The relationship between children’s digital literacy and parents’ mental health: evidence from China

Abstract

Background

Understanding how to improve the mental health of middle-aged and elderly people is an important issue that needs to be addressed urgently to promote healthy ageing. Moreover, children’s digital literacy has become critical for families to adapt to the digital age. However, few studies have investigated the relationship between children’s digital literacy and parents’ mental health.

Methods

Based on data from the China Family Panel Studies (CFPS) dataset, this study investigates the relationship between children’s digital literacy and parents’ mental health and its mediating mechanism. Empirical analyses are conducted using the double/debiased machine learning (DML) model and causal mediation analysis (CMA) model.

Results

(1) Increased digital literacy among children improves parents’ mental health, and the effect is more pronounced for mothers, while there is no significant effect for fathers. These results have been verified by endogeneity and robustness tests, further confirming the reliability of the above findings; (2) these effects are heterogeneous at the regional, household, and individual levels and are more pronounced in the eastern and central regions, in urban areas, in areas with high household socioeconomic status, and among parents with poorer health outcomes; and (3) household income and emotional support are key mediating pathways through which children’s digital literacy improves parents’ mental health.

Conclusion

This study highlights the importance of intergenerational digital spillover in addressing the mental health of middle-aged and older adults in China’s rapidly digitizing society. It provides novel insights for strengthening intergenerational digital support to promote healthy ageing.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-025-25128-3.

Introduction

The global population is ageing at an accelerated rate. According to the World Health Organization, the world’ s population aged 60 years and older will reach 1.4 billion by 2030, accounting for one-sixth of the global population. Notably, approximately 14% of those aged 60 years and older will experience mental health problems [1]. China’s middle-aged and elderly population is also gradually increasing. By the end of 2023, the proportion of people aged 50 and above in China’s total population had reached 37.8% [2], while nearly 26% of older adults had varying degrees of mental health problems [3]. Therefore, understanding how to improve the mental health of middle-aged and elderly people is an important issue that needs to be addressed urgently to promote healthy ageing.

With the development of the digital economy, the use of digital technology can clearly improve the mental health problems of middle-aged and elderly people [4], which means that the use of digital technology is a new way to alleviate the mental health problems of these age groups. However, older adults often face barriers, such as age and educational level, which make access to digital health benefits more challenging [5]. With the deepening of digitalization, young people have become the most important users of digital technology. For instance, the number of internet users aged 40 and under in China will increase to 48.8% by the end of 2024 [6]. Moreover, children are increasingly acting as digital intermediaries for their parents. In this shift, children’s digital literacy, which is the ability to use digital technology to find information, communicate and solve problems, has become critical for families to adapt to the digital age. Therefore, exploring the impact of children’s digital literacy on parents’ mental health is crucial for formulating policies aimed to enhancing intergenerational digital support to promote healthy ageing.

Previous studies on digital literacy and mental health have been focused primarily on the impact of an individual’s own level of digital literacy on mental health. For example, Yang et al. reported that digital literacy has a positive effect on an individual’s mental health and that the mechanisms of impact include increased income, improved quality of employment, and increased informal social support [7]. Li et al. also found that digital literacy notably enhances the mental health of respondents, particularly among those aged over 60, those with a high school education or above, and those with moderate to high incomes across all regions of China [8]. However, research on the intergenerational impact of digital literacy on mental health is lacking. In addition, the literature is dominated by traditional linear regression methods in terms of research methodology. The traditional linear regression model relies on linear assumptions when causal inferences are made, ignoring the nonlinear relationship between the variables [9]. For instance, the results from the literature show a nonlinear relationship between social media use and mental health [10]. Moreover, mental health is affected by many factors, and it is necessary to control for as many other factors as possible. When high-dimensional control variables are introduced, the traditional linear regression model faces the “curse of dimensionality”, which leads to questions about the accuracy of the causal estimation results [11]. In addition, the existing studies have been focused primarily on traditional mediation analysis methods for analyzing the mediation effect. This approach leads to biased estimates of the mediation effect, as it fails to account for the influence of confounding factors on the mediating and explanatory variables [12, 13].

This study uses double/debiased machine learning (DML) and causal mediation analysis (CMA) to investigate the relationship between children’s digital literacy and parents’ mental health and its mediating mechanisms. The core objectives include (1) using DML to identify the relationship between children’s digital literacy and parents’ mental health; (2) analyzing the heterogeneity of the aforementioned impact; and (3) using CMA to reveal the mechanism underlying the mediating effect of children’s digital literacy on parents’ mental health. The main contributions of this study are as follows: First, it provides empirical evidence for the intergenerational spillover effects of digital literacy on mental health, helping to address the research gap on how digitalization affects health outcomes within families. Second, methodologically, it enriches the relevant literature by applying DML to overcome limitations of traditional linear models in addressing nonlinear relationships and high-dimensional control variables. Meanwhile, based on the causal inference framework, CMA is applied to test the mediating mechanism, which takes into account the influence of confounding factors and relaxes the modelling assumptions. Third, the findings of this study can provide references for policy-making in China and other comparable countries around the world, facilitating the resolution of mental health issues among middle-aged and elderly populations through intergenerational digital support.

The rest structure of this paper is as follows. “Theoretical analysis” section builds the theoretical foundation of the study. “Data and methodology” section introduces the methods in detail. “Results” section presents the research results. “Discussion” section discusses the results. The conclusion and implications are presented in the last section.

