When adult children migrate: Intergenerational solidarity, family functioning, and subjective wellbeing among older parents left behind in rural China

Highlights

• •Migration intensity was associated with poorer family functioning, largely through eroded associational solidarity.
• •Filial-piety norms were preserved even as behavioral contact eroded.
• •Remittances offset the subjective-wellbeing cost of adult-child migration but not the family-functioning cost.
• •Community cohesion buffered the indirect effect of migration on family functioning by dampening the associational mechanism.
• •Older mothers showed larger associational solidarity penalties than older fathers.

.

Abstract

China's sustained rural-to-urban labor migration has produced a population of older parents whose adult children work at a substantial distance from the home village. We examined whether the out-migration of adult children (respondents' grown offspring who had left the household for work) was associated with family functioning and subjective well-being among older parents remaining in three rural Chinese provinces (Sichuan, Anhui, and Henan), and which intergenerational solidarity pathways accounted for those associations. We conducted a cross-sectional, questionnaire-based survey of 479 rural older parents aged 55 or older, using a four-stage cluster sampling design and a 71-item, culturally adapted, closed-response instrument administered face-to-face by trained enumerators. Hierarchical multiple regression, parallel mediation, moderated mediation, multi-group analyses, and sensitivity analyses were conducted. Migration intensity was associated with poorer family functioning (β = -.21, p < .001), which was mediated by reduced associational solidarity (indirect effect = -0.142, 95% CI [-0.207, -0.082]). For subjective well-being, remittance adequacy was the strongest predictor (β = .25, p < .001), with functional solidarity as a compensatory mediator (indirect = 0.066, 95% CI [0.024, 0.114]). Normative filial-piety endorsement was preserved despite eroded contact (a-path = -0.05, ns). Community cohesion moderated the associational pathway (index = 0.058, 95% CI [0.012, 0.110]). Mothers carried larger associational penalties than fathers (β = -.36 vs. -.19; χ² = 4.81, p = .028). Reliability ranged from α = .77 to .88; CFA fit was acceptable (CFI = .964, RMSEA = .054). Because the design was cross-sectional and the sample comprised only households with at least one migrant adult child, estimates describe associations within migrant households rather than causal effects of migration. The offsetting pattern between relational decline and material compensation has direct implications for rural elder-care policy and the international evidence base on left-behind older parents in migration-sending societies.

1. Introduction

International labor migration and large-scale internal rural-to-urban movement have reshaped intergenerational family life in low- and middle-income countries (LMICs) over the past two decades (Duminy et al., 2023). Migration disrupts caregiving systems for children left behind, but the same migration flows reorganize the lives of the older parents who remain; many of whom assume primary caregiving responsibility for grandchildren (Carranza, 2022; Dakua & Lhungdim, 2025; Fauk et al., 2025). Literature shows that this reorganization is rarely benign for the older generation. Sarkar and colleagues (2023) reported in the Longitudinal Aging Survey in India that empty-nested older parents had a significantly higher prevalence of poor self-rated health and depression than co-residing peers. Muhammad, Sulaiman, and Srivastava (2022) showed in the same dataset that older women with migrant sons faced 76% higher odds of major depression than older men with migrant sons, and Mandal and Paul (2024) reported that remittances compensated for only about 55% of migration's adverse effect on left-behind older parents' self-rated health. Tabassum and Lalji (2025) extended this to physical health and showed that women in Indian migrant households had lower body mass index than those in non-migrant households. The pattern that emerges is one of partial material compensation alongside relational deficit, with women bearing the larger burden. China's experience sharpens this pattern. Li and colleagues (2020) showed that adult-child migration predicted poorer self-rated health, lower body mass index, reduced ADL capacity, and elevated depressive symptoms among older parents. Liu and colleagues (2021) replicated these associations among older parents with chronic disease, and Zhang, Lyu, Cheng, and Zhang (2022) decomposed the overall effect into an income channel operating through remittances and a time-allocation channel operating through reduced caregiving, concluding that the time-allocation channel outweighed the income channel. The cultural backdrop is also shifting. Wu and colleagues (2024) documented a measurable cohort-level decline in offspring-based concepts of eldercare, and Liu, Zhan, and Qiu (2021) reported in mixed-methods data from Anhui, Henan, and Sichuan that filial piety is no longer the sole foundation of old-age financial security among migrant-worker families.

The intergenerational solidarity framework set out by Roberts, Richards, and Bengtson (1991) provides the appropriate analytical lens for these findings, organizing parent–adult-child relationships across affectual, associational, functional, normative, consensual, and structural dimensions. The framework matters here because migration disrupts each dimension differently: structural opportunity declines mechanically, functional exchange may intensify with remittances, and associational contact depends on whether mediated communication can substitute for daily presence (Baykara-Krumme & Fokkema, 2019). Hwang, Fu, Brown, and Silverstein (2024) demonstrated that even strong digitally connected solidarity profiles only partially compensated for in-person contact during the COVID-19 pandemic. Despite the framework's relevance, three gaps remain. First, most studies examine a single outcome and a single solidarity dimension, leaving the dimensional structure untested as it operates simultaneously across both family functioning and subjective well-being. Second, community-level moderators are seldom modeled alongside individual- and family-level predictors despite their prominence in older-adult mental-health work more broadly (Sun, 2025). Third, gender differences have been documented at the population level in India but not formally tested through multi-group path comparisons in Chinese rural samples.

A further gap concerns how migration itself is conceptualized. Much of the literature treats adult-child migration as a binary exposure, yet migration is a configuration of interdependent elements: how many children have left and for how long, where they have gone, how often they return or call, whether and how adequately they remit, and how household roles reorganize in their absence, often along gender lines (Baykara-Krumme & Fokkema, 2019; Zhang et al., 2022). These elements are embedded in household socioeconomic conditions that shape both the decision to migrate and its consequences for those who stay (C. Liu et al., 2021). We therefore measured migration exposure as a set of related indicators covering intensity, duration, destination, contact frequency, remittance receipt and adequacy, and return visits, rather than as a single dichotomy, and we restricted the sample to migrant households so that variation across these elements, rather than migration versus non-migration, is the object of analysis.

The present study addresses these gaps. We examined how adult-child migration relates to family functioning and subjective well-being, and which solidarity pathways account for these associations. The two outcomes were treated in parallel because they occupy different levels of analysis: family functioning describes the household system, whereas subjective well-being is an individual evaluative state (see Analytical Strategy). Five hypotheses were tested. H1: Higher migration exposure is associated with poorer family functioning and lower subjective well-being. H2: These associations are mediated by intergenerational solidarity, principally through associational and affectual dimensions, while normative solidarity is preserved. H3: Functional solidarity offsets the effect of migration on subjective well-being but not on family functioning. H4: Community cohesion moderates the migration–outcome relationship by dampening the underlying mediating mechanism. H5: Older mothers show larger associational-solidarity penalties than older fathers.

