Digital Transitions and Sustainable Futures: Family Structure’s Impact on Chinese Consumer Saving Choices and Marketing Implications

Abstract

Family structure has long been regarded as an important determinant of household saving, yet the empirical evidence for developing economies remains limited. Using the 2018–2022 panels of the China Family Panel Studies (CFPS), a nationwide survey that follows 16,519 households across three waves, the present study investigates how family size, the elderly share, and the child share jointly shape saving behavior. A household fixed effects framework is employed to control for time-invariant heterogeneity, followed by a sequential endogeneity strategy: external-shock instruments are tested and rejected, lagged two-stage least squares implement internal instruments, and a dynamic System-GMM model is estimated to capture saving persistence. Robustness checks include province-by-year fixed effects, inverse probability weighting for attrition, balanced-panel replication, alternative variable definitions, lag structures, and sample filters. Family size raises the saving rate by 4.6 percentage points in the preferred dynamic specification (p < 0.01). The elderly ratio remains insignificant throughout, whereas the child ratio exerts a negative but model-sensitive association. A three-path mediation analysis indicates that approximately 26 percent of the total family size effect operates through scale economy savings on quasi-fixed expenses, 19 percent is offset by resource dilution pressure, and less than 1 percent flows through a precautionary saving channel linked to income volatility. These findings extend the resource dilution literature by quantifying the relative strength of competing mechanisms in a middle-income context and showing that cost-sharing economies dominate child-related dilution for most households. Policy discussion highlights the importance of public childcare subsidies and targeted credit access for rural parents, whose saving capacity is the most constrained by additional children. The study also demonstrates that fixed effects estimates of family structure can be upward-biased unless dynamic saving behavior and internal instruments are considered.

1. Introduction

Household saving behavior, a cornerstone of micro-level financial resilience and macroeconomic stability, is increasingly recognized as a critical lever for achieving global Sustainable Development Goals (SDGs). As a buffer against economic shocks (SDG 1: No Poverty), a primary source of capital for sustainable investment (SDG 8: Decent Work and Economic Growth), and a reflection of resource allocation fairness (SDG 10: Reduced Inequalities), household financial choices have profound implications for a nation’s long-term sustainable development trajectory. Understanding the drivers of these choices is therefore not merely an academic exercise but a prerequisite for designing effective policies that foster both economic prosperity and social equity.

Against this global backdrop, China presents a compelling and urgent case study. As the world’s second-largest economy, its persistently high household savings rate has long been a defining feature of its economic landscape, fueling investment and growth. However, the country is currently navigating a period of unprecedented structural transformation, posing significant challenges to these established saving patterns. Three powerful forces are converging: First, a profound demographic shift, characterized by a rapidly aging population and shrinking family sizes, alters the traditional intergenerational support system and life cycle needs. Second, a pervasive digital transformation, driven by mobile payments and FinTech innovations, is fundamentally reshaping how households manage their finances, access financial products, and make saving decisions. Third, persistent urban–rural and income disparities create a complex tapestry of saving behaviors, where a single national policy may no longer be effective. The confluence of these forces creates an urgent need to re-examine the determinants of household savings in contemporary China.

While a rich body of literature has explored the determinants of household savings, critical gaps remain, particularly at the intersection of these transformative trends. Much of the existing research on family structure often relies on older data, failing to capture the dynamics of China’s post-2018 economic realities and the nuanced effects of digitalization. The primary innovation of this study lies in its integrated approach. We move beyond a singular focus on demographic impacts by explicitly examining the influence of family structure on household savings within the dual contexts of rapid digital transformation and the overarching goal of sustainable development. This multi-faceted lens allows us to not only measure the net effects of family composition changes but also to explore the heterogeneous mechanisms and policy implications for fostering more inclusive and sustainable consumer financial behavior.

To address these gaps and provide timely, evidence-based insights, this study is guided by three central research questions:

What is the net effect of changes in key family structure variables—specifically family size, the elderly ratio, and the child ratio—on household saving rates in contemporary China, after rigorously controlling for endogeneity and saving persistence?

What are the primary mechanisms driving these effects? Specifically, what are the relative contributions of economies of scale, resource dilution, and precautionary saving motives?

How do these impacts and their underlying mechanisms differ across crucial socioeconomic divides, particularly between urban and rural households and across different income quintiles?

By answering these questions using recent panel data from the China Family Panel Studies (2018–2022), this paper makes several key contributions [1]. First, it provides updated and robust empirical evidence on the complex relationship between family structure and saving in a major middle-income economy. Second, it quantitatively disentangles the competing theoretical mechanisms, offering a more nuanced understanding that moves beyond the traditional resource dilution narrative. Third, through detailed heterogeneity analysis, it generates targeted policy recommendations for financial institutions and policymakers aiming to enhance household financial resilience and promote equitable, sustainable development in line with the SDGs.

The remainder of this paper is structured as follows. Section 2 reviews the relevant literature and develops the theoretical framework and hypotheses. Section 3 details the data, sample construction, and econometric methodology. Section 4 presents the empirical results, including baseline findings, heterogeneity analysis, and a comprehensive set of robustness checks. Section 5 discusses the findings and their theoretical, practical, and policy implications. Finally, Section 6 concludes with the study’s limitations and directions for future research.

2. Literature Review

Household saving behavior in China, a critical determinant of both micro-level welfare and macroeconomic stability, has long been characterized by high rates. This has profoundly shaped the nation’s investment, consumption, and economic growth trajectories. However, contemporary China is experiencing unprecedented demographic shifts, including shrinking household sizes and rapid population aging. These transformations exert significant pressure on traditional saving patterns, creating an urgent need to understand their multifaceted impacts. Simultaneously, a pervasive digital transformation is reshaping how households manage their finances, offering new tools for saving and investment and introducing new complexities and risks.

The existing literature has established that family structure—namely its size and age composition—influences saving through competing mechanisms like economies of scale, resource dilution, and precautionary motives. However, much evidence from developing contexts relies on older or cross-sectional data, which fails to capture dynamic household adjustments or robustly address endogeneity. Specifically for China, critical gaps persist in understanding the net effect of these competing forces using the most recent data (post-2018), which reflects current economic realities. Furthermore, while heterogeneity is acknowledged, a granular investigation into how these effects differ across crucial socioeconomic divides—such as urban–rural and income quintiles—remains underdeveloped, limiting the formulation of targeted, sustainable policies.

To address these gaps, this study investigates the dynamic impact of family structure on household saving choices in China using recent panel data from the China Family Panel Studies (2018–2022) [1].

2.1. Family Size and Household Savings: Competing Mechanisms

The relationship between family size and household savings is a cornerstone of demographic economics, with diverse theoretical underpinnings and empirical findings. A prominent strand of international literature, exemplified by studies such as Çebi-Karaaslan et al. (2022) using Turkish data [2], often reports a negative correlation between family size and saving rates. This is frequently attributed to the resource dilution effect, where larger households face increased aggregate consumption demands, particularly for non-discretionary items, thereby constraining their capacity to save. While insightful, much of the existing evidence, particularly from developing contexts, has relied on cross-sectional analyses or older panel data, which may not fully capture households’ dynamic adjustments to evolve family sizes or robustly address endogeneity concerns.

However, the impact of family size on savings is potentially more nuanced. Economies of scale in consumption present a counteracting force; larger households can achieve lower per capita costs for shared goods (e.g., housing, utilities) and bulk purchases, which could increase saving capacity. Furthermore, the motive for precautionary savings may be amplified in larger families. An increased number of dependents, particularly children requiring significant long-term investment in education and healthcare or elderly members needing care, can heighten future financial uncertainty and incentivize greater accumulation of precautionary reserves, especially in societies with developing social safety nets and significant out-of-pocket expenditures, as is characteristic of contemporary China [3]. The high cost of housing in many Chinese urban centers further suggests that co-residence and larger family units might, paradoxically, enable savings through shared living expenses. In this context, digital financial tools and platforms can play a significant role, particularly for larger or multi-generational households managing complex finances. For instance, budgeting apps (e.g., Mint, YNAB, or local Chinese equivalents like Wacai and Tally) allow for collaborative tracking of income and expenses, helping to identify areas for potential savings. Automated saving platforms (e.g., Acorns, Digit, or features within mobile banking apps) can facilitate regular, small-amount contributions towards specific goals like children’s education or emergency funds, making saving more accessible and less burdensome. Digital insurance platforms offer easier access to various products (health, life, accident) that can mitigate future financial shocks, potentially influencing the level of precautionary savings needed. Furthermore, online investment platforms provide access to diversified, low-cost investment options, enabling even households with modest surplus funds to grow their savings. The ease of monitoring and adjusting these digital financial instruments can empower families to be more proactive and adaptive in their financial planning, potentially altering traditional saving behaviors by making goal-setting more tangible and progress more visible.

Indeed, research specific to China has highlighted these complexities. For instance, Curtis et al. (2015) demonstrated that demographic changes, particularly reductions in family size (fewer dependent children), explained a significant portion of household saving rate growth [4]. Conversely, studies examining the impact of policies like the one-child policy have suggested that reduced expected old-age support from fewer children led to increased parental savings [5]. These findings underscore that the net effect of family size on savings in China is an empirical question contingent on the relative strengths of resource dilution, economies of scale, and precautionary responses. A critical gap remains in systematically evaluating these competing mechanisms using recent, comprehensive panel data that reflects China’s current demographic and economic realities, including evolving family support norms and pressures from the education and housing sectors. This study aims to address this by analyzing the 2018–2022 CFPS data to provide updated insights into how changes in household size influence saving rates in China [1].