Theoretical analysis

Relationships between children’s digital literacy and parents’ mental health

The theory of intergenerational solidarity emphasizes the significance of parent-child interactions for families [14]. A large body of literature has confirmed that intergenerational solidarity substantially affects parents’ mental health [15–17]. In recent years, with the advancement of digital technologies, digital literacy has played a crucial role in shaping intergenerational solidarity. Digital communication methods such as text messages, social media, and video calls have become vital for children to maintain contact with their parents. These approaches help overcome geographical distances, increase the frequency of interactions among family members [18, 19], and thus strengthen the intergenerational solidarity between parents and children. More importantly, as children’s digital literacy improves, digital solidarity has emerged as a new dimension of intergenerational solidarity, exerting a positive influence on parents’ mental health [20, 21]. In addition, socioemotional selectivity theory posits that individuals’ social goals change as their perception of time changes [22]. With increasing age, parents tend to prioritize intergenerational emotional interactions as their primary social goal, and mothers have greater emotional needs than fathers do [23]. This characteristic results in a stronger impact of children’s digital literacy on mothers’ mental health than on fathers’ mental health.

Mediating mechanisms between children’s digital literacy and parents’ mental health

Household income

Household income is a key factor affecting parents’ mental health. In general, higher household income provides parents with more sufficient economic support and more comprehensive living security, which to a large extent contributes to improving their mental health [24–26]. Children’s digital literacy plays an important role in increasing household income, which is reflected in the following aspects. On the one hand, from the perspective of human capital theory, an individual’s skills and knowledge can be converted into economic returns [27]. As a new type of human capital, digital literacy not only helps enhance children’s information-searching abilities and expand their employment opportunities but also improves their work efficiency, ultimately enabling them to achieve higher incomes [28, 29]. On the other hand, improving children’s digital literacy can not only increase their income but can also help family members obtain employment and increase their income by conveying employment information and policy subsidies to them through digital tools [20, 30]. Additionally, enhanced digital literacy among children reduces their intergenerational income dependence on their parents, promoting the upward mobility of household income and thereby further increasing overall household income [31, 32].

Household healthcare expenditure

Health capital theory suggests that an individual’s health status depends on health investments, with household healthcare expenditure serving as an important form of such investment [33]. It constitutes a significant pathway influencing parents’ mental health [34, 35]. Children’s digital literacy can influence household healthcare expenditure by reducing the cost of accessing health information and enhancing family members’ health awareness [36, 37]. Notably, the greater the children’s digital literacy is, the greater the likelihood that they will purchase fitness and sports-related equipment as well as health products through online platforms, thereby increasing household healthcare expenditures [38].

Children’s intergenerational support

Social support theory emphasizes that social resources such as emotional and economic support are effective ways to alleviate psychological stress [39]. In terms of emotional support, children with high digital literacy are more likely to maintain frequent interactions with their parents through social media, video calls, and other means [40], reducing parents’ sense of loneliness and neglect. This emotional support is particularly important for parents and can significantly reduce the risk of depression [41]. In terms of economic support, improved digital literacy among children enhances their economic independence, thereby increasing economic support for their parents. Economic support not only relieves parents’ economic pressure but also conveys children’s care and can enhance parents’ mental health [42, 43].

Data and methodology

Data sources

The data for this study are derived from the China Family Panel Studies (CFPS), which is launched by the Institute of Social Science Research at Peking University in 2010. The CFPS is a nationally representative survey using multistage, implicitly stratified, and proportional population-based sampling methodology conducted every two years. The dataset contains detailed information on Chinese residents, including health, income, healthcare consumption, health behaviors, and other individual characteristics [44]. This study primarily uses data from this dataset from 2016, 2018, 2020, and 2022. The study subjects are selected as parents with at least one child, who are no older than 80 years old and required to be included in all four survey waves (2016, 2018, 2020, and 2022). Ultimately, 5732 parent-child pairs are included in this study.

Variables

Dependent variable: parents’ mental health

In this study, the Center for Epidemiologic Studies Depression Scale (CES-D) test score is used to measure parents’ mental health, which has been used in previously published literature [45, 46]. The CFPS provides two versions of the CES-D scores: the 20-item version (CES-D20) and the 8-item abbreviated version (CES-D8). Both scales are rooted in the original CES-D framework to evaluate an individual’s depression level, with the CES-D8 being a streamlined adaptation of the CES-D20, reducing the number of items from 20 to 8 in order to increase response rates. The items in the CES-D20 and CES-D8 include statements such as “I feel depressed” and “I find it hard to do anything”, with each question rated on a 4-point scale (from 1 to 4). To ensure comparability between the CES-D20 and CES-D8 scores, a reciprocal conversion was applied using the equipercentile equating method, which yields analytically comparable results. In terms of scoring ranges, the CES-D20sc spans from 20 to 80, while the CES-D8 ranges from 8 to 32. Consistently, higher scores on either version of the CES-D indicate more severe depression and poorer mental health among parents. Notably, the CFPS dataset includes CES-D20 data for 2016, 2018, 2020, and 2022, whereas CES-D8 data are available for only 2018, 2020, and 2022. To retain a larger sample size, this study uses CES-D20 scores for baseline regression analyses and CES-D8 scores for robustness checks1.

