A Coordinated Analysis of Bidirectional Associations between Life Satisfaction and Cognitive Function

Abstract

Life satisfaction is an important component of wellbeing to consider in relation to cognitive function due to its modifiability, potential utility in public health policy, and associations with cognitive health outcomes. However, little is known about the directionality of the association between life satisfaction and cognitive function, and past longitudinal work has yielded mixed findings. Using a coordinated data analytic approach, the current research used data from five longitudinal studies with at least three co-occurring waves of life satisfaction and cognitive function assessments, including over 60,000 individuals aged 50 years and older across multiple countries. Bivariate latent growth curve models and random intercept cross-lagged panel models were used to model between- and within-person associations and to test the potentially bidirectional association between life satisfaction and cognitive function over time. Findings revealed modest between- and within-person associations between life satisfaction and cognitive function. At the between-person level, individuals with higher life satisfaction also had higher cognitive function and the two constructs changed together across time. However, initial levels of one did not predict long-term change trajectories in the other. At the within-person level, declines in life satisfaction predicted subsequent declines in cognitive function, and vice versa. The current research advances our understanding of the relationship between life satisfaction and cognitive function, suggesting that these constructs change together in the long-term and predict changes in each other in the short-term. Findings provide observational evidence for the potential utility of life satisfaction in promoting healthy cognitive aging.

Introduction

Alzheimer’s Disease and related dementias (ADRD) are a significant public health concern, with around 50 million people living with dementia worldwide (Patterson, 2018). This number is expected to almost triple by the year 2050 (Patterson, 2018), highlighting a critical need for prevention efforts. Life satisfaction, or people’s subjective perceptions that their life is going well, may be a promising target for prevention given its modifiability (Koydemir et al., 2021) and evidence for its associations with cognitive health (e.g., Beck et al., 2024; Peitsch et al., 2016; Zhu et al., 2022) and multiple modifiable dementia risk factors (Livingston et al., 2020; see Willroth et al., 2024 for a review). However, the literature on life satisfaction and cognitive function is newly emerging, and little is known about the directionality of this association.

Several psychological theories lend support for the relationship between life satisfaction and cognition. For example, Fredrickson’s (2001) broaden-and-build theory suggests that positive emotions broaden people’s thought-action repertoires, cognition, and attention, thereby enabling flexible and creative thinking. Over time, this broadening is thought to build physical, cognitive, psychological, and social resources, creating an upward spiral that leads to a more enduring state of wellbeing (Fredrickson, 2001; Fredrickson & Joiner, 2002). As positive emotions are closely tied to life satisfaction (Cohn et al., 2009), individuals with higher life satisfaction may be more likely to exhibit the short-term broadening of cognition and long-term cognitive resources proposed by the broaden-and-build theory. Moreover, the broaden-and-build theory suggests that these more enduring biopsychosocial resources can function as reserves to be drawn upon in the future (Fredrickson, 2001). Similarly, cognitive reserve theories suggest that biopsychosocial resources built throughout the lifespan create a reserve that protects individuals against the effects of neuropathology and age-related cognitive decline (Richards & Deary, 2005). Other theoretical models, such as the Engine Model of Wellbeing (Jayawickreme et al., 2012), further demonstrate how external and internal resources enable wellbeing. Together, these theories highlight how life satisfaction can contribute to biopsychosocial resources that are protective for cognition, and vice versa.

However, past longitudinal research on the association between life satisfaction and cognitive health outcomes has yielded mixed findings. Using samples from Canada, Singapore, and Korea, several studies have found that life satisfaction predicts lower risk of mild cognitive impairment and dementia (Peitsch et al., 2016; Rawtaer et al., 2017; Zhu et al., 2022). However, two studies found that associations between life satisfaction and cognitive outcomes were no longer statistically significant after controlling for other risk factors such as depressive symptoms (Sutin et al., 2018; Willroth et al., 2023), and other studies have failed to find an association between life satisfaction and cognitive outcomes altogether (Kim et al., 2021; Komura et al., 2023). Meta-analyses reaffirm these mixed findings (Beck et al., 2024; Bell, Singham, Saunders, Buckman, et al., 2022; Bell, Singham, Saunders, John, et al., 2022). For example, results of an individual participant data meta-analysis provided evidence for a protective effect of life satisfaction for incident dementia diagnosis, though these effects were not consistent across datasets (Beck et al., 2024).

In addition to this relatively mixed literature, little is known about the directionality of the association between life satisfaction and cognitive function. One possibility is that life satisfaction prevents or reduces cognitive decline. There are several potential mechanisms that may underly such a protective effect, including improved cardiovascular functioning (Sin, 2016), increased social activity (e.g., Pinquart & Sörensen, 2000; Rafnsson et al., 2015), and more positive health behaviors (e.g., not smoking, physical exercise; Grant et al., 2009). The Lancet Commission on Dementia Prevention, Intervention, and Care has identified 12 modifiable dementia risk factors: physical inactivity, social isolation, smoking, depression, hypertension, diabetes, hearing loss, higher body weight, alcohol use, education, traumatic brain injury, and air pollution (Livingston et al., 2020). At least nine of these risk factors showed consistent links to wellbeing (Willroth et al., 2024), making these risk factors plausible mechanisms of the association between life satisfaction and cognitive function. For example, life satisfaction may lead to better cognitive health outcomes by promoting social engagement and physical activity and reducing depression, smoking, and alcohol consumption. Life satisfaction therefore has the potential to influence multiple modifiable dementia risk factors at once, making it a promising target for prevention.

In addition, cognitive decline may lead to later declines in life satisfaction, for example, by reducing older adults’ ability to participate in the activities that they enjoy and value. If so, this would highlight the importance of developing psychosocial interventions to reduce participation barriers and support life satisfaction among individuals experiencing cognitive decline (see Britton et al., 2024 for a review). Finally, several biological, lifestyle, and socioeconomic factors may influence both life satisfaction and cognitive function and thus may lead to spurious between-person associations between life satisfaction and cognitive function. To distinguish among these possibilities, it is critically important to disentangle between-person associations from potentially causal within-person effects. Between-person associations can demonstrate whether people with higher life satisfaction experience better cognitive function relative to those with lower life satisfaction, and controlling for time-invariant covariates in the estimation of these between-person associations can rule out potential confounders. By contrast, within-person effects can identify potentially causal effects of changes in life satisfaction on later changes in cognitive function, and vice versa. Disentangling these between-person and within-person effects is an important first step in distinguishing between potential causal explanations for the association between life satisfaction and cognitive function.