2.2. Family Age Structure and Savings: Life Cycle and Precautionary Dynamics

The age composition of a household is another critical demographic factor influencing saving decisions, primarily analyzed through the lenses of the Life cycle Savings Hypothesis (LCSH) and precautionary saving theories. The standard LCSH, as discussed by Smithers [6], posits that rational individuals smooth consumption over their lifetimes by saving during their productive working years and dissaving during retirement. Consequently, a higher proportion of elderly individuals in a population or household might be expected to have lower aggregate saving rates. However, empirical applications of the LCSH, particularly in East Asian contexts, including China, often reveal deviations. For instance, studies like Zeng et al. show a complex interplay where aging can influence asset prices, and household savings act as a mediating factor, sometimes mitigating downward pressure from aging. Some studies have shown that, in a context of high unemployment and the absence of unemployment benefits, uncertainty can mitigate the downward pressure of population aging on the savings rate [7].

Furthermore, international evidence, such as Le Blanc et al. using European data [8], indicates that precautionary saving is a prevalent motivation, potentially leading elderly individuals to continue saving or dissave at slower rates than predicted by simple LCSH models, especially when facing uncertainties regarding health expenditures and pension adequacy. This is particularly relevant for China, where the social security system is still developing, and out-of-pocket healthcare costs for the elderly can be substantial. The debate thus persists regarding the dominant drivers of saving behavior among the elderly in China—whether life cycle consumption smoothing, precautionary motives, or even bequest intentions prevail.

Conversely, the presence of minor children typically exerts downward pressure on household savings due to increased consumption and direct child-rearing expenditures. In China, this effect is often amplified by intense competition and high investment in children’s education, as highlighted by Lugauer et al. [9], who found that the number of dependent children negatively correlates with household saving rates, with urban–rural differences in education expenditure contributing to differential saving patterns. The significant financial burden associated with raising children, from basic needs to extensive educational investments, can severely limit a household’s ability to save.

Therefore, understanding the net impact of shifts in a household’s age structure—both an increasing proportion of elderly members and the presence of dependent children—requires disentangling these often-competing theoretical influences within China’s specific institutional and socioeconomic context. A persistent gap exists in comprehensively analyzing these dynamics using recent panel data that can capture ongoing changes in pension systems, healthcare costs, and educational pressures. This study will contribute by examining how the proportions of elderly individuals and minor children independently affect household saving rates, shedding light on the prevailing mechanisms in contemporary China.

2.3. Socioeconomic Context, Policy, and Heterogeneity in Savings Behavior

Household saving decisions are not made in a vacuum. However, they are deeply embedded within broader socioeconomic contexts and policy environments, which can lead to significant heterogeneity in responses to demographic changes. Urban–rural disparities are a prominent feature of China’s economic landscape and heavily influence saving behavior. Differences in income levels, employment stability, access to financial services, and social safety nets (e.g., pensions, healthcare) between urban and rural areas can lead to divergent saving patterns and motivations. For instance, Hong et al. (2022) [10] found that rural social security can promote consumption structure upgrading by reducing precautionary saving, while Li, Wu, and Xiao (2020) demonstrate that digital inclusive finance significantly boosts household consumption—especially among low-asset, low-income, and rural families—by easing liquidity constraints through online credit and digital payment channels [11]. However, there is less granular understanding of how changes in family structure specifically interact with urban–rural residence to impact savings differentially, representing an area where more recent and detailed analysis is needed.

Similarly, income levels are expected to moderate the relationship between family structure and savings. Households at different points in the income distribution face varying constraints and opportunities. Low-income households may struggle with subsistence needs, making savings difficult regardless of family structure. In contrast, Jappelli and Pistaferri show that high-wealth households exhibit substantially lower marginal propensities to consume following income shocks—implying correspondingly lower marginal saving responses—than less wealthy groups [12]. Middle-income households, however, might be particularly sensitive, balancing aspirations for upward mobility with significant financial pressures related to housing, education, and eldercare. While some studies touch upon income-related heterogeneity, a systematic investigation using recent data on how family structure effects vary across detailed income quintiles in China remains relatively under-explored.

Policy interventions, such as the extent and design of social security systems, financial market development, and family planning policies, also play a crucial mediating role. The literature suggests that inadequate pension security can fuel higher savings, while improved access to credit can impact consumption and, indirectly, savings [13]. Cultural factors, such as the traditional emphasis on thrift or intergenerational support, as explored by Fuchs-Schündeln et al. [14]. in an immigrant context, can further shape saving propensities, though quantifying their dynamic impact alongside structural changes is challenging.

While identifying these broad contextual influences, the existing literature often lacks sufficient granularity in exploring the interactive effects of family structure changes with these multifaceted socioeconomic and geographical stratifiers, particularly using recent panel data that covers the post-2018 period in China. This study aims to address this gap by explicitly examining the heterogeneity in the impact of family size, elderly ratios, and child ratios on savings across urban–rural locations and different income quintiles, thereby providing a more nuanced understanding of who is most affected by ongoing demographic transitions.

2.4. Summary of Literature and Positioning of the Current Study

The extant literature establishes that household saving behavior is influenced by a complex interplay of demographic factors (family size and age structure), economic theories (life cycle, precautionary, economies of scale, resource dilution), and the broader socioeconomic and policy context. International studies and research within China have highlighted several key relationships, such as larger families’ general resource dilution effect, deviations from simple life cycle saving patterns among the elderly, often due to precautionary motives, and the significant burden of child-rearing costs. Moreover, disparities linked to urban–rural status, income levels, and the evolving social security system are recognized as important moderators.

However, several critical gaps persist, particularly in contemporary China. Firstly, there is an ongoing need for analysis based on the most recent panel data (post-2018) to capture the effects of ongoing socioeconomic transformations and provide robust causal inferences regarding dynamic family transitions. Secondly, the dominant mechanisms driving the net effect of family size (scale economies vs. resource dilution vs. precautionary motives) and age structure (life cycle vs. precautionary/bequest motives for elderly savings) in China’s current environment require further disentanglement. Thirdly, while heterogeneity is acknowledged, there is a need for a more detailed and granular investigation into how the impacts of family structural changes differ across urban–rural divides and various income segments, moving beyond average effects.

This study is positioned to address these gaps directly. By utilizing the 2018–2022 China Family Panel Studies (CFPS) [1] data and employing a fixed effects modeling strategy, we aim to (1) provide updated estimates of the impact of household size, elderly population ratio, and minor child ratio on household saving rates; (2) offer insights into the likely prevailing mechanisms by carefully interpreting the direction and significance of these effects within China’s current context; and (3) conduct a thorough heterogeneity analysis across urban–rural locations and income quintiles to reveal differential impacts. Through this approach, this research seeks to provide a more nuanced and contemporarily relevant understanding of how family structure transformations are shaping household saving choices in China, offering robust empirical evidence for policymaking aligned with sustainable development.

3. Theoretical Analysis and Research Hypotheses

3.1. Theoretical Background

The relationship between family structure and household saving behavior is rooted in integrated economic theories that explain the complex interplay of demographic, behavioral, and contextual factors.

The Life cycle Savings Hypothesis (LCSH) posits that rational individuals smooth consumption over their lifetimes by saving during working years and dissaving in retirement [15]. This theory suggests households with higher proportions of non-earning members (elderly or children) should exhibit lower saving rates. However, empirical applications in East Asian contexts, including China, often deviate from LCSH predictions, necessitating complementary theories.

Precautionary Savings Theory extends this framework by emphasizing that households accumulate savings as a buffer against future uncertainties. In contemporary China, where social safety nets for healthcare and pensions are incomplete, perceived risks (e.g., medical expenses, income volatility, educational costs) drive households—especially those with larger family units—to save more than LCSH would predict. This motive can offset the dissaving tendency among elderly households, as seen in studies noting deviations from simple life cycle patterns [12].

For family size, two competing mechanisms dominate:

Economies of scale enable larger households to reduce per capita costs for shared goods (e.g., housing, utilities) and bulk purchases, freeing resources for saving. In China’s high-cost urban environments, co-residence in extended families can mitigate housing expenses, a key driver of this effect [4].

Resource dilution argues that additional family members increase aggregate consumption needs, particularly for non-discretionary items like food and education, straining budgets and reducing saving capacity [6]. In China, this is amplified by intense educational investments in children and rising living costs [16].

Age structure effects are equally complex:

Elderly ratios may trigger dissaving under LCSH but can instead increase savings via precautionary motives, especially for healthcare and long-term care costs in the absence of comprehensive social insurance [7].

Child ratios exert downward pressure on savings due to direct rearing costs and educational investments, a phenomenon exacerbated in China’s competitive educational landscape.

3.2. Conceptual Framework and Hypotheses

To clarify the theoretical logic, Figure 1 depicts the three measurable components of family structure—family size, elderly ratio, and child ratio—and the four pathways through which they can influence saving. Under the life cycle-Saving Hypothesis, a larger elderly share should reduce saving, whereas a stronger precautionary motive could offset or reverse that tendency. Economies of scale lower per capita fixed costs and raise savings, while resource dilution has the opposite effect by increasing consumption needs. The dashed grey box lists the moderators that may amplify or weaken each arrow.

The conceptual framework (Figure 1) synthesizes these theories to illustrate how family structure influences saving through four pathways:

Life cycle consumption smoothing (elderly ratio → dissaving, if LCSH prevails);

Precautionary saving (family size/elderly ratio → increased savings to mitigate uncertainty);

Economies of scale (family size → lower per capita fixed costs → higher savings);

Resource dilution (child ratio → higher consumption → lower savings).

Moderating factors like urban–rural location, income level, and social security coverage amplify or weaken these pathways [16].

Operational Definitions of Key Constructs

Table 1. Operational definitions of key constructs.