Independent variable: children’ s digital literacy

Digital literacy encompasses an individual’s ability to use digital tools skillfully to search, access, manage, integrate, evaluate, and analyse digital resources within a specific life context [47, 48]. Drawing on studies by Ji et al. and Yang et al., this study will measure individual digital literacy in terms of digital device use, digital applications, and digital dependency based on the entropy method [7, 49]. The entropy method, which has the advantages of objectively determining weights and making full use of data, has been used to measure digital literacy [7] (see Supplementary 1 for details). The specific measurement indices are shown in Table 1.

Table 1.

Digital literacy indicator system

Primary index	Secondary index	Index description
Digital equipment use	Whether using a mobile device to access the internet	Yes = 1, No = 0
	Whether using a computer to access the internet	Yes = 1, No = 0
Digital applications	Whether using the internet almost every day for learning	Yes = 1, No = 0
	Whether using the internet for work	Yes = 1, No = 0
	Frequency of using the internet for socializing	Number of uses per week
	Whether using the internet almost every day for entertainment	Yes = 1, No = 0
	Whether shopping online almost every day (e.g. Taobao, Jingdong)	Yes = 1, No = 0
Digital dependency	The importance of internet use for work	Importance rating 1–5
	The importance of internet use for entertainment	Importance rating 1–5
	The importance of internet use for socializing	Importance rating 1–5
	The importance of internet use for learning	Importance rating 1–5
	The importance of internet use to daily life	Importance rating 1–5

Control variables

Based on previous research [7, 8, 25, 45, 46], the control variables in this study include individual-level, household-level, and district-level control variables. Individual-level controls include age, gender, marital status, and education; household-level controls include family size, sibling, household net assets, average annual household commercial insurance consumption, whether anyone in the household receives a pension or old-age pension, and whether the household receives government benefits; and regional-level controls include the economic status of the province in which the household is located, the endowment of healthcare resources in the province in which the household is located (number of healthcare institutions, number of healthcare technicians, and number of beds), urban-rural status, and geographic location. Table 2 presents the descriptive statistics for all the variables.

Table 2.

Descriptive statistics of the variables

Variable		Definition of variable	Mean	Std. dev.
Dependent variable	Parents’ mental health	/	32.6374	7.8917
Independent variable	Children’ s digital literacy (CDL)	/	0.3706	0.1842
Control variables	Age	Age of individuals	54.1543	8.2490
	Gender	Male = 1, Female = 0	0.4721	0.4993
	Marriage	Married or cohabiting = 1, Single = 0	0.9405	0.2365
	Education	Primary and below = 0, Junior = 1, Senior = 2, University and above = 3	0.2503	0.5919
	Family size	Number of household members	4.7206	1.5512
	Sibling	Number of siblings within the household	0.1751	0.4236
	Household net assets	Logarithm of household net assets	12.9729	1.2663
	Expenditure on commercial insurance	Logarithm of average annual household expenditure on commercial insurance	3.7845	4.1992
	Pension	Yes = 1, No = 0	0.4112	0.4921
	Government subsidy	Yes = 1, No = 0	0.4670	0.4990
	GDP	Logarithm of gross domestic product	10.9403	0.3997
	Health institution	Number of health institutions per 10,000 people	8.1409	2.6478
	Health technician	Number of health technicians per 1,000 people	7.1104	1.1256
	Bed	Number of beds per 10,000 people	64.0116	9.6450
	Urban	Urban = 1, Rural = 0	0.5340	0.4989
	Area	Eastern = 1, central = 2, Western = 3	1.8697	0.8330

Methodology

Double/Debiased machine learning (DML) model

In this study, a double/debiased machine learning (DML) model is used to test the impact of children’s digital literacy on parents’ mental health. Compared with the traditional linear model, DML has the following advantages [50]: First, the regularization algorithm of DML can automatically screen high-dimensional control variables to identify those with greater predictive accuracy. This not only enables control over high-dimensional variables but also effectively avoids the impact of the “curse of dimensionality” and multiple covariance on the estimation accuracy. Second, the relationships between mental health and its related influencing factors are not strictly linear. The estimations based on a linear regression model may have a modelling bias, whereas DML can effectively address nonlinear data to prevent modelling bias and ensure the accuracy of causality estimations. In addition, DML not only ensures the unbiasedness of the processing effect coefficients but also eliminates problems such as regularization bias and overfitting in machine learning through cross-fitting and orthogonalization. The steps of DML model construction are as follows:

First, partial linear models are constructed, as shown in Eq. (1):

1

2

In Eqs. (1) and (2), is parents’ mental health and is the level of their children’ s digital literacy. is the effect of children’ s digital literacy on parents’ mental health, which is estimated using a machine learning method, and is the set of control variables. and are nonlinear functions, and affects via function and via function . denotes the year fixed effect, denotes the household-level fixed effect, and and are random error terms.