The existing literature on longitudinal associations between life satisfaction and cognitive function is limited in its ability to make these important distinctions, as it primarily consists of prospective cohort studies that examine whether life satisfaction predicts cognitive outcomes at a later time point without accounting for prior levels of cognitive function and without disentangling between- and within-person associations (e.g., Kim et al., 2021; Peitsch et al., 2016; Sutin et al., 2018). One notable exception is a study conducted by Zainal & Newman (2022), which found initial evidence for bidirectional associations between life satisfaction and cognitive function using latent change scores across 23 years within a single sample. Further replication is needed, particularly in light of the broader mixed literature on life satisfaction and cognitive health.

To address these open questions, the current study uses coordinated data analysis (CDA) to examine the potentially bidirectional relationship between life satisfaction and cognitive function in five longitudinal studies of aging. CDA uses tools from meta-analysis to test the same research question across several independent datasets. By synthesizing results across multiple large-scale studies, CDA increases confidence in study findings and the replicability and generalizability of results. Given that existing single study papers on life satisfaction and cognitive function have yielded inconsistent findings, the current research makes important contributions to the field’s overall understanding and confidence in this association using a CDA approach.

Using five longitudinal datasets, the current CDA addresses two primary aims. In Aim 1, we used bivariate latent growth curve models to investigate how levels and change in life satisfaction were associated with levels and change in cognitive function across older adulthood. We hypothesized that lower initial levels of life satisfaction would be associated with lower initial cognitive function (H1.1) and that less positive life satisfaction trajectories would be associated with steeper cognitive decline (H1.2). In Aim 2, we used random intercept cross-lagged panel models to test the between-person associations (H2.1), bidirectional within-person associations (i.e., cross-lagged effects; H2.1.2 and H2.3), and concurrent within-person associations of life satisfaction and cognitive function. We predicted that life satisfaction and cognitive function would be positively associated at the between-person level (H2.1), that within-person declines in life satisfaction would predict subsequent declines in cognitive function (H2.2), that within-person declines in cognitive function would predict subsequent declines in life satisfaction (H2.3), and that life satisfaction and cognitive function would be positively associated concurrently at the within-person level (H2.4). Table 1 outlines our hypotheses and their associated models and parameters.

Table 1.

Mapping of Hypotheses onto Models and Parameters

Hypothesis	Model	Parameter	Prediction	Support for Prediction	Meta-analytic Estimates
1.1. Lower initial levels of life satisfaction will be associated with lower initial cognitive function	BLGCM	Intercept-intercept correlation	Positive, 95% CI does not contain 0	Yes; Statistically significant in 100% of studies (i.e., 5 out of 5)	r = 0.19, 95% CI [0.07, 0.32]
1.2. Less positive life satisfaction trajectories will be associated with more cognitive decline	BLGCM	Slope-slope correlation	Positive, 95% CI does not contain 0	Yes; Statistically significant in 80% of studies (i.e., 4 out of 5)	r = 0.23, 95% CI [0.13, 0.32]
2.1. Life satisfaction and cognitive function will be positively associated at the between-person level	RI-CLPM	Intercept-intercept correlation	Positive, 95% CI does not contain 0	Yes; Statistically significant in 100% of studies (i.e., 5 out of 5)	r = 0.187, 95% CI [0.08, 0.30]
2.2. Higher life satisfaction will prospectively predict higher cognitive function at the within-person level	RI-CLPM	Cross-lagged effect of life satisfaction on cognitive function	Positive, 95% CI does not contain 0	Yes; Statistically significant in 80% of studies (i.e., 4 out of 5)	B = 0.060, 95% CI [0.04, 0.08]a
2.3. Higher cognitive function will prospectively predict higher life satisfaction at the within-person level	RI-CLPM	Cross-lagged effect of cognitive function on life satisfaction	Positive, 95% CI does not contain 0	Yes; Statistically significant in 60% of studies (i.e., 3 out of 5)	B = 0.101, 95% CI [0.08, 0.12]a
2.4. Life satisfaction and cognitive function will be positively associated concurrently at the within-person level	RI-CLPM	Covariance of the residuals	Positive, 95% CI does not contain 0	Yes; Statistically significant in 80% of studies (i.e., 4 out of 5)	r = 0.053, 95% CI [0.03, 0.08]

^aContinuous time meta-analytic cross effect.

Method

Transparency and Openness

The current study design, hypotheses, and analytic plan were preregistered on the Open Science Framework (OSF): https://osf.io/hjv8g/?view_only=2efd0f6af9ec4b7ebed0a4efe03e8fff. Data cleaning and analytic code are available on OSF at the same link (Willroth et al., 2025). Using guidelines for transparent preregistration deviation reporting (Willroth & Atherton, 2024), we identified one deviation and three unregistered steps in our preregistration (see Supplemental Table 1 for their descriptions and justifications). The current study involved secondary analysis of five existing datasets. Two of the datasets (i.e., Health and Retirement Study; Survey of Health, Ageing, and Retirement in Europe) are publicly available, and the remaining three datasets are available through protected access. Instructions for data access are available in Supplemental Table 2. This research was approved by the Washington University in St. Louis Institutional Review Board (protocol #202207090, “Well-being and Dementia).

Samples

The Integrative Analysis of Longitudinal Studies of Aging (IALSA) research network (https://www.maelstrom-research.org/network/ialsa) and the Gateway to Global Aging Data platform (https://g2aging.org/) were used to find longitudinal datasets that met the study-level inclusion criterion of a minimum of three co-occurring waves of life satisfaction and cognitive function measures. Study-level criteria related to assessment measures are outlined in the Measures section below. Based on these criteria, five existing longitudinal datasets were included in the present research. Participant-level inclusion criteria required that participants be 50 years and older and have at least one concurrent measurement occasion of life satisfaction and cognitive function. Table 2 presents a summary of the datasets included in our analyses, and Table 3 presents descriptive statistics for variables in each dataset.

Table 2.

