the UK (England, Scotland, Wales and Northern Ireland). One of the largest surveys of its kind, the Understanding Society panel is representative of the UK population of all ages and has been designed to ensure ethnic minorities are adequately represented. Complete details on the survey design and sample are reported elsewhere (Buck and McFall, Reference Buck and McFall2012) and all study protocols have been scrutinised and authorised by a number of research committees to ensure compliance with the Helsinki Declaration of 1975/2008. Beginning in 2009–2010 (wave 1), households have been visited every 24 months to capture changes in circumstances over time. To capture relative level of background social disadvantage, we combined measures of parents’ education and occupation from the first two waves (2009–2011). Measures of cognitive ability and mental health were all taken from data collected at wave 3 (2011–2012).
Measures
Social disadvantage
In order to make claims about the contribution of SES to mental health, it is important that social disadvantage is assessed before participants have gained substantial experience in the labour force given that mental health contributes substantially to employment and income levels (e.g. Chatterji et al., Reference Chatterji, Alegría, Lu and Takeuchi2007). Across waves 1 and 2 of the UKHLS, participants reported the highest educational qualifications their father and mother attained (1 = did not go to school, 2 = left school with no qualifications, 3 = left school with some qualifications/certificates, 4 = gained further qualifications after leaving school, 5 = gained university or higher degree). Responses were reverse-coded so higher numbers reflect greater disadvantage. Participants also reported both parents’ job titlesFootnote 1 when they were aged 14 which were subsequently coded to the ONS Standard Occupational Classification 2010 (ONS, 2010). This is the current standard occupational classification for the UK by which jobs are classified on the basis of their title and subsequently given a coded ranking (range = 111–927, where 111–125 = managers, directors and senior officials and 911–927 = elementary occupations) on the basis of both skill level and specialisation. Coded scores were ranked in line with the ordinal nature of this variable. Table 1 reports the frequency of each parental education category as well as mean occupational ranking for both parents for the final sample. A large proportion of mother's occupational data is missing because participants reported that their mothers were not working when they were 14 (39.3% of sample, compared with 5.4% of fathers).
Table 1. Descriptive statistics for all key variables and the measures employed to comprise the measure of social disadvantage
Standard deviations in parentheses. Weighted estimates are shown for individuals for whom data on the following variables were available: gender, ethnicity, age, cognitive ability, social disadvantage and at least one outcome variable.
aRanging from 111–125 = managers, directors and senior officials to 911–927 = elementary occupations.
Each of the four measures – mother's and father's highest educational qualification, mother's and father's ranked occupational score – were standardised to have a mean of zero and standard deviation of one, before being averaged and subsequently re-standardised to produce a normally distributed measure of social disadvantage with an adequate level of internal reliability (Cronbach's α = 0.75). To maximise the sample size, cohort members were included in the analyses if they provided data on at least two of the four key measures [of the 27 985 included in the final study, complete data were provided on average for 3.16 of the four measures (s.d. = 0.78)].
Cognitive ability
In wave 3, participants completed a cognitive ability module at home as guided by the interviewer. For the current analysis, data on performance in word recall, numeric ability and verbal fluency tasks were employed. In the word recall tasks (episodic memory), participants listened to a list of 10 words and were asked to immediately recall these words and again later after a short delay. The measure of verbal fluency required participants to name as many animals as they could within 1 min. Performance on such semantic category fluency tasks depends on a set of cognitive functions including searching of semantic memory for category extensions, working memory updating and self-monitoring to keep track of performance (Henry and Crawford, Reference Henry and Crawford2004; Daly et al., Reference Daly, McMinn and Allan2014). To test numeric ability, participants answered five questions graded in complexity designed to reflect the use of numbers in everyday life (e.g. calculating discounted prices, correct change and compound interest). Scores from each of the three tasks (see Table 1) were standardised and the subsequent z-scores were averaged to create a composite measure of cognitive ability, with acceptable internal consistency (Cronbach's α = 0.69).