Second, based on the above partially linear models, the DML model is implemented primarily through the following three steps:

(1) Nuisance function estimation. Machine learning algorithms are used to estimate the nonlinear effects of the control variable on and to obtain and , respectively. and represent the conditional expectation estimates of the control variable on and , respectively, which are expressed as follows:

3

4

In this study, the gradient boosting algorithm is adopted to estimate and . The key hyperparameters such as learning_rate, max_depth, n_estimators, and subsample are optimized through random search. The search space range is learning_rate: [0.01, 0.05, 0.1, 0.2], max_depth: [3–8], n_estimators: [100, 200, 300, 400, 500], and subsample: [0.6, 0.8, 1.0]. The main optimization goal is to minimize the mean square error (MSE), and R² is used as an auxiliary indicator. Based on the above process, the optimal parameter combination for the estimation model is ultimately determined as learning_rate = 0.1, max_depth = 5, n_estimators = 300, and subsample = 1.0; the optimal parameter combination for the estimation model is learning_rate = 0.1, max_depth = 4, n_estimators = 200, and subsample = 1.0.

To prevent overfitting, this study implements K-fold cross-validation (K = 5) with sample splitting for estimation. Specifically, the sample is randomly divided into 5 parts, one of which is selected as the test set, and the remaining 4 parts are used as the training set to train and . The above process is repeated 5 times, with each subset serving as the test set once. The reasons for selecting 5-fold cross-validation are as follows: first, the literature indicates that 5-fold cross-fitting has better model performance in simulations and empirical studies [50]; second, this study compares the MSE values of model estimation under different K-fold cross-fitting scenarios (K = 2, 3…, 10), with the bias value serving as an auxiliary judgement indicator. As shown in Fig S1 (see Supplementary 2 for details), the comparison results indicate that when K = 5, the MSE value is the smallest and has a smaller bias value. Therefore, K = 5 is selected as the optimal number of folds in this study.

(2) Residual calculation. The outcome residual and the treatment residual are calculated through residual orthogonalization. The specific calculation formulas are as follows:

5

6

Among them, the outcome residual and the treatment residual represent the residuals after the influence of the control variables is removed. The Neyman orthogonality condition is: .

(3) Final linear regression. Linear regression is performed on the outcome residual and the treatment residual to obtain the estimated value of the causal effect. The specific linear regression model is as follows:

7

Causal mediation analysis (CMA) model

Based on the DML model, this study further uses the causal mediation analysis (CMA) model to analyse the mechanisms through which digital literacy affects health inequality, which was proposed by Imai et al. [51]. The CMA model incorporates mediation analysis into the causal inference framework in a way that traditional mediation analysis methods cannot [52, 53] and helps to address the endogeneity problem present in traditional mediation analysis [54].

The steps of causal mediation analysis are as follows: first, define clear target parameters (i.e., causal effects) using the potential outcome framework [12]; then introduce reasonable identification assumptions to transform unobservable target parameters into observable statistical parameters (i.e., causal identification); and finally, estimate the statistical parameters using the data (see Supplementary 3 for details).

Results

Baseline result

The results of the baseline regressions are shown in Table 3. Columns (1), (3), and (5) respectively present the effects of children’s digital literacy on parents’ mental health, fathers’ mental health, and mothers’ mental health, respectively, without considering the control variables. Columns (2), (4), and (6) report the same effects after including the control variables at the individual level, family level, and regional level. The results show that the effect of children’s digital literacy on parents’ mental health, especially mothers’ mental health, is significantly negative at the 1% level, whereas the effect on fathers’ mental health is not significant when the control variables are not taken into account. When individual-level, family-level and regional-level control variables are added, the impact of children’s digital literacy on parents’ mental health and mothers’ mental health is significantly negative at the 5% level, and the impact on fathers’ mental health is also not significant. These results suggest that increasing children’s digital literacy improves the mental health of mothers but not that of fathers.

Table 3.

The results of the baseline regressions

	Parents’ mental health		Father’s mental health		Mother’s mental health
	(1)	(2)	(3)	(4)	(5)	(6)
CDL	−2.4880***	−1.3494**	−0.2004	0.0195	−3.1759***	−1.8954**
	(0.6393)	(0.6826)	(0.8399)	(0.8415)	(0.8344)	(0.9033)
Individual-level control variables	No	Yes	No	Yes	No	Yes
Household-level control variables	No	Yes	No	Yes	No	Yes
Regional-level control variables	No	Yes	No	Yes	No	Yes
Year FE	Yes	Yes	Yes	Yes	Yes	Yes
Household FE	Yes	Yes	Yes	Yes	Yes	Yes
N	5732	5732	2656	2656	3076	3076

** p < 0.05, *** p < 0.01; robust standard errors are reported in parentheses

Endogeneity test

This study needs to address some endogeneity issues. First, there may be a reverse causal relationship between digital literacy and mental health. According to previous studies [7], individuals’ health needs can increase the frequency of access to and use of digital skills, which in turn has an impact on children’s digital literacy. Second, there may also be unobservable factors that affect mental health. This study primarily uses the instrumental variables approach to address reverse causality and unobservable omitted variables.

The selection of the instrumental variable needs to fulfill the two conditions of being correlated with the endogenous variables and at the same time uncorrelated with the model error term. In this study, “the average digital literacy level of children from other households in the same residential area (IV)” is selected as instrumental variables for digital literacy, as has been done in previous studies [55]. In terms of relevance, due to peer effects, children in a given household are influenced by the digital literacy of children from other households. In terms of exogeneity, the digital literacy of children from other households does not directly affect the mental health of the household’s parents. Furthermore, this study testes for potential weak instrumental variable problem using the two-stage least squares (2SLS) method. The results indicate that the first stage F-statistic are much greater than 10, rejecting the null hypothesis of a weak instrumental variable and thus confirming the validity of the selected instrumental variable in this study. Based on the DML instrumental variable model proposed by Chernozhukov et al. [50], the estimation results are shown in Table 4. These results show that the significance and direction of the effects of children’s digital literacy on parents’ mental health are consistent with those of the baseline model.

