The Effects of Education on Cognition in Older Age: Evidence from Genotyped Siblings¹

Introduction

Cognitive health is one of the most critical aspects of health as people begin to enter the later stages of life (Evans et al., 2018). As individuals get older, they begin to face subtle changes to how they function and their overall cognitive ability. There is a great deal of variation in how cognition levels and trajectories vary in old age. Some people may experience and retain a high level of cognition in older age, whereas others in late life experience a decline (Gow et al., 2007). This has many implications because the cognitive ability that a person has in their older age has the potential to influence their autonomy, well-being, and quality of life (Díaz-Venegas et al., 2019). As populations continue to get older, it is imperative to look at specific indicators of cognition in later life, and consider how early life factors, such as educational attainment, continue to shape population-level cognition and health in old age.

There is a myriad of social, economic, and health-related consequences for older adults tied to their cognition. Failing cognition could lead to increased dependence on other people to complete tasks that were once simple (Gill et al., 2010), increased costs of potential treatments or adjustments to lifestyle practices (Vossius et al., 2011), and lead to disability, increased hospital visits, other morbidities and even death (Robertson, Savva, & Kenny, 2013). Research has shown that cognitive decline may be indicative for early phases of dementia or Alzheimer’s disease in later life, thus leading to greater consequences (Ritchie et al., 2016).

Cognition in later life is influenced by a multitude of predictors throughout many stages of the life course. Social connections that an individual has are an indicator of better cognitive function later in life, whereas social isolation is associated with many disadvantages (Evans et al., 2018; Kuiper et al., 2016). Level of cognition has also been positively influenced by lack of illness and mental health problems (Ritchie et al., 2016), greater amounts of exercise and physical activity throughout life (Sofi et al., 2011), being in a relationship with a significant other (Boss, Kang, & Branson, 2015), specific personality traits (Luchetti et al., 2016), and more favorable neighborhood contexts (Ailshire, Karraker, & Clarke, 2017). Conversely, cognition has been negatively associated with greater prevalence of illness (Ritchie et al., 2016), depression (van den Kommer et al., 2013; Zaninotto et al., 2018), lack of exercise or relationships (Sofi et al., 2011; Boss et al., 2015), disadvantaged neighborhood contexts (Ailshire et al., 2017), and age (Plassman et al. 2010). All these factors have implications not just for cognition, but other outcomes that involve health as well. However, despite all these factors that have the potential to influence cognition in later stages of the life course, one that gets the most attention is the role of education.

The influence of education in later life outcomes is well established in many disciplines and is the focus of many health outcomes (Galama, Lleras-Muney, & Kippersluis, 2018). People with higher levels of education tend to be far healthier on average, have greater longevity, and have fewer instances of morbidity throughout the life course (Davies et al., 2018; Lager & Torssander, 2012). This extends to the association that education has with cognition. Despite this, there is a debate about the relationship of education with many outcomes in later life, particularly if education influences these outcomes, or if the outcomes are a result of behavioral or genetic differences. While education has been shown to have positive effects on economic outcomes later in the life course, chiefly income (Clouston et al., 2012), there is a question of how strong the role is regarding cognition.

Research on educational attainment in early life and later cognition is vast (Chen et al. 2019; Clouston et al., 2012; Clouston et al., 2020; Lövdén et al., 2020; Schneewei, Skirbekk, & Winter-Ebmer, 2014; Wilson et al., 2019). Understanding this relationship is critical, since one’s cognitive ability plays a vital role in quality of life towards the end of the life course, and other outcomes such as lifestyle choices and social interactions (Langa et al., 2009; Ritchie et al., 2016). However, there have been some disputes regarding the determinants of cognition among adults in later life, particularly when it comes to the role of education. Some studies have shown that educational attainment is positively associated with cognition in later life (Richards & Hatch, 2011; Schneewei, et al., 2014; Yount, 2008). Chen and colleagues (2019) found that higher education among individuals in early life served as a defensive factor in aging, which in turn helped delay potential cognitive decline later on (Stern, 2009; Stern 2012). Conversely, other research has shown little to no link between education and cognitive abilities later in life (Glymour, Tzourio, & Dufouil, 2012; Zahodne, et al., 2011).

A major component of this debate in educational attainment’s role in cognition is due to whether education is a correlational or causal factor (Schneewei et al., 2014). For example, the association between these two factors could be a result of reverse causation, where high levels of cognitive ability in early life may influence educational attainment. However, education is something that could be influenced by a myriad of unobserved characteristics, which in turn could impact cognition in later life. A large literature has focused on the causal role education plays in childhood cognition, which ultimately could influence these same characteristics later in life (Clouston et al., 2020; Davies et al., 2018; Deary & Johnson, 2010; Richards & Sacker, 2011). Furthermore, there are several suggestions that higher levels of education in life not only affect cognitive ability but also may attenuate aging-associated declines in cognition, although there is doubt regarding the latter (Lövdén et al., 2020).