2. Method

2.1. Design and Setting

We conducted a cross-sectional questionnaire-based survey across three provinces in central and western China (Sichuan, Anhui, and Henan) between October 2025 and December 2025. These provinces were selected because they are among the principal sending regions for internal labor migration in China and are explicitly identified as such in the migrant-worker fieldwork conducted by Liu, Zhan, and Qiu (2024). Reporting follows the STROBE statement for cross-sectional observational studies (von Elm et al., 2007)

2.2. Participants and Sampling

Eligible participants were rural residents aged 55 years or older with at least one biological adult child who had lived outside the household for work for at least 6 consecutive months in the preceding 5 years. Exclusion criteria comprised cognitive or hearing impairment, acute bereavement within the prior 30 days, and refusal to consent. We used a four-stage cluster sampling design. First, three provinces were selected purposively. Second, two prefecture-level municipalities per province were drawn through simple random sampling, yielding six municipalities (Mianyang and Nanchong in Sichuan; Fuyang and Lu'an in Anhui; Zhumadian and Xinyang in Henan). Third, three administrative villages were drawn at random within each municipality (18 villages). Fourth, approximately 27 households per village were sampled using systematic random-walk procedures starting from the village center. The requirement of at least one migrant adult child was a deliberate design decision: the research questions concern how the configuration of migration, in its intensity, duration, destination, and accompanying flows of contact and money, relates to outcomes among parents already exposed to it. The design, therefore, does not estimate the consequences of migration relative to non-migration. Households experiencing migration may differ systematically from non-migrant households in socioeconomic conditions, family dynamics, and social capital (C. Liu et al., 2021), so all inferences are conditional on membership in a migrant household, and the interpretive language used throughout the manuscript has been moderated accordingly.

2.3. Sample Size

Sample size was determined a priori using G*Power 3.1 (Faul et al., 2009), and anchored to the moderated regression for H4 because moderation is conventionally the lowest-powered test in this family of analyses. G*Power returned n = 395 with α = .05, power = .80, a small-to-medium effect size of f² = 0.025, and 15 predictors in the regression model. We then raised this figure by approximately 16% to buffer anticipated refusals, post-screening exclusions, and incomplete outcome data (395 × 1.16 ≈ 458), giving a rounded operational recruitment target of n = 460. The cluster sampling design was addressed in analysis rather than through sample inflation. Using the realized average cluster size (479 / 18 ≈ 26.6) and the village-level intracluster correlation for the primary outcome (ρ = 0.03), the Kish design effect for a cluster mean is DEFF = 1 + (26.6 − 1) × 0.03 = 1.77 (Kish, 1965). Because ρ fell below the 0.05 threshold pre-specified for multilevel re-estimation (Snijders & Bosker, 2012), and because a mean-based design effect is not the appropriate adjustment for the regression coefficients that carry H1–H5, clustering was handled through bias-corrected bootstrap inference and the pre-specified sensitivity rule, with every model estimated at a case-to-parameter ratio above 30:1. The realized analytic sample of n = 479 exceeded the operational target.

2.4. Measures

The 71-item structured questionnaire covered seven domains. Intergenerational solidarity was assessed using a 13-item adaptation of the Bengtson Intergenerational Solidarity Scale, drawing on the dimensional structure (Roberts et al., 1991), and covering the four dimensions most directly affected by migration: affectual (4 items), associational (3 items), functional (3 items), and normative (3 items). Community cohesion was measured with the 10-item Conjoint Community Resiliency Assessment Measure (CCRAM-10) developed and validated by Leykin, Lahad, Cohen, Goldberg, and Aharonson-Daniel (2013). Family functioning was defined, following the McMaster model, as the capacity of the family system to accomplish its core tasks of communication, problem solving, role allocation, affective expression, and behavior control (Epstein et al., 1983). It was measured with a 5-item adaptation of the General Functioning orientation of the McMaster Family Assessment Device, revised for use with older parents in migration-sending households; each of the five items (listed in Table 3) represents one of these task domains, and higher scores indicate healthier perceived functioning.(Epstein et al., 1983) Subjective well-being was measured with a 4-item scale covering contentment, future orientation, purposeful daily activity, and perceived value to family and community; the single-item ladder of life satisfaction was administered alongside it. The 4-item scale was the sole well-being measure entered in the regression, mediation, and moderation models; the Cantril ladder served only as a descriptive and convergent validity check and does not appear in any inferential model.(Cantril, 1965) Role adaptation was measured with five items developed for this study.

Table 3.

Content Validity Index Results for Adapted and Original Items (Six-Expert Panel).

Item / Scale	I-CVI relevance	I-CVI clarity	Modifid κ	Decision
Family Functioning (5 items)
FF1: Family communication about migrant child	1.00	1.00	1.00	Retained
FF2: Joint problem-solving across distance	0.83	1.00	0.81	Retained
FF3: Affective expression in household	1.00	0.83	1.00	Retained
FF4: Role responsibility allocation	0.83	0.83	0.81	Retained
FF5: Behavioral control of grandchildren	1.00	1.00	1.00	Retained
Role Adaptation (5 items)
RA1: Taking on tasks previously held by adult child	1.00	1.00	1.00	Retained
RA2: Adapting to a smaller daily household	0.83	1.00	0.81	Retained
RA3: Confidence managing finances independently	1.00	0.83	1.00	Retained
RA4: Reorganizing daily routines after departure	0.83	0.83	0.81	Retained
RA5: Sense of life purpose after departure	0.83	1.00	0.81	Retained
Bengtson-adapted Solidarity items
AFF-adapted: Affectual closeness across distance	1.00	0.83	1.00	Retained
ASS-adapted: Associational substitution by digital contact	0.83	0.83	0.81	Retained
FUN-adapted: Adequacy of remittance flow	1.00	1.00	1.00	Retained
NOR-adapted: Felt obligation balance migrant vs. local child	0.83	0.83	0.81	Retained
Items dropped at content-validity stage
DROP1: Migrant child's perceived guilt	0.50	0.83	0.45	Dropped (low I-CVI)
DROP2: Frequency of village gossip about migrant child	0.67	1.00	0.63	Dropped (low I-CVI)
Scale-level summary (14 retained items)
S-CVI/Average (relevance)	0.93	—	—	Above 0.90 threshold
S-CVI/Average (clarity)	0.90	—	—	At threshold
S-CVI/Universal Agreement	0.72	—	—	Reported for transparency

I-CVI = item-level Content Validity Index (proportion of experts rating an item 3 or 4 on a 1–4 scale). Modified κ = chance-corrected agreement (Polit et al., 2007). Items with I-CVI < 0.78 dropped per pre-specified criteria.

Migration exposure was captured through eight structured items covering migration intensity, duration of the longest absence, destination type (within-province, inter-provincial, or international), contact frequency, remittance receipt and adequacy, return-visit frequency, and reason for migration. Nine demographic and household covariates included age, gender, marital status, education, household monthly income, household size, distance to the nearest urban center, self-rated physical health, and number of grandchildren in the household. All items used closed-response formats, predominantly 5-point Likert scales; six items were reverse-coded prior to analysis to control for acquiescence. The English version of the questionnaire is attached as Supplementary File 1. Migration intensity, the focal exposure, was computed as the proportion of the respondent's living adult children currently residing outside the household for work (observed range 0.17–1.00). The remaining migration items were retained as separate covariates rather than collapsed into a composite, because duration, destination, contact, and remittances represent temporal, spatial, relational, and material dimensions of the migration configuration that the analyses were designed to keep distinct.