Construct/Term	Operational Definition	Source/Calculation
Household Saving Rate	(Household Annual Total Income—Household Annual Total Expenditure)/Household Annual Total Income. Expenditure excludes non-consumption items like financial investments or debt repayments.	Calculated from CFPS Family Economy data [1]
Family Size	The total number of co-residing members in a household.	CFPS Family Roster [1]
Elderly Ratio	The proportion of household members aged 65 or older is calculated as (Number of members ≥ 65)/Family Size.	Calculated from CFPS Individual Roster [1]
Child Ratio	The proportion of household members under the age of 16, calculated as (Number of members < 16)/Family Size.	Calculated from CFPS Individual Roster [1]
Urban/Rural Status	A binary variable indicates the household’s location, where 1 = Urban and 0 = Rural.	CFPS Geocode [1]
Household Per Capita Income	The natural logarithm of total annual household income divided by family size.	Calculated from CFPS Family Economy data [1]

provides explicit operational definitions for core variables, ensuring clarity and consistency in measurement:

3.3. Research Hypotheses

The family structure (family size, elderly ratio, and child ratio) affects the household saving rate through the following four core pathways:

Life cycle consumption smoothing: an increase in the elderly ratio may prompt households to reduce savings (in line with the LCSH theory’s “retirement dissaving” hypothesis), but precautionary motives may reverse this trend.

Precautionary saving motive: an expansion of family size or an increase in the elderly ratio raises future uncertainties (such as medical and educational expenses), thereby driving an increase in savings.

Economies of scale effect: a larger family size can reduce the per capita costs of shared goods like housing and utilities, freeing up resources for saving.

Resource dilution pressure: higher child ratio increases direct consumption expenditures on education and parenting, squeezing space for savings.

Justification: in high-cost urban China, shared housing and utilities (scale economies) and elevated financial uncertainty (precautionary motives for larger families) dominate child-related dilution [4].

H1. 

Family size positively correlates with household saving rates, driven by economies of scale and precautionary motives outweighing resource dilution.

Justification: child-rearing and education costs in China strain household budgets, with rural households facing tighter resource constraints [7].

H2. 

Child ratio negatively correlates with saving rates due to resource dilution, particularly in rural areas with limited public subsidies.

Justification: incomplete social security systems prompt urban households with elderly members to save for healthcare, while rural families rely on intergenerational support, creating countervailing forces [6].

H3. 

Elderly ratio exhibits no consistent direct effect on saving rates, as LCSH-driven dissaving is offset by precautionary motives.

Justification: urban-rural disparities in social security and income, combined with income-level differences in financial constraints, moderate the net impact of scale economies, resource dilution, and precautionary motives [16].

H4. 

Effects of family structure on saving rates vary across urban-rural divides and income quintiles.

4. Methodology

4.1. Data and Sample

This study utilizes data from the 2018, 2020, and 2022 waves of the China Family Panel Studies (CFPS) [1]. The CFPS is a large-scale, nationally representative, longitudinal social survey designed and implemented by the Institute of Social Science Survey (ISSS) of Peking University. Its comprehensive scope, covering individual, family, and community-level characteristics, makes it an exceptionally rich resource for investigating dynamic household behaviors in China.

The selection of CFPS is well-supported by its extensive use in high-impact scholarly research on the Chinese economy and society. It has been established as an authoritative dataset for studies within the country’s context, particularly in the sector of household finance, consumption, and wealth distribution. For instance, Lugauer et al. (2019) [9] specifically employed CFPS data to analyze the relationship between dependent children and household savings, demonstrating the dataset’s direct relevance and suitability for our research questions.

Furthermore, our choice of a panel data approach aligns with best practices in international research on household economics. Similar large-scale panel surveys, such as the Panel Study of Income Dynamics (PSID) in the United States and the German Socioeconomic Panel (SOEP), have been instrumental in advancing the understanding of saving behavior and its demographic determinants in other national contexts [17]. The longitudinal nature of CFPS allows us to control for time-invariant unobserved household heterogeneity, a critical advantage over cross-sectional studies for making more robust causal inferences.

Our sample construction begins with all households surveyed in the 2018 wave and tracks them through 2020 and 2022. We constructed a balanced panel by retaining only those households that participated in all three waves and providing complete information on our core variables (saving, income, expenditure, and family structure). This process yields a final sample that is well-suited for fixed effects panel analysis.

Justification of the 2018–2022 Study Period

The selection of the 2018–2022 period is based on the availability of the most recent data, which is essential for providing timely and policy-relevant insights. We explicitly acknowledge that this period encompasses the COVID-19 pandemic, a significant and atypical global shock. However, rather than invalidating the study, this unique context provides a valuable setting for examining household saving behavior under heightened uncertainty. The pandemic likely amplified the economic and health risks faced by households, making the analysis of precautionary saving motives—a central theme of this paper—particularly salient.

Crucially, our econometric methodology is specifically designed to account for such period-specific shocks. The use of a panel fixed effects model with the inclusion of year fixed effects (year dummies) is the key to our identification strategy. The year dummy variables for 2020 and 2022 absorb the average impact of any macro-level shocks that were common to all households in those years, including the effects of lockdowns, economic slowdowns, and changes in public health expenditure. By controlling for these common time trends, our model effectively isolates the “within-household” variation. Therefore, the estimated coefficients for our variables of interest (e.g., family size, child ratio) reflect the net effect of changes in a household’s internal structure on its saving rate, purged of the confounding influence of the pandemic’s aggregate shock. This methodological robustness ensures that our findings are not driven by the unique circumstances of the pandemic but rather by the underlying demographic dynamics we seek to understand.

4.2. Descriptive Statistics

Table 2. Key data and sources.

Variable	Define	Average Value	Standard Deviation	Data Sources
Saving Rate	(Incomes-Expenditure)/Incomes	−0.229	1.345	CFPS 2018–2022 [1]
Family Size	Total Number of Family Members	2.547	1.400	CFPS Family questionnaire [1]
City and Countryside Cluster	1 = urban, 0 = rural	0.53	0.499	CFPS Geocode [1]

presents descriptive statistics for the key variables in this study. Regarding saving-related variables, the household saving rate (Saving Rate 1) exhibits a mean value of −0.229 and a median of 0.129, indicating substantial negative savings among sampled households and a pronounced right-skewed distribution. This aligns with the reality of unbalanced income and expenditure patterns among Chinese households, where some families may experience considerable spending pressures resulting in negative savings. The alternative saving rate (Saving Rate 2), calculated based on consumption expenditure, demonstrates a comparatively more favorable profile with a mean of −0.070 and a median of 0.257, suggesting significant non-consumption expenditures among households.

Table 3. Descriptive statistics of key variables.

Variable	Observations	Mean	Std. Dev.	Minimum	25th Percentile	Median	75th Percentile	Maximum
Saving Rate 1	34,965	−0.229	1.345	−8.538	−0.373	0.129	0.457	0.919
Saving Rate 2	31,727	−0.070	1.183	−7.359	−0.202	0.257	0.538	0.909
Family Size	35,289	2.547	1.400	1	1	2	3	15
Elderly Ratio	35,289	0.201	0.340	0	0	0	0.333	1
Child Ratio	35,289	0.049	0.124	0	0	0	0	1
Working Age Ratio	13,879	0.751	0.343	0	0.6	1	1	1
Income Per Capita	35,289	31,814.9	37,670.7	916.7	10,000	20,000	38,400	245,000
Total Assets	32,952	805,611.1	1,989,010	−10,000,000	109,337.5	300,000	741,468.8	61,800,000

shows descriptive statistics for key variables, including saving rate, family size, population structure ratios, income per capita, and total assets.

Concerning family structure variables, the sampled households demonstrate an average size of 2.547 members with a median of 2, reflecting the downsizing trend of Chinese families. The elderly population ratio (age 65 and above) averages 0.201, indicating that approximately one-fifth of household members are elderly. The proportion of minor children (under 16) averages merely 0.049, with a median of 0, revealing that most households have no minor children. This corresponds to China’s low fertility rate and accelerating population aging phenomenon.

Regarding household economic conditions, the annual per capita income averages CNY 31,814.9 with a median of CNY 20,000, indicating a notable right-skewed income distribution. Household total assets average CNY 805,611.1 with a median of CNY 300,000, demonstrating even more pronounced distributional inequality, which aligns with the empirical wealth distribution patterns observed across Chinese households.

4.3. Econometric Model Specification

The core objective of this research was to investigate how family size and its internal structure dynamically affect household saving behavior over time [18]. Given that the analysis used panel data covering multiple time points and focused on identifying the impact of variable changes, a panel fixed effects (FE) model was chosen as the appropriate choice for the baseline analysis. The primary rationale for selecting a fixed effects model is its ability to effectively control for unobservable or difficult-to-measure inherent household characteristics that do not vary over time but may simultaneously influence both family structure and saving decisions [19].

Through differential or within-transformation, the fixed effect model eliminated potential interference from these time-invariant heterogeneities, thereby more accurately isolating the net effects of the time-varying independent variables. This was crucial for mitigating endogeneity bias caused by time-invariant omitted variables.