Study Descriptions

Study	Location	N	Analytic Baseline Year	Age Range	# of MOs	Interval between MOs in Years	Life Satisfaction Measure	Cognitive Function Domains
MAP/MARS	United States	1,702	2008	65–101	Up to 10	1	SWLS	Episodic Memory (immediate & delayed recall), Semantic Memory, Working Memory, Perceptual Orientation, Perceptual Speed
HRS	United States	11,961	2006	51–104	Up to 4	4	SWLS	Episodic Memory (immediate & delayed recall), Mental Status
ELSA	England	8,606	2005	50–90	Up to 7	2a	SWLS	Episodic Memory (immediate & delayed recall), Verbal Fluency
SHARE	28 European countries and Israel	35,218	2006	50–104	Up to 7	2b	Single item	Episodic Memory (immediate & delayed recall), Verbal Fluency
LASA	The Netherlands	4,410	1992–1993 (Cohort 1); 2002–2003 (Cohort 2); 2012–2013 (Cohort 3)	54–85	Up to 6	3–4	Two items (current life, life as a whole)	Episodic Memory (immediate & delayed recall), Processing Speed

Note. MO = maximum number of measurement occasions per participant. SWLS = Satisfaction with Life Scale.

^aELSA wave 6 was not included in the current research because not all cognitive function measures were collected during this wave, leading to a 4-year interval between waves 5 and 7.

^bSHARE wave 3 was not included in the current research because life satisfaction data is not available during this wave, leading to a 4-year interval between waves 2 and 4.

Table 3.

Descriptive Statistics of Key Study Variables

	MAP/MARS	HRS	ELSA	SHARE	LASA
Variable	M (SD) or %	M (SD) or %	M (SD) or %	M (SD) or %	M (SD) or %
Life Satisfaction
Wave 1	7.48 (1.75)	6.76 (2.45)	7.09 (2.06)	7.54 (1.80)	7.26 (1.38)
Wave 2	7.56 (1.68)	6.60 (2.49)	6.70 (2.14)	7.86 (1.66)	7.26 (1.42)
Wave 3	7.56 (1.73)	6.89 (2.41)	6.86 (2.04)	7.75 (1.72)	7.30 (1.36)
Wave 4	7.63 (1.66)	7.01 (2.38)	6.95 (2.11)	7.80 (1.66)	7.26 (1.39)
Wave 5	7.62 (1.64)	-	7.03 (2.05)	7.69 (1.67)	7.24 (1.29)
Wave 6	7.71 (1.68)	-	6.99 (2.02)	7.92 (1.56)	7.25 (1.28)
Wave 7	7.65 (1.71)	-	6.99 (1.99)	7.97 (1.53)	-
Wave 8	7.71 (1.66)	-	-	-	-
Wave 9	7.58 (1.67)	-	-	-	-
Wave 10	7.67 (1.62)	-	-	-	-
Cognitive function
Wave 1	0.14 (0.61)	0.00 (1.00)	−0.01 (0.85)	−0.02 (0.84)	−0.01 (0.85)
Wave 2	0.17 (0.65)	−0.10 (1.03)	0.00 (0.91)	0.06 (0.89)	0.11 (0.90)
Wave 3	0.13 (0.72)	−0.10 (1.06)	0.02 (0.92)	0.09 (0.90)	−0.05 (0.88)
Wave 4	0.06 (0.74)	0.03 (1.05)	0.02 (0.94)	0.03 (0.90)	0.23 (0.92)
Wave 5	0.05 (0.80)	-	0.00 (0.98)	−0.01 (0.92)	0.12 (0.91)
Wave 6	0.01 (0.87)	-	0.02 (1.01)	0.01 (0.90)	−0.18 (0.85)
Wave 7	−0.01 (0.89)	-	0.02 (1.02)	−0.01 (0.90)	-
Wave 8	−0.02 (0.95)	-	-	-	-
Wave 9	−0.02 (0.99)	-	-	-	-
Wave 10	−0.04 (0.95)	-	-	-	-
Age	81.07 (7.04)	69.99 (8.96)	65.83 (9.68)	64.94 (9.86)	65.87 (8.36)
% Men/male	24%	41%	44%	45%	48%
% Women/female	76%	59%	56%	55%	52%
% White	75%	83%	-	-	-
% Black/African American	24%	13%	-	-	-
% Other Racial Identity	1%	4%	-	-	-
APOE genotype (presence of e4)	25%	-	-	-	28%
Years of Education	15.51 (3.14)	12.54 (3.12)	11.59 (2.65)	10.61 (4.32)	9.80 (3.53)
Neuroticism	2.93 (1.41)	6.58 (2.02)	-	-	1.89 (1.88)
Depression	1.06 (1.61)	1.73 (2.37)	1.92 (2.41)	2.01 (2.64)	1.28 (1.26)

Note. Means and standard deviations are presented for the final versions of all variables used in the main analyses (i.e., life satisfaction, depression, and neuroticism after POMP scoring; z-scored cognitive function composites), with the exception of age which is presented before centering.

Rush Memory and Aging Project and Minority Aging Research Study

The Memory and Aging Project (MAP) and Minority Aging Research Study (MARS) samples were used from the Rush Alzheimer’s Disease Center (Bennett et al., 2012, 2018). MAP is a longitudinal cohort study of adults 65 years and older recruited from retirement communities and subsidized senior housing facilities in Chicagoland and northeastern Illinois starting in 1997. MARS is a longitudinal cohort study of adults 65 years and older who self-identify as Black or African American. Participants for MARS were recruited starting in 2004 in the metropolitan Chicago area and outlying suburbs. Each year, participants in both MAP and MARS completed a battery of self-report surveys and a cognitive function assessment. The life satisfaction measure was first administered in 2008 in both cohorts and thus 2008 is the earliest possible analytic baseline. Because MAP/MARS participants completed the same study procedures, the two samples were combined for analysis. The Rush MAP/MARS analytic sample size that met inclusion criteria included 1,702 participants (M_age = 81.07, SD_age = 7.04).

Health and Retirement Study

The Health and Retirement Study (HRS; Juster & Suzman, 1995; Sonnega et al., 2014) is an ongoing, nationally representative longitudinal panel study of adults over age 50 in the United States and their spouses. Data collection began in 1992, with measurement occasions that include cognitive function assessments every two years. Half of the HRS respondents complete a survey that includes a life satisfaction measure at each measurement occasion, with the other half receiving it in the next occasion. The 2006 wave was the first wave including life satisfaction and cognitive function and was therefore used as the first study-level time point. The HRS analytic sample size that met inclusion criteria included 11,961 participants (M_age = 69.99, SD_age = 8.96).