SF-12 mental component summary
Participants completed the short-form (SF-12) measure of general health from which separate physical and mental component summary (MCS) scales can be derived (Ware et al., Reference Ware, Kosinski and Keller1996). The MCS is a screening tool able to detect depression and anxiety across a variety of populations (Gill et al., Reference Gill, Butterworth, Rodgers and Mackinnon2007; Vilagut et al., Reference Vilagut, Forero, Pinto-Meza, Haro, de Graaf, Bruffaerts, Kovess, de Girolamo, Matschinger, Ferrer and Alonso2013). The MCS scores reported here are derived from six questions asking participants how much time over the past 4 weeks (1 = all of the time, 5 = none of the time) they had a lot of energy, found physical health or emotional problems interfered with social activities, accomplished less than they would like as a result of emotional problems, did work or other activities less carefully than usual, felt calm and peaceful as well as downhearted and depressed. A multi-step process (including creation of indicator variables, weighting of indicator variables and standardising: for complete details see Ware et al., Reference Ware, Kosinski, Turner-Bowker and Gandek2001) converts responses to a single mental health functioning score from 0 (low) to 100 (high). The mean MCS score in the current sample was 49.39 (s.d. = 9.78). Prior to all analyses, SF-12 MCS scores were reverse coded with higher scores representing poorer mental health to facilitate comparison with other variables.
General Health Questionnaire-12
Participants completed the 12-item version of the GHQ-12, used to detect psychiatric cases in the general population by comparing the respondent's current state with their usual state (Goldberg et al., Reference Goldberg, Gater, Sartorius, Ustun, Piccinelli, Gureje and Rutter1997). The GHQ-12 is a short yet well-validated scale often used as the gold standard in psychiatric case classification (Kelly et al., Reference Kelly, Dunstan, Lloyd and Fone2008). Participants responded on a 4-item scale (where 1 = more so than usual, 2 = about the same as usual, 3 = less so than usual and 4 = much less so than usual) to questions about their general happiness, confidence, ability to face problems, overcome difficulties, make decisions and enjoy normal day-to-day activities. GHQ-12 responses were employed here in two ways. First, we generated a continuous variable by totalling responses to give a Likert-like score from 0 to 36 where higher scores represent poorer mental health. The mean GHQ-Likert in the current sample was 11.07 (s.d. = 5.48). In a second step, we coded ‘more so…’ and ‘about the same as usual’ responses as 0, whilst ‘less so…’ and ‘much less so than usual’ responses were recoded as 1. Recoded scores were summed to give a score of 12 where 0 represents least distressed and 12 most distressed. In line with accepted convention (Goldberg et al., Reference Goldberg, Gater, Sartorius, Ustun, Piccinelli, Gureje and Rutter1997), participants with a score of 3 or more were termed as achieving ‘psychiatric caseness’ (i.e. likely to present with psychiatric disorder). In total, 23.63% of the sample met this criterion for caseness on this second GHQ outcome variable.
Diagnosis of clinical depression
At each wave, participants indicated whether a doctor or other health professional had ever told them that they have clinical depression. Data from across the three waves were re-coded into a binary variable reflecting whether or not participants had received a diagnosis of this kind from their doctor. This was the case for 8.94% of the sample.