2.5. Translation and Validation

We followed the six-stage forward–back translation protocol described by Beaton, Bombardier, Guillemin, and Ferraz (2000). Two independent bilingual translators produced forward English-to-Mandarin translations, which were synthesized through facilitated review; two blinded translators back-translated to English, with iterative revision until equivalence was achieved. An expert committee reviewed the final items, and standardized oral renderings in regional dialects (Sichuanese, Anhui Mandarin, and Henan Mandarin) were prepared for enumerator use, given variable literacy among older rural residents. Cognitive pretesting with 12 older adults followed the procedures set out (Willis, 2005). Six content experts (three rural sociologists and three gerontologists) rated each item on relevance, clarity, and cultural appropriateness using the Content Validity Index (CVI) procedure with retention thresholds of I-CVI ≥ 0.78 and S-CVI/Average ≥ 0.90 (Polit et al., 2007). A pilot study (n = 32) in non-sampled villages established preliminary reliability before the main fieldwork. We then used the split-sample approach (Boateng et al., 2018): exploratory factor analysis (EFA) with Promax rotation and parallel analysis on a random calibration subsample, and confirmatory factor analysis (CFA) with diagonally weighted least squares on the held-out validation subsample, using the conventional fit criteria of CFI ≥ 0.95, RMSEA ≤ 0.08, and SRMR ≤ 0.08. The decision to impose a confirmatory structure at the validation stage, rather than repeating an exploratory analysis, followed directly from this split-sample logic: the EFA on the calibration subsample established the dimensional structure empirically (reported in full in Table 5), and the CFA on the held-out validation subsample then tested whether a structure specified a priori from Bengtson's framework, and recovered by the calibration EFA, replicated in independent cases (Boateng et al., 2018). Fitting a second EFA to the validation data would have answered a different question and provided no test of the hypothesized four-factor structure. Evidence bearing on the discriminant validity of the resulting factors and on common method bias is reported in the Results.

Table 5.

Exploratory Factor Analysis of Adapted Solidarity Items (Calibration Subsample, n = 239).

Item	F1 Affectual	F2 Associational	F3 Functional	F4 Normative	h²
AFF1: Emotional closeness with migrant child	0.81	0.10	0.08	0.12	0.69
AFF2: Mutual understanding across distance	0.78	0.13	0.09	0.14	0.65
AFF3: Perceived caring from migrant child	0.74	0.16	0.18	0.10	0.61
AFF4: Trust in migrant child's intentions	0.71	0.11	0.14	0.15	0.57
ASS1: Frequency of phone/video contact	0.14	0.79	0.10	0.07	0.66
ASS2: Frequency of in-person visits	0.16	0.76	0.11	0.05	0.62
ASS3: Joint family decision-making frequency	0.12	0.69	0.13	0.16	0.54
FUN1: Remittances received from migrant child	0.09	0.11	0.78	0.06	0.64
FUN2: Practical help during illness or need	0.13	0.15	0.74	0.10	0.60
FUN3: Adequacy of financial transfers	0.17	0.09	0.71	0.11	0.55
NOR1: Felt obligation to support migrant child	0.13	0.07	0.10	0.74	0.58
NOR2: Acceptance of migrant child's path	0.15	0.10	0.09	0.69	0.52
NOR3: Endorsement of filial duty norm	0.10	0.14	0.07	0.65	0.47
Eigenvalues and variance
Eigenvalue	4.42	1.94	1.56	1.18	Sum = 9.10
% variance explained	30.5%	13.4%	11.3%	9.6%	Cumulative = 64.8%

Principal axis factoring with Promax (oblique) rotation. h² = communality. CFA validation: CFI = 0.964, TLI = 0.951, RMSEA = 0.054, SRMR = 0.051.

2.6. Data Collection and Ethics

Twelve master's-level enumerators (six female, six male) completed five days of training before independent fieldwork. Inter-rater reliability across enumerators on a 20-case standardization exercise was required to exceed Cohen's κ = 0.75 prior to deployment. Gender-concordant enumerator assignment was used for older mother participants where preferred, in line with regional norms on female privacy. Quality control included daily supervisor debriefs and back-checking the re-administration of selected questionnaire items in 10% of completed cases. Written informed consent was obtained after a standardized script was read in Mandarin or the relevant regional dialect; respondents with limited literacy provided thumbprint signatures alongside the script-read consent in line with the contextual-adaptation provisions of the Declaration of Helsinki (Goodyear et al., 2007).

2.7. Analytical Strategy

Analyses were conducted in R 4.3.1, with mediation and moderation models estimated using the PROCESS macro v4.2 in SPSS 27.0 for cross-checking. We assessed missing-data patterns using Little's MCAR test as implemented in the procedure (Jamshidian et al., 2014) and applied multiple imputation (m = 20) under a Missing at Random assumption where appropriate; the magnitude and pattern of missingness are reported at the start of the Results. Sample characteristics were summarized using descriptive statistics. Item-level response distributions for each scale were tabulated to support transparent reporting (Table 4). H1 was tested through hierarchical multiple regression with demographic controls entered in Step 1 and the migration block in Step 2. H2 and H3 were tested via parallel mediation (PROCESS Model 4) with 5,000 bootstrap samples and bias-corrected 95% confidence intervals. H4 was tested through moderated regression (PROCESS Model 1); we additionally fit Model 7 to test moderated mediation. H5 was tested through multi-group regression with χ² difference tests for path equality across gender. Sensitivity analyses included multilevel re-estimation if the village-level intracluster correlation exceeded 0.05 (Snijders & Bosker, 2012).

Table 4.

Item-Level Response Distributions for All Scales.

Scale / Item	1 (%)	2 (%)	3 (%)	4 (%)	5 (%)	Mean ± SD
Family Functioning
FF1. Family discusses important matters openly	3.5	8.4	21.3	44.7	22.1	3.74 ± 0.99
FF2. Family solves problems together despite distance	4.6	11.7	23.4	39.5	20.8	3.60 ± 1.06
FF3. Affection is expressed in our household	2.9	7.7	19.0	47.0	23.4	3.80 ± 0.97
FF4. Roles are clearly understood at home	3.1	9.2	22.5	43.6	21.6	3.71 ± 1.00
FF5. Behavior of children/grandchildren is well-managed	4.0	10.4	24.6	41.1	19.9	3.62 ± 1.04
Scale total						3.71 ± 0.71
Associational Solidarity
ASS1. Phone or video contact with migrant child is frequent	5.6	16.7	26.5	31.7	19.5	3.42 ± 1.13
ASS2. Migrant child visits home often enough	9.0	23.4	32.8	24.6	10.2	3.08 ± 1.12
ASS3. Joint family decision-making remains regular	6.7	18.4	27.8	29.6	17.5	3.34 ± 1.16
Scale total						3.28 ± 0.83
Normative Solidarity
NOR1. Children should support parents in old age	1.0	2.7	8.4	33.6	54.3	4.38 ± 0.83
NOR2. I accept my child's decision to migrate for the family's good	1.7	3.8	12.5	39.2	42.8	4.18 ± 0.91
NOR3. Filial duty remains central to family life	0.8	2.1	7.9	32.4	56.8	4.42 ± 0.81
Scale total						4.31 ± 0.61
Functional Solidarity
FUN1. I receive financial transfers from migrant child	5.6	10.6	15.0	39.7	29.0	3.76 ± 1.13
FUN2. Migrant child provides practical help when I need it	4.4	11.5	20.5	41.5	22.1	3.65 ± 1.06
FUN3. The financial flow is adequate for my needs	3.8	9.4	21.5	44.1	21.3	3.69 ± 1.02
Scale total						3.70 ± 0.74
Subjective Wellbeing
SWB1. Most days I feel content with my life	4.6	12.1	25.5	39.7	18.2	3.55 ± 1.06
SWB2. I look forward to the days ahead	5.6	13.8	26.7	37.6	16.3	3.45 ± 1.08
SWB3. I have purposeful activities in daily life	5.0	11.5	23.6	41.3	18.6	3.57 ± 1.06
SWB4. I feel valued by my family and community	3.5	9.0	22.5	44.5	20.5	3.69 ± 1.00
Scale total						3.56 ± 0.77

Family functioning and subjective well-being were treated as parallel outcomes rather than placed in a causal chain. The two constructs occupy different levels of analysis: family functioning is an appraisal of the household system, whereas subjective well-being is an individual evaluative state, and the hypotheses concern how migration relates to each, not how one shapes the other. Their zero-order association (r = .53; Table 7) is reported for transparency, and no directional claim about the relationship between the two outcomes is made or tested.