Table 4.

Endogeneity test results

	2SLS		DML	2SLS		DML	2SLS		DML
	CDL	Parents’ mental health	Parents’ mental health	CDL	Parents’ mental health	Parents’ mental health	CDL	Parents’ mental health	Parents’ mental health
CDL		−2.0547***	−2.5308*		1.9958	0.0222		−4.0343***	−3.6908***
		(0.7911)	(1.4082)		(1.5551)	(1.4481)		(1.3549)	(1.3526)
IV	1.4202***			1.4666***			1.5790***
	(0.0236)			(0.0440)			(0.0425)
Control variables	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Household FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
N	5732	5732	5732	2656	2656	2656	3076	3076	3076
F	2903.29			593.87			716.42

* p < 0.1, *** p < 0.01; robust standard errors are reported in parentheses

Robustness test

Replacing the independent variable

This study employs the coefficient of variation method to recalculate digital literacy. As shown in Table 5, both the direction of the effect of children’s digital literacy on parents’ mental health and its significance remain consistent with the baseline results, indicating that the baseline results are robust.

Table 5.

Results of the robustness test

		Parents’ mental health	Father’s mental health	Mother’s mental health
Replacing the independent variable (Coefficient of variation)		−1.4700*	−0.1004	−1.8782*
		(0.8050)	(0.9956)	(1.0599)
Replacing the dependent variable (CES 8)		−0.8327**	0.0607	−1.2777**
		(0.3719)	(0.4625)	(0.5016)
Resetting the machine learning model
Changing the cross-validation	K-fold = 3	−1.0622*	−0.1032	−1.4775*
		(0.6332)	(0.8419)	(0.8623)
	K-fold = 8	−1.1335*	0.1895	−1.6418*
		(0.6774)	(0.8459)	(0.8788)
Replacement of machine learning algorithms	OLS	−1.1363*	0.1234	−4.4604***
		(0.6779)	(0.9126)	(0.9078)
	Lasso	−1.1259*	0.8082	−1.5300*
		(0.6638)	(0.8317)	(0.8468)
	Random forest	−1.2178*	0.2987	−2.2517***
		(0.6267)	(0.8656)	(0.8764)
Dropping the samples
Samples with physical complaints within two weeks		−1.4688**	−0.4945	−1.7554*
		(0.7213)	(0.9042)	(1.0249)
Samples from regions with developed digital infrastructure		−1.2958*	0.4946	−1.9696**
		(0.7188)	(0.8860)	(0.9559)
Samples with chronic illness within six months		−1.2558*	−0.1664	−1.8013*
		(0.6793)	(0.9270)	(0.9636)
Control variables		Yes	Yes	Yes
Year FE		Yes	Yes	Yes
Household FE		Yes	Yes	Yes

* p < 0.1, ** p < 0.05, *** p < 0.01; robust standard errors are reported in parentheses

Replacing the dependent variable

This study uses the CES-8 instead of the CES-20 to measure parents’ mental health. As shown in Table 5, the direction and significance of the effect of children’s digital literacy on parents’ mental health remain consistent with the baseline results, confirming the robustness of the latter.

Resetting the machine learning model

First, the cross-validation of the DML model is changed from 5-fold to 3-fold and 8-fold to explore the possible impact of the sample allocation ratio on the conclusions. Second, the machine learning algorithms are changed by replacing the gradient boosting used previously with OLS, lasso, and random forest to explore the possible impact of the prediction algorithms on the conclusions. The results in Table 5 show that neither the sample partition ratio of the DML model nor the machine learning algorithm used for prediction affects the conclusions, which are sufficiently robust to validate the benchmark findings.

Dropping the samples

Owing to the different economic development statuses of different regions in China, the level of digital infrastructure development differs across regions. To mitigate the influence of such disparities on regression results, this study excludes six regions with well-developed digital infrastructure (Beijing, Shanghai, Guangdong, Jiangsu, Zhejiang, and Fujian) and re-estimates the regression.

Additionally, considering the impact of chronic diseases and physical discomfort on mental health, samples with chronic diseases within six months and physical discomfort within two weeks are removed, followed by re-estimation. As shown in Table 5, even after excluding these samples, the effects of children’s digital literacy on parents’ mental health and mothers’ mental health are still significantly negatively correlated, further confirming the robustness of the baseline results.

Heterogenous analysis

Heterogeneity by region

The inconsistent level of digital infrastructure development in different regions of China has led to individual differences in digital literacy. To analyse the regional heterogeneity of the effects of children’s digital literacy on parents’ mental health, this study conducts a regression estimation of the sample according to three different dimensions: eastern, central, and western regions. As shown in Table 6, the estimation results indicate that children’s digital literacy has a more significant effect on parents’ mental health, especially mothers’ mental health, in the eastern and central regions.

Table 6.