Neuroticism
Neuroticism refers to a personality trait characterised by high emotionality and sensitivity to stress, which is known to be a robust correlate of depressive symptoms and self-reported lifetime mental disorder (Jylhä and Isometsä, Reference Jylhä and Isometsä2006; Hakulinen et al., Reference Hakulinen, Elovainio, Pulkki-Råback, Virtanen, Kivimäki and Jokela2015) and has been used as an indirect measure of risk for depression (Navrady et al., Reference Navrady, Ritchie, Chan, Kerr, Adams, Hawkins, Porteous, Deary, Gale, Batty and McIntosh2017). In the current survey, neuroticism was measured as part of a 15-item questionnaire designed to capture the Big Five personality traits (Donnellan and Lucas, Reference Donnellan and Lucas2008; Lang et al., Reference Lang, John, Lüdtke, Schupp and Wagner2011). Participants were asked to rate three statements, relating to whether they saw themselves as ‘someone who gets nervous easily’, ‘someone who is relaxed [and] handles stress well’ or ‘someone who worries a lot’. For each item, participants responded on a seven-point response scale (where 1 = does not apply to me at all and 7 = applies to me perfectly). After reverse scoring ‘someone who is relaxed [and] handles stress well’ responses were averaged to produce a measure ranging from low (1) to high (7) neuroticism (M = 3.56, s.d. = 1.44).
Data analysis
First, we aimed to determine how social disadvantage and cognitive ability combine to predict mental health (indexed via the SF-12 MCS and GHQ-Likert scores) and neuroticism (also modelled as a continuous variable) in linear regression analyses. Each of these variables was standardised to have a mean of zero and standard deviation of one across the entire sample to facilitate comparability across analyses. In a series of two models, we adjusted first for key background covariates [model 1: adjusted for age, age-squared, sex, ethnic identity (white/non-white)] and then added the interaction between our standardised measures of cognitive ability and social disadvantage to the initial model (model 2). Comparable logistic regressions were conducted to test whether disadvantage and cognitive ability were associated with reaching the threshold for ‘caseness’ using the GHQ-12 and the likelihood of receiving a diagnosis of depression by a doctor (model 1) and to identify whether the disadvantage × cognitive ability interaction term also predicted the binary outcomes examined (model 2). Simple slopes were then generated using the Stata margins command and compared to determine the association between social disadvantage and mental health at low (−1 s.d.), medium (mean) and high (+1 s.d.) levels of cognitive ability. Finally, we used the Stata marginsplot command to present the simple slopes from fully adjusted regression models graphically.
Model 1: Mental healthi = β 0i + β 1social disadvantagei + β 2cognitive abilityi + β 3femalei + β 4whitei + β 5agei + β 6age2i + ε i
Model 2: Mental healthi = β 0i + β 1social disadvantagei + β 2cognitive abilityi + β 3social disadvantagei × cognitive abilityi + β 4femalei + β 5whitei + β 6agei + β 7age2i + ε i
As part of the wave 3 self-completion survey (when demographic characteristics, cognitive ability and mental health were measured), 39 296 adults were administered at least one mental health measure and had survey weights available. From this group, 27 985 individuals had available covariate data and were included in the current sample. Participants were excluded primarily because they were missing background social disadvantage data (see Table S1) chiefly due to survey routing in the first two waves of Understanding Society. Specifically, due to survey length constraints, the full set of parental social background questions (i.e. father/mother education and occupation) were only put to participants during the first 6 months of sampling at wave 1 and the final 18 months of wave 2. In addition, we excluded new entrants to the adult survey from the youth panel and those joining households (e.g. through marriage) who did not have the opportunity to complete the social disadvantage questions in waves 1 and 2. The excluded group was therefore notably younger than the existing sample on average, as shown in columns 1 and 2 of Table S1.
In contrast, the included and excluded groups differed little in their levels of cognitive ability and mental health. Nevertheless, to account for differences between the included and excluded groups on key covariates, we produced inverse probability weights. To do this, we used available covariate data (i.e. age, gender, ethnicity, cognitive ability measures) to predict the probability of inclusion (v. exclusion) in the current sample in a logistic regression model. The results of this model were used to generate inverse probability weights which were subsequently combined with an existing survey weight designed to take account of non-response, probability of selection and possible sampling error (Knies, Reference Knies2017). When the final weighting variable was applied, the background characteristics, cognitive ability and mental health of the current study sample aligned very closely with the full representative wave 3 sample (see columns 3 and 4 of Table S1), thus helping ensure that the generalisability of the study results to the population was maintained. Finally, we estimated unweighted regression models which were minimally different to the main results presented below (see Table S2).
