Skip to main content
NCBI home page
The Journals of Gerontology Series B: Psychological Sciences and Social Sciences logoLink to The Journals of Gerontology Series B: Psychological Sciences and Social Sciences
. 2022 Mar 21;77(10):1779–1790. doi: 10.1093/geronb/gbac055

Identifying Racial and Rural Disparities of Cognitive Functioning Among Older Adults: The Role of Social Isolation and Social Technology Use

Kaileigh A Byrne 1,✉, Reza Ghaiumy Anaraky 2
Editor: Alyssa Gamaldo
PMCID: PMC9535781  PMID: 35312775

Abstract

Objectives

Social isolation is associated with poorer cognitive outcomes among older adults. The use of online social technology platforms may provide a means to reduce social isolation. However, research examining whether social technology can mitigate the negative effects of social isolation on cognitive functioning is limited. This study investigates the interaction between social isolation and social technology use on cognitive functioning among older adults and seeks to identify racial and rural–urban differences in this relationship.

Method

Data were obtained from the Health and Retirement Study 2014–2018 waves (N = 5,358). Participants (aged 50–102) completed self-report measures of social isolation, loneliness, and frequency of online social communication and completed the modified Telephone Interview for Cognitive Status, which assesses cognitive functioning. Examinations of race focused on differences between Black/African American and White/Caucasian groups; rurality was operationalized using Beale Rural–Urban Continuum Codes. Data were analyzed using structural equation models.

Results

Social technology use moderated the negative relationship between social isolation and cognitive functioning, controlling for age, education, gender, wealth, and general computer usage. Greater social technology use was associated with better cognitive functioning among socially isolated older adults. Results showed evidence of racial, but not rural–urban, differences in the relationship between social technology use and cognitive functioning. Regardless of the degree of social isolation, frequent social technology use was associated with improved cognitive functioning in Black/African American older adults but not White/Caucasians older adults.

Discussion

Social technology may represent a way to mitigate cognitive decline, particularly among Black/African American older adults.

Keywords: Cognition, Rurality, Social Interaction, Social Networks, Technology

Social isolation refers to an objective lack of meaningful social connections, and loneliness refers to the subjective feeling of being alone and psychologically distant from others (Donovan & Blazer, 2020; Mund et al., 2020). There is strong evidence that social isolation and loneliness are associated with greater risk of cognitive decline (Griffin et al., 2020; Shankar et al., 2013; Sutin et al., 2020). The onset of normal age-related declines in cognitive performance typically begins around age 50 (Verhaeghen & Salthouse, 1997). However, more pronounced cognitive decline, called cognitive impairment, can interfere with daily activities and may represent a precursor of dementia (Wilson et al., 2011). While cognitive impairment is uniformly negative in terms of health outcomes, it affects select communities disproportionately: rural-dwelling older adults and Black/African American minorities (Mehta & Yeo, 2017; Weden et al., 2018). It is therefore important to understand factors that may exacerbate or temper cognitive decline, particularly among high-risk groups.

One such factor that may mitigate the detrimental effect of social isolation on cognitive decline is reliance on social technology used to maintain contact with loved ones or forge new connections. Social technology refers to online social networking platforms, such as Skype, Zoom, or Facebook/Facebook Messenger, that allow for synchronous video, voice, and instant messaging communication (Kaplan & Haenlein, 2010; Zhang et al., 2021). This means of communication is particularly promising as technology use among senior citizens has increased substantially over the past 20 years. While only 14% of older adults reported using the Internet in 2000, in 2017 that percentage increased to 67% (Anderson & Perrin, 2017). Furthermore, frequent technology communication among individuals aged 65 and older is associated with greater social support and diminished feelings of loneliness (Zhang et al., 2021). Social technology use may, therefore, represent a means to temper the negative effects of social isolation and loneliness on cognitive declines. In the present study, we first examine how the frequency of such social technology use influences the relationship between social isolation and cognitive functioning, and second, focus on characterizing racial and rural disparities within this putative relationship.

Substantial evidence suggests that social isolation and loneliness predict a heightened risk of cognitive decline and dementia among older adults (Griffin et al., 2020; Shankar et al., 2013; Sutin et al., 2020). One study using the Health and Retirement Study (HRS) data established that loneliness is associated with a steeper decline in cognitive functioning over a 10-year period and a 40% increased risk of dementia, even when controlling for demographics, clinical conditions, and genetic risk factors (Sutin et al., 2020). Similarly, other work with the HRS data has shown that social isolation is associated with poorer cognitive functioning among older adults (Griffin et al., 2020). Meta-analysis findings have provided further evidence that social isolation is associated with poorer cognitive functioning, including memory, executive functioning, and global cognition, among older adults aged 50 and older (Evans et al., 2019).

In contrast to the negative effects of social isolation and loneliness, technology use among older adults may be beneficial to both social well-being and cognitive functioning. Social technology use is associated with higher subjective well-being, lower depressive symptoms, and lower levels of loneliness among older adults (Chopik et al., 2016; Zhang et al., 2021). Moreover, greater computer use is associated with better verbal reasoning and executive functioning, when controlling for age, gender, and education (Calhoun & Lee, 2019; Tun & Lachman, 2010). An experimental study further demonstrated that social network-naïve older adults aged 75–86 who were assigned to an eight-week Facebook-using intervention exhibited greater working memory updating at post-test compared to both an active control and waitlist control group (Myhre et al., 2017). This work suggests that social technology may be linked with improved cognitive functioning in older adults.

However, specific marginalized populations may encounter additional challenges, particularly in access to technology and risk of developing cognitive impairment. The prevalence of cognitive impairment is 2.7 times higher in Black/African American older adults aged 55 and older compared to White/Caucasian older adults (Langa et al., 2010). Furthermore, individuals living in rural regions experience greater rates of cognitive impairment and dementia than those living in urban regions (Weden et al., 2018). Such racial and rural disparities in cognitive impairment underscore the importance of identifying ways to mitigate cognitive decline in these populations. While technology can have beneficial effects on cognitive functioning in older adults, racial and rural minorities often face structural difficulties in Internet access. Rural regions are more likely to have limited access to broadband internet (Douthit et al., 2015), and Black and Hispanic minorities are more likely to live in socioeconomically disadvantaged neighborhoods with poorer technological infrastructure, which can limit Internet access (Ray et al., 2017). Despite the potential cognitive and psychosocial benefits of using social technology, there has been limited work examining racial and rural disparities in older adults’ social technology use. It is therefore unclear how social technology use influences cognitive functioning in racial minorities and rural older adults, yet understanding such relationships are critically important in these groups that are at high-risk for cognitive impairment.