Table 7.

Means, Standard Deviations, and Zero-Order Correlations Among Study Variables (N = 479).

Variable	M	SD	1	2	3	4	5	6	7	8	9	10	11
1. Migration intensity	0.46	0.25	—
2. Duration (years)	7.40	4.80	.41***	—
3. Destination (1=inter-prov)	0.55	0.50	.21***	.26***	—
4. Contact frequency	6.40	4.90	-.24***	-.19***	-.14**	—
5. Affectual solidarity	4.16	0.66	-.18***	-.15**	-.09*	.48***	—
6. Associational solidarity	3.28	0.83	-.31***	-.25***	-.18***	.62***	.55***	—
7. Functional solidarity	3.70	0.74	.26***	.21***	.22***	.20***	.34***	.30***	—
8. Normative solidarity	4.31	0.61	-.07	-.05	-.03	.29***	.46***	.39***	.36***	—
9. Community cohesion	3.49	0.70	-.08	-.05	-.04	.17**	.22***	.30***	.18***	.25***	—
10. Family functioning	3.71	0.71	-.23***	-.17**	-.13**	.37***	.43***	.51***	.12*	.38***	.33***	—
11. Subjective wellbeing	3.56	0.77	-.15**	-.11*	-.09*	.26***	.39***	.35***	.31***	.33***	.27***	.53***	—

*p < .05, **p < .01, ***p < .001 (two-tailed). Destination coded 0 = within-province, 1 = inter-provincial or international.

Model complexity was managed in relation to the sample size. No single omnibus structural model combining all mediators, the moderator, and both outcomes was estimated; each hypothesis was tested in a separate model containing at most 15 estimated predictors, keeping the ratio of cases to estimated parameters above 30:1 in every model, and bias-corrected bootstrap intervals were used in place of asymptotic standard errors for all indirect effects. We nonetheless acknowledge that estimating multiple mediators, interaction terms, and conditional indirect effects from cross-sectional data of moderate size carries a risk of overfitting, and the moderated-mediation and subgroup estimates, in particular, should be treated as exploratory, pending replication.

Because all constructs were reported by the same respondents on one occasion, common method variance was addressed both procedurally and statistically. Procedural remedies included reverse-coded items, separation of scale blocks within the interview, standardized enumerator scripts, and assurances of confidentiality. Statistical checks comprised Harman's single-factor test, average variance extracted, composite reliability, and heterotrait–monotrait ratios of correlations; results are reported in the Results section.

3. Results

3.1. Sample Flow and Recruitment

Of 542 eligible households approached across the 18 sampled villages, 53 declined to participate (refusal rate 9.8%), and 9 were excluded post-screening (6 for cognitive impairment and 3 for hearing impairment). One additional case was dropped from analysis after fieldwork for incomplete data on the primary outcome scales. The final analytic sample comprised 479 left-behind older parents, exceeding the a priori recruitment target of n = 460, which was set from the G*Power requirement of 395 with a buffer for anticipated non-response. Refusal patterns did not differ significantly across provinces, χ² (2) = 3.41, p = .182.

3.2. Missing Data, Pilot Reliability, and Inter-Rater Agreement

Missing data were minimal across primary variables, ranging from 0.4% to 3.4% per item, with a total of 2.3% across the analytic dataset. Little's MCAR test was non-significant, χ² (184) = 201.8, p = .172. This supported the Missing at Random assumption; multiple imputation (m = 20) was applied to the primary analyses. The pilot study (n = 32) yielded acceptable preliminary reliability across all seven scales, with Cronbach's α ranging from .69 (role adaptation, 5 items) to .86 (affectual solidarity, 4 items); all values met or exceeded the a priori thresholds of α ≥ .70 for adapted scales and α ≥ .65 for newly developed items. Inter-rater reliability across the twelve enumerators on the 20-case standardization exercise reached Cohen's κ = 0.83 (range 0.78–0.89), exceeding the pre-specified 0.75 threshold required for independent fieldwork. Missingness was concentrated in the household-income item (3.4%) and the remittance-adequacy item (2.9%); all remaining items were below 2.0%. Inspection of the missingness pattern showed no monotone structure and no association between item-level missingness and age, gender, or province (all p > .10), consistent with the MCAR result. The imputation model included all analysis variables, along with the demographic covariates, as auxiliaries.

3.3. Sample Characteristics

The 479 participants had a mean age of 64.7 years (SD = 7.3, range 55–86); 58.0% were male, most were married (88.9%), and 71.2% had primary schooling or less. Participants reported a mean of 1.8 migrant adult children (SD = 0.9) and a mean migration intensity of 0.46 (SD = 0.25). Inter-provincial migration to coastal cities was the dominant destination (54.7%), and 79.3% of participants received remittances. The demographic and migration profile is reported in Table 1.

Table 1.

Sample Demographic and Migration Characteristics (N = 479).

Variable	Category	n	%	M (±SD)
Age (years)	55–86	479	100.0	64.7 (7.3)
Gender
Male	—	278	58.0	—
Female	—	201	42.0	—
Marital status
Married	—	426	88.9	—
Widowed	—	45	9.4	—
Separated/divorced	—	8	1.7	—
Education
No formal schooling	—	156	32.6	—
Primary	—	185	38.6	—
Middle / Matric	—	102	21.3	—
Senior secondary or above	—	36	7.5	—
Monthly household income (CNY)
< 1,500	—	69	14.4	—
1,500–3,500	—	199	41.5	—
3,501–6,500	—	163	34.0	—
> 6,500	—	48	10.0	—
Province (and prefecture-level cities)
Sichuan (Mianyang, Nanchong)	—	171	35.7	—
Anhui (Fuyang, Lu'an)	—	158	33.0	—
Henan (Zhumadian, Xinyang)	—	150	31.3	—
Number of migrant adult children	1–6	479	100.0	1.8 (0.9)
Migration intensity	0.17–1.00	479	—	0.46 (0.25)
Destination
Within province (urban)	—	185	38.6	—
Inter-provincial (coastal cities)	—	262	54.7	—
International	—	32	6.7	—
Duration of longest absence (years)	0.5–25	479	—	7.4 (4.8)
Receives remittances (yes)	—	380	79.3	—
Contact frequency (calls/month)	0–30	479	—	6.4 (4.9)
Self-rated physical health (1–5)	1–5	479	—	3.1 (0.9)

CNY = Chinese yuan. Migration intensity = proportion of adult children currently living outside the household for work.

3.4. Reliability and Descriptive Statistics

The four adapted Bengtson subscales reached α between .77 and .86, with affectual solidarity at the high end and normative solidarity at the low end. The CCRAM-10 community cohesion scale reached α = .88. The McMaster-derived family functioning scale reached α = .81, and the subjective wellbeing scale α = .79. Among the four solidarity dimensions, normative solidarity recorded the highest mean (M = 4.31, SD =± 0.61) and associational solidarity the lowest (M = 3.28, SD = ±0.83), consistent with the dimensional asymmetry that solidarity theory predicts when adult children migrate. Table 2 reports scale-level reliability and descriptive statistics.

Table 2.

Reliability and Descriptive Statistics of Study Scales (N = 479).

Scale	k	M	SD	α	95% CI for α
Affectual solidarity	4	4.16	0.66	.86	[.83, .88]
Associational solidarity	3	3.28	0.83	.80	[.76, .83]
Functional solidarity	3	3.70	0.74	.82	[.78, .85]
Normative solidarity	3	4.31	0.61	.77	[.73, .80]
Community cohesion (CCRAM-10)	10	3.49	0.70	.88	[.85, .90]
Family functioning	5	3.71	0.71	.81	[.77, .84]
Subjective wellbeing	4	3.56	0.77	.79	[.75, .82]
Cantril ladder (single item, 0–10)	1	5.96	1.81	—	—
Role adaptation	5	3.55	0.80	.75	[.71, .79]

k = number of items; α = Cronbach's alpha. Response scale = 1–5 except the Cantril ladder (0–10). 95% CIs computed via Feldt's method.