English Longitudinal Study of Ageing

The English Longitudinal Study of Ageing (ELSA; Banks et al., 2024) is a longitudinal cohort survey of adults 50 years and older living in England. Data collection began in 2002 and was completed every two years. New participants were added in 2007, 2009, 2013, and 2015. The second wave of ELSA completed in 2005 was the first wave at which life satisfaction and cognitive function were assessed and was therefore used as the first study-level time point. Wave 6 (2013) was not included in the present research because not all cognitive function measures were collected during this wave. The ELSA analytic sample size that met inclusion criteria included 8,606 participants (M_age = 65.83, SD_age = 9.68).

Survey of Health, Ageing, and Retirement in Europe

The Survey of Health, Ageing, and Retirement in Europe (SHARE; Bergmann et al., 2019; Börsch-Supan et al., 2013) is the largest pan-European social science panel study. SHARE has been conducted biannually since 2004 and includes adults ages 50 and over. SHARE includes data from 20 European countries (Austria, Belgium, Croatia, Czech Republic, Denmark, Estonia, France, Germany, Greece, Hungary, Italy, Luxembourg, the Netherlands, Poland, Portugal, Slovenia, Spain, Sweden, and Switzerland) plus Israel up to Wave 6. Eight additional countries were added in Wave 7: Bulgaria, Cyprus, Finland, Latvia, Lithuania, Malta, Romania, and Slovakia. The current study used Wave 2 (2006) and Wave 4 (2010) through Wave 9 (2021) of SHARE. Wave 3 (2008) was not included, as life satisfaction was not assessed during this wave. The SHARE analytic sample size that met inclusion criteria included 35,218 participants (M_age = 64.94, SD_age = 9.86).

Longitudinal Aging Study Amsterdam

The Longitudinal Aging Study Amsterdam (LASA; Hoogendijk et al., 2016, 2020) started in Amsterdam in 1992 to study the physical, emotional, cognitive, and social functioning of older adults in the Netherlands. LASA includes three cohorts of adults 55 years and older, with measurement occasions approximately every three years. Data collection started for the first cohort of participants born between 1908 and 1937 in 1992–1993 (Wave B), with the second cohort of participants born between 1938 and 1947 added 10 years later in 2002–2003 (Wave 2B) and the third cohort of participants born between 1948 and 1957 added in 2012–2013 (Wave 3B). Each cohort’s study baseline (e.g., Wave B for first cohort, Wave 2B for second cohort) was set as their analytic baseline in the current research. The final LASA analytic sample size included 4,410 participants (M_age = 65.87, SD_age = 8.36).

Measures

Life Satisfaction

To be included in the current study, life satisfaction needed to be assessed continuously and preferably with some version of the Satisfaction with Life Scale (SWLS; Diener et al., 1985). Single-item measures of life satisfaction were also allowed, as they have been found to be valid and to perform similarly to the SWLS (Cheung & Lucas, 2014). We excluded studies that used the Life Satisfaction Index (LSI; Neugarten et al., 1961) because many of the items capture changes in life satisfaction relative to prior developmental periods (e.g., “I am just as happy as I was when I was younger”), which alters the meaning compared to the other measures of life satisfaction included in the current research and complicates interpretations of life satisfaction trajectories.

To assess life satisfaction, Rush MAP/MARS, HRS, and ELSA used the 5-item SWLS (Diener et al., 1985). Response options ranged from 1 (Strongly Disagree) to 7 (Strongly Agree) in all three datasets except for Wave 8 (2006) of HRS, which used a 6-point scale omitting the “neither agree nor disagree” option; differences in Likert-scales were handled through rescaling, described in more detail below. In LASA, life satisfaction was operationally defined using two questions (Van Zonneveld, 1961). The first question asked about satisfaction with current life, and the second question asked about satisfaction with life as a whole. Each question had five response categories, ranging from ‘very dissatisfied’ to ‘very satisfied.’ Finally, SHARE assessed life satisfaction with the single item: “On a scale from 0 to 10 where 0 means completely dissatisfied and 10 means completely satisfied, how satisfied are you with your life?”

For all five datasets, life satisfaction composites were calculated by taking the average of the life satisfaction items at each time point. Because the studies used different response scales, life satisfaction composites were then percent of maximum possible (POMP) scored and divided by 10 (Cohen et al., 1999). The resulting scores ranged from 0–10, with higher scores representing higher life satisfaction.

Cognitive Function

For the current research, we included studies with task-based measures of cognitive function. Table 2 includes information on the cognitive function domains assessed in each of the samples. Detailed information on the measures used to assess cognitive function in each sample can be found in Supplemental Table 3. For all five datasets, we computed z-scores for each cognitive test based on baseline means and standard deviations and then computed the mean of those z-scored tests to derive the final cognitive function score for each participant. Each study used a different set of cognitive tests that assessed different domains of cognitive function. Thus, we did not aim to statistically harmonize cognitive function across datasets. Coordinated data analysis allows us to preserve this natural measurement heterogeneity because analyses are conducted separately within each dataset.

Covariates

Covariates included age, sex, race, education, Apolipoprotein E (APOE) genotype, depressive symptoms, and neuroticism. Previous studies have linked sociodemographic characteristics such as age, sex, race, and education to life satisfaction (e.g., Zhang et al., 2017) and cognitive function (e.g., Quiñones et al., 2022; van Hooren et al., 2007). Low levels of depression and neuroticism represent cognitive resilience factors (Graham et al., 2021; Wilson et al., 2014), and APOE genotype was controlled for as a primary genetic risk factor for dementia (Yu et al., 2017). All covariates were assessed at analytic baseline, except for relatively time-invariant covariates that were only assessed at earlier timepoints (e.g., sex, education).

Age.

For all datasets, age was assessed at analytic baseline and was centered at 65 years. We chose age 65 for centering because this age was represented in all samples. That is, 65 was the minimum age of participants in Rush MAP/MARS, and close to the mean age in the remaining samples.

Sex.

Sex was assessed as a dichotomous variable in all five studies and was coded as 0 = female/women and 1 = male/men for the current analyses.