While there is evidence that social isolation increases the risk of cognitive impairment (Griffin et al., 2020; Shankar et al., 2013; Sutin et al., 2020), the role of social technology use in moderating this relationship is unclear. This work seeks to bridge that gap in knowledge. Using HRS data, this study first aims to assess whether using online technology for social purposes can reduce the detrimental effects of social isolation on cognitive functioning among adults aged 50 and older. Furthermore, the second aim of the study is to characterize rural–urban disparities and racial differences between Black/African American and White Caucasian racial groups in the interaction between social isolation and social technology use on cognitive functioning. We predict that greater social isolation will be associated with poorer cognitive functioning, but greater social technology use will reduce the negative impact of social isolation on cognitive functioning. It is further expected that the detrimental effects of social isolation on cognitive functioning will be greater in Black/African American older adults and rural-living individuals compared to White/Caucasian and urban dwellers, respectively. Because access to technology can be particularly challenging in underserved populations (Douthit et al., 2015; Ray et al., 2017) and cognitive impairment is also more prevalent in these populations (Langa et al., 2010; Weden et al., 2018), a three-way interaction is expected; lower social technology use and greater social isolation may be associated with poorer cognitive functioning in Blacks/African Americans compared to Whites/Caucasians and rural dwellers compared to urban dwellers.

Method

Sample

This research investigation received ethics approval from the university’s Institutional Review Board before study procedures began. Data were analyzed from the HRS, a nationally representative longitudinal study of Americans aged 50 and above that includes demographics, health, psychosocial, and cognitive measures. Beginning in 1992, the Institute for Social Research at the University of Michigan began collecting survey data from study participants every 2 years (Juster & Suzman, 1995; Sonnega et al., 2014). In 2006, the HRS introduced the Participant Psychosocial and Lifestyle Questionnaire to a random subsample of half the study participants who complete this survey during every biannual survey wave (Smith et al., 2013). For this study, data were primarily obtained from the core data set and the Psychosocial and Lifestyle Questionnaire for the 2014, 2016, and 2018 waves. Additionally, rurality information from these participants was obtained by pooling data from the 2013 and 2003 HRS Cross-Wave Census Region/Division data waves (using the 2003 response if a 2013 response was missing). The final sample size was comprised of observations that had cognitive functioning data for at least one-wave/timepoint and did not have missing values for all independent variables and covariates.

Measures

Cognitive functioning

The HRS data set includes the modified Telephone Interview for Cognitive Status (TICS-m), a valid and reliable measure of cognitive functioning that has strong sensitivity and specificity for identifying dementia (Castanho et al., 2014; Seo et al., 2011). In line with previous research (Sutin et al., 2020), four assessments from the HRS TICS-m were used to measure cognitive functioning: (a) immediate memory recall of 10 items (range 0–10 points), (b) delayed memory recall of 10 items (range 0–10 points), (c) a backward count test of attention in which participants count backward from 20 (range 0–2 points), and (d) the serial sevens subtraction task in which participants count down from 100 by increments of seven (range 0–5 points; Crimmins et al., 2011). Thus, the HRS core data include a subset of items from the TICS-m, rather than the entire TICS-m. This total TICS-m score in the HRS ranges from 0 to 27 points with higher scores reflect higher cognitive functioning. Previous research demonstrates that measuring psychopathology, including dementia, along a continuous scale provides more reliable and valid psychometric information than discrete categorical scales (Markon et al., 2011). Therefore, the present study examines cognitive functioning across a continuous scale.

Demographic information

Self-reported age, biological sex, education (in years), race, wealth, and general computer use at baseline were included in analyses. The race was classified as non-Hispanic White/Caucasian, non-Hispanic Black/African American, American Indian/Alaskan Native, Asian, Hawaiian Native/Pacific Islander, and other/unknown. Following previous research with this data set (Bosworth & Smart, 2009; Gillen et al., 2020), wealth was operationalized as the sum of all financial (e.g., stocks, bonds, and savings) and housing assets. General computer use indicates the frequency of using a computer for email, Internet, or other tasks; higher numbers reflect greater frequency of computer use.

Rurality

The nine-category Beale Rural–Urban Continuum Code categorization system, developed by the United States Department of Agriculture (USDA), was utilized to define rurality (USDA ERS, 2013). Beale Rural–Urban Continuum Codes reflect a county’s degree of metropolitan characteristics based on each county’s population size and distance from a metropolitan region. The HRS data condense these nine categories into three groups, such as a code of 1 indicates an urban county (metropolitan area with population >1,000,000), a code of 2 indicates a suburban county (metropolitan area with a population of 250,000–1,000,000), and codes ≤3 are categorized as rural counties (nonmetropolitan counties with population <250,000).

Social technology use

Participants indicated the frequency in which they communicated using Skype, Facebook, or other social media with any of their children, other family members, or friends (not including individuals who live with them). Participants responded to this question separately for each of the three relationships (children, other family, and friends) using a 1 (3+ Times a Week) to 6 (Hardly Ever or Never) Likert scale (α = 0.87). Previous research has validated this measure with older adults, demonstrating high internal consistency (α = 0.87; Zhang et al., 2021). We reversed the coding on these three items so that higher scores on this construct reflect greater levels of social technology communication with family and friends.

Loneliness

The 11-item shortened version (Hughes et al., 2004) of the UCLA Loneliness Scale (Russell et al., 1996) was used to assess subjective feelings of loneliness and social isolation. This shortened version has been shown to have high internal consistency in older adult populations (α = 0.87; Lee & Cagle, 2017). Participants indicate the frequency with which they experience each of the items (e.g., feel left, alone, or isolated from others) using a three-item response scale ranging from never/hardly ever to often. After reverse-coding necessary items, sum scores (ranging from 11 to 33) were calculated such that higher scores reflect greater loneliness (α = 0.88).

Cohabitation status

Cohabitation status was defined based on whether the participant lived with someone (i.e., spouse or partner) or lived alone.