3.5. Content Validity

Six content experts rated 16 candidate items on the family functioning, role adaptation, and Bengtson-adapted scales for relevance and clarity using a 1–4 ordinal scale. Fourteen items met the I-CVI ≥ 0.78 retention threshold and were carried forward; two items fell below threshold and were dropped. Scale-level S-CVI/Average reached 0.93 for relevance and 0.90 for clarity, both at or above the pre-specified 0.90 acceptance criterion. Modified kappa values for retained items ranged from 0.81 to 1.00, indicating strong agreement after correction for chance. Table 3 reports the full content validity results.

3.6. Item-Level Response Distributions

Two patterns are notable. First, the visit-frequency item of the associational solidarity scale (ASS2) showed the lowest item mean across the four solidarity dimensions (M = 3.08, SD = ±1.12), with 32.4% of respondents disagreeing or strongly disagreeing that their migrant child visited home often enough. Second, normative solidarity items showed a pronounced rightward skew, with mean endorsement reaching 4.42 (SD = 0.81) for the filial-duty item, and 56.8% of respondents strongly agreeing that filial duty remains central to family life. Family functioning (M = 3.71, SD = ±0.71) and functional solidarity (M = 3.70, SD = ±0.74) clustered around the upper-middle of the scale, while subjective wellbeing (M = 3.56, SD = ±0.77) sat slightly lower. Item-level percentage endorsement, means, and standard deviations for all five focal scales are reported in Table 4.

3.7. Factor Structure of the Adapted Solidarity Items

The Kaiser-Meyer-Olkin measure of sampling adequacy was 0.84, and Bartlett's test of sphericity was significant, χ²(78) = 1714.3, p < .001, confirming that the data were suitable for factor analysis. Parallel analysis supported a four-factor solution explaining 64.8% of total variance. The four factors corresponded to Bengtson's affectual, associational, functional, and normative dimensions. All retained items loaded above 0.60 on their primary factor with cross-loadings below 0.25. Confirmatory factor analysis on the held-out validation subsample (n = 240) using diagonally weighted least squares estimation yielded acceptable fit, CFI = 0.964, TLI = 0.951, RMSEA = 0.054 (90% CI [0.038, 0.071]), SRMR = 0.051. Table 5 reports the rotated factor loadings.

3.8. Discriminant Validity and Common Method Bias

Average variance extracted (AVE) met the .50 benchmark for every construct except normative solidarity, which fell marginally below it (AVE = .48); composite reliability ranged from .77 to .89. Under the Fornell–Larcker criterion, the square root of AVE for each construct exceeded that construct's correlations with all other constructs. Heterotrait–monotrait ratios ranged from .15 to .66, all below the conservative .85 threshold (Henseler et al., 2015); the largest values arose for the affectual–associational (.66), family functioning–subjective well–being (.65), and associational solidarity–family functioning (.63) pairs, indicating related but empirically separable constructs. Harman's single-factor test showed the first unrotated factor accounting for 27.4% of common variance, well below the 40% level conventionally taken to signal dominant method variance, and a confirmatory model loading all items on a single factor fit poorly (CFI = .61, RMSEA = .14). These checks cannot exclude method variance entirely, but they give no indication that construct overlap or common method bias accounts for the reported associations (Podsakoff et al., 2024). Table 6 reports AVE, composite reliability, and the full HTMT matrix.

Table 6.

Discriminant Validity Statistics: Average Variance Extracted, Composite Reliability, and Heterotrait–Monotrait Ratios (Validation Subsample, n = 240) Validation subsample used for confirmatory factor analysis (n = 240).

Construct	AVE	CR	1	2	3	4	5	6	7
1. Affectual solidarity	.61	.86	—
2. Associational solidarity	.58	.81	.66	—
3. Functional solidarity	.55	.82	.41	.36	—
4. Normative solidarity	.48	.77	.57	.49	.45	—
5. Community cohesion	.52	.89	.26	.35	.21	.30	—
6. Family functioning	.53	.82	.52	.63	.15	.46	.40	—
7. Subjective wellbeing	.51	.80	.48	.43	.38	.40	.33	.65	—

AVE = average variance extracted; CR = composite reliability. Off-diagonal entries are heterotrait–monotrait (HTMT) ratios of correlations; all values fall below the .85 threshold (Henseler et al., 2015). All discriminant-validity statistics reported here (AVE, CR, and the off-diagonal HTMT ratios) were computed on the held-out validation subsample (n = 240) used for the confirmatory factor analysis; all other analyses used the full analytic sample (N = 479).

3.9. Bivariate Associations

Migration intensity correlated negatively with family functioning (r = -.23, p < .001) and with subjective wellbeing (r = -.15, p < .01). Among the four solidarity dimensions, associational solidarity showed the strongest negative correlation with migration intensity (r = -.31, p < .001) and the strongest positive correlation with family functioning (r = .51, p < .001). Functional solidarity reversed the pattern: it correlated positively with migration intensity (r = .26, p < .001) and more strongly with subjective wellbeing (r = .31, p < .001) than with family functioning (r = .12, p < .05). Table 7 reports the full matrix of zero-order correlations.

3.10. Hypothesis 1: Migration Exposure and Outcomes

For family functioning, demographic controls accounted for 9% of variance, R² = .09, F(5, 473) = 9.36, p < .001. Adding the migration block accounted for an additional 17% of variance, ΔR² = .17, p < .001. Within the migration block, migration intensity (β = -.21, p < .001), inter-provincial destination (β = -.11, p = .024), and contact frequency (β = .22, p < .001) reached significance. For subjective wellbeing, the coefficient profile reorganized: migration intensity dropped to non-significance (β = -.10, p = .081), contact frequency remained significant (β = .15, p = .003), and remittance adequacy emerged as the strongest positive predictor (β = .25, p < .001). Hypothesis 1 was supported in full for family functioning and partially for subjective wellbeing. Table 8 reports the full hierarchical regression results.

Table 8.

Hierarchical Regression of Migration Exposure on Family Functioning and Subjective Wellbeing.

Predictor	B (FF)	SE	β	p	B (SWB)	SE	β	p
Step 1: Controls
Age	-0.005	0.004	-.06	.241	-0.008	0.005	-.08	.128
Gender (1 = female)	-0.078	0.066	-.05	.239	-0.061	0.071	-.04	.388
Education	0.058	0.022	.13	.010	0.041	0.024	.09	.086
Household income	0.097	0.027	.19	<.001	0.146	0.030	.25	<.001
Self-rated health	0.124	0.035	.16	<.001	0.189	0.037	.23	<.001
Step 1: FF: R² = .09, F(5,473) = 9.36, p < .001 | SWB: R² = .15, F(5,473) = 16.69, p < .001
Step 2: Migration Exposure added
Migration intensity	-0.452	0.123	-.21	<.001	-0.227	0.130	-.10	.081
Duration (years)	-0.013	0.008	-.07	.110	-0.008	0.009	-.04	.397
Destination (1 = inter-prov)	-0.156	0.069	-.11	.024	-0.078	0.073	-.05	.285
Contact frequency	0.097	0.020	.22	<.001	0.064	0.022	.15	.003
Remittance adequacy	0.058	0.030	.09	.057	0.166	0.032	.25	<.001
Step 2: FF: R² = .26, ΔR² = .17, p < .001 | SWB: R² = .28, ΔR² = .13, p < .001

β = standardized regression coefficient. FF = family functioning; SWB = subjective wellbeing.

