Caregiver mental health is associated with early childhood language outcomes and perception bias in rural China

Abstract

Language development and the home language environment during early childhood are critical for long-term child outcomes. Caregiver mental health may influence early language outcomes directly, but it can also introduce perception bias, which refers to the discrepancies between caregiver self-assessments and the actual status of child language outcomes. This study examines the associations between caregiver mental health symptoms and (1) child language development and home language environment, and (2) caregiver perception bias in self-report assessments of child language development and home language environment. The study recruited 137 rural Chinese households with children aged 16–24 months. Objective measures of child language development and the home language environment were collected using Language Environment Analysis (LENA) technology. Caregiver perception biases were measured by the discrepancies between the objective and caregiver self-report measurements. Results show that caregiver anxiety and stress symptoms were linked to poor child language development, while symptoms of depression and anxiety were associated with less stimulating home language environment. Caregivers with depressive and anxiety symptoms tended to overestimate their children’s language development, and those with depressive symptoms also overestimated their own verbal inputs. These findings call for caution when implementing self-report assessments of early childhood development.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-39674-2.

Introduction

Language development during the first three years of life is a fundamental component of early childhood development that lays the groundwork for future language, cognitive, and learning outcomes. Delays in early language development can have lifelong consequences, such as low educational attainment and diminished employment performance^1–3. Unfortunately, many children in low- and middle-income countries (LMICs) are at risk of early language delays. In rural China, for example, a recent meta-analysis revealed that 46% of young children failed to reach their language development potential⁴. Growing evidence indicates that poor home language environments, characterized by insufficient verbal interactions between children and their caregivers, is a main predictor of early language developmental delay in LMICs^5–7. Identifying the factors that hinder an interactive, stimulating home language environment is crucial for improving early language development in low-income regions.

Caregiver mental health substantially influences child development outcomes, but its specific impact on language development remains unclear. One in five mothers in LMICs reported depressive symptoms in LMICs during the post-partum period⁸. A survey of 2541 participants suggested that 39% of primary caregivers (mothers and grandmothers) of children under age three living in rural China reported depressive, anxiety, or stress symptoms⁹. A substantial body of research links caregiver depressive symptoms to impaired child cognitive, motor, and social-emotional development, as well as disturbances in early attachment and increased behavior problems^10–12. A Lancet review suggests that depressive symptoms contribute to these developmental adversities by altering the early caregiving environment, particularly the caregiver’s parenting practices and interactions with the child¹³. However, the much smaller body of literature has focused on early childhood language development and other mental health symptoms beyond depression.

Research on caregiver mental health and early language development is further complicated by the perception bias associated with caregiver mental health symptoms. Perception bias, the discrepancies between caregiver self-assessments and the actual status of child language outcomes, occurs when caregivers experiencing mental health symptoms misinterpret or overlook their own and their children’s behaviors, making distorted assessments. When perception bias results in negative attributions, caregivers could misinterpret typical child behaviors as problematic or deliberately disobedient rather than situationally driven^14,15. Caregiver mental health symptoms could also impair their capacity to pay attention to the child¹⁶, thus reduce the accuracy of caregiver-reported measurements. Despite most widely implemented measurements of child development rely