Race.

We included variables that assessed race using the Office of Management and Budget’s five minimum categories for data on race. Participants who selected more than one race were included in a ‘multiracial’ category. Race was treated as a categorical variable, and we included all categories that comprised greater than 5% of the sample. We coded race as missing for groups comprising less than 5% of the sample, but these participants were still included in analyses through the use of full information maximum likelihood estimation (FIML). LASA and SHARE datasets do not contain information on racial identity, and race was not included for ELSA because the race variable included only ‘White’ and ‘Non-White’ categories. As such, race was only included in analyses for Rush MAP/MARS and HRS. In these samples, race was coded such that 0 = White/European American, 1 = Black/African American, and all other races were coded as missing due to insufficient sample sizes.

Education.

Education was assessed in years. Rush MAP/MARS, HRS, and SHARE included a continuous ‘years of education’ variable. For LASA and ELSA, categorical education variables were recoded to approximate number of years (e.g., high school graduate = 12 years).

APOE Genotype.

APOE genotype was coded as 0 = absence of e4 allele and 1 = presence of e4 allele, a genetic risk factor for ADRD (Yu et al., 2017). APOE genotype was included for MAP/MARS and LASA, as this information was publicly available in these datasets.

Depression.

Depression was assessed using either the 8-item (HRS, ELSA, and SHARE), 10-item (Rush MAP/MARS), or 20-item (LASA) version of the Center for Epidemiological Studies Depression Scale (CES-D; Radloff, 1977). With the exception of LASA, which used a 0 (Rarely or never) to 3 (Mostly or always) scale, all datasets used a yes/no response scale. Participants rated their depressive symptoms over the past week in all studies. Depression scores were POMP scored and divided by 10. The resulting scores ranged from 0–10, with higher scores representing more depression symptoms.

Neuroticism.

Neuroticism was assessed using 12 items from the NEO Five-Factor Inventory (Costa & McCrae, 1992) in Rush MAP/MARS. For HRS, neuroticism was assessed by averaging the items calm (reversed), moody, worrying, and nervous. LASA assessed neuroticism using 15 items from the Dutch Personality Questionnaire (Luteijn et al., 1975). Neuroticism was not available at study baseline for ELSA or SHARE. When available, neuroticism scores were POMP scored and divided by 10. The resulting scores ranged from 0–10, with higher scores representing higher neuroticism.

Analytic Plan

All data analyses were conducted in R (Version 4.3.1; R Core Team, 2022) using the lavaan (Rosseel, 2012), psych (Revelle, 2017), ctSEM (Driver et al., 2017), metafor (Viechtbauer, 2010), and CoTiMA (Dormann & Homberg, 2024) packages. Prior to the main analyses, between- and within-person correlations were computed for the life satisfaction and cognitive function variables. Alpha was set at .05 to determine significance. Analytic baseline for the current research was defined as the first wave at which both life satisfaction and cognitive function were assessed. Because each statistical test reflects a substantively different hypothesis, we did not correct for multiple tests. Standardized effect sizes were evaluated relative to other effects in psychology using Funder and Ozer’s (2019) guidelines. Cross-lagged effects in the RI-CLPMs were evaluated using average cross-lagged effect sizes from RI-CLPMs estimated by Orth et al. (2024), with .02 corresponding to the 25^th percentile (i.e., “small” effects), .05 corresponding to the 50^th percentile (i.e., “medium” effects), and .11 corresponding to the 75^th percentiles (i.e., “large” effects). We conducted analyses separately within each of the five datasets, and then used random-effects meta-analysis to synthesize the results. Baseline age, sex, race (when available), and APOE genotype (when available) were included as covariates in main analyses.

Bivariate Latent Growth Curve Models

To address our first research aim, we used bivariate latent growth curve models (BLGCMs; see Figure 1). BLGCMs can estimate trajectories of two variables (e.g., life satisfaction and cognitive function) while simultaneously modeling correlations between their initial level and change parameters (Muniz-Terrera et al., 2017). To test our hypotheses, first-order BLGCMs were used to specify the correlation between initial levels of life satisfaction and initial levels of cognitive function (H1.1), as well as the correlation between the slopes of life satisfaction and the slopes of cognitive function (H1.2). Models were scaled by years in study, centered at a participant’s analytic baseline, and linear trajectories were modeled such that a unit of 1 equals 1 year for the latent slope rather than 1 wave between studies. This approach ensured estimates across studies could be directly compared and were not impacted by time between measurement occasions. Full Information Maximum Likelihood (FIML) estimation was used to handle missing data.

Figure 1.

Bivariate Latent Growth Curve Models and Random Intercept Cross-Lagged Panel Models for Life Satisfaction and Cognitive Function

Note. Models are presented for (a) bivariate latent growth curve models (BLGCMs) and (b) random intercept cross-lagged panel models (RI-CLPMs) for life satisfaction and cognitive function used in main analyses. Figure 1A illustrates a BLGCM for a dataset with yearly assessments, and Figure 1B illustrates a discrete time RI-CLPM.

Random Intercept Cross-Lagged Panel Models

For our second aim, we used random intercept cross-lagged panel models (RI-CLPMs; see Figure 1) to examine between-person and both prospective and concurrent within-person bidirectional associations between life satisfaction and cognitive function. FIML was used to handle missing data. Within each model, we constrained the autoregressive paths, cross-lagged effects, and covariances to be equal across measurement occasions.

RI-CLPMs are discrete time models, which means that they assume time lags between measurements are equal and that all participants are assessed at the same time lag. Thus, the cross-lagged effects from studies with one-year time lags (e.g., Rush MAP/MARs) cannot be directly compared to the cross-lagged effects from studies with four-year time lags (e.g., HRS). To facilitate comparison and synthesis of cross-lagged effects across studies with different time lags between measurement occasions, we conducted a second set of continuous time structural equation models within each dataset and then conducted a continuous time meta-analysis using the ctsem (Driver et al., 2017) and CoTiMA (Dormann & Homberg, 2024) packages in R. We modeled individual varying continuous time intercepts to account for between-person differences in life satisfaction and cognitive function. We initially ran into model convergence problems and implausible estimates. After consulting with the creator of the CoTiMA package, we resolved these issues and used the ctmaOptimizeFit() function from the CoTiMA R package. This function iteratively tries different time scaling values to optimize model fit (Dormann & Homberg, 2024). This approach reduces the likelihood of out-of-range estimates, which can happen when the fitting algorithm uses random start values that result in a locally but not globally optimal fit.