Research on policies and laws that are implemented at key points in the life course has been shown to have causal effects on later-life cognition. Banks and Mazzonna (2012) studied when compulsory school laws were implemented, specifically the age at which students could leave school. They found that educational attainment increases memory test scores in old age for students and, additionally, executive functioning in male students. Similarly, Glymour and colleagues (2008) discovered that increases in mandatory schooling of students led to improvements in performance on cognition tests, particularly memory scores, decades after they completed their schooling. However, they found that it may in fact not be causal due to other factors, such as genetics. Other studies have shown that higher educational attainment can aid in coping with age-related brain deterioration, which allows people to better handle cognitive tasks in later life (Lenehan et al., 2015).

In addition to issues of reverse causality, potential confounding between cognition and educational attainment has also been raised as an important empirical issue. One typical exploration of the role of confounding due to family background and genetics is to compare siblings. Data limitations have not allowed this technique in general, although some studies have examined the question in early life. These studies have shown that genetics appears to explain a large share of the similarity in cognition between siblings (Moorman, Carr, & Greenfield, 2018; Reynolds et al., 2014). Other research has studied sibling differences with regard to educational attainment and cognition but focused on how education is influenced by cognition, not the inverse (Polderman et al., 2015). Differential investments by parents may also lower levels of sibling similarity (Baier, 2019).²

This paper seeks to examine how sibling differences in educational attainment are related to differences in cognition in older age. This design, which is novel in the study of old age cognition, allows family background and shared genetics to be controlled. We add measured genetics that is unshared between siblings to the sibling difference design to further probe the possibility of genetic confounding between educational attainment and later-life cognition.

Data

We used data from the UK Biobank (UKB) project. The participants, aged between 37 and 74 years, were originally recruited between 2006 and 2010.³ Cognition (fluid intelligence⁴) is measured in the UKB by summing the number of 13 logic puzzles that the participants could answer correctly in two minutes (Davies et al. 2018). This measure is only available for ∼200,000 of the 500,000 UKB participants, as it was added as a module toward the end of the recruitment window.

Educational levels of the UK Biobank participants were measured by mapping each major educational qualification that can be identified from the survey measures to an International Standard Classification of Education (ISCED) category and imputing a years-of-education equivalent for each ISCED category (Lee et al., 2018).

Although siblings are not identified in the survey, respondents’ genetics can be used to measure genetic relatedness among all pairs of respondents. We first use the UKB provided kinship file, listing all pairwise kinships among 100,000 pairs in the sample of nearly 500,000 individuals. We first choose pairs with kinship >0.2, which reflects first-degree biological relatives (parents/siblings). We then choose the remaining pairs who are <13 years apart in age, leaving ∼22,000 sibling dyads. We then chose to keep only one dyad from any family with more than one dyad, leaving ∼17,600 dyads. The number of dyads who also have non-missing cognition scores (only available for ∼1/3 of the full UKB sample) and educational attainment information is 4,138 (8,276 respondents). We include only respondents of European ancestry in our analysis.

We constructed polygenic scores (PGS) for three traits for which large genome-wide association studies (GWAS) are publicly available and do not contain UK Biobank samples: Alzheimer’s disease (Kunkle et al. 2019), cognition (Rietveld et al. 2014), and educational attainment (Lee et al., 2018). We clumped GWAS summary statistics by PLINK(42), using 1,000 Genome Project Phase III European samples as a reference for linkage disequilibrium (LD). We used an LD window size of 1Mb and a pairwise r² threshold of 0.1. We did not apply any p-value thresholding to select variants (Choi & O’Reilly, 2019). PRSice-2 was used to calculate PGS. PGS of each trait was standardized to have a mean of zero and a standard deviation of one.

Table 1 presents descriptive statistics for our sample. Panel 1 shows the sample in the UK Biobank who have European ancestry, showing over 400,000 observations and 34,000 observations of siblings. Panel 2 limits this sample to those with a cognitive test available, where we have over 160,000 observations including 13,000 observations of individuals who have a sibling in the sample. Panel 3 restricts the sample to those sibling pairs where each member also has a cognitive test as well as educational attainment information available, providing nearly 8,300 observations. Comparing our analysis sample to the full sample, we see a high level of similarity among the variables in our analysis.