Results of the heterogeneity analysis

		Parents’ mental health	Father’s mental health	Mother’s mental health
Heterogeneity by eastern, central and western regions	Eastern	−2.0717**	−1.0355	−2.3225*
		(0.9486)	(1.2041)	(1.3729)
	Central	−0.5780	1.9950	−3.0176*
		(1.1555)	(1.5464)	(1.6083)
	Western	−1.3906	−0.0689	−1.9649
		(1.2325)	(1.7673)	(1.6650)
Heterogeneity by urban and rural areas	Urban	−16,736**	−0.8784	−1.9174*
		(0.8551)	(1.1582)	(1.1606)
	Rural	−0.5141	0.9673	−1.0725
		(0.9047)	(1.2668)	(1.2736)
Heterogeneity by household socioeconomic status	Low household socioeconomic status	0.7418	1.7079	0.9678
		(1.1031)	(1.3562)	(1.3009)
	High household socioeconomic status	−1.7711*	−1.4446	−2.3685**
		(0.9393)	(1.0591)	(1.1611)
Heterogeneity by individual health status	Poor health	−2.3534**	0.3174	−2.5692*
		(1.1670)	(1.7343)	(1.4493)
	Moderate health	−1.3353	−0.2643	−1.6655
		(0.9300)	(1.1615)	(1.3001)
	Good health	1.2200	1.7081	2.5164
		(1.1732)	(1.5254)	(1.6538)
Control variables		Yes	Yes	Yes
Year FE		Yes	Yes	Yes
Household FE		Yes	Yes	Yes

* p < 0.01, ** p < 0.05; robust standard errors are reported in parentheses

Digital literacy also varies significantly between urban and rural areas in China. This study further conducts regression estimations by dividing the samples along the urban-rural dimension to explore the impact of urban-rural heterogeneity in children’s digital literacy on parents’ mental health. As shown in Table 6, the estimation results indicate that digital literacy has a more significant impact on parents’ mental health, especially mothers’ mental health, in urban areas.

Heterogeneity by household socioeconomic status

This study examines the differences in digital literacy among different household socioeconomic status groups. With reference to existing studies [56, 57], this study uses the International Socioeconomic Index (ISEI) to measure socioeconomic status. Based on the individual ISEI values provided by the CFPS database, the mean value of parents’ ISEI is used to measure household socioeconomic status in this study. Household socioeconomic status is then categorized into high and low groups based on the median value: those with scores greater than or equal to the median are classified as high household socioeconomic status, and those with scores below the median as low household socioeconomic status. The results in Table 6 show that the effect of children’ s digital literacy on mothers’ mental health is more significant in household with high socioeconomic status, suggesting that improving children’ s digital literacy can improve the mental health of mothers in household with high socioeconomic status.

Heterogeneity by individual health status

This study further analyses the impact of children’s digital literacy on parents’ mental health across health conditions. Parents’ health status is categorized as good, moderate, or poor on the basis of self-assessed health, and the regression results are shown in Table 6. The results show that children’s digital literacy significantly affects the mental health of parents with poor health status, especially mothers with poor health status.

Mechanism analysis

Household income

In this study, per capita household income is used to examine whether children’s digital literacy affects parents’ mental health through household income. The results in Table 7 indicate that the mediating effect of household income is significant at the 1% level. Specifically, 43.4% of the total impact of children’s digital literacy on parents’ mental health is transmitted through household income, whereas this percentage is 35.3% when mothers’ mental health is considered. These findings confirm that household income is a pathway through which children’s digital literacy affects parents’ mental health and that increased digital literacy helps improve family economic ability, which in turn improves parents’ mental health.

Table 7.

Results of causal mechanism analysis and sensitivity analysis

		Parents’ mental health					Mother’s mental health
		Causal mechanism analysis	Sensitivity analysis				Causal mechanism analysis	Sensitivity analysis
		Estimate					Estimate
Household income
ADE		−0.7830	−0.2	0.04	0.3949	0.1320	−1.7660**	−0.2	0.04	0.3907	0.1214
ACME		−0.6000***					−0.9570***
Total effect		−1.3820**					−2.7320***
Proportion of ACME (%)		43.4					35.2
Household healthcare expenditure
ADE		−1.4054**					−2.3903**
ACME		0.0230					−0.0068
Total effect		−1.3825**					−2.3970***
Proportion of ACME (%)		/					/
Emotional support
ADE		−1.3870	−0.25	0.0625	0.0272	0.1453	−2.3609*	−0.3	0.09	0.0479	0.1545
ACME		−0.3500***					−0.6252***
Total effect		−1.7380*					−2.9862**
Proportion of ACME (%)		20.2%					20.9%
Economic support
ADE		−1.4185**					−2.4325***
ACME		−0.0062					−0.0307
Total effect		−1.4247**					−2.4632***
Proportion of ACME (%)		/					/

* p < 0.1, ** p < 0.05, *** p < 0.01; ACME is the average causal mediation effect, ADE is the average direct effect, and total effect = ADE + ACME;is the value when ACEM = 0;andrepresent the proportions of residual variance in the mediation model and outcome model as explained solely by unobserved confounders, respectively;is the value when ACEM = 0;andrepresent the coefficients of determination of the mediation model and the outcome model, respectively

Household healthcare expenditure

Annual per capita household healthcare consumption (including fitness and exercise and the purchase of related products, equipment, healthcare products, etc.) is used to measure household healthcare expenditure in this study. The results in Table 7 show that the household healthcare expenditure does not have a mediating effect on the impact of children’s digital literacy on parents’ mental health and mothers’ mental health. The impact of children’s digital literacy on both parents’ and mothers’ mental health is dominated by direct effects.