Sensitivity Analyses

A series of additional covariates were added in sensitivity analyses. In the BLGCMs, we included the covariates as predictors of the latent intercepts and slopes. In the RI-CLPMs, we included the covariates as predictors of the random intercepts. As noted above, the following potential confounders were included as covariates in main analyses: baseline age, sex, race (when available), and APOE genotype (when available). These Step 1 covariates represent unambiguous confounders for which reverse causality is unlikely or implausible. We based our key hypothesis tests on these models.

Sensitivity analyses controlled for additional covariates, for which the causal relation is less clear. In Step 2, we added baseline education because education is a potential confounder but it may also be a mediator. Finally, baseline neuroticism (when available) and depression were added in Step 3 because these covariates could be confounders (i.e., variables that causally impact both variables of interest), mediators (i.e., variables that are intermediate in the causal chain), or colliders (i.e., variables that are causally influenced by both variables of interest) and because they share considerable conceptual and empirical overlap with life satisfaction. We did not include lifestyle factors, such as physical or social activity, as covariates because we conceptualize these as likely time-varying mediators of the association between life satisfaction and cognitive function. For consistency, we also conducted the above analyses controlling for only the covariates that were available across all five studies (i.e., age, sex, education, depression).

Meta-analysis

Random-effects meta-analysis was conducted for the key hypothesis tests. We used standardized effects in the meta-analysis. The estimated standard errors were used to estimate the sampling variance for the meta-analytic aggregations. Discrete time cross-lagged effects from the RI-CLPMs could not be directly compared across studies with different timescales. Thus, to compute meta-analytic effects for the cross-lagged effects, we used continuous time meta-analysis. Although statistical significance can be interpreted in the traditional way, continuous time cross effects were transformed from continuous time units into discrete time units to facilitate the interpretation of effect sizes (Driver et al., 2017). Due to the smaller number of studies, we did not consider study-level moderators of the meta-analytic effects. However, any differences in findings are discussed qualitatively based on sample-specific characteristics and measures.

Results

Correlations and ICCs

Prior to main analyses, we first examined between- and within-person correlations and intraclass correlation coefficients (ICCs) for life satisfaction and cognitive function (Supplemental Table 4). Across samples, the ICCs ranged from .44 to .66 for life satisfaction and from .62 to .74 for cognitive function, indicating moderate within-person variability in both constructs. Life satisfaction and cognitive function were weakly correlated at both the between-person and within-person levels, with between-person correlations ranging from .05 to .30 across samples and within-person correlations ranging from .03 to .08 across samples.

Bivariate Latent Growth Curve Models

For brevity, we present forest plots of study-specific and meta-analytic effects for each model in the main manuscript. Figure 2 presents the forest plots for BLGCMs testing our first research question. A summary of results and meta-analytic estimates is presented in Table 1. Tables and summaries of complete individual study results are included in Supplemental Tables 5–38. Consistent with our hypothesis (H1.1), lower initial levels of life satisfaction were associated with lower initial levels of cognitive function in all five studies. The meta-analytic estimate indicated a medium association between life satisfaction and cognitive function intercepts (r = 0.19, 95% CI = [0.07, 0.32]). Lending support for H1.2, less positive life satisfaction trajectories were significantly associated with greater cognitive decline in four out of five studies (i.e., all studies except LASA). The meta-analytic summary revealed a medium association between life satisfaction and cognitive function slopes (r = 0.23, 95% CI = [0.13, 0.32]).

Figure 2.

Forest Plots for Bivariate Latent Growth Curve Models for Life Satisfaction and Cognitive Function

Note. Results of BLGCMs for life satisfaction and cognitive function in models accounting for age, sex, race, and APOE genotype (when available). Point estimates for each study are represented by squares and error bars indicate 95% confidence intervals (CI) around those effects. RE Model = random effects meta-analytic estimate. Between-study heterogeneity statistics (Q, I²) are shown above each forest plot.

We also examined life satisfaction intercepts predicting cognitive function slopes, and vice versa, though we did not make specific hypotheses about these parameters. The meta-analytic association between the life satisfaction intercept and the cognitive function slope was very near zero and statistically non-significant (r = 0.01, 95% CI [−0.05, 0.07]). Similarly, the meta-analytic association between the cognitive function intercept and the life satisfaction slope was very near zero and statistically non-significant (r = 0.02, 95% CI [−0.01, 0.06]). Thus, initial levels of life satisfaction did not predict the rate of cognitive decline, and initial levels of cognitive function did not predict life satisfaction change.

Random Intercept Cross-Lagged Panel Models

Figure 3 presents the forest plots for study-specific and meta-analytic effects for RI-CLPMs testing our second research question. Because the time lag between measurement occasions differed across studies, it would be inappropriate to directly compare the effect sizes from the cross-lagged effects obtained from traditional discrete time RI-CLPMs. Thus, meta-analytic estimates and heterogeneity statistics are not reported for these parameters. Instead, we conducted continuous time versions of the RI-CLPMs and used continuous time meta-analysis to compute meta-analytic cross effects (see Figure 4 and Supplemental Tables 39–40 for complete continuous time results).

Figure 3.

Forest Plots for Random Intercept Cross-Lagged Panel Models for Life Satisfaction and Cognitive Function

Note. Results of RI-CLPMs for life satisfaction and cognitive function in models accounting for age, sex, race, and APOE genotype (when available). Point estimates for each study are represented by squares and error bars indicate 95% confidence intervals (CI) around those effects. RE Model = random effects meta-analytic estimate. Between-study heterogeneity statistics (Q, I²) are shown above each forest plot.

Figure 4.

Continuous Time Plots for the Cross Effect of Life Satisfaction on Cognitive Function and the Cross Effect of Cognitive Function on Life Satisfaction at Different Time Intervals

Note. The solid grey lines depict the individual study estimates (1= ELSA; 2 = Rush MAP/MARS; 3= LASA; 4 = HRS; 5 = SHARE). The dashed line depicts the meta-analytic estimate.

Consistent with H2.1, life satisfaction and cognitive function were positively associated at the between-person level in all five studies. The meta-analytic effect was small to medium and statistically significant (r = 0.19, 95% CI = [0.08, 0.30]).