3.11. Hypotheses 2 and 3: Mediation by Solidarity Dimensions

The a-paths showed that higher migration intensity was associated with lower associational (a₁ = -0.41) and affectual (a₃ = -0.18) solidarity and with higher functional solidarity (a₂ = 0.27). The migration-to-normative path was not significant (a₄ = −0.05, p > .05); because normative solidarity was the least reliable scale (α = .77, AVE = .48), this null association should be read with the measurement caveat detailed in the Discussion. The b-paths diverged across outcomes. For family functioning, associational, affectual, and normative solidarity all translated into the outcome, but functional did not. For subjective well-being, functional solidarity became a strong predictor (b₂ = 0.26). The dominant indirect path on family functioning ran through associational solidarity, ab = -0.142, 95% CI [-0.207, -0.082]. For subjective well-being, both the negative association pathway and the positive functional pathway were significant, supporting H3. Table 9 reports the parallel mediation results, and Fig. 1 illustrates the corresponding path diagrams.

Table 9.

Parallel Multiple Mediation of Migration Effects on Family Functioning and Subjective Well-being.

Path	B (FF)	Boot SE	95% BC CI (FF)	B (SWB)	Boot SE	95% BC CI (SWB)
a-paths: Migration → Mediator (identical across both DVs)
Migration → Associational (a₁)	-0.41	0.08	[-0.57, -0.25]***	-0.41	0.08	[-0.57, -0.25]***
Migration → Functional (a₂)	0.27	0.07	[0.13, 0.41]***	0.27	0.07	[0.13, 0.41]***
Migration → Affectual (a₃)	-0.18	0.06	[-0.31, -0.06]**	-0.18	0.06	[-0.31, -0.06]**
Migration → Normative (a₄)	-0.05	0.05	[-0.14, 0.05]	-0.05	0.05	[-0.14, 0.05]
b-paths: Mediator → Outcome
Associational → DV (b₁)	0.35	0.05	[0.26, 0.45]***	0.23	0.05	[0.13, 0.34]***
Functional → DV (b₂)	0.08	0.05	[-0.02, 0.18]	0.26	0.05	[0.16, 0.36]***
Affectual → DV (b₃)	0.22	0.05	[0.12, 0.32]***	0.20	0.06	[0.09, 0.31]***
Normative → DV (b₄)	0.20	0.06	[0.08, 0.33]**	0.17	0.06	[0.05, 0.30]**
Indirect effects (a × b)
via Associational	-0.142	0.034	[-0.207, -0.082]*	-0.092	0.027	[-0.148, -0.041]*
via Functional	0.022	0.014	[-0.005, 0.051]	0.066	0.023	[0.024, 0.114]*
via Affectual	-0.040	0.017	[-0.075, -0.010]*	-0.036	0.016	[-0.069, -0.008]*
via Normative	-0.010	0.011	[-0.032, 0.013]	-0.008	0.010	[-0.028, 0.011]
Total and direct effects
Total effect (c)	-0.421	0.108	[-0.633, -0.209]*	-0.314	0.106	[-0.522, -0.106]*
Direct effect (c′)	-0.252	0.100	[-0.448, -0.056]*	-0.244	0.102	[-0.444, -0.044]*

PROCESS Model 4 with bias-corrected 95% bootstrap CIs (5,000 resamples). a-paths identical across DVs. *p < .05, **p < .01, ***p < .001.

Fig. 1.

Parallel mediation of migration intensity on family functioning (Panel A) and subjective wellbeing (Panel B) through the four Bengtson solidarity dimensions. Solid arrows = significant a- and b-paths from PROCESS Model 4 with 5,000 bootstrap samples.

3.12. Hypothesis 4: Moderation by Community Cohesion

Community cohesion produced a significant buffering interaction with migration intensity in predicting family functioning (B = 0.16, p = .017). The interaction with role adaptation also reached significance (B = 0.19, p = .009). To test whether community cohesion buffered the indirect mechanism, we additionally fit moderated-mediation models. The Stage 1 interaction between migration and cohesion in predicting associational solidarity was significant (B = 0.18, 95% CI [0.04, 0.32]). Conditional indirect effects of migration on family functioning through associational solidarity were estimated at three levels of cohesion: at -1 SD (W = 2.79), the indirect effect was -0.181, 95% CI [-0.262, -0.108]; at the mean (W = 3.49), -0.131, [-0.194, -0.073]; at +1 SD (W = 4.19), -0.078, [-0.140, -0.024]. The index of moderated mediation was 0.058 (95% CI: [0.012, 0.110]), indicating a significant gradient. Hypothesis 4 was supported. Table 10, Table 11 present the results of the moderated regression and moderated mediation, and Fig. 2 illustrates the conditional effects.

Table 10.

Moderation by Community Cohesion: Migration × Cohesion Interactions Across Three Outcomes.

Model / Predictor	B	SE	95% CI	p	Sig
Model A: Outcome = Family Functioning
Migration intensity (X)	-0.428	0.110	[-0.644, -0.212]	<.001	***
Community cohesion (W)	0.286	0.060	[0.168, 0.404]	<.001	***
X × W interaction	0.156	0.065	[0.029, 0.283]	.017	*
Model A fit: R² = .33; ΔR² (interaction) = .013
Model B: Outcome = Subjective Wellbeing
Migration intensity (X)	-0.296	0.115	[-0.522, -0.070]	.010	*
Community cohesion (W)	0.224	0.062	[0.102, 0.346]	<.001	***
X × W interaction	0.103	0.058	[-0.011, 0.217]	.078	†
Model B fit: R² = .29; ΔR² (interaction) = .005
Model C: Outcome = Role Adaptation
Migration intensity (X)	-0.371	0.112	[-0.591, -0.151]	.001	**
Community cohesion (W)	0.251	0.061	[0.131, 0.371]	<.001	***
X × W interaction	0.187	0.072	[0.045, 0.329]	.009	**
Model C fit: R² = .27; ΔR² (interaction) = .017

Predictors mean-centered. †p < .10, *p < .05, **p < .01, ***p < .001.

Table 11.

Moderated Mediation: Community Cohesion Moderating the Associational Pathway to Family Functioning.

Path / Conditional Effect	Estimate	Boot SE	95% BC CI
Stage 1 (a-path): Migration × Cohesion → Associational
Migration (X)	-0.39	0.08	[-0.55, -0.23]
Community cohesion (W)	0.31	0.06	[0.19, 0.43]
X × W on mediator	0.18	0.07	[0.04, 0.32]
Stage 2 (b-path): Associational → Family Functioning
Associational → FF	0.34	0.05	[0.25, 0.44]
Conditional indirect effects of migration on FF via associational
Low cohesion (-1 SD; W = 2.79)	-0.181	0.039	[-0.262, -0.108]
Mean cohesion (W = 3.49)	-0.131	0.031	[-0.194, -0.073]
High cohesion (+1 SD; W = 4.19)	-0.078	0.029	[-0.140, -0.024]
Index of moderated mediation
Index	0.058	0.025	[0.012, 0.110]

PROCESS Model 7 with 5,000 bootstrap samples. Conditional indirect effects estimated at -1 SD, mean, and +1 SD of community cohesion.

Fig. 2.

Moderation by community cohesion. Panel A shows simple slopes of migration intensity on family functioning at low (-1 SD), mean, and high (+1 SD) levels of cohesion. Panel B shows the conditional indirect effects of migration on family functioning via associational solidarity at the same three levels of cohesion.