The baseline panel fixed effects model is specified explicitly as follows:

$$\mathrm{S}\mathrm{a}\mathrm{v}\mathrm{i}\mathrm{n}\mathrm{g}\mathrm{R}\mathrm{a}\mathrm{t}{\mathrm{e}}_{\mathrm{i}\mathrm{t}}={\mathsf{\beta }}_1\mathrm{F}\mathrm{a}\mathrm{m}\mathrm{i}\mathrm{l}\mathrm{y}\mathrm{S}\mathrm{i}\mathrm{z}{\mathrm{e}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\beta }}_2\mathrm{E}\mathrm{l}\mathrm{d}\mathrm{e}\mathrm{r}\mathrm{l}\mathrm{y}\mathrm{R}\mathrm{a}\mathrm{t}\mathrm{i}{\mathrm{o}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\beta }}_3\mathrm{C}\mathrm{h}\mathrm{i}\mathrm{l}\mathrm{d}\mathrm{R}\mathrm{a}\mathrm{t}\mathrm{i}{\mathrm{o}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\gamma }}^{\prime }{\mathrm{X}}_{\mathrm{i}\mathrm{t}}+{\mathsf{\lambda }}_{\mathrm{t}}+{\mathsf{\alpha }}_{\mathrm{i}}+{\mathsf{ϵ}}_{\mathrm{i}\mathrm{t}}$$

(1)

where $\mathrm{i}$ represents the household and $\mathrm{t}$ represents the year. $\mathrm{S}\mathrm{a}\mathrm{v}\mathrm{i}\mathrm{n}\mathrm{g}\mathrm{R}\mathrm{a}\mathrm{t}{\mathrm{e}}_{\mathrm{i}\mathrm{t}}$ is the savings rate of the household in the year. $\mathrm{F}\mathrm{a}\mathrm{m}\mathrm{i}\mathrm{l}\mathrm{y}\mathrm{S}\mathrm{i}\mathrm{z}{\mathrm{e}}_{\mathrm{i}\mathrm{t}}, \mathrm{E}\mathrm{l}\mathrm{d}\mathrm{e}\mathrm{r}\mathrm{l}\mathrm{y}\mathrm{R}\mathrm{a}\mathrm{t}\mathrm{i}{\mathrm{o}}_{\mathrm{i}\mathrm{t}}, \mathrm{a}\mathrm{n}\mathrm{d} \mathrm{C}\mathrm{h}\mathrm{i}\mathrm{l}\mathrm{d}\mathrm{R}\mathrm{a}\mathrm{t}\mathrm{i}{\mathrm{o}}_{\mathrm{i}\mathrm{t}}$ are the core explanatory variables: family size, elderly population ratio, and minor child ratio. The corresponding coefficients ${\mathsf{\beta }}_1,{\mathsf{\beta }}_2 \mathrm{a}\mathrm{n}\mathrm{d} {\mathsf{\beta }}_3$ are the parameters of primary interest in this study, measuring the average marginal impact of each unit change in these structural variables on the household saving rate. ${\mathrm{X}}_{\mathrm{i}\mathrm{t}}$ is a vector containing other time-varying control variables, such as the household per capita income, $\mathrm{L}\mathrm{n}\_\mathrm{I}\mathrm{n}\mathrm{c}\mathrm{o}\mathrm{m}\mathrm{e}\mathrm{P}\mathrm{e}\mathrm{r}\mathrm{C}\mathrm{a}\mathrm{p}\mathrm{i}\mathrm{t}\mathrm{a}$, and the logarithm of household total assets. $\mathrm{L}\mathrm{n}\_\mathrm{T}\mathrm{o}\mathrm{t}\mathrm{a}\mathrm{l}\mathrm{A}\mathrm{s}\mathrm{s}\mathrm{e}\mathrm{t}\mathrm{s}{\mathsf{\gamma }}^{\prime }$ is the coefficient vector corresponding to these control variables. ${\mathsf{\lambda }}_{\mathrm{t}}$ represents year fixed effects, capturing the impact of the macroeconomic environment, policy shocks, or other year-specific factors that all households commonly face over time, and ${\mathsf{\alpha }}_{\mathrm{i}}$ represents unobservable, time-invariant household individual fixed effects, which absorb all time-invariant household characteristics. ${\mathsf{ϵ}}_{\mathrm{i}\mathrm{t}}$ is the random disturbance term, assumed to satisfy standard assumptions after controlling for the aforementioned factors.

4.4. Model Advantages and Limitations

The fixed effects model addresses endogeneity from time-invariant omitted variables. However, it has key constraints: time-invariant variables (e.g., cultural norms) cannot be estimated as household fixed effects absorb their effects. Measurement errors in income variables’ household fixed effects absorb their effects. However, the fixed effects model does not fully resolve all endogeneity concerns. Significant residual endogeneity risks persist from (a) unobserved time-varying confounders, such as shifts in local economic conditions not captured by year fixed effects or dynamic changes in household head’s health or risk preferences that correlate with both life cycle savings and family structure changes, and (b) reverse causality, for instance, where past savings levels might influence subsequent decisions about family size or labor market participation. While our approach mitigates bias from time-invariant unobservables, these sources of dynamic endogeneity remain a challenge for causal interpretation and ideally would be addressed through instrumental variable techniques in future research, which are beyond the scope of the current study’s primary focus on panel data estimation with available CFPS variables. Furthermore, a notable limitation concerning covariates is the inability to incorporate time-varying characteristics of the household head (such as dynamic changes in their education, health status, or employment) due to data merging challenges. While fixed effects control for time-invariant unobserved heterogeneity, the omission of these potentially influential time-varying head-level factors could introduce omitted variable bias and affect the precision of our estimates. This constraint also limits our ability to explore heterogeneity along these important dimensions. To offset this data gap, we now (i) control for province-by-year shocks and (ii) verify the results on a balanced panel. Both checks point to the same qualitative conclusion, yet we acknowledge that future work equipped with complete head-level panels could probe this issue more definitively.

Because the public CFPS panel does not track the household head’s education and health status across all three waves, we carried out two additional robustness checks. First, we re-estimate every specification with province-fixed effects alongside the existing year dummies, thereby absorbing regional shocks such as province-level schooling reforms or epidemics. Second, we replicate the baseline regression on a balanced-panel subsample—56 percent of households that appear in every wave. In both exercises, the family size coefficient remains stable in sign and magnitude (

Table 4. Attrition robustness checks.

	(1) Logit: Attrition	(2) IPW FE	(3) Balanced FE
Family size	−0.189 *** (0.014)	0.047 *** (0.012)	0.045 *** (0.012)
Observations	16,109	24,550	19,906
R-squared (within)	—	0.262	0.249

Note: Column 1 reports coefficients from a logistic regression where the dependent variable equals one if the household attrits before 2022. Columns 2–3 report fixed effects estimates of Equation (1) with clustered (household-level) robust standard errors in parentheses. Column 2 is weighted by inverse-probability weights w = 1/($1-\widehat{p}$); column 3 restricts the sample to the balanced panel (7046 households observed in all three waves). Significance: *** p < 0.01. All regressions control for elderly_ratio, child_ratio, ln (income per capita), total assets, and year dummies.

), suggesting that unobserved head-level dynamics do not materially bias our findings.

Robustness Protocol: To mitigate serial correlation and heteroscedasticity, we cluster standard errors at the household level, allowing arbitrary within-household error correlation. Subsequent analyses—baseline regressions, heterogeneity tests (urban–rural/income groups), and robustness checks (alternative variable measurements)—adhere to this specification.

4.5. Attrition and External Validity

Across the three CFPS waves, 16,519 baseline households are observed at least once. Of these, 10,382 households (63%) remain in 2022, while 6137 (37%) lost. A logistic model of attrition (Table 2, column 1) indicates that larger families (β = −0.189, p < 0.01) and higher-income households (β = −0.135, p < 0.01) are less likely to drop out, whereas urban residence is not predictive. To gauge the impact on our estimates, we take two complementary steps. First, we construct inverse probability weights (IPWs) from the predicted drop-out probabilities and re-estimate the baseline model. Second, we replicate the analysis on the balanced panel of 7046 households observed in every wave. As shown in Table 4, the family size coefficient is 0.052 in the unweighted regression, 0.047 with IPW, and 0.045 on the balanced panel—all positive and significant at the 1% level. Because attrition is somewhat higher among larger and lower-income families, our findings are most directly applicable to the 63% of households that remain in the panel; we revisit this boundary in Section 4.4.

5. Empirical Results and Analysis

This chapter applies the theoretical framework, variable system, and econometric model developed in the preceding sections to the three-wave panel data (2018, 2020, 2022) from the China Family Panel Studies (CFPS) [1]. Its primary objective was to systematically examine how dynamic changes in three core family structure variables—household size, elderly dependency ratio, and child dependency ratio—affected household savings rates. To ensure robust and generalizable findings, the analysis proceeded in three stages: first, a baseline model estimation using fixed effects regression; second, a heterogeneity analysis across urban areas and income groups; and third, a series of comprehensive robustness checks.

5.1. Baseline Regression Results

5.1.1. Model Specification Test: Fixed Effects vs. Random Effects

Our analysis began with a critical choice between two standard panel data models: fixed effects (FE) and random effects (RE) models. The FE model is adept at controlling for unobserved, time-invariant household characteristics (e.g., a family’s inherent thriftiness or cultural background) that could influence family structure and saving behavior. Focusing on changes within each household over time provides a more rigorous estimate of the causal effect. The RE model, in contrast, assumes these unobserved characteristics are uncorrelated with the explanatory variables, a strong assumption that, if violated, can lead to biased results.

To formally decide between the two, we conducted a Hausman test. This test is a diagnostic tool to determine which model is more appropriate for our data. The test results were highly significant (p < 0.001), strongly rejecting the null hypothesis required for the RE model. This indicates that the RE model would produce unreliable estimates. Consequently, we adopted the FE model for our baseline analysis, as it provides more robust and credible results by accounting for stable, unobserved household differences.

5.1.2. Sequential Regression and Baseline Model Presentation

To demonstrate the robustness of our findings, we present our baseline results using a sequential regression approach in

Table 5. Baseline regression results of family structure on household saving rate.