Consistent with H2.2, the cross-lagged effect of changes in life satisfaction on subsequent changes in cognitive function was positive and statistically significant in four out of five studies (i.e., all studies except LASA), suggesting that decreases in life satisfaction at one time point predict decreases in cognitive function at the next. Autoregressive effects can be found in the individual study results presented in Supplemental Tables 22–38. The continuous time meta-analytic cross effect, which is analogous to the cross-lagged effect in a traditional RI-CLPM, was statistically significant (B = 0.06, 95% CI = [0.04, 0.08]) with a medium effect size relative to average cross-lagged effects from RI-CLPMs in psychology (Orth et al., 2024). After transforming into discrete time units, the continuous time meta-analysis showed that the cross effect was maximal at a time lag of 1.2 years with a small effect size (β = 0.03).

Consistent with H2.3, the cross-lagged effect of changes in cognitive function on subsequent changes in life satisfaction was positive in all five studies and statistically significant in three out of five studies (i.e., all studies except LASA and HRS), suggesting that decreases in cognitive function at one time point predict decreases in life satisfaction at the next. The continuous time meta-analytic cross effect was statistically significant (B = 0.10, 95% CI = [0.08, 0.12]) with a medium effect size compared with average cross-lagged effects from RI-CLPMs in psychology (Orth et al., 2024). After transforming into discrete time units, the continuous time meta-analysis showed that the cross-lagged effect was maximal at a time lag of 1.2 years with a small effect size (β = 0.05).

For most individual studies, the optimal time lag to maximize the cross effects (depicted in Figure 4) was shorter than the shortest time lag observed in the study. This is consistent with Dormann & Griffin (2015) who demonstrated that the optimal time lag is often shorter than the time lag used in most longitudinal panel studies and that controlling for an unmeasured third variable (in this case the random intercepts) can result in an optimal time lag that is shorter than the shortest time lag available in a given study.

Finally, consistent with H2.4, residual variances for life satisfaction and cognitive function were positively associated at the within-person level in four of the five studies (i.e., all studies except LASA), suggesting that individuals’ life satisfaction was higher than their personal average at time points in which their cognitive function was also higher than average, after accounting for their between-person association and autoregressive and cross-lagged effects. The meta-analytic effect was small and statistically significant (r = 0.05, 95% CI = [0.03, 0.08]).

Sensitivity Analyses

Results of sensitivity analyses including baseline education as an additional Step 2 covariate and baseline neuroticism (when available) and depression as additional Step 3 covariates are presented for BLGCMs in Supplemental Tables 10–14 (unstandardized estimates) and 16–20 (standardized estimates) and for RI-CLPMs in Supplemental Tables 27–31 (unstandardized estimates) and 33–37 (standardized estimates). Analyses controlling for only the covariates that were available across all five studies (i.e., age, sex, education, depression) are presented for MAP/MARS, LASA, and HRS in Supplemental Tables 15, 21, 32, and 38. Corresponding analyses for ELSA and SHARE can be found in Model 3 of the above sensitivity analyses. Minimally adjusted models (adjusting for age only) are also presented for BLGCMs in Supplemental Tables 5–9 and for RI-CLPMs in Supplemental Tables 22–26.

The overall pattern of results remained the same across main analyses and sensitivity analyses including education as an additional Step 2 covariate. After the addition of Step 3 covariates, fewer studies revealed significant effects for the following parameters: BLGCM intercept-intercept correlation (H1.1; statistically significant in 4 out of 5 studies), RI-CLPM intercept-intercept correlation (H2.1; statistically significant in 2 out of 5 studies), and RI-CLPM cross-lagged effect of cognitive function on life satisfaction (H2.3; statistically significant in 4 out of 5 studies). Overall, our results were highly consistent across sensitivity analyses, with only minor changes after the inclusion of covariates, highlighting the robust nature of our findings.

Results Summary

In sum, we found meta-analytic support for all hypotheses, and all hypothesized results were found in at least three of the five studies (see Table 1). Of the five studies, LASA most frequently yielded nonsignificant results, which may reflect methodological differences related to LASA’s sample size, use of a 2-item life satisfaction measure, longer time lag, or geographic location in The Netherlands. At the between-person level, individuals with higher life satisfaction also had higher cognitive function, and the two constructs changed together across time. However, initial levels of one did not predict long-term change trajectories in the other. At the within-person level, declines in life satisfaction predicted subsequent declines in cognitive function, and vice versa.

Discussion

The current research used CDA to examine associations between life satisfaction and cognitive function in five longitudinal studies of older adults across multiple countries. As the prior literature has yielded mixed findings on the association between life satisfaction and cognitive function (e.g., Beck et al., 2024; Bell, Singham, Saunders, Buckman, et al., 2022; Bell, Singham, Saunders, John, et al., 2022), the ability to examine this association across multiple large-scale studies using CDA is a major strength of the current research. Using five longitudinal studies, our results are based on data collected across almost three decades from over 60,000 participants in over 30 countries. Two analytic approaches were used to model between- and within-person associations and to test the potentially bidirectional association between life satisfaction and cognitive function. First, bivariate latent growth curve models examined whether and how life satisfaction and cognitive function change together across older adulthood. Second, random intercept cross-lagged panel models tested between-person, prospective within-person, and concurrent within-person associations to explore whether changes in life satisfaction precede changes in cognitive function, and vice versa. Together, this approach provides critical information on how life satisfaction and cognitive function are linked in older adulthood.

Overall, our findings provide evidence for moderate between-person and modest within-person associations between life satisfaction and cognitive function. All hypotheses were supported, with all hypothesized results found in at least three of the five studies as well as in random-effects meta-analyses. At the between-person level, life satisfaction and cognitive function were positively associated. Moreover, life satisfaction and cognitive function were found to change together across older adulthood, such that more positive life satisfaction trajectories were associated with less cognitive decline. However, initial levels of one did not predict change trajectories of the other. At the within-person level, cross-lagged effects suggested that within-person decreases in life satisfaction predicted subsequent decreases in cognitive function, and vice versa. Compared to average cross-lagged effects in the published literature, our effect sizes ranged from small to large (Orth et al., 2024). Finally, residual covariances highlighted that, at time points in which individuals’ life satisfaction was lower than usual, their cognitive function was also worse than usual. These findings support a reciprocal relationship between life satisfaction and cognitive function, whereby higher life satisfaction may prevent or reduce cognitive decline, and cognitive decline may negatively impact future life satisfaction. This reciprocal relationship aligns with the “upward spiral” proposed by broaden-and-build theory (Fredrickson & Joiner, 2002), in that enhanced wellbeing and cognitive resources may mutually reinforce each other.