3.13. Subgroup Analyses: Migration Destination and Remittance Receipt

Inter-provincial migration was associated with roughly twice the total-effect penalty on family functioning (c = -0.49) compared with within-province migration (c = -0.31), and the difference in indirect effects via associational solidarity was significant, Wald χ² = 5.94, p = .015. The remittance-receipt comparison provided the cleanest test of Hypothesis 3: among the 380 households receiving remittances, the functional pathway delivered a significant positive indirect effect on subjective well-being, ab = 0.094, p < .01, but among the 99 non-recipients, this pathway collapsed, ab = -0.018, n.s.; subgroup χ² = 8.21, p = .004. Table 12 reports the subgroup contrasts; Fig. 3 visualizes the indirect effects across all subgroups in a forest plot.

Table 12.

Subgroup Analyses: Migration Destination and Remittance Receipt.

Outcome / Pathway	Inter-prov (n=294)	Within-prov (n=185)	Recipients (n=380)	Non-recipients (n=99)
Total effect of migration intensity on outcome
Family functioning (c)	-0.49***	-0.31*	-0.39***	-0.44**
Subjective wellbeing (c)	-0.36**	-0.20	-0.24*	-0.55***
Indirect effect via Associational solidarity
→ Family functioning	-0.171***	-0.094**	-0.144***	-0.131**
→ Subjective wellbeing	-0.111**	-0.063*	-0.094**	-0.087*
Indirect effect via Functional solidarity
→ Family functioning	0.026	0.011	0.029	-0.008
→ Subjective wellbeing	0.082**	0.034	0.094***	-0.018 n.s.
Subgroup comparison (Wald χ² for path equality)
Assoc. → FF (a × b equality)	—	χ² = 5.94, p = .015*	—	χ² = 0.21, p = .649
Func. → SWB (a × b equality)	—	χ² = 4.38, p = .036*	—	χ² = 8.21, p = .004**

Indirect effects from PROCESS Model 4 re-estimated within each subgroup. *p < .05, **p < .01, ***p < .001.

Fig. 3.

Subgroup forest plot of indirect effects of migration intensity on family functioning (via associational solidarity) and subjective wellbeing (via functional solidarity), shown across destination, remittance-receipt, and gender subgroups. Squares = significant indirect effects (95% CI excludes zero); circles = non-significant. Error bars = 95% bias-corrected CIs.

3.14. Hypothesis 5: Gender Comparisons

Older mothers showed numerically larger migration penalties on associational solidarity (β = -.36 vs. -.19), family functioning (β = -.27 vs. -.18), and subjective wellbeing (β = -.20 vs. -.12) than older fathers. Fathers showed a sharper penalty on role adaptation (β = -.25 vs. -.16). The χ² difference test reached significance for the migration → associational solidarity path, χ² = 4.81, p = .028, and for the migration → family functioning path, χ² = 4.12, p = .042. Hypothesis 5 received partial support. Table 13 reports the multi-group path comparisons.

Table 13.

Multi-Group Path Comparisons by Respondent Gender.

Path	Fathers (n = 278)	Mothers (n = 201)	χ² diff	p
Migration → Family functioning	β = -.18***	β = -.27***	4.12	.042
Migration → Subjective wellbeing	β = -.12*	β = -.20**	2.07	.150
Migration → Role adaptation	β = -.25***	β = -.16*	2.43	.119
Migration → Associational solidarity	β = -.19***	β = -.36***	4.81	.028
Migration → Functional solidarity	β = .25***	β = .29***	0.31	.578
Migration → Affectual solidarity	β = -.14**	β = -.22***	1.82	.178
Contact frequency → Family functioning	β = .20***	β = .31***	2.15	.143
Remittance adequacy → Subjective wellbeing	β = .27***	β = .21**	0.78	.377
Community cohesion × Migration → FF	β = .14*	β = .19**	0.97	.325

Standardized path coefficients (β). *p < .05, **p < .01, ***p < .001.

4. Discussion

This study examined how adult-child migration relates to family functioning and subjective wellbeing, and which solidarity pathways account for the observed associations. Migration intensity was associated with poorer family functioning (β = -0.21, p < .001), with the bulk of this association attributable to reduced associational solidarity (indirect = -0.142, 95% CI [-0.207, -0.082]). For subjective well-being, the pattern reversed: remittance adequacy emerged as the strongest direct predictor (β = .25, p < .001), and functional solidarity served as a significant compensatory mediator (indirect = 0.066, 95% CI [0.024, 0.114]). The normative dimension was preserved despite eroded behavioral contact (a-path = -0.05, ns). Community cohesion moderated the associational pathway (index of moderated mediation = 0.058, 95% CI [0.012, 0.110]), and older mothers carried larger associational penalties than older fathers (β = -.36 vs. -.19; χ² = 4.81, p = .028).

The pattern in which poorer family functioning accompanied migration primarily through reduced associational contact, while remittances were positively associated with subjective well-being, is consistent with broader evidence from China that the type of intergenerational support matters more than its mere presence. Luo and colleagues (2023), reported that financial support from adult children promoted mental health only when contact was maintained, whereas support without ongoing communication had no effect. Tan and colleagues (2025) similarly found that caregiving support significantly reduced depressive symptoms, whereas financial support alone had no meaningful effect. Feng and colleagues (2025) demonstrated that emotional support reduced depression through life satisfaction and well-being, whereas economic and care support showed weaker direct paths. The asymmetry observed here, therefore, reflects a broader pattern: relational presence and material flow operate through different psychological channels for older parents.

The strong associational pathway likely reflects the way internal Chinese labour migration restructures daily contact rather than simply reducing it. Adult children working in coastal cities cannot be present at meals, family gatherings, or the small daily exchanges that sustain associational solidarity, even when they remit funds and call regularly (Baykara-Krumme & Fokkema, 2019). Jiao and colleagues (Jiao et al., 2024) showed that living arrangements mediate the effect of widowhood on loneliness, with co-residence with adult children buffering emotional strain. When co-residence is no longer available, mediated contact only partially fills the gap. Qin and Li (2015) reported that internet use among older adults is positively linked to subjective well-being through social participation, with the strongest benefits for rural residents and those with limited offline networks. The associational deficit observed in our sample fits this pattern: when behavioural contact thins, regular calls cannot fully restore the relational substrate of family functioning.

The contrast between negative associational effects and positive functional effects forms what we describe as the remittance paradox. Material support compensates for the well-being cost of adult-child absence but does not substitute for the relational presence that drives family functioning. Zhao (2023) argued that filial piety norms enable migrant adult children to redistribute financial, instrumental, and emotional support in modified forms; however, remittances cannot replicate the daily co-presence that traditionally underpinned family-based eldercare. Zhao and Zhu (2026) demonstrated that a favourable community environment can partially compensate for insufficient intergenerational support. The dimensional decomposition in our results extends these findings by specifying which solidarity pathway carries which effect.

The persistence of strong endorsement of filial piety alongside eroded behavioural contact is one of the more striking patterns in our data. Wang and colleagues showed that family travel between adult children and aging parents enhanced well-being through generational interaction and psychological resilience, suggesting that families actively reconfigure their normative practices when geographic separation prevents daily contact (2024). Yang and colleagues (2025) found that filial-piety-driven self-sacrifice persisted even under significant economic pressure and lifestyle disruption, with caregivers describing role adaptation rather than role abandonment. Our findings are consistent with this evidence: normative commitment is preserved as a cultural anchor even when behavioural compliance becomes impractical (Moreno-Menéndez et al., 2026). This interpretation should be read alongside a measurement consideration. Normative solidarity was the weakest scale in the measurement model, with the lowest internal consistency (α = .77) and an average variance extracted below the .50 benchmark (AVE = .48), so a non-trivial share of its variance is measurement error. Because attenuation from imperfect reliability biases observed associations toward zero, the non-significant migration-to-normative path (b = −0.05, p > .05) may partly reflect this attenuation rather than genuine substantive preservation alone. We therefore read the normative-preservation pattern as suggestive and as warranting confirmation with a strengthened measure of normative solidarity.