	Model (1)	Model (2)	Model (3)
Explained Variable	Saving Rate 1	Saving Rate 1	Saving Rate 1
Methods of Estimation	FE	FE	FE
Family Size	0.0871 *** (0.0122)	0.0516 *** (0.0106)	0.0507 *** (0.0106)
Elderly Ratio	−0.1834 *** (0.0696)	0.0811 (0.0618)	0.0656 (0.0616)
Child Ratio	−0.3622 *** (0.1068)	−0.1289 (0.0897)	−0.1183 (0.0890)
Ln Per Capita Income		1.0374 *** (0.0245)	1.0224 *** (0.0250)
Ln Total Assets			−0.0467 *** (0.0107)
Family Size Squared
Year Fixed Effects	Yes	Yes	Yes
Constant	−0.4522 *** (0.0349)	−10.5713 *** (0.2425)	−9.8228 *** (0.2714)
Observations	32,822	32,822	31,694
Number of Groups	16,005	16,005	15,751
R-squared (within)	0.005	0.250	0.247

Note: standard errors in parentheses are clustered at the household level (fid). *** p < 0.01.

. This method involves building the model in steps:

Model (1) includes only our core family structure variables to show their raw association with the saving rate.

Model (2) adds a key economic control—household per capita income—to see if the initial effects persist after accounting for income levels.

Model (3) adds household total assets to our full baseline specification.

This step-by-step presentation aims to assess the stability of the coefficients for our main variables of interest. A significant and stable coefficient across these models is less likely to be a spurious finding, thus strengthening our confidence in the result. Model (3) is considered our primary baseline model for subsequent interpretation.

Considering the presence of covariance, covariance analysis was performed in this study, and it was tested that there is no covariance between the models, so model III was established as the main baseline model. (The Variance Inflation Factor is abbreviated as VIF. Generally, the higher the VIF, the higher the likelihood of covariance. In practice, variables with VIF > 10 are often defined as having multicollinearity.)

To quantitatively analyze the impact of family structure and economic factors on household savings behavior,

Table 6. Table of results of Multiple Linear Regression Analysis.

Model	B	Std. Error	Beta	1	Sig.	Tolerance	VIF
(Constant)	−6.190	0.073		−84.360	0.000
Elderly Ratio	0.128	0.020	0.034	6.554	0.000	0.926	1.080
Child Ratio	−0.175	0.057	−0.016	−3.077	0.002	0.847	1.181
Ln Total Assets	−0.104	0.005	−0.123	−21.498	0.000	0.742	1.348
Family Size	0.177	0.005	0.190	34.365	0.000	0.798	1.254
Ln Per Capita Income	0.689	0.008	0.541	91.067	0.000	0.687	1.456

Note: dependent variable: savings rate 1: (Total income − total expenditure)/Total income.

shows the results of multiple linear regression analysis. The table summarizes the relationship between core explanatory variables (such as the proportion of elderly population, proportion of children, and household size) and the dependent variable (savings rate 1, calculated as (total income − total expenditure)/total income), while controlling for economic indicators such as total assets and per capita income.

Table 5 shows that sequentially adding control variables significantly impacts the coefficient estimates of core explanatory variables. In Model (1), without controlling for economic factors, the coefficients of family size (family size), elderly population ratio (elderly ratio), and minor child ratio (child ratio) are all significant at the 1% level. However, after adding household per capita income (Model 2) and total assets (Model 3), the coefficients of the elderly population ratio (elderly ratio) and the minor child ratio (child ratio) become insignificant. In contrast, the coefficient of family size (family size), though somewhat reduced, remains significantly positive at the 1% level. This indicates that economic conditions influence household saving decisions and moderate the effects of family structure variables. Therefore, we will use Model (3) as the primary baseline for detailed interpretation.

5.1.3. Interpretation of the Family Size Effect

Interpreting the Positive Family Size Effect:

The baseline regression results (Table 5, Model 3) reveal a statistically significant and positive coefficient for family size (β ≈ 0.051, p < 0.01), indicating that, on average, an increase in household size is associated with a higher household saving rate during the 2018–2022 period. This finding aligns with our revised theoretical framework and hypothesis H1, suggesting that the combined forces of economies of scale in consumption and heightened precautionary saving motives likely dominate the resource dilution effects within this context.

While our fixed effects model effectively controls for time-invariant household characteristics, it does not permit a direct empirical disentanglement of the precise quantitative contributions of scale economies versus precautionary motives to the observed positive effect of family size on savings. Nevertheless, drawing upon established economic theory and considering the specific socioeconomic landscape of contemporary China, both mechanisms emerge as highly plausible, albeit not directly tested in isolation by our current model, drivers of this phenomenon.

It is conceivable that persistent economies of scale, particularly in crucial expenditure areas such as housing—where co-residence in larger family units can significantly reduce per capita costs amidst high market prices—and through the shared use of fixed household assets and services, could offer a tangible pathway to increased per capita saving capacity. Concurrently, it is also reasonable to infer that the heightened anticipated future financial pressures faced by larger families, stemming from significant investments in children’s education, potential healthcare needs for a greater number of members, and prospective eldercare responsibilities, alongside pervasive general economic uncertainties, might provide strong incentives for increased precautionary accumulation. Therefore, our empirical finding of a positive association between family size and savings is consistent with the hypothesis that these scale and precautionary factors collectively contribute to this outcome, although their relative explanatory power warrants further, more direct investigation using alternative methodologies.

The results of the nonlinear analyses suggest that the positive effect of household size on savings rates is most pronounced for medium-sized households (4–5 persons). In comparison, it becomes weaker or statistically insignificant for huge households (≥6 persons). This inverted-U (or saturation) pattern suggests a potential dynamic: for small- to medium-sized households, the benefits of economies of scale and possibly preventive savings spikes driven by a sense of responsibility may be maximized relative to resource needs. As households become very large, however, resource dilution effects—especially the high costs of supporting numerous children or other dependents—may begin to offset or even overwhelm the factors that promote saving, resulting in a net positive impact on the savings rate that flattens or declines [20].

Thus, household size’s observed positive average effect is primarily driven by expanding households to medium size, where the convergence of economies of scale and precautionary needs appears to be most conducive to higher savings. At the same time, resource constraints become the dominant factor in huge households.

5.2. Addressing Endogeneity: IV and Dynamic FE Estimation

Although the fixed effects (FE) model removes bias from time-invariant unobservables, it cannot eliminate bias from time-varying shocks or reverse causality. For example, a household’s ability to save may influence later fertility decisions. We, therefore, reinforce identification in two steps.

Instrumental variables. External-shock candidates (provincial two-child policy intensity and the urban–rural COVID-19 differential) pass the relevance threshold (first-stage F = 14.3) but fail the Sargan–Hansen over-identification test (p = 0.002). They are consequently discarded. We then use internal instruments: first lags of the family structure variables. Their first-stage F-statistic of 282.3 rules out weak-IV bias, yet the 2SLS estimate of family size (−0.010, SE 0.039) is insignificant, suggesting that the baseline FE coefficient is upward-biased.

Dynamic fixed effects. Because saving behavior is highly persistent, we estimate a dynamic FE model that includes the first leg of the saving rate (FE-LDV). This specification captures state dependence, although some Nickell bias may remain in a three-wave panel.

The contrasting results in

Table 7. Comparison of estimation results: FE, Lagged-IV, and dynamic FE.

Variable	(1) FE (Baseline)	(2) IV-Panel (L1 Lag)	(3) Dynamic FE (FE-LDV)
Saving_Rate1	Coef. (Std. Err.)	Coef. (Std. Err.)	Coef. (Std. Err.)
family_size	0.0530 *** (0.0107)	−0.0100 (0.0390)	0.0371 ** (0.0152)
elderly_ratio	0.0882 (0.0623)	0.4083 (0.5332)	0.0408 (0.0975)
child_ratio	−0.1320 (0.0901)	0.1822 (0.1778)	0.0454 (0.1394)
L. saving_rate1	—	—	−0.3611 *** (0.0218)
Control Variables	Included	Included	Included
Observations	32,822	17,291	17,291
Number of groups	16,005	11,148	11,148

Notes: Cluster-robust standard errors (household level) in parentheses. ***, ** denote significance at the 1%, 5% levels. All models include the full control set and year dummies. Column 3 controls for state dependence but may retain modest Nickell bias because the panel spans only three waves.

tell a nuanced story. The simple FE model (Model 1) has a positive endogeneity bias. While addressing this, the standard IV approach (Model 2) may be sensitive to the significant reduction in sample size. The dynamic FE model (Model 3) offers a compelling middle ground. It reveals that the positive effect of family size re-emerges once the persistence in saving behavior is accounted for (the lagged saving rate is highly significant). This suggests that ignoring the dynamic nature of savings is another critical source of bias. While the DFE model itself can have limitations (i.e., Nickell bias), its result provides our most robust evidence, showing a significant positive effect. After accounting for both dynamics and (to some extent) endogeneity, an increase in household size is indeed associated with a higher saving rate, lending credible support to our central hypothesis. Bias is attenuated here because T = 3, and the lagged coefficient is well below 1.

5.3. Heterogeneity Analysis

The baseline regression model reveals the average effect of family size, elderly, and minor child ratio on household saving rates. However, considering China’s vast regional differences, significant urban–rural dual structure, and continuously widening income gap, these average effects may mask heterogeneity in household behavior across different socioeconomic backgrounds. This section aims to test research hypothesis H4, exploring whether the impact of family structure transitions on household saving rates differs significantly based on households’ geographic location (urban–rural) and economic status (income level). We can understand how influence mechanisms manifest differently across various groups through heterogeneity analysis, thus providing a basis for developing more targeted policies.

5.3.1. Urban–Rural Differences

First, we examine whether the effects of family structure variables on saving rates exhibit urban–rural differences. We divide the total sample into urban and rural subsamples based on household location classification (urban). We apply the same panel fixed effects model as the baseline model (Table 5, Model 3) to each subsample. The results are shown in

Table 8. Results of urban–rural heterogeneity analysis.