Results
Descriptive statistics
The final sample size was 27 985 (56.21% female) and the sample was 87.36% white and the mean age was 47.97 (range 18–103). Table 1 presents the descriptive statistics for this sample including both cognitive ability measures as well as the measures of parental SES used to index background disadvantage. There was a moderate negative correlation between social disadvantage and cognitive ability (r = −0.31, p < 0.001, N = 27 985). The three continuous mental health measures (SF-12 MCS, GHQ-Likert and neuroticism) were positively correlated with one another (average r = 0.57, all ps < 0.001).
Regressions
Table 2 reports the outcomes of the key regressions for the two models and for each of the five adult mental health measures. Higher levels of social disadvantage predicted raised levels of psychological distress as gauged using the SF-12 MCS (β = 0.026, s.e. = 0.008, p < 0.001) and the GHQ-Likert measure (β = 0.040, s.e. = 0.007, p < 0.001) and disadvantage also predicted slightly increased neuroticism levels (β = 0.015, s.e. = 0.007, p < 0.05). A 1 s.d. increase in background disadvantage was associated with a significant increase in the likelihood of crossing the case criterion of the GHQ (OR = 1.08, 95% CI 1.04–1.12) but did not predict an increased risk of being diagnosed with depression. There were statistically significant associations between cognitive ability and mental health across all measures such that higher levels of cognitive ability were associated with fewer mental health problems.
Table 2. Outcomes of linear regressions assessing the interaction between cognitive ability and social disadvantage in predicting mental health in the Understanding Society cohort
Standard errors/95% confidence intervals in parentheses. Analyses are weighted using the Understanding Society wave 3 self-completion weight and an inverse probability weight generated to account for selection bias in the sample with available data for this study.
a Variable is standardised to a mean of zero and standard deviation of one. SF-12 MCS, Short-Form 12 mental component summary; GHQ, General Health Questionnaire. §p < 0.1, *p < 0.05, **p < 0.01, ***p < 0.001.
Table 2 also presents the outcomes of the key regressions employed to test our main hypothesis that cognitive ability would moderate the relationship between background disadvantage and current mental health. Across three linear regression models, we identified statistically significant interactions between cognitive ability and social disadvantage in predicting adult mental health (SF-12 MCS: β = −0.059, s.e. = 0.008, p < 0.001; GHQ-Likert: β = −0.034, s.e. = 0.008, p < 0.001; neuroticism: β = −0.025, s.e. = 0.008, p < 0.01). An examination of the simple slopes confirmed that the interaction effects identified were in the direction anticipated. As shown in the lower panel of Table 2, the link between social disadvantage and mental health was substantially stronger at low (−1 s.d.), compared with medium or high (+1 s.d.) levels of cognitive ability. On average across the two distress measures (SF-12 MCS and GHQ), a 1 s.d. increase in disadvantage predicted a 0.089 s.d. increase in distress at low cognitive ability, a 0.043 s.d. increase at mean levels and a −0.004 s.d. decrease at high levels of cognitive ability. Figure 1 illustrates this relationship across the distribution of social disadvantage where moving from very low (−2 s.d.) to very high (+2 s.d.) levels of social disadvantage was associated with a 0.36 s.d. increase in psychological distress at low levels of cognitive ability, whereas no adverse effects of disadvantage were observed among those with high cognitive ability. A similar, albeit weaker interaction effect, was observed between background disadvantage and cognitive ability in predicting neuroticism as shown in Table 2 and illustrated in Fig. 1.
Fig. 1. Association between social disadvantage and mental health at low (−1 s.d.), medium (mean) and high (+1 s.d.) levels of cognitive ability. Left panel: SF-12 MCS; middle panel: GHQ-12 Likert; right panel: neuroticism.