Contact with social network

This nine-item social isolation measure assesses the frequency with which participants (a) meet up, (b) talk on the phone, or (c) write/email with their children, other family members, or friends (α = 0.71; Zhang et al., 2021). The response scale ranged from 1 (3+ Times a Week) to 6 (Hardly Ever or Never), and all items were reverse-coded and averaged so that higher values indicate greater contact with family and friends.

Perceived social support

Perceived social support was assessed through 12 separate questions regarding how well the participant feels their partner/spouse, children, other family members, and friends (a) understand the way they feel, (b) can be relied upon if they have a serious problem, and (c) they can open up to and talk about their worries (Bertera, 2005; Walen & Lachman, 2000). The response scale ranges from 1 (A Lot) to 4 (Not at All; α = 0.81). All items were reverse coded and averaged so that higher scores reflect greater perceived social support.

Social engagement

Social engagement was defined as the frequency of participating in the following seven activities: (a) working with children or young people, (b) doing activities with grandchildren, nieces/nephews, or neighborhood children, (c) volunteering, (d) attending educational or training courses, (e) going to a sport, social, or other club, (f) participating in a local community arts group, and (g) attending meetings of nonreligious organizations (Amano et al., 2020). Response options ranged from 1 (Daily) to 7 (Never; α = 0.66). Items were reverse-coded and averaged; higher scores indicate more frequent social engagement.

Data Analysis

The categorical variable of the race was categorized as White/Caucasian, Black/African American, or other racial group, and the categorical variable Rurality was operationalized as Urban, Rural, or Suburban. In all models, the independent variables were specified as social technology use and social isolation, and cognitive functioning was the outcome variable. Gender, age, level of education, wealth, and general computer use frequency were included in all models as covariates.

Loneliness, contact with social network, social support, social engagement, and cohabitation were conceptualized to be latent variable indicators of social isolation. In order to confirm these latent variables, we first carried out a Confirmatory Factor Analysis (CFA). CFA is a common statistical method that tests the factor structure of a measurement instrument and can assess its validity (Harrington, 2009). The latent variables resulting from the CFA along with the covariates were used in a Structural Equation Modeling (SEM) framework to predict cognitive functioning based on TICS-m scores. SEM is a statistical approach for studying the relationships between observed (e.g., age and race) and latent (e.g., social isolation and social technology use) variables (Hoyle, 1995). The scores for cognitive functioning were measured in 2014, 2016, and 2018.

We first sought to determine whether social isolation was predictive of changes in cognitive functioning over a four-year period using latent growth analyses in SEM framework where there were records of cognitive functioning from 2014, 2016, and 2018. However, there was no significant difference between the scores measured across these three time periods (p = .905). Because there was no significant difference, a latent growth model would be disadvantageous as studying the slope term would be inconsequential and only reduce the power of analyses. Therefore, instead of a latent growth model, we conducted a multilevel SEM model with random intercepts to account for repeated measures of cognitive functioning scores from each participant. TICS-m cognitive functioning scores were regressed on the independent variables—social technology use and social isolation—and the covariates. To test the hypothesis that greater social technology use would reduce the negative impact of social isolation on cognitive functioning, a two-way interaction between social isolation and social technology use was included in the model.

To assess the effects of race and rurality, we carried out Latent Class Analyses (LCA), a statistical approach for studying qualitatively different clusters (groups) of categorical data. In this case, these clusters were race (Model 2) and rurality (Model 3). Because race and rurality were observed variables and not latent variables, the “Knownclass” option in Mplus was utilized. Three different racial and rurality groups were analyzed as different clusters. LCA enables us to test if any of effects in the overall SEM model (Model 1) are different per racial or rurality group. Two-way and three-way interactions were included to test the remaining study hypotheses.

While social technology use was used as a continuous variable for all statistical analyses, a median split was performed for this variable to facilitate graphic interpretation of the interaction effects. Values below the median were dichotomized as “low social technology use” (coded as 0), and values above the median were dichotomized as “high social technology use” (coded as 1). All analyses were carried out in Mplus 8.4, with a robust maximum likelihood estimator. This estimator is robust to any non-normality issues.

Results

Sample Demographics

After pooling data across waves, the data set used for analyses consisted of 5,358 observations (Mage = 70.54, SDage = 9.92, 58.9% female). The initial 2014 wave sample included 5,443 observations, and there were 85 participants who were lost to follow-up from 2014 to 2018. For analytic purposes, we stratified race as non-Hispanic White/Caucasian (71.8%), non-Hispanic Black/African American (20.5%), and members of other racial/ethnic backgrounds (7.7%). The other racial/ethnicity group was combined into a single category because of small sample sizes for each of these individual racial/ethnic groups (American Indian/Alaskan Native, Asian, Native Hawaiian/Pacific Islander, and Other/Unknown). The sample is shown in Table 1.

Table 1.

Descriptive Information for Primary Study Variables

Stratification by race
Black/African American (N = 1,099) Other racial group (N = 410) White/Caucasian (N = 3,849) Total (N = 5,358) p Value
Variable Mean (SD) Mean (SD) Mean (SD) Mean (SD)
Age 67.42 (8.75) 66.80 (9.20) 72.19 (9.97) 70.54 (9.92) <.001
Education (years) 12.72 (2.72) 11.74 (4.21) 13.26 (3.00) 13.03 (3.08) <.001
TICS-m cognitive function 14.25 (3.68) 14.99 (3.72) 16.12 (3.43) 15.66 (3.58) <.001
Social technology use 1.24 (1.72) 1.43 (1.72) 1.11 (1.50) 1.16 (1.56) .162
General computer usage 4.37 (2.70) 4.73 (2.67) 4.70 (2.70) 4.64 (2.70) .002
Wealth (in US dollars) −13,145 (89.3 k) 10,091 (202 k) 155,399 (1.69 mil) 109,709 (1.43 mil) <.001
Biological sex n n n n <.001
 Female 748 (68.1%) 234 (57.1%) 2,173 (56.5%) 3,155 (58.9%)
 Male 351 (31.9%) 176 (42.9%) 1,676 (43.5%) 2,203 (41.1%)
Stratification by rurality
Rural (N = 1,352) Suburban (N = 1,181) Urban (N = 2,823) Total (N = 5,356) p Value
Variable Mean (SD) Mean (SD) Mean (SD) Mean (SD)
Age 71.88 (9.66) 71.68 (9.82) 69.92 (10.01) 70.54 (9.92) <.001
Education (years) 12.83 (2.83) 12.80 (3.38) 13.23 (3.06) 13.04 (3.08) <.001
TICS-m cognitive function 15.47 (3.66) 15.48 (3.62) 15.82 (3.53) 15.66 (3.58) .004
Social technology use 0.94 (1.49) 1.23 (1.51) 1.24 (1.61) 1.16 (1.56) .022
General computer usage 4.52 (2.73) 4.60 (2.72) 4.71 (2.68) 4.64 (2.70) .094
Wealth 77,832 (293 k) 110,465 (575 k) 124,705 (1.93 mil) 109,709 (1.43 mil) .614
Biological sex n n n n .168
 Female 821 (60.7%) 702 (59.4%) 1,630 (57.7%) 3,155 (58.9%)
 Male 531 (39.3%) 479 (40.6%) 1,193 (42.3%) 2,203 (41.1%)
Open in a new tab