Children’s intergenerational support

Emotional support. The parent-child relationship is used to measure emotional support. The parent-child relationship is a dummy variable, it takes a value of 0 if the relationship is categorized as “not very close”, “not too close”, or “average” in the dataset, and 1 if it is “close” or “very close”. The results in Table 7 show that emotional support is a transmission mechanism for the influence of children’s digital literacy on parents’ mental health and mothers’ mental health, with respective mediating effects of 20.2% for parents overall and 20.9% specifically for mothers. These results imply that children’s digital literacy can improve parents’ mental health, especially mothers’ mental health, by enhancing parent-child relationships.

Economic support. In this study, whether children provide economic assistance to parents is used to measure emotional support, which is a dummy variable that takes a value of 0 if the children do not provide economic assistance to the parents and 1 otherwise. The results in Table 7 show that economic support does not mediate the effects of children’s digital literacy on parents’ mental health or mothers’ mental health.

Sensitivity analysis

The causal mediation effect relies on the sequential ignorability assumption, which implies that the correlation coefficient () between the error terms of the mediation model and the outcome model equals zero. If an unobserved confounder results in a nonzero , this indicates a violation of the sequential ignorability assumption. This study employs the sensitivity analysis framework developed by Imai et al. to test for such violations [51]. The core of this method involves first calculating the value of at which the mediating effect is zero and then examining whether unobserved confounders could yield such a value.

Table 7 presents the range of values of when the mediating effect is 0. The results show that the mediating effect of household income on parents’ mental health and mothers’ mental health is 0 when =−0.2, whereas the mediating effect of emotional support on parents’ mental health and mothers’ mental health is 0 when =−0.25 and =−0.3, respectively. To test whether unobserved confounders could produce these values, this study follows Imai et al.’ s method, which defines . Here, and represent the proportions of residual variance in the mediation model and outcome model when explained solely by unobserved confounders. Taking the mediation effect of household income on parents’ mental health (with = −0.2) as an example, if such confounders exist, their explanatory proportion of residual variance in at least one model must not exceed 0.2. The results in Table 7 show that the coefficients of determination for the mediation model () and the outcome model (), reflecting the explanatory power of included control variables for total variance, are 0.3949 and 0.1320, respectively. The residual variance proportions for the two models mentioned above are 0.6051 and 0.8680, which are multiplied by 0.2 to obtain the alone explanatory power of the confounding factors for the model’s total variance of 0.1210 and 0.1736, respectively. This finding indicates that to offset the mediation effect fully, an unobserved confounder would need to explain approximately 17.36% of the total variance in the outcome model independently. Notably, the proportion of the total variance of the outcome model that is explained by all the existing observed control variables together in this study is only 13.20%. Given that this study has incorporated as many known important control variables as possible, the likelihood of an unobserved confounder with greater explanatory power than the sum of all the observed variables is extremely low, indicating that the mediating effect of household income is robust. Similarly, the robustness of the following mediating effects is confirmed: household income on mothers’ mental health (ρ= −0.2), emotional support on parents’ mental health (ρ= −0.25), and emotional support on mothers’ mental health (ρ= −0.3).

Discussion

This study employs the China Family Panel Studies (CFPS) dataset, with double/biased machine learning (DML) and causal mediation analysis (CMA), to examine the relationship between children’s digital literacy and parents’ mental health. The key findings are as follows:

First, the enhancement of children’s digital literacy positively affects parents’ mental health, and the effect is more pronounced for mothers, while there is no significant effect for fathers. These results are verified by endogeneity and robustness tests, further confirming the reliability of the above findings. These findings contribute to the literature by expanding the research scope on the intergenerational spillover effects of digitalization on health outcomes. While prior studies have focused predominantly on downwards spillover effects [58, 59], i.e., the influence of parents’ digital skills on their children’s health, this study shifts the focus to the upward spillover effects, specifically examining how children’s digital literacy affects parents’ mental health. This perspective contributes to the understanding of intergenerational digital spillover effects. Furthermore, the findings highlight children’s digital literacy as an effective way to alleviate parents’ mental health challenges in the digital age. Notably, the more substantial benefit observed among mothers can be contextualized within the framework of gender roles in Chinese families. Mothers typically assume greater responsibilities in child-rearing and household management, and children’s enhanced digital literacy may alleviate mothers’ stress by reducing the burden of information acquisition in areas such as educational support and health management [30, 60]. In contrast, fathers, who often play traditional breadwinning roles, may rely less on their children’s digital support, resulting in the nonsignificant effect observed here. Additionally, compared with fathers, mothers, who are primary maintainers of intergenerational emotional bonds within families, tend to exhibit greater intergenerational emotional needs, which may further amplify the positive impact of children’s digital literacy on their mental health [22, 23].