Next, we showed that cognitive ability moderated the link between social disadvantage and both GHQ-caseness and being diagnosed with depression. An examination of the simple slopes showed that a 1 s.d. increase in social disadvantage was linked to a 20% increase in the odds of GHQ-caseness for those with low cognitive ability, a 10% increase for those of medium ability and was unrelated to elevated GHQ at high cognitive ability levels, as shown in Table 2. Figure 2 plots this interaction effect and indicates that moving from very low (−2 s.d.) to very high (+2 s.d.) levels of social disadvantage was linked with an increase in GHQ-caseness of 14 percentage points (from 21% to 35%) at low levels of cognitive ability, whereas no discernible increase was found for those with high cognitive ability. Finally, the simple slopes analysis revealed marginally significant associations between disadvantage and rates of diagnosis of clinical depression at low and high cognitive ability levels (see Table 2). Whilst social disadvantage was linked to a greater likelihood of depression diagnosis at low levels of cognitive ability as expected, disadvantage predicted a reduced likelihood of being diagnosed with depression at high levels of cognitive ability.
Fig. 2. Association between social disadvantage and mental health at low (−1 s.d.), mean and high (+1 s.d.) levels of cognitive ability. Left panel: GHQ-12 caseness; right panel: diagnosis of depression.
Discussion
Adults from disadvantaged backgrounds were at elevated risk of psychological distress and neuroticism relative to those from more affluent families. However, as anticipated, those characterised by high levels of cognitive ability showed a pattern consistent with being buffered from the potential detrimental psychological effects of even impoverished backgrounds. Similar patterns of results were identified when we examined clinically significant levels of emotional problems and diagnoses of depression: those with higher cognitive ability levels showed no adverse mental health effect of a disadvantaged upbringing. In contrast, background social circumstances appeared to matter substantially for those with low levels of cognitive ability who showed notable social gradients in distress and mental health problems. The potential stress-buffering benefits of cognitive ability appeared to take a graded form: as intelligence levels increased the link between social disadvantage and adverse mental health outcomes diminished. These data represent an important qualification when considering socioeconomic inequalities in psychological distress: social gradients in mental health may not affect all individuals equally, instead they may occur chiefly among those with low-to-average levels of cognitive ability.
Resilience research has previously highlighted a buffering effect of cognitive ability on adolescent depressive symptoms (Riglin et al., Reference Riglin, Collishaw, Shelton, McManus, NG-Knight, Sellers, Thapar, Frederickson and Rice2016), but the present report is the first to show that this protective pattern extends to psychological distress in adulthood in a sample that is representative of the entire adult population across all ages. The suggestion that psychological resources may protect individuals from the adverse impacts of early-life disadvantage on mental and physical health is key to a number of influential life-course models of social gradients (Matthews and Gallo, Reference Matthews and Gallo2011), yet cognitive ability has been somewhat overlooked as a protective resource of this kind. This oversight is intriguing given the diverse ways in which cognitive ability might lessen the negative impact of social disadvantage on psychological health. Greater cognitive resources may, for example, enable greater emotional regulation in the face of stressors (Schmeichel and Tang, Reference Schmeichel and Tang2015) or allow individuals to successfully limit the negative financial and/or social consequences of distressing incidents.