Note: mil = million; SD = standard deviation; TICS-m = modified Telephone Interview for Cognitive Status.

To assess potential collinearity issues, correlations between all covariates and social technology were performed. Education and general computer use were correlated (r = 0.17, p < .01); however, as both represent distinctive constructs, both were kept in the subsequent models. No other significant associations were observed.

Confirmatory Factor Analysis

Table 2 reports the results of the CFA. The social isolation construct was created using loneliness, social support, contact with social network, social engagement, and cohabitation status. Social engagement and cohabitation status items were removed from the model due to their poor factor loadings for the social isolation construct. After removing these two items, we calculated the average variance extracted (AVE) for each of the latent variables that had good factor loadings (loneliness, social support, social contact, and the three items assessing social technology). AVE shows the amount of variance captured by the items in relation to the variance by measurement error. The observed AVEs were well above acceptable thresholds (Hair et al., 2010). Furthermore, the CFA model had an acceptable fit overall (Standardized Root Mean Square [SRMR] = 0.068, Tucker–Lewis Index [TLI] = 0.916, and Comparative Fit Index [CFI] = 0.955). Although the TLI value is slightly below the good fit threshold, the SRMR and CFI values both suggest a good model fit (Hu & Bentler, 1999). Therefore, social isolation and social technology were used in the SEM model as the latent variables.

Table 2.

Confirmatory Factor Analysis Results

Items Loadings
Social isolation AVE = 0.896
UCLA loneliness 0.663
Social support −0.754
 Contact with social network −0.565
Social engagement* −0.334
Cohabitation status* −0.197
Social technology use AVE = 0.834
How often do you communicate by Skype, Facebook, or other social media to communicate with any of your children? 0.873
How often do you communicate by Skype, Facebook, or other social media to communicate with family members? 0.930
How often do you communicate by Skype, Facebook, or other social media to communicate with your friends? 0.886
Open in a new tab

Note: AVE = average variance extracted. Starred items represent variables with poor factor loadings that were removed from the analyses.

Structural Equation Models

Overall SEM model (Model 1)

A model without the interaction term between social technology use and social isolation constructs was first conducted to use as the baseline model (Table 3). We then created a model with the interaction term and compared it to this baseline model. Results show a significant fit improvement (χ 2(1) = 4.696, p = .030). Therefore, we report the model with the interaction term below (Figure 1).

Table 3.

SEM Model Results

Model 0: The baseline SEM β SE p Value
Social technology use 0.065 0.029 *
Social isolation −0.106 0.020 ***
Age −0.212 0.011 ***
Education (years) 0.336 0.012 ***
Gender −0.071 0.011 ***
General computer usage 0.050 0.011 ***
Wealth 0.024 0.018 p = .176
Model 1: The overall SEM
Social technology use 0.095 0.028 **
Social isolation −0.089 0.020 ***
Social technology use × Social isolation 0.109 0.030 ***
Age −0.214 0.011 ***
Education (years) 0.335 0.012 ***
Gender −0.073 0.011 ***
General computer usage 0.050 0.011 ***
Wealth 0.024 0.018 p = .167
Model 2: Race
Social technology use χ 2(2) = 6.322 *
 White 0.046 0.27 p = .086
 Black 0.147 0.072 *
 Other 0.223 0.079 **
Social isolation −0.089 0.020 ***
Social technology use × Social isolation 0.076 0.025 p = .548
Age χ 2(2) = 15.319 ***
 White −0.285 0.012 ***
 Black −0.159 0.027 ***
 Other −0.236 0.038 ***
Education (Years) χ 2(2) = 18.108 ***
 White 0.315 0.014 ***
 Black 0.321 0.028 ***
 Other 0.357 0.042 ***
Gender −0.095 0.012 ***
General computer usage χ 2(2) = 11.606 **
 White 0.039 0.013 **
 Black 0.053 0.026 *
 Other 0.006 0.040 p = .886
Wealth χ 2(2) = 15.503 ***
 White 0.016 0.011 p = .157
 Black 0.074 0.025 **
 Other 0.086 0.031 **
Model 3: Rurality
Social technology use 0.125 0.038 **
Social isolation −0.038 0.031 p = .218
Social technology use × Social isolation 0.100 0.032 **
Age −0.202 0.011 ***
Education (years) χ 2(2) = 12.006 **
 Urban 0.322 0.016 ***
 Rural 0.372 0.023 ***
 Suburb 0.313 0.027 ***
Gender χ 2(2) = 12.675 **
 Urban −0.050 0.015 ***
 Rural −0.135 0.020 ***
 Suburb −0.069 0.033 ***
General computer usage 0.050 0.011 ***
Wealth 0.030 0.019 p = .120
Open in a new tab

Notes: SEM = Structural Equation Modeling.

*Indicates significance at p < .05,

**indicates significance at p < .01, and

***indicates p < .001.

Figure 1.

Figure 1.

Open in a new tab

Path model results with social technology use and social isolation predicting cognitive functioning, measured by scores on the modified Telephone Cognitive Interview Status, as moderated by race and rurality. Ovals show components of the latent variables (i.e., social technology use and social isolation). The numbers associated to each item is the loading of that item on the relevant latent variable. Other values are standardized regression coefficients. **indicates significance at p < .01, and ***indicates p < .001.