By using longitudinal data, controlling for key confounders, and employing statistical techniques that can establish temporal precedence, the current work moves research on life satisfaction and cognitive function closer to causality. However, observational methods are inherently limited, and several threats to causal inference remain. For example, the within-person cross-lagged effects may reflect causal relations under the assumption of no unobserved time-varying confounders. However, the presence of unobserved time-varying confounders could lead to spurious effects of life satisfaction on cognitive function, of cognitive function on life satisfaction, or both. For example, lifestyle factors and coping resources, such as health behaviors or social support, may represent time-varying confounders that influence these associations (Willroth et al., 2024). Making causal inferences about relations between wellbeing and health is challenging and typically necessitates triangulation across multiple causal inference strategies (Rohrer & Lucas, 2020). Thus, although the present research builds on past work by disentangling between- from within-person associations, more research is needed using other types of study designs (e.g., experimental methods, Mendelian randomization) to test causal relations between life satisfaction and cognitive function. Nevertheless, in the current research, the use of longitudinal models that adjust for relevant covariates and disentangle between-person effects from within-person effects is an important first step toward supporting causal inference.

It is important to establish whether the observed bidirectional associations between life satisfaction and cognitive function are causal because, if so, this would have important implications for intervention and policy. At the individual level, positive psychological interventions have been found to promote life satisfaction and wellbeing (Carr et al., 2021; Koydemir et al., 2021), and if life satisfaction causally impacts cognitive function, these same interventions may be beneficial for preventing or slowing cognitive decline. Future work is needed to evaluate the feasibility and scalability of life satisfaction interventions and determine their impact on cognitive decline. Relative to other psychosocial individual differences that have been associated with cognitive health (e.g., personality, sense of purpose), life satisfaction is an especially promising potential protective factor to investigate because of its utility in public health policy. Life satisfaction is broadly assessed in large representative national and international surveys, and it is increasingly being used to evaluate the effectiveness of public policy (Durand, 2018; Helliwell, 2008; World Health Organization, 2013). For example, the Gallup World Poll measures life satisfaction annually in more than 130 countries, enabling leaders and policymakers to evaluate policies with respect to their expected or observed influence on life satisfaction (Diener & Tay, 2015; Helliwell, 2008). If the association between life satisfaction and cognitive function is causal, life satisfaction may be a promising target for policy-level prevention strategies aimed at reducing or slowing cognitive decline.

Likewise, if the effect of cognitive decline on later decreases in life satisfaction is causal, this has implications for interventions and policies designed to promote quality of life among people living with cognitive decline and dementia. Although few studies have investigated life satisfaction among people living with dementia, one systematic review and meta-analysis identified several factors associated with quality of life in people living with dementia, including social engagement, functional ability, caregiver wellbeing, and physical and mental health factors (Martyr et al., 2018). Thus, interventions and policies focused on these factors may be beneficial for supporting life satisfaction among individuals experiencing cognitive decline.

Limitations and Future Directions

Along with its strengths, the current research also has limitations that should be addressed in future work. For example, although the current research makes use of multiple large-scale studies, these studies largely reflect WEIRD (White, Educated, Industrialized, Rich, Democratic; Henrich et al., 2010) samples. Future research should therefore expand this work to more diverse samples to determine whether our findings are generalizable to other sociodemographic and cultural groups, including in low- and middle-income countries where dementia prevalence is highest (Wimo et al., 2017). Additionally, while continuous time meta-analytic cross effects were examined to address the differing time scales in these studies, between-person components are dependent on study design and capture stable factors with respect to a study’s time interval (Hamaker, 2023). As such, it should be noted that the interpretation of the intercepts and residual covariances in our RI-CLPM analyses depends on the time interval of the studies. It is also important to note that five datasets is likely the lower bound to estimate a random-effects meta-analysis.

Further, although the current research provides valuable information on the link between life satisfaction and cognitive function, additional work is needed examining other components of wellbeing and specific cognitive function domains. As wellbeing is a multifaceted construct (e.g., Linton et al., 2016; Willroth, 2023), additional work is needed to determine which components of wellbeing are most strongly tied to cognitive function. Another study examining the bidirectional relations between wellbeing and cognitive function found that effects were stronger for eudaimonic wellbeing than for life satisfaction (Pfund et al., 2025), highlighting the importance of examining other wellbeing components in future coordinated analyses. Finally, one limitation of the current research is its focus on global cognition rather than specific cognitive domains (e.g., episodic memory, processing speed). However, past research suggests that differences in associations between wellbeing and cognitive function may be more similar than different across cognitive domains (Pfund et al., 2025).

Finally, the current research lays the foundation for further work investigating the complex and dynamic relationship between life satisfaction and cognitive function. As self-reports, informant reports, and objective assessments of cognitive decline can provide unique information (e.g., Milanovic et al., 2023; Ryu et al., 2020), future work on life satisfaction and cognitive function may consider including and comparing multiple types of assessments. Additionally, future research would benefit from using multistate models to examine how life satisfaction influences transitions between normal cognition, mild cognitive impairment (MCI), and dementia (Lewis et al., 2024). To address the dynamic nature of wellbeing, future research should also consider interactive processes between positive and negative aspects of wellbeing and how the experience of mixed emotions relates to cognitive function. Together with the current CDA findings, this body of work would offer valuable insights into the impact and application of life satisfaction across stages of cognitive function.

Conclusion

The current research advances our understanding of the relationship between life satisfaction and cognitive function by using CDA to examine this association across five longitudinal studies of older adults from multiple countries. These constructs appear to change together in the long-term and predict changes in each other in the short-term. However, initial levels of one did not predict rates of change in the other. Overall, these findings point to the promise of targeting life satisfaction in both individual- and policy-level interventions to promote healthy cognitive aging. Moreover, these results highlight the importance of monitoring and supporting life satisfaction among older adults experiencing cognitive decline.