	Urban (1)	Rural (2)
Methods of Estimation	FE	FE
Family Size	0.0554 *** (0.0128)	0.0385 ** (0.0166)
Elderly Ratio	0.1676 ** (0.0754)	0.0130 (0.0992)
Child Ratio	0.1230 (0.1087)	−0.3024 * (0.1470)
Ln Per Capita Income	0.9218 *** (0.0377)	1.1144 *** (0.0365)
Ln Total Assets	−0.0375 *** (0.0138)	−0.0728 *** (0.0191)
Year Fixed Effects	Yes	Yes
Constant	−9.1814 *** (0.4235)	−10.0262 *** (0.3829)
Observations	16,446	14,351
Number of Groups	8749	7420
R-squared (within)	0.232	0.256
FE_province	Yes	Yes
FE_year	Yes	Yes

Note: standard errors in parentheses are clustered at the household level (fid). * p < 0.1, ** p < 0.05, *** p < 0.01.

.

The results in 4 reveal significant urban–rural differences.

Urban–Rural Heterogeneity in Family Structure Effects on Savings.

Key Findings: Family size is positively associated with savings in both urban (β = 0.055, p < 0.01) and rural areas (β = 0.039, p < 0.05). The observed stronger positive association in urban areas may reflect more pronounced economies of scale in higher-cost urban living environments where shared housing and utilities can yield greater proportional savings. Additionally, urban households might possess a greater capacity to translate these scale efficiencies into actual savings due to, on average, higher and more stable incomes and better access to diverse financial instruments and saving vehicles, a notion consistent with theories on financial market development and income effects on saving. Higher elderly ratios correlate significantly with urban savings (β = 0.168, p < 0.05). This finding strongly aligns with precautionary saving theories, suggesting that urban households with more elderly members increase savings in anticipation of substantial future healthcare expenditures and potential long-term care costs, particularly given the ongoing development of China’s formal eldercare support system and a potentially weaker reliance on traditional multi-generational co-residence for care in urban settings. The absence of a significant correlation in rural areas (β = 0.013) could be attributed to the enduring strength of informal, community-based, and intergenerational family support systems for the elderly, as frequently documented in sociological studies of rural China. Such informal insurance mechanisms might mitigate the imperative of individual rural households for heightened precautionary accumulation, specifically for elder care, thus diverging from urban patterns. Conversely, higher child ratios are correlated with sharply lower rural savings (β = −0.302, p < 0.05). This pronounced negative impact in rural settings is consistent with theories emphasizing the substantial burden of human capital investment—particularly in education—under tighter resource constraints. For rural households, where average incomes are generally lower and access to public subsidies for child-rearing may be less comprehensive, the direct costs associated with raising children can significantly crowd out potential savings. The statistically insignificant effect observed in urban areas (β = 0.123) may be attributable to higher average urban incomes providing a more substantial buffer against these child-related expenditures, potentially different public service provisions, or varying intra-household resource allocation strategies that do not necessitate a direct trade-off with savings to the same extent as in rural areas.

A further reason for the sharper child-related dilution in rural areas is the more limited reach of public childcare and compulsory education subsidies, which forces rural parents to finance a larger share of education out-of-pocket—an observation consistent with the “private cost crowd-out” mechanism in human capital models of household saving. By contrast, urban households benefit from deeper credit markets and a denser social security net, allowing scale economy gains to dominate and cushioning the immediate cash-flow impact of additional children.

Policy Implications: To advance SDGs, prioritize rural education subsidies (SDG1,4) and urban long-term care insurance (SDG 3) to mitigate savings disparities. Central investments in rural childcare infrastructure (SDG 10) can narrow urban–rural welfare gaps, enhancing equitable development.

5.3.2. Income-Level Differences

Next, we explore whether the impact of family structure on saving rates varies with household income levels. We divided the sample into five groups based on household per capita income quintiles (Income Quintile), where Q1 represents the lowest income group, and Q5 is the highest income group. We conducted panel fixed effects regression for each group separately. The results are presented in

Table 9. Analysis of income-level heterogeneity.

	Q1 (Lowest)	Q2	Q3	Q4	Q5 (Highest)
Methods of Estimation	FE	FE	FE	FE	FE
Family Size	0.0087 (0.0418)	0.0208 (0.0258)	0.0721 *** (0.0237)	0.0573 ** (0.0242)	0.0256 (0.0200)
Elderly Ratio	0.0397 (0.2164)	−0.0467 (0.1375)	0.2562 * (0.1504)	−0.0205 (0.0919)	0.1865 (0.1183)
Child Ratio	−0.4437 (0.3895)	−0.0343 (0.1687)	−0.0586 (0.1593)	−0.1769 (0.1841)	0.1891 (0.1441)
Ln Per Capita Income	2.1968 *** (0.1014)	0.5747 *** (0.1350)	0.4719 *** (0.1361)	0.6403 *** (0.0978)	0.4329 *** (0.0331)
Ln Total Assets	−0.0391 (0.0368)	−0.0663 ** (0.0272)	−0.0208 (0.0248)	−0.0545 *** (0.0179)	−0.0190 (0.0143)
Year Fixed Effects	Yes	Yes	Yes	Yes	Yes
Constant	−19.2138 ** (0.9126)	−4.8078 *** (1.3038)	−4.6018 *** (1.3365)	−5.9915 *** (1.0158)	−4.4547 *** (0.4053)
Observations	6067	6372	6384	6487	6384
Number of Groups	4302	4878	4986	4907	4294
R-squared (within)	0.301	0.018	0.024	0.034	0.065
FE_province	Yes	Yes	Yes	Yes	Yes
FE_year	Yes	Yes	Yes	Yes	Yes

Note: standard errors in parentheses are clustered at the household level (fid). * p < 0.1, ** p < 0.05, *** p < 0.01.

.

Results of Income Heterogeneity Analysis

Family size effect: a significant positive association is concentrated in the middle-income group (Q3–Q4: β significant, p < 0.01).

This finding could be interpreted as middle-income households likely surpassing binding subsistence constraints that might dominate saving decisions for low-income groups. However, they still face considerable future financial uncertainties (e.g., for children’s education, eldercare, housing upgrades) that fuel precautionary motives. They may also be better positioned to realize economies of scale than low-income households. For low-income (Q1–Q2) households, the non-significant effect might indicate that any potential economies of scale are overwhelmed by immediate consumption needs and resource dilution from additional members, leaving little to no capacity for increased savings. Conversely, for high-income groups (Q5), wealth accumulation strategies may be more diversified and less dependent on marginal household consumption efficiencies [21]. Alternatively, they may have already achieved optimal consumption levels where family size changes have a negligible impact on their high saving rates or sophisticated investment portfolios.

Age structure effect: The elderly ratio shows no significant correlation with savings across all income groups, with only the Q3 group showing a weak positive correlation (p < 0.1). This general lack of a strong, uniform effect across income strata might suggest that life cycle considerations and precautionary motives related to the elderly are either counterbalanced by other income-specific saving drivers or manifest too heterogeneously within broad income quintiles to yield consistent aggregate results. The weak positive effect in Q3 could hint at a particular alertness or strategic saving response in this group concerning future eldercare costs, given their aspirational status and potential for intergenerational wealth transfer planning. The child ratio effect is also largely unstable across income quintiles, with only the Q4 (upper-middle income) group showing a weak adverse effect (p < 0.1). This instability in the child ratio effect across income groups likely reflects the complex and varying interplay of factors such as diverse household investment strategies in children’s human capital, differing abilities to absorb child-related costs without impacting savings, and varying reliance on or provision of intergenerational financial support, making a single theoretical prediction difficult without more granular data on these mediating factors.

The concentration of positive family size effects in Q3–Q4 aligns with the buffer stock saving framework: middle-income households have moved beyond subsistence constraints yet remain liquidity-constrained, so they actively convert scale economy savings into precautionary buffers. For the top quintile, diversified asset portfolios and easier access to capital markets reduce the marginal utility of such cost savings, explaining the attenuated coefficient in Q5.

Control variable: The positive effect of per capita income (Ln_SncomePerCapita) on the savings rate decreases with increasing income (low-income groups have stronger elasticity). The negative impact of total assets (Ln_TotalAssets) is significant in the Q2 and Q4 groups, reflecting the substitution effect of middle-income household asset allocation on savings.

Conclusion: Income heterogeneity supports hypothesis H4, but the impact pattern is more complex than urban–rural differences, with significant nonlinear features (Stata interaction model validation). The middle-income group is most sensitive to changes in family structure, highlighting their systemic vulnerability in China’s economic transformation. This heightened sensitivity underscores the potential challenges for achieving inclusive and sustainable growth.

5.3.3. Other Heterogeneity

It should be noted that because this study could not successfully merge household heads’ time-varying personal characteristic information during the data construction phase, it is impossible to test heterogeneity based on the head’s education level, age, or other personal characteristics as initially planned. These household head-level variables, especially the head’s age and education level, are theoretically important factors influencing household saving decisions and the ability to respond to family structure changes. This study’s inability to examine heterogeneity along these dimensions is a limitation.

5.3.4. Summary of Heterogeneity Analysis

While the preceding sections explored heterogeneity based on urban–rural location and income levels, it is important to reiterate a limitation stemming from data availability. This study could not examine heterogeneity based on time-varying household head characteristics, a point discussed further in our overall model limitations.

Heterogeneity Findings

Key Conclusions for H4:

Urban–rural divide: the elderly ratio increases savings in urban households (healthcare cost expectations), while the child ratio reduces savings in rural households (education expenditure pressures).

Income stratification: family size effects concentrate in middle-income groups, reflecting their systemic vulnerability during demographic transitions.

Synergy: These targeted measures enhance household financial resilience while advancing equitable socioeconomic development. Given our sample’s higher retention of relatively stable households, the findings should be extrapolated to highly mobile or rapidly changing families with due caution.

5.4. Robustness Checks

This section will conduct a series of robustness checks to test the reliability of the aforementioned baseline regression results and evaluate their sensitivity to model specifications, variable measurement methods, or sample selection. These checks can enhance confidence in the credibility of core findings. The checks will cover multiple dimensions, including alternative, dependent variables, alternative, independent variable measurement methods, comparison of different panel models, introducing variable lags, and adjustment of sample ranges.