Results revealed a significant two-way interaction between social isolation and social technology use predicting cognitive functioning (Supplementary Figure 1). Using social technology is more effective in improving cognitive functioning for individuals with higher levels of social isolation (β = 0.109, p < .001). Furthermore, a main effect of social technology was also observed such that for every one standard deviation increase in social technology use, cognitive functioning increases by 0.095 standard deviations units (β = 0.095, p < .01). In addition, a main effect of social isolation emerged such that greater social isolation was predictive of poorer cognitive functioning (β = −0.089, p < .001). In terms of covariates, older age (β = −0.214, p < .001) and male sex (β = −0.073, p < .001) were associated with lower levels of cognitive functioning, while higher education (β = 0.335, p < .001) and more frequent general computer use (β = 0.050, p < .001) were associated with higher levels of cognitive functioning. We did not find any effects of wealth on cognitive functioning (β = 0.024, p = .167). This model accounts for 21.7% of the variance in cognitive functioning.

Latent class analysis model with race (Model 2)

To assess racial disparities in the relationship between social isolation, social technology use, and cognitive functioning, the same SEM model was conducted as a latent class model using “Knownclass,” because race is an observed variable. The three-way interaction between race, social technology use, and social isolation, and the two-way interaction between race and social isolation did not reach the level of significance (ps > .05). Furthermore, with the addition of race in this model, the interaction term between social technology use and social isolation became nonsignificant (β = 0.076, p = .548) However, a significant two-way interaction between race and social technology use (χ 2(2) = 6.322, p < .05) showed that social technology use was associated with improved cognitive functioning for Black/African American individuals (β = 0.147, p < .05) and members of other races (β = 0.223, p < .001), but the effect of social technology use on cognitive functioning was not significant for White/Caucasian individuals (p = .086). Figure 2A shows these results.

Figure 2.

Figure 2.

Open in a new tab

Relationship between social isolation and TICS-m cognitive scores and social technology use (0 = low, 1 = high) by (top) race and (bottom) rurality. To facilitate graphic visualization of the results, we conducted a median split in social technology use scores (0 = low social technology use; 1 = high social technology use). On the x-axis, social isolation refers to the composite score of the standardized social variable variables (loneliness, social contact, and social support). On the y-axis, TICS-m score refers to the standardized TICS-m scores in the study population. TICS-m = modified Telephone Cognitive Interview Status.

In addition, race moderated the effect of age on cognitive functioning (χ 2(2) = 9.181, p < .05) such that age is a stronger predictor of cognitive functioning for White/Caucasian individuals (β = −0.285, p < .001) followed by members of other races (β = −0.236, p < .001) and Black/African American individuals (β = −0.159, p < .001). The race also moderated the effect of education on cognitive functioning (χ 2(2) = 18.108, p < .001) such that higher education has the strongest effect on cognitive functioning for members of other races (β = 0.357, p < .001) followed by White/Caucasian individuals (β = 0.315, p < .001), and Black/African American individuals (β = 0.321, p = .001). Furthermore, the effect of general computer usage on cognitive functioning was moderated by race (χ 2(2) = 11.606, p < .01). General computer use predicts cognitive functioning for White and Black races (βs = 0.039, 0.053, ps < .05) but not for other races (β = 0.006, p = .886). Finally, we found that the effect of wealth on cognitive functioning was moderated by race (χ 2(2) = 15.503, p < .001). While wealth does not predict cognitive functioning for White individuals (β = 0.016, p < .157), wealthier individuals who are Black/African American (β = 0.074, p < .01), or a member of other races (β = 0.086, p < .01) have higher levels of cognitive functioning. To avoid repetition, effects that are not moderated by race are not reported. Supplementary Figure 2 shows the full model.

Latent class analysis model with rurality (Model 3)

The LCA model with rurality revealed similar results to the main SEM model and did not show a three-way interaction effect between rurality, social technology use, and social isolation on cognitive functioning. Furthermore, we did not find two-way interactions between rurality and social isolation or rurality and social technology use predicting cognitive functioning (Figure 2B). However, rurality did moderate the effect of education and gender covariates on cognitive functioning (education: χ 2(2) = 12.066, p < .01, gender: χ 2(2) = 12.675, p < .01). The effect of education on cognitive functioning was highest for rural residents (β = 0.372, p < .001), followed by urban (β = 0.322, p < .001) and suburban (β = 0.313, p < .001) residents. Furthermore, males living in rural areas have the lowest cognitive function (β = −0.135, p < .001), followed by suburban (β = −0.069, p < .001), and urban (β = −0.50, p < .001) residents. As other effects were similar to the main SEM model, we do not report them here (Supplementary Figure 3 shows all the effects).

Discussion

This study examined the interaction between social isolation and social technology use on cognitive functioning and identified rural–urban and racial disparities in this relationship. Consistent with our hypotheses, results demonstrated that greater social technology use attenuated the negative effects of social isolation on cognitive functioning among older adults aged 50 and above, controlling for education, age, gender, wealth, and frequency of general computer use. Furthermore, racial, but not rural, differences were observed in the relationship between social technology use and cognitive functioning. Greater social technology use predicted better cognitive functioning among minority racial groups, but not among White/Caucasian older adults. This finding suggests that, regardless of one’s level of social isolation, using online social technology platforms to communicate with family and friends may be particularly beneficial for cognitive functioning in Black/African American older adults and members of other racial/ethnic backgrounds.

Consistent with prior studies on social isolation and loneliness (Griffin et al., 2020; Shankar et al., 2013; Sutin et al., 2020), this study showed that greater social isolation and loneliness were associated with poorer cognitive functioning and can negatively affect memory and attention processes. In addition, prior research using the HRS data demonstrated that higher social technology use among older adults is linked with better physical health and mental well-being (Chopik et al., 2016). The present study extends this work by showing that greater use of social technology to communicate with family and friends is also associated with better cognitive functioning.