Results

Table 2 presents our main results. We include separate indicator variables for each educational category, with “no credential” as the omitted group. Column 1 shows the associations for the sample of respondents with European ancestry and non-missing cognition scores and allows a comparison with our analysis sample of siblings. The results are nearly identical and suggest large differences in cognition by educational category. For example, respondents with A-level credentials are over 2.3 points higher (1 SD) than those without credentials. Column 3 adds PGS for educational attainment, Alzheimer’s disease, and cognition. One standard deviation in the AD-PGS is associated with a 0.08 point reduction in cognition. Notably, these genetic controls only reduce the associations between educational attainment and cognition by ∼10%, suggesting a modest role of genetic confounding.

Columns 4 and 5 present our sibling difference specifications. In general, the associations between educational attainment and cognition shrink by ∼40-50% compared to the results that do not control for sibling fixed effects; adding sibling differences in PGS to these models does not change these results. Overall, the results suggest a considerable level of confounding by family background and genetics but continue to point to remaining large associations between educational attainments and cognition in older age.

We next consider whether there is evidence for gene-environment interactions (GxE) between educational attainment (EA) and later-life cognition (Cog). Table 3 first explores the possibility of selection into education, which would suggest an important gene-environment correlation, where polygenic scores would predict both the “environment” of educational attainment and the old age cognition outcome of interest. Column 1 presents associations predicting educational attainment based on sibling differences in polygenic scores, with controls for age fixed effects, sex, and genetic principal components. Even between siblings, we do find evidence that EA-PGS predicts educational attainment, as found in other studies (Fletcher et al., 2020; Lee et al., 2018). In contrast, we find no association between Alzheimer’s disease (AD)-PGS and educational attainment, which mirrors earlier work focusing on APOE that suggests these variants do not impact decisions early in the life course (Cook & Fletcher, 2015). Thus, the AD-PGS appears to be a reasonable candidate to consider in GxE analysis.

With the findings from Column 1 suggesting no evidence of gene-environment correlation for the AD-PGS, Columns 2 and 3 examines gene-environment interactions. The results show evidence that the impacts of educational attainment on later-life cognition are reduced for individuals with a higher genetic risk of Alzheimer’s disease (higher AD-PGS scores) and that this finding is not affected by also controlling for interactions between educational attainment and EA-PGS and Cog-PGS. In order to further explore these findings, Table 1A in the appendix examines gene-environment interactions for each level of educational attainment rather than treating attainment as a continuous measure. The results continue to show evidence that individuals with higher AD-PGS have smaller old age cognitive benefits from educational attainment than do individuals with lower AD-PGS.

Discussion

Previous studies have sought to tie the role that educational attainment plays in cognition later in life. However, much of the literature has found mixed results, including positive and negative associations, with this relationship (Glymour et al., 2012; Richards & Hatch, 2011; Yount, 2008; Zahodne, et al., 2011). Additionally, research has focused attention on whether the relationship between education and cognition in later life is causal or correlational as a result of social, economic, behavioral, or genetic factors (Clouston et al., 2020; Davies et al., 2018; Schneewei et al., 2014). This paper applied a new research design to these questions by examining how sibling differences, as well as genetic predictors, influence the effects of educational attainment on cognition in old age. The analyses conducted indicate that there are indeed associations between educational attainment and cognition in older age, even controlling for confounding by family background and genetic factors.

There are important limitations to this study. First, the analysis is done with sibling dyads, which does not allow an analysis of individuals with no siblings. Another limitation is that this study does not include an examination of individuals of different racial and ethnic backgrounds. Future research should seek to examine how sibling and genetic differences impact the association between educational attainment and cognition by racial groups, as research has shown that there are racial differences at the population level in cognition and Alzheimer’s disease in later life (Barnes et al., 2015; Howell et al., 2017). A third limitation is the lack of experimental or quasi-experimental variation in educational attainments. We instead leverage sibling differences in educational attainment (controlling for genetic factors) to examine sibling differences in cognition. A fourth limitation is that only a subsample of the UKB has information on cognition, which reduced the power of our analysis, although we still use one of the largest datasets of siblings that measure genetics and cognition in older age.

Despite the limitations outlined above, this research is the first to use sibling fixed-effect models and genetic predictors to estimate the effects of education on cognition in older age. Results point to important associations between education and cognition in older age. These findings largely support previous work done on how education can aid in coping with age-related deterioration (Lenehan et al., 2015), and thus act as a vehicle to maintain or improve cognition in later life. We also provide novel evidence of an interaction between educational attainment and the PGS for AD in predicting cognition in later life, suggesting those at higher genetic risks receive smaller benefits from education than those at lower genetic risks. This highlights the need to examine the intersection of both social and genetic factors when studying outcomes in late life.