5.4.1. Alternative Dependent Variables

Results using alternative saving definitions (SavingRate2 based on consumption, Ln_SavingAmount1, Asinh_SavingAmount1) confirm the robust positive association between family size and savings. The Elderly Ratio association remains insignificant. The child ratio shows a negative association that becomes significant under specific alternative measures, particularly those capturing saving levels (

Table 10. Regression results with alternative dependent variables.

	(1)	(2)	(3)	(4)
Methods of Estimation	FE	FE	FE	FE
Family Size	0.0507 *** (0.0106)	0.0471 *** (0.0096)	0.1315 *** (0.0139)	0.5345 *** (0.0877)
Elderly Ratio	0.0656 (0.0616)	−0.0027 (0.0551)	0.0279 (0.0743)	0.5459 (0.4639)
Child Ratio	−0.1183 (0.0890)	−0.1419 * (0.0792)	0.0026 (0.1388)	−1.6218 ** (0.7835)
Ln Per Capita Income	1.0224 *** (0.0250)	0.9250 *** (0.0226)	1.3559 *** (0.0268)	6.9418 *** (0.1240)
Ln Total Assets	−0.0467 *** (0.0107)	−0.0338 *** (0.0095)	−0.0102 (0.0130)	−0.4363 *** (0.0784)
Year Fixed Effects	Yes	Yes	Yes	Yes
Constant	−9.8228 *** (0.2714)	−8.8328 *** (0.2424)	−3.8753 *** (0.3040)	−62.3800 *** (1.4431)
Observations	31,694	29,337	18,480	31,694
Number of Groups	15,751	15,156	11,624	15,751
R-squared (within)	0.247	0.273	0.353	0.163
FE_province	Yes	Yes	Yes	Yes
FE_year	Yes	Yes	Yes	Yes

Note: standard errors in parentheses are clustered at the household level (fid). * p < 0.1, ** p < 0.05, *** p < 0.01.

).

5.4.2. Alternative Independent Variable Measurements

Using absolute counts (Elderly Count, Child Count) instead of ratios or employing categorical family-type dummies yields consistent results: the positive association of family size and the insignificant association of the elderly dimension persist. The negative association with children (Child Count or households with children) remains apparent, sometimes reaching significance (

Table 11. Regression results with alternative independent variable measurements.

	(1)	(2)	(3)
Methods of Estimation	FE	FE	FE
Family Size	0.0507 *** (0.0106)	0.0504 *** (0.0119)
Elderly Ratio	0.0656 (0.0616)
Child Ratio	−0.1183 (0.0890)
Elderly Count		0.0296 (0.0240)
Child Count		−0.0455 * (0.0264)
Family Type (ref = Type 1)
Type 2: Pure elderly households			−0.0747 (0.0460)
Type 3: Households with minor children			−0.1370 *** (0.0525)
Type 4: Working-age members only			−0.1284 *** (0.0493)
Ln Per Capita Income	1.0224 *** (0.0250)	1.0222 *** (0.0250)	1.0253 *** (0.0250)
Ln Total Assets	−0.0467 *** (0.0107)	−0.0471 *** (0.0107)	−0.0453 *** (0.0107)
Year Fixed Effects	Yes	Yes	Yes
Constant	−9.8228 *** (0.2714)	−9.8144 *** (0.2704)	−9.6190 *** (0.2717)
Observations	31,694	31,694	31,694
Number of Groups	15,751	15,751	15,751
R-squared (within)	0.247	0.247	0.246
FE_province	Yes	Yes	Yes
FE_year	Yes	Yes	Yes

Note: standard errors in parentheses are clustered at the household level (fid). * p < 0.1, *** p < 0.01.

).

5.4.3. Considering Lagged Effects

Including one-period lagged family structure variables shows no significant persistent association with current saving rates after controlling for current economic conditions, suggesting contemporaneous factors are primary drivers (

Table 12. Regression results with lagged variables.

	(1)	(2)
Methods of Estimation	FE	FE
Family Size	0.0507 *** (0.0106)
Elderly Ratio	0.0656 (0.0616)
Child Ratio	−0.1183 (0.0890)
L. Family Size		−0.0026 (0.0166)
L. Elderly Ratio		−0.0748 (0.1016)
L. Child Ratio		−0.0958 (0.1472)
Ln Per Capita Income	1.0224 *** (0.0250)	1.0590 *** (0.0420)
Ln Total Assets	−0.0467 *** (0.0107)	−0.0436 ** (0.0194)
Year Fixed Effects	Yes	Yes
Constant	−9.8228 *** (0.2714)	−10.0598 *** (0.4567)
Observations	31,694	16,182
Number of Groups	15,751	10,598
R-squared (within)	0.247	0.257
FE_province	Yes	Yes
FE_year	Yes	Yes

Note: standard errors in parentheses are clustered at the household level (fid). ** p < 0.05, *** p < 0.01.

).

5.4.4. Sample Limitation Analysis

Re-estimating the model after excluding potential outliers (based on standard deviations or absolute saving rate bounds) confirms the robustness of the positive family size association and the insignificant elderly ratio association. The negative child ratio association again shows sensitivity, becoming significant in some restricted samples (

Table 13. Results of sample limitation analysis.

	(1)	(2)	(3)
Methods of Estimation	FE	FE	FE
Family Size	0.0507 *** (0.0106)	0.0437 *** (0.0070)	0.0264 *** (0.0054)
Elderly Ratio	0.0656 (0.0616)	0.0430 (0.0379)	0.0457 (0.0294)
Child Ratio	−0.1183 (0.0890)	−0.1215 ** (0.0617)	−0.0588 (0.0451)
Ln Per Capita Income	1.0224 *** (0.0250)	0.6142 *** (0.0128)	0.4584 *** (0.0105)
Ln Total Assets	−0.0467 *** (0.0107)	−0.0409 *** (0.0067)	−0.0299 *** (0.0050)
Year Fixed Effects	Yes	Yes	Yes
Constant	−9.8228 *** (0.2714)	−5.7392 *** (0.1423)	−4.2325 *** (0.1149)
Observations	31,694	30,992	28,618
Number of Groups	15,751	15,587	15,057
R-squared (within)	0.247	0.201	0.171
FE_province	Yes	Yes	Yes
FE_year	Yes	Yes	Yes

Note: standard errors in parentheses are clustered at the household level (fid). ** p < 0.05, *** p < 0.01.

).

5.4.5. Summary of Robustness Checks

Synthesizing across all checks, the positive association between family size and household saving rates, and the general lack of a significant association for the elderly ratio demonstrates high robustness. While directionally consistent, the negative association linked to the child ratio appears less robust and is sensitive to specific model and measurement choices [22]. However, it frequently emerges as statistically significant, suggesting a genuine underlying pressure exists. The fixed effects model’s appropriateness and the contemporaneous variables’ primary relevance are confirmed.

5.4.6. Proxy for Household Head’s Age

Incorporating a proxy for household head’s age dynamics (birth year interacted with survey year) does not alter the main conclusions regarding the positive family size association or the insignificant elderly ratio association, suggesting age-related confounding is unlikely to be the primary driver of these findings.

5.4.7. Nonlinear Effects of Family Size

Exploring nonlinearities reveals an inverted U-shaped pattern: the positive association of family size with savings is strongest for medium-sized households (4–5 members) and diminishes for huge families, suggesting competing mechanisms are at play.

5.4.8. Rigorous Tests for Urban–Rural Heterogeneity

Propensity Score Matching (PSM) on balanced samples confirms that rural households remain significantly more vulnerable to negative savings associated with higher child ratios than urban households. A placebo test assigning random urban status further validates that the observed heterogeneity is not driven by chance.

Because panel attrition is somewhat higher among larger and lower-income families, the results documented in this section most directly apply to the 63 percent of households remaining in the sample throughout all three waves.

5.5. Mechanism Analysis

To clarify how family size influences household saving, we conduct a three-path mediation analysis that traces (i) scale economy gains from shared fixed costs, (ii) resource dilution pressure arising from additional children, and (iii) a precautionary saving motive linked to income volatility.

Table 14. Mediation (mechanism) analysis.

Pathway/Effect	Direct Effect (${\mathit{\beta }}_{\mathit{m}}$)	Indirect Effect (${\mathit{\beta }}_{\mathit{m}}$)	Share of Total (%)
Scale economy (fixed-cost per capita)	0.0456 * (0.0120)	0.0162 (0.0061)	26.3%
Resource dilution (consumption pressure)	0.0456 * (0.0120)	−0.0120 (0.0054)	−19.4%
Precautionary motive (income volatility)	0.0456 * (0.0120)	0.0004 (0.0018)	0.6%

Notes: Indirect effects equal a × b in the standard mediation formula; “Share of total” = ${\beta }_m$/(${\beta }_d$ + ${\beta }_m$). Robust standard errors (clustered at the household level) appear in parentheses. Significance: * p < 0.10. All equations include the same controls and year fixed effects as the baseline specification.

reports the direct effect of family size on the saving rate and the indirect contribution of each pathway.

The figures show that roughly 26 percent of the total family size effect on saving operates through scale economies: larger households spread quasi-fixed expenses (housing, utilities) over more members, lowering per capita cost and boosting saving. A smaller but statistically significant share—about 19 percent in the opposite direction—flows through resource dilution, as additional children raise consumption pressure and dampen saving. The precautionary saving channel contributes less than one percent and is statistically indistinguishable from zero, indicating that risk smoothing is secondary in this context.

The positive net impact of family size on household saving is driven primarily by cost-sharing economies that more than offset the child-related dilution effect.

5.6. Summary of Findings and Answers to Research Questions

This section synthesizes the empirical results to the research questions in the indirect production directly.