The positive association between social technology use and cognitive functioning is especially pronounced among Black/African Americans and members of other racial groups. Because Black/African American older adults are at higher risk for cognitive impairment and dementia than White/Caucasian older adults (Langa et al., 2010), social technology may be a particularly useful method to slow or temper cognitive decline in this population. Social technology use may foster opportunities for both social engagement and cognitive stimulation, which may have positive downstream effects on age-related cognitive functioning (Kamin & Lang, 2020; Khoo & Yang, 2020; Myhre et al., 2017). However, the relationship between social technology use and cognitive functioning among minority groups has been a largely overlooked research area. Recent work suggests that changes in age-related cognitive functioning may follow a different pattern in Black/African American older adults from White/Caucasian older adults (Tan et al., 2021). Current cognitive interventions and social programs may not be tailored to address the unique challenges of diverse populations. However, social technology can offer a flexible, customizable way for these individuals to engage with their loved ones, many of whom share their culture and identity, in a supportive and cognitively engaging way.

While we anticipated that there would be an interaction between both race and social isolation and rurality and social isolation on cognitive functioning, the results did not support this prediction. The negative effects of social isolation on cognitive functioning appear ubiquitous across racial groups and rural and urban-dwelling older adults. Furthermore, study results did not reveal any rural–urban differences in the effect of social technology use on cognitive functioning. Rural older adults tend to rely more on smaller, close-knit social groups (Henning-Smith et al., 2019). It is possible that rural older adults may have less frequent social contacts, but such relationships may be perceived as more meaningful and reliable compared to urban older adults. We also predicted that lower social technology use and greater social isolation may be associated with poorer cognitive functioning in Blacks/African Americans and rural dwellers. However, the three-way interactions were nonsignificant, and thus, these hypotheses were not supported.

Ancillary findings showed that wealth had a positive effect on cognitive functioning for older adults from minority backgrounds compared to White/Caucasian older adults, which suggests that minorities from higher socioeconomic backgrounds have better age-related cognitive outcomes. Years of education had more beneficial effects on cognitive functioning in rural older adults than urban older adults. Given that rural older adults are more vulnerable to cognitive impairment and dementia than urban older adults (Weden et al., 2018), it is important to identify targeted ways to slow cognitive decline in rural older adults; education appears to be such a factor.

Limitations

Certain study limitations should be considered. First, the associations examined in this study rely on a correlational design, and the causal direction of these relationships cannot be established. It is possible that individuals who have better cognitive capabilities are more open to using social technology to communicate due to a self-selection bias (Zhang et al., 2021). Randomized controlled trials are needed to corroborate such a causal relationship. Second, we note that operationalization of social technology use was based on the frequency of communication, rather than the quality or satisfaction of relationships maintained using technology. Future research should examine how the nature or meaningfulness of such relationships maintained through social technology may impact cognitive functioning.

In examining racial differences, we merged non-White/Caucasian and non-Black/African American participants from other racial and ethnic groups into a single category because of sample size constraints. Future research focused on differences among Hispanic/Latino ethnic groups or Asian, Native American, or Pacific Islander racial groups are needed to better understand ways to improve cognitive functioning using technology in these ethnic and racial groups. In investigating rural differences, we note that the categorization of rural counties was based on populations that had less than 250,000 residents. This limitation can be attributed to restrictions in data availability in the data set, which transformed the nine-item Beale Rural–Urban Continuum Code categorization system into a three-item code; thus, information that would have allowed for a more fine-grained examination of rurality was not available. This categorization of rural counties may introduce heterogeneity in the potential level of community development and availability of services, and it is possible that analysis of rurality using a smaller-population threshold may yield different results of the effect of rurality on the relationship between cognitive functioning and social technology use.

Implications and Future Directions

Social isolation is a modifiable risk factor that occurs late in life and has rapid, detrimental effects on cognitive decline (Griffin et al., 2020; Shankar et al., 2013). This work has implications for how social determinants of cognitive health can improve intervention strategies in underserved racial groups. One possible way to mitigate this risk is through access to social technology platforms, which may both mitigate the negative effects of social isolation on cognitive functioning in older adults and reduce racial disparities in cognitive decline. Future research should be aimed at developing and implementing such interventions, especially among Black/African American older adults, who are both at higher risk of cognitive decline and glean greater benefits of social technology use on cognitive functioning.

Furthermore, this study examined cognitive functioning in terms of the TICS-m short-term memory, long-term memory, and basic attention items. It remains unclear how social technology use may influence the relationship between social isolation and other cognitive functions. Future studies may benefit from examining additional measures of cognition, such as working memory, executive functioning, and decision-making, to understand the relationship between social technology use and social isolation on a broader range of cognitive functions.

Conclusions

This work provides evidence of racial, but not rural–urban, differences in the relationship between online social communication and cognitive functioning among older adults. Furthermore, greater online social communication is associated with better cognitive functioning among individuals experiencing high levels of social isolation. Methods to increase social technology use among older adults, especially among racial minorities, may be beneficial in mitigating cognitive decline.

Supplementary Material

gbac055_suppl_Supplementary_Material
Click here for additional data file. (191.8KB, pdf)

Acknowledgments

The data source is the Health and Retirement Study (HRS), which is publicly available here: https://hrs.isr.umich.edu/. This study was not preregistered. We thank the CCADMR Leadership team, including Sue Levkoff, PhD, Cheryl Dye, PhD, Lucy Ingram, PhD, Marvella Ford, PhD, and Daniella Friedman, PhD for their guidance throughout this research.

Contributor Information

Kaileigh A Byrne, Clemson University, Clemson, South Carolina, USA.

Reza Ghaiumy Anaraky, Clemson University, Clemson, South Carolina, USA.

Funding

This research was supported by the Carolina Center for Alzheimer’s Disease and Minority Research (CCADMR) Pilot Research Grant through the National Institute on Aging (NIA 1P30AG059294-01). The HRS is sponsored by the NIA (U01AG009740) and is conducted by the University of Michigan.

Conflict of Interest

None declared.

Author Contributions

K.A.B. conceptualized the study and wrote the first draft of the manuscript. R.G.A. performed all statistical analyses, wrote the results section, and contributed to editing the manuscript.