Second, the heterogeneity analysis reveals regional-level, household-level, and individual-level differences in the impact of children’s digital literacy on parents’ mental health. The effects are more significant in the eastern and central regions, urban areas, households with high socioeconomic status, and parents with poor health status. The stronger effects in eastern and urban areas reflect better digital infrastructure and higher internet penetration, facilitating children’s digital literacy development and its practical application [61]. In contrast, rural and western areas may suffer from “the digital divide”, limiting children’s access to high-quality digital resources [62]. Policy interventions in rural and western regions should prioritize narrowing infrastructure gaps and providing digital literacy training for both children and parents. Children in households with higher socioeconomic status have greater access to digital devices and higher levels of digital literacy, which can enhance upward income mobility for households [32]. For example, digital skills can enhance educational opportunities and part-time job prospects for children, directly improving family income. With respect to parents in poor health, more digitally literate children are more likely to use digital tools to strengthen their emotional connection with their parents and to help them access health information and telemedicine support [63], reducing psychological stress related to health management.

Third, the mechanism analysis identifies household income and emotional support as key mediating pathways through which children’s digital literacy improves parents’ mental health. Regarding the mediating role of household income, this finding supports the existing conclusion that income mediates the relationship between digital literacy and mental health [7], and this study further emphasizes the mediating effect of income in intergenerational digital spillover. The prominent role of household income underscores the connection between digital literacy and labor market outcomes. Children with strong digital skills are more likely to engage in high-value digital activities (e.g., online education and remote work), thereby increasing their personal income [29]. Moreover, household’s collaboration through digital tools strengthens the employability of household members and enhances their employment opportunities [64]. These aforementioned employment effects can increase household income, ease parents’ financial burdens, and consequently alleviate their mental health issues. Beyond household income, emotional support also mediates the relationship between children’s digital literacy and parents’ mental health. This result is similar to the study in South Korea, which found social support (including emotional support) plays an intermediary role between digital literacy and mental health among older women [65]. However, the difference lies in that the research in this study focuses on exploring the mediating effect of emotional support in intergenerational digital spillover. The improvement of children’s digital literacy enhances their ability to use digital tools (e.g., social media and video calls). These digital tools can overcome geographical and temporal constraints and increase the frequency of communication between parents and children, which in turn strengthens the emotional support [66, 67]. This strengthened emotional support effectively improves parents’ mental health. In contrast, neither household healthcare expenditures nor economic support has significant mediating effects. This phenomenon can be attributed to the following factors. On the one hand, the current prevalence of e-health behaviors among residents is still low, and there is an obvious digital divide. In addition, most residents are cautious about the rapid development of internet hospitals, and these factors together lead to a nonsignificant mediating effect on household healthcare expenditures [68]. On the other hand, China’s social security system has gradually improved, and formal support systems have begun to address the issue of care security for elderly people [69], thereby reducing their financial dependence on their children. In addition, as parents age, their need for emotional support tends to surpass that for economic support [70], which further accounts for the nonsignificant mediating effect of economic support.

Conclusion and implications

This study provides strong evidence that children’s digital literacy can improve the mental health of parents, especially mothers, through increased household income and emotional support. The findings highlight the importance of intergenerational digital spillover in addressing the mental health of middle-aged and older adults in China’s rapidly digitizing society.

The results of this study have several important policy implications. First, the government should further improve policies to enhance children’s digital literacy. These policies should include increasing the supply of digital resources, offering training to boost digital skills for learning, work and daily life, and strengthening intergenerational digital solidarity. Second, the government should adopt targeted digital literacy promotion policies based on regional differences. For western and rural areas with underdeveloped digital infrastructure, more investment should be made in infrastructure construction and the supply of digital resources should be increased. In digital skills training, rural regions should prioritize programs focused on digital agriculture and agricultural product e-commerce, while western regions should emphasize digital employment skills training to enhance their employment opportunities and ability to increase household income. For low-income households, financial subsidies should be increased to improve the digital participation capacity of households with lower socioeconomic status. Third, regarding intergenerational digital solidarity, the government should develop user-friendly digital platforms for intergenerational communication to strengthen children’s emotional support for their parents.​.

This study has several limitations. First, in the selection of indicators for assessing digital literacy, owing to the availability of data, indicators such as cognitive, evaluative, or ethical components of digital competence are not included. Future research should further include the above dimension indicators to develop a more comprehensive and reasonable digital literacy evaluation system. In addition, the measurement of emotional support in future studies needs to be more granular. Second, the analysis of mediating mechanisms in this study focuses on intergenerational economic support and emotional support, while unobserved factors such as parents’ digital anxiety or intergenerational technology use conflicts, are not included in the study. Future investigations could expand into these areas to deepen our understanding of the mediating mechanisms. Third, more attention should be paid to the risk of multiple testing arising from heterogeneity analysis in future research. It is necessary to integrate existing multiple test correction methods with machine learning approaches to enhance the robustness of heterogeneity analysis results. Additionally, more advanced machine learning techniques could be employed to present findings in a more intuitive manner.

Authors’ contributions

Yun Ye conceptualized, researched, and wrote the manuscript. The manuscript has been read and approved by the author.

Funding

This study is funded by the Research Fund of Hainan Medical University (No. RZ2300006014).

Data availability

The data used in this paper is public available secondary dataset, which can be available at the official website: https://www.isss.pku.edu.cn/cfps/.

Declarations

Ethics approval and consent to participate

Not applicable. The paper is exempt from ethics approval since it is an analysis of secondary data.

Consent for publication

Not Applicable.

Competing interests

The authors declare no competing interests.