What is the net effect of family structure on saving?

Our analysis reveals that family size has a robust, positive net effect on household saving rates in China after controlling for endogeneity and household-specific effects. Conversely, the elderly ratio has no significant aggregate effect, while the child ratio exerts a negative pressure on savings, an effect that is particularly pronounced in specific subgroups.

What are the primary mechanisms?

A formal mediation analysis demonstrated that the positive family size effect is primarily driven by economies of scale, which contribute a positive 26.3% to the total effect. This force is strong enough to offset children’s negative resource dilution pressure (−19.4%). Precautionary motives linked to income volatility play a negligible role in this dynamic.

How do these effects differ across groups?

The effects are highly heterogeneous. The financial burden of children falls almost exclusively on rural households. In contrast, precautionary saving for eldercare is a distinctly urban phenomenon. Furthermore, the positive impact of family size is most concentrated among middle-income households, who appear most sensitive to changes in their financial capacity.

6. Conclusions and Perspectives of the Study

6.1. Theoretical Significance

This study achieves a triple breakthrough in household savings behavior theory by integrating dynamic panel analysis and mechanism decomposition, significantly advancing the frontiers of resource dilution theory, the life cycle hypothesis, and precautionary savings research. The research findings and theoretical contributions are as follows.

Reconstructing the theoretical framework of the household size effect: quantifying the dynamic equilibrium between economies of scale and resource dilution. Traditional household production models often emphasize the dominant role of the resource dilution effect, arguing that an increase in household size suppresses savings by increasing consumption demand. However, this study, based on dynamic panel data from the China Family Panel Studies (CFPS) from 2018 to 2022 [1], quantifies the relative strengths of the two mechanisms for the first time in a middle-income context: the economic scale effect contributes 26.3% of the positive impact of household size on savings, while resource dilution offsets only 19.4% of it. This finding reveals a unique pathway for multigenerational cohabiting households in Chinese cities, where high housing costs and strong public service sharing prevail, to enhance savings through per capita sharing of quasi-fixed costs such as housing and utilities (as shown in Figure 1).

Unlike the negative correlation conclusions in developing countries such as Turkey [2], this study confirms that the economic scale effect reaches its peak when household size expands to 4–5 people, while resource dilution becomes dominant only after exceeding six people. This nonlinear relationship revises Becker’s “quantity-quality trade-off” theory, indicating that in China, where education expenditure is highly marketized, medium-sized families can offset child-rearing costs through economies of scale, while extremely large families face savings difficulties due to surging variable expenditures on education and healthcare [20].

Challenging the universality of the life cycle hypothesis: the reversal effect of precautionary motives on elderly savings.

The classic Life cycle Savings Hypothesis (LCSH) predicts that an increase in the proportion of the elderly population will lead to a decline in the savings rate. However, this study found that in urban Chinese households, a 1% increase in the proportion of the elderly population resulted in a 0.168% increase in the savings rate (p < 0.05), while rural households showed no significant effect. This contradiction can be explained by the precautionary savings theory: in a context where social medical insurance coverage is only 68% (2022 data), urban households tend to accumulate reserves for chronic disease treatment (such as hypertension and diabetes, with annual out-of-pocket expenses averaging CNY 8760) and long-term care (private nursing homes with monthly fees averaging CNY 4500) for elderly members [3].

In contrast to the phenomenon of slow dissaving among the elderly in European studies [8], the “excessive savings” behavior of urban households in China reflects the dual pressures of the collapse of the informal care system and the lack of formal social security systems. Notably, rural households, which still rely on intergenerational support from their children (72% of rural elderly receive financial assistance from their children), do not exhibit significant savings behavior influenced by the elderly. This urban–rural divide provides a theoretical basis for improving rural pension security [11].

Expanding the institutional context of preventive savings theory: mechanism differentiation from a heterogeneity perspective.

This study systematically verifies the moderating role of the institutional environment on household savings motives for the first time.

Urban–rural dimension: In rural households, a 1% increase in the proportion of children reduces the savings rate by 0.302% (p < 0.05), while urban households show no significant effect. This stems from the “private crowding-out” effect of education expenses in rural areas—rural households bear 73% of children’s education costs (41% in urban areas), forcing parents to reduce savings [7];

Income Dimension: The positive effect of household size on savings is most pronounced among middle-income groups (Q3–Q4) (β = 0.072, p < 0.01), while high-income households, due to diversified asset allocation (financial assets account for 38% of total assets), see a diminished marginal contribution from economies of scale. This finding supports the “buffer stock savings” theory, which posits that middle-income households, having escaped subsistence constraints yet facing uncertainty over large expenditures such as housing and education, are more sensitive to changes in household structure.

6.2. Practical and Marketing Implications

The findings offer actionable insights for financial service providers and marketers aiming to serve Chinese consumers better.

Product Segmentation by Family Structure: The significant behavioral differences based on family composition call for targeted product design. Financial institutions should develop and promote affordable education savings plans and family health insurance for rural households with children facing the greatest saving pressure. For urban households with elderly members, products addressing healthcare and long-term care costs, such as critical illness insurance top-ups and annuities, should be prioritized.

Focus on Middle-Income Households: as the segment most sensitive to family size changes, middle-income households represent a key market for flexible financial products that adapt to evolving family needs, such as modular insurance plans or investment portfolios with adjustable risk levels.

Leveraging Digital Channels for Tailored Services: Financial service providers should utilize digital platforms to deliver personalized financial advice based on household structure and income. AI-driven robo-advisors, for instance, can offer customized budgeting tools and saving plans for families at different life stages, thereby fostering greater financial resilience.

6.3. Policy Implications for Sustainable Development

Our findings carry significant policy implications, particularly for fostering equitable development in line with the UN’s Sustainable Development Goals (SDGs).

Targeted Support for Rural Families (SDG 1, 4, 10): The severe impact of child-rearing costs on rural savings points to a critical area for intervention. Policymakers should prioritize expanding subsidies for rural childcare and education and investing in rural public infrastructure. Such measures can alleviate direct financial burdens, reduce inequality, and enhance human capital development.

Enhancing Eldercare Security in Urban Areas (SDG 3, 10): The link between higher elderly ratios and increased urban savings signals significant anxiety over future healthcare costs. Expanding affordable long-term care insurance and accessible public eldercare services is crucial. This can reduce excessive precautionary saving, improve the well-being of the elderly, and unlock household consumption potential.

Supporting Middle-Income Household Resilience (SDG 10): Given their heightened sensitivity to demographic shifts, policies should focus on enhancing middle-income households’ financial literacy and capability. This includes improving their access to diverse and sustainable saving and investment options, potentially leveraged through digital financial tools, to mitigate financial pressures from housing and education.

7. Limitations and Future Research Avenues

7.1. Study Limitations

Data Constraints on Household Head Characteristics: The analysis lacks time-varying data on household heads (e.g., education, health status, employment), which could moderate the impact of family structure on savings. For instance, a head’s education level may influence financial literacy and saving strategies, while health shocks could alter precautionary saving motives. The absence of such variables may introduce omitted variable bias, particularly in heterogeneous subgroups (e.g., rural households or low-income families).

Dynamic Endogeneity and Causal Inference: Although fixed effects models control for time-invariant unobservables, residual endogeneity persists from time-varying confounders (e.g., local economic shocks, changes in risk preferences) and reverse causality (e.g., savings influencing fertility decisions). The instrumental variable approach using lagged family structure variables revealed potential upward bias in baseline estimates, indicating that causal interpretations require caution.

Scope of Mechanism Analysis: The mediation analysis quantified scale economies, resource dilution, and precautionary saving but overlooked other pathways, such as intergenerational financial transfers or social capital effects. For example, extended family co-residence may facilitate informal risk-sharing, which could mediate the relationship between family size and savings.

Generalizability to Rapidly Changing Populations: The sample retention rate (63%) was higher for stable households, potentially underrepresenting transient families (e.g., urban migrants or young couples). Additionally, the study period (2018–2022) includes the COVID-19 pandemic, which may have amplified precautionary saving motives atypically, limiting generalizability to non-crisis periods.

7.2. Future Research Directions

Micro-level Heterogeneity with Panel Data: Longitudinal data tracking household heads’ dynamic characteristics (e.g., education, occupation, health) would enable a more nuanced analysis of heterogeneity. For instance, examining whether highly educated heads mitigate resource dilution through higher income or financial planning.

Causal Identification with Policy Shocks: Exploiting exogenous policy changes (e.g., the three-child policy or social security reforms) as natural experiments could strengthen causal inference. Such designs could isolate the impact of family structure from confounding trends, particularly in urban–rural comparisons.

Expanded Mechanism Analysis: Incorporating measures of social capital (e.g., community networks), digital financial tool usage, or intergenerational transfers would enrich the understanding of mediating pathways. For example, testing whether mobile banking apps enhance scale economies in large families by improving budget tracking.

Long-term Effects and Policy Interventions: Longitudinal studies following households through demographic transitions (e.g., children entering higher education or elderly parents requiring care) could assess the durability of family structure effects. This would inform policies like targeted childcare subsidies or eldercare insurance, which may alter saving behavior dynamically.

Furthermore, future research should prioritize integrating household saving dynamics with UN Sustainable Development Goals (SDGs), such as the following:

SDG 1 (No Poverty): analyzing how family structure affects savings among low-income households to design inclusive social safety nets.

SDG 3 (Good Health and Well-being): exploring the link between elderly ratios, healthcare costs, and savings to optimize public health financing.

SDG 10 (Reduced Inequalities): investigating urban–rural disparities in saving responses to family structure changes to inform equitable policy design.

By addressing these limitations and avenues, subsequent studies can deepen the understanding of how family structure shapes household economic behavior in China’s rapidly evolving socioeconomic landscape.