References

  1. Amano, T., Morrow-Howell, N., & Park, S. (2020). Patterns of social engagement among older adults with mild cognitive impairment. The Journals of Gerontology, Series B: Psychological Sciences and Social Sciences, 75(7), 1361–1371. doi: 10.1093/geronb/gbz051 [DOI] [PubMed] [Google Scholar]
  2. Anderson, M., & Perrin, A. (2017). Technology use among seniors. Pew Research Center for Internet & Technology. [Google Scholar]
  3. Bertera, E. M. (2005). Mental health in U.S. adults: The role of positive social support and social negativity in personal relationships. Journal of Social and Personal Relationships, 22(1), 33–48. doi: 10.1177/0265407505049320 [DOI] [Google Scholar]
  4. Bosworth, B., & Smart, R. (2009). Evaluating micro-survey estimates of wealth and saving. Available at SSRN 1553165. [Google Scholar]
  5. Calhoun, D., & Lee, S. B. (2019). Computer usage and cognitive capability of older adults: Analysis of data from the Health and Retirement Study. Educational Gerontology, 45(1), 22–33. doi: 10.1080/03601277.2019.1575026 [DOI] [Google Scholar]
  6. Castanho, T. C., Amorim, L., Zihl, J., Palha, J. A., Sousa, N., & Santos, N. C. (2014). Telephone-based screening tools for mild cognitive impairment and dementia in aging studies: A review of validated instruments. Frontiers in Aging Neuroscience, 6, 16. doi: 10.3389/fnagi.2014.00016 [DOI] [PMC free article] [PubMed] [Google Scholar]
  7. Chopik, W. J. (2016). The benefits of social technology use among older adults are mediated by reduced loneliness. Cyberpsychology, Behavior and Social Networking, 19(9), 551–556. doi: 10.1089/cyber.2016.0151 [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Crimmins, E. M., Kim, J. K., Langa, K. M., & Weir, D. R. (2011). Assessment of cognition using surveys and neuropsychological assessment: The health and retirement study and the aging, demographics, and memory study. The Journals of Gerontology, Series B: Psychological Sciences and Social Sciences, 66(Suppl. 1), 162–1 71. doi: 10.1093/geronb/gbr048 [DOI] [PMC free article] [PubMed] [Google Scholar]
  9. Donovan, N. J., & Blazer, D. (2020). Social isolation and loneliness in older adults: review and commentary of a National Academies report. The American Journal of Geriatric Psychiatry, 28(12), 1233–1244. doi: 10.1016/j.jagp.2020.08.005 [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Douthit, N., Kiv, S., Dwolatzky, T., & Biswas, S. (2015). Exposing some important barriers to health care access in the rural USA. Public Health, 129(6), 611–6 20. doi: 10.1016/j.puhe.2015.04.001 [DOI] [PubMed] [Google Scholar]
  11. Evans, I. E. M., Martyr, A., Collins, R., Brayne, C., & Clare, L. (2019). Social isolation and cognitive function in later life: A systematic review and meta-analysis. Journal of Alzheimer’s Disease, 70(s1), S119–S144. doi: 10.3233/JAD-180501 [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Gillen, M., Yang, H., & Kim, H. (2020). Health literacy and difference in current wealth among middle-aged and older adults. Journal of Family and Economic Issues, 41(2), 281–299. doi: 10.1007/s10834-019-09648-w [DOI] [Google Scholar]
  13. Griffin, S. C., Mezuk, B., Williams, A. B., Perrin, P. B., & Rybarczyk, B. D. (2020). Isolation, not loneliness or cynical hostility, predicts cognitive decline in older Americans. Journal of Aging and Health, 32(1), 52–60. doi: 10.1177/0898264318800587 [DOI] [PubMed] [Google Scholar]
  14. Hair, J. F., Black, W., Babin, B., & Anderson, R. (2010). Multivariate data analysis (7th ed.). Upper Saddle River, NJ: Prentice-Hall. [Google Scholar]
  15. Harrington, D. (2009). Confirmatory factor analysis. Oxford university press. [Google Scholar]
  16. Henning-Smith, C., Moscovice, I., & Kozhimannil, K. (2019). Differences in social isolation and its relationship to health by rurality. The Journal of Rural Health, 35, 540–549. doi: 10.1111/jrh.12344 [DOI] [PubMed] [Google Scholar]
  17. Hoyle, R. H. (1995). The structural equation modeling approach: Basic concepts and fundamental issues. In Hoyle R. H. (Ed.), Structural equation modeling: Concepts, issues, and applications (pp. 1–15). Sage Publications, Inc. [Google Scholar]
  18. Hu, L. T., & Bentler, P. M. (1999). Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives. Structural Equation Modeling: A Multidisciplinary Journal, 6(1), 1–55. doi: 10.1080/10705519909540118 [DOI] [Google Scholar]
  19. Hughes, M. E., Waite, L. J., Hawkley, L. C., & Cacioppo, J. T. (2004). A short scale for measuring loneliness in large surveys: Results from two population-based studies. Research on Aging, 26(6), 655–672. doi: 10.1177/0164027504268574 [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Juster, F. T., & Suzman, R. (1995). An overview of the health and retirement study. Journal of Human Resources, 7–56. doi: 10.2307/146277 [DOI] [Google Scholar]
  21. Kamin, S. T., & Lang, F. R. (2020). Internet use and cognitive functioning in late adulthood: Longitudinal findings from the Survey of Health, Ageing and Retirement in Europe (SHARE). The Journals of Gerontology: Series B, 75(3), 534–539. doi: 10.1093/geronb/gby123 [DOI] [PubMed] [Google Scholar]
  22. Kaplan, A. M., & Haenlein, M. (2010). Users of the world, unite! The challenges and opportunities of social media. Business Horizons, 53(1), 59–68. doi: 10.1016/j.bushor.2009.09.003 [DOI] [Google Scholar]
  23. Khoo, S. S., & Yang, H. (2020). Social media use improves executive functions in middle-aged and older adults: A structural equation modeling analysis. Computers in Human Behavior, 111, 106388. doi: 10.1016/j.chb.2020.106388 [DOI] [Google Scholar]
  24. Langa, K. M., Kabeto, M., & Weir, D. (2010). Report on race and cognitive impairment using HRS in 2010 Alzheimer’s disease facts and figures. Retrieved July, 12. doi: 10.1016/j.jalz.2010.01.009 [DOI]