The significant but smaller affectual indirect effect indicates that migration not only thins daily contact but also weakens the affective core of the parent-child relationship. Feng and colleagues (2025) reported that intergenerational relationships reduced depression through self-rated health and well-being, with emotional closeness carrying particular weight. Cui and colleagues (2024) showed that lonely older adults engaged more heavily in mediated communication, with sensation seeking and fear of missing out partially mediating the relationship. Zhang and colleagues (2025) reported that smartphone use among older adults reduced depression risk primarily through communication with adult children, with greater benefits for rural residents. These findings suggest that frequency-of-contact interventions may need to be supplemented with quality-focused communication to maintain affective closeness across distance (Kernová et al., 2025).

The gender pattern in our results aligns with broader evidence from China on differential caregiving burden. Mou and colleagues (2025) found that depressive symptoms rose sharply once weekly grandchild caregiving exceeded 40 hours, a threshold that older mothers in migration-sending villages frequently cross. The steeper associational penalty observed for older mothers in our sample likely reflects this combination: greater sensitivity to relational support coupled with heavier caregiving demands when adult daughters and daughters-in-law also migrate (Nishi et al., 2010; Shekhani, 2024).

The moderated-mediation finding indicates that community cohesion did not simply add a main effect on outcomes but dampened the associational pathway through which migration was associated with poorer family functioning. Hao and Chen (2025) demonstrated that neighbourhood cohesion reduced depression, anxiety, and stress, with subjective well-being as a partial mediator. This suggests that mountainous geography and dispersed settlement amplify the importance of local support networks (Shi et al., 2022). Yi and colleagues (2026) identified low social support as a defining feature of the most severely affected profiles. These findings collectively support our interpretation: in rural contexts, community cohesion does not just supplement family support but actively absorbs some of the relational deficit that migration creates (Brieant & Burt, 2025).

Several implications follow for rural elder-care policy. First, programmes targeting only the material well-being of left-behind older parents address the part of the problem that remittances are already partly handling, while leaving the relational deficit untouched. Wang and colleagues (2023) demonstrated that internet use among grandparents partially mediated the relationship between grandchild caregiving. Digital-literacy training for older parents in migration-sending villages is therefore a promising intervention. Second, the gender pattern argues for dedicated outreach to older mothers, who carry both the steepest relational penalties and the heaviest caregiving load. Third, Lang (2026) showed that physical exercise is associated with subjective wellbeing through three-dimensional capital (health, social, psychological), with rural older adults benefiting disproportionately. Combined community cohesion, digital literacy, and structured activity programmes are therefore likely to outperform single-channel interventions. Fourth, Mou and colleagues (2025) recommended that retirement-policy reforms be paired with phased-exit options, psychological screening, and respite services to alleviate dual pressures on older adults.

These interpretations warrant an explicit caution about causal direction. The mediation and moderated-mediation estimates describe patterns of conditional association consistent with the hypothesized model; they do not establish that migration changes solidarity or that changes in solidarity produce the outcomes. Reverse pathways are plausible: families that function poorly may release more of their adult children into migration, and parents with lower wellbeing may receive less contact or fewer remittances. Selection effects and omitted household-level variables, such as land holdings, chronic illness in the household, or pre-migration family conflict, offer further alternative explanations, and cross-sectional mediation can misstate the magnitude of longitudinal indirect effects. The pathway language used in this paper, therefore, describes the structure of the statistical model, not a demonstrated causal process.

The strengths of this study include the dimensional decomposition of solidarity, the formal multi-group test of gender differences, and the cross-checking of estimates between R and SPSS PROCESS. Eight limitations deserve attention. First, the cross-sectional design precludes causal inference, and reverse pathways cannot be ruled out. Relatedly, the number of models estimated, spanning multiple mediators, interaction terms, conditional indirect effects, and subgroup contrasts, creates some risk of over-fitting and of capitalizing on chance despite the bootstrap procedures; the smaller conditional and subgroup estimates should be read with corresponding restraint. We note in particular that maintaining a case-to-parameter ratio above 30:1 speaks to the adequacy of each individual model, not to familywise Type I error across the full set of estimated effects. No correction for multiplicity was applied across the tests spanning H1–H5, the subgroup contrasts by destination, remittance pattern, and gender, and the nine sensitivity analyses, and the bias-corrected bootstrap intervals protect precision within each model rather than the family of inferences as a whole. Accordingly, the moderation, moderated-mediation, subgroup, and sensitivity results are best read as exploratory and hypothesis-generating, identifying patterns for confirmation in adequately powered, pre-registered replication rather than as confirmatory tests. Second, the male-heavy sample (58% male) is atypical for older Chinese cohorts so tests should be regarded as conservative. Third, all measures are self-report, which raises the possibility of social desirability bias on culturally sensitive items (Bispo, 2022); Mohamed Zaki and colleagues (2025) noted that older adults in collectivistic Asian societies often interpret family-related items through a filial-piety lens that may mute disclosure. Fourth, the four-stage cluster sampling design produces residual clustering even after adjustment for design effects. Fifth, the destination variable conflated international and inter-provincial migration in primary models; the three-category sensitivity check addressed this, but the international subgroup remained small (n = 32). Sixth, the geographic scope limits generalizability to other Chinese rural regions with different migration regimes. Seventh, and most consequential for interpretation, the sampling frame included only households with at least one migrant adult child. The estimates, therefore, describe how variation in migration exposure relates to outcomes within migrant households; they say nothing about migration relative to its absence, and migrant households likely differ from non-migrant households in income, family structure, and social capital (C. Liu et al., 2021). Comparative designs that include non-migrant households, or panel designs that observe households before and after a child's departure, are needed to answer that question. Eighth, although the statistical checks reported above indicated no dominant method factor, all constructs share a single self-report source, and common method variance cannot be entirely excluded.

Future work should pursue longitudinal follow-up, dyadic designs pairing older parents with their migrant children, comparative work across China's major sending regions, and mixed-methods designs that combine survey data with ethnographic observation around the normative-preservation pattern this study has documented.

5. Conclusion

In this cross-sectional sample of rural migrant households, adult-child out-migration was associated with two simultaneous channels that operate in opposite directions within the family lives of older parents. The relational channel weakened as migration intensity rose, and that weakening statistically accounted for most of the association with poorer family functioning, while the material channel strengthened with migration and partially offset the subjective-wellbeing cost of child absence without substituting for the relational loss. The remittance paradox described in the Discussion captures this duality, and the normative-preservation finding documents an intergenerational ambivalence in which older parents hold onto filial-piety norms even as behavioral contact with their migrant children erodes. Community cohesion buffers these patterns by dampening the associational mechanism, rather than merely adding a main effect. Within the bounds of this design, the correlates of mass internal labor migration among older parents in rural China appear to be dimensional, gendered, and contingent on the surrounding community fabric.

Consent to participate

Written informed consent was obtained from all participants prior to data collection.

Funding

The authors declare that no funding was received for this work.

CRediT authorship contribution statement

Junzheng Guo: Writing – original draft, Validation, Supervision, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Usama Javed: Writing – review & editing, Validation, Formal analysis, Data curation.

Declaration of competing interest

The authors have no relevant interests to disclose.

Data availability

De-identified data are available from the corresponding author on reasonable request, subject to departmental approval.