Although future research is needed to uncover the exact mechanisms by which intelligence uncouples the association between background disadvantage and psychological distress, the moderating effect reported here speaks for the inclusion of cognitive ability in diathesis–stress approaches to distress. These posit that psychological distress is not an inevitable consequence of exposure to stress, but is also dependent upon certain vulnerability factors which may or may not predispose an individual to become distressed (Monroe and Simon, Reference Monroe and Simon1991). An interactive perspective of this kind may also be useful for better understanding the role of cognitive reserve as outlined in prior reports (e.g. Martin et al., Reference Martin, Kubzansky, LeWinn, Lipsitt, Satz and Buka2007; Koenen et al., Reference Koenen, Moffitt, Roberts, Martin, Kubzansky, Harrington, Poulton and Caspi2009). In this literature, high cognitive ability is proposed to increase resilience to the emergence of psychiatric symptoms despite the presence of underlying morphologic brain alterations. More efficient use of brain networks may underlie this ability to cope with disease (Stern, Reference Stern2002). In the context of the current study, it may be the case that superior cognitive function modifies the impact of stressors, broadly defined to encapsulate either environmental or neurological ‘insults’, on symptom expression (Barnett et al., Reference Barnett, Salmond, Jones and Sahakian2006). We suggest that in the case of social deprivation, cognitive function could: (i) buffer the effect of exposure to SES-related stress on the development of neuropathology and neuropsychiatric disorders leading to fewer expressed symptoms or, (ii) weaken the expression of symptoms within disorders to a greater extent among disadvantaged individuals who have fewer alternative (social, economic) resources to draw on than their more affluent counterparts.
As is the case in the field of resilience, the current findings do not necessarily translate directly into clinical recommendations (Rutter, Reference Rutter2013), but may generate important strategic implications for the development of preventative measures in mental health care. They highlight an at-risk subset of the population, especially vulnerable to developing psychological difficulties and who may be particularly likely to gain from effective support and early intervention. This support could take several forms including tackling contextual concerns (e.g. marital conflict/relationship instability, parental depression; Klebanov et al., Reference Klebanov, Brooks-Gunn and Duncan1994), material deprivation (Leventhal and Brooks-Gunn, Reference Leventhal and Brooks-Gunn2003; Engle et al. Reference Engle, Fernald, Alderman, Behrman, O'Gara, Yousafzai, de Mello, Hidrobo, Ulkuer, Ertem and Iltus2011), parenting strategies (e.g. Nowak and Heinrichs, Reference Nowak and Heinrichs2008) and child social–emotional functioning (Engle et al., Reference Engle, Fernald, Alderman, Behrman, O'Gara, Yousafzai, de Mello, Hidrobo, Ulkuer, Ertem and Iltus2011; Heberle and Carter, Reference Heberle and Carter2015). Supporting socioeconomically disadvantaged individuals with low-to-average cognitive ability in this way may better equip them to negotiate stressors and alleviate some of the distress associated with disadvantage.
The principal divergence from the predicted pattern of results arose for diagnoses of clinical depression, which did not show an association with background disadvantage. One explanation for this divergence is a reduction in power for this variable due to a smaller available sample size and the lower prevalence of depression compared with GHQ-caseness (see Table 1), which may impact the likelihood of detecting a relationship. Nonetheless, we do observe a significant interaction for depression diagnoses, such that the direction of the typical positive association between disadvantage and depression is reversed at high levels of cognitive ability. This may arise because, unlike the remaining self-report measures, a diagnosis of depression is contingent on access to and engagement with a doctor or other healthcare professional. If those from more advantaged neighbourhoods and those with high cognitive ability tend to be more likely to access diagnosis/treatment for mental health problems (e.g. Beier and Ackerman, Reference Beier and Ackerman2003; Delgadillo et al., Reference Delgadillo, Asaria, Ali and Gilbody2016) then our proposed interaction effect for clinical depression (as diagnosed by a doctor or healthcare professional) would be markedly diminished. In future, researchers should examine whether and how disadvantage and cognitive ability individually and interactively predict treatment-seeking.
Intriguingly, this apparent reversal of the social gradient at high levels of cognitive ability was also identified when mental health was measured by the SF-12 MCS: higher ability individuals from more socially disadvantaged backgrounds showed significantly better mental health than their more socially advantaged counterparts. Differences in the self-reported SF-12 measure cannot be attributed to differential access to treatment and healthcare services and together these patterns appear to further enrich the interactive relationship between cognitive ability and social disadvantage and psychological distress, insofar as they allow for the possibility that better socioeconomic circumstances can in fact be associated with poorer mental health for the above-average intelligent. It is important not to overinterpret this finding given it was not anticipated and was only significant in one of the mental health measures reported here. Nonetheless, it is interesting to note that the SF-12 has been found to be significantly better at discriminating people with and without mental diagnoses than the GHQ-12 (Gill et al., Reference Gill, Butterworth, Rodgers and Mackinnon2007), and thus may be a more sensitive tool for detecting specific effects of this kind.