  25. Lee, J., & Cagle, J. G. (2017). Validating the 11-Item Revised University of California Los Angeles Scale to assess loneliness among older adults: An evaluation of factor structure and other measurement properties. American Journal of Geriatric Psychiatry, 25(11), 1173–1183. doi: 10.1016/j.jagp.2017.06.004 [DOI] [PubMed] [Google Scholar]
  26. Markon, K. E., Chmielewski, M., & Miller, C. J. (2011). The reliability and validity of discrete and continuous measures of psychopathology: A quantitative review. Psychological Bulletin, 137(5), 856–879. doi: 10.1037/a0023678 [DOI] [PubMed] [Google Scholar]
  27. Mehta, K. M., & Yeo, G. W. (2017). Systematic review of dementia prevalence and incidence in United States race/ethnic populations. Alzheimer’s and Dementia, 13(1), 72–83. doi: 10.1016/j.jalz.2016.06.2360 [DOI] [PubMed] [Google Scholar]
  28. Myhre, J. W., Mehl, M. R., & Glisky, E. L. (2017). Cognitive benefits of online social networking for healthy older adults. The Journals of Gerontology, Series B: Psychological Sciences and Social Sciences, 72(5), 752–760. doi: 10.1093/geronb/gbw025 [DOI] [PubMed] [Google Scholar]
  29. Mund, M., Freuding, M. M., Möbius, K., Horn, N., & Neyer, F. J. (2020). The stability and change of loneliness across the life span: A meta-analysis of longitudinal studies. Personality and Social Psychology Review, 24(1), 24–52. doi: 10.1177/1088868319850738 [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Ray, R., Sewell, A. A., Gilbert, K. L., & Roberts, J. D. (2017). Missed opportunity? Leveraging mobile technology to reduce racial health disparities. Journal of Health Politics, Policy and Law, 42(5), 901–924. doi: 10.1215/03616878-3940477 [DOI] [PubMed] [Google Scholar]
  31. Russell, D. W. (1996). UCLA Loneliness Scale (Version 3): Reliability, validity, and factor structure. Journal of Personality Assessment, 66(1), 20–40. doi: 10.1207/s15327752jpa6601_2 [DOI] [PubMed] [Google Scholar]
  32. Seo, E. H., Lee, D. Y., Kim, S. G., Kim, K. W., Kim, D. H., Kim, B. J., Kim, M. -D., Kim, S. Y., Kim, Y. H., Kim, J. -L., Kim, J. W., Moon, S. W., Park, J. H., Ryu, S. -H., Yoon, J. C., Lee, N. J., Lee, C. U., Jhoo, J. H., Choo, L. H., & Woo, J. I. (2011). Validity of the telephone interview for cognitive status (TICS) and modified TICS (TICSm) for mild cognitive imparment (MCI) and dementia screening. Archives of Gerontology and Geriatrics, 52(1), e26–e 30. doi: 10.1016/j.archger.2010.04.008 [DOI] [PubMed] [Google Scholar]
  33. Shankar, A., Hamer, M., McMunn, A., & Steptoe, A. (2013). Social isolation and loneliness: Relationships with cognitive function during 4 years of follow-up in the English Longitudinal Study of Ageing. Psychosomatic Medicine, 75(2), 161–1 70. doi: 10.1097/PSY.0b013e31827f09cd [DOI] [PubMed] [Google Scholar]
  34. Smith, J., Fisher, G., Ryan, L., Clarke, P., House, J., & Weir, D. (2013). Psychosocial and lifestyle questionnaire. Survey Research Center, Institute for Social Research. [Google Scholar]
  35. Sonnega, A., Faul, J. D., Ofstedal, M. B., Langa, K. M., Phillips, J. W., & Weir, D. R. (2014). Cohort profile: The health and retirement study (HRS). International Journal of Epidemiology, 43(2), 576–585. https://doi.org/ 10.1093/ije/dyu067 [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Sutin, A. R., Stephan, Y., Luchetti, M., & Terracciano, A. (2020). Loneliness and risk of dementia. The Journals of Gerontology: Series B, 75(7), 1414–1422. doi: 10.1093/geronb/gby112 [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Tan, S. C., Gamaldo, A. A., Brick, T., Thorpe, R. J., Allaire, J. C., & Whitfield, K. E. (2021). The effects of selective survival on black adults’ cognitive development. The Journals of Gerontology, Series B: Psychological Sciences and Social Sciences, 76, 1489–1498. doi: 10.1093/geronb/gbab003 [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Tun, P. A., & Lachman, M. E. (2010). The association between computer use and cognition across adulthood: Use it so you won’t lose it? Psychology and Aging, 25(3), 560–568. doi: 10.1037/a0019543 [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. USDA Economic Research Service. (2013). Rural–Urban Continuum Codes.https://www.ers.usda.gov/data-products/rural-urban-continuum-codes/
  40. Verhaeghen, P., & Salthouse, T. A. (1997). Meta-analyses of age-cognition relations in adulthood: Estimates of linear and nonlinear age effects and structural models. Psychological Bulletin, 122(3), 231–49. doi: 10.1037/0033-2909.122.3.231 [DOI] [PubMed] [Google Scholar]
  41. Walen, H. R., & Lachman, M. E. (2000). Social support and strain from partner, family, and friends: Costs and benefits for men and women in adulthood. Journal of Social and Personal Relationships, 17(1), 5–30. doi: 10.1177/0265407500171001 [DOI] [Google Scholar]
  42. Weden, M. M., Shih, R. A., Kabeto, M. U., & Langa, K. M. (2018). Secular trends in dementia and cognitive impairment of U.S. rural and urban older adults. American Journal of Preventive Medicine, 54(2), 164–172. doi: 10.1016/j.amepre.2017.10.021 [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Wilson, R. S., Leurgans, S. E., Boyle, P. A., & Bennett, D. A. (2011). Cognitive decline in prodromal Alzheimer disease and mild cognitive impairment. Archives of Neurology, 68(3), 351–356. doi: 10.1001/archneurol.2011.31 [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Zhang, K., Kim, K., Silverstein, N. M., Song, Q., & Burr, J. A. (2021). Social media communication and loneliness among older adults: the mediating roles of social support and social contact. The Gerontologist, 61(6), 888–896. doi: 10.1093/geront/gnaa197 [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

gbac055_suppl_Supplementary_Material
Click here for additional data file. (191.8KB, pdf)

Articles from The Journals of Gerontology Series B: Psychological Sciences and Social Sciences are provided here courtesy of Oxford University Press

RESOURCES