The current study has some limitations. We relied on self-report measures of mental health rather than clinician-guided interviews and assessments. Future work should redress this and also seek to determine whether the buffering effect of cognitive ability extends to other clinical diagnoses known to vary with SES such as anxiety disorders (Stansfield et al., Reference Stansfield, Clark, Rodgers, Caldwell and Power2011). Second, due to the concurrent timing of our moderator and outcome measures, we cannot infer that cognitive ability levels were not affected by the emergence of mental health problems which could result in inconsistent estimates (Boyce et al., Reference Boyce, Wood, Delaney and Ferguson2017). Whilst the issue of measurement timing cannot be directly addressed with the Understanding Society data, we note that this interpretation does not apply to previous reporting of this key interaction using longitudinal panel data in which cognitive ability was measured during childhood and many years prior to measurements of adult health (Bridger and Daly, Reference Bridger and Daly2017). Further, our reliance on between-person comparisons makes it difficult to rule out alternative explanations for these data particularly given the tendency of multiple risk factors for poor health (including risk of low cognitive ability) to cluster in disadvantaged environments (Melchior et al., Reference Melchior, Moffitt, Milne, Poulton and Caspi2007; Belsky et al., Reference Belsky, Moffitt, Corcoran, Domingue, Harrington, Hogan, Houts, Ramrakha, Sugden, Williams, Poulton and Caspi2016). To better understand this potential confound, future studies should test whether the link between within-person changes in SES and mental health is modified by cognitive ability levels.
Finally, we utilised retrospective measures of social disadvantage prior to entry into the labour market in order to rule out reverse causality (whereby adult SES partially reflects the impact of one's mental health; Chatterji et al., Reference Chatterji, Alegría, Lu and Takeuchi2007; Egan et al., Reference Egan, Daly and Delaney2016). Such retrospective measures of parental occupation and education are open to concerns about recall bias given that individuals may not always correctly remember or report earlier life events. Although recall bias can vary with age and factors such as complexity of childhood circumstances, accounts from datasets in which it is possible to directly compare retrospective and prospective measures from the same cohorts report a substantial degree of consistency, licensing the use of retrospective measures where necessary (Brown, Reference Brown2014; Jivraj et al., Reference Jivraj, Goodman, Ploubidis and de Oliveira2017). Moreover, it has been argued that retrospective reports may in some circumstances lead to an under-reporting of the association between SES and disease outcomes (Batty et al., Reference Batty, Lawlor, Macintyre, Clark and Leon2005). However, it is worth noting that despite this potential attenuation, we report here a substantial difference in distress levels (0.36 s.d.) between those with low cognitive ability raised in the poorest compared with the most affluent families.
Conclusions
The current results provide further support for the hypothesis that cognitive ability may buffer the long-term association between childhood socioeconomic disadvantage and poor mental health in adulthood. Across all measures available, including self-reported measures of psychological distress, trait neuroticism and clinical diagnoses of depression, the disadvantage-mental health link was strongest for individuals with below average levels of cognitive ability and was no longer detectable among those with above average levels of ability. Future work should seek to identify the precise pathways through which cognitive ability may protect individuals in this way.
Acknowledgements
The authors would like to thank the UK Data Service for providing these data as well as those teams responsible for conducting and enabling the Understanding Society panel study, including its principal funders the Economic and Social Research Council (ESRC), the Centre for Longitudinal Studies and the UKHLS participants themselves.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S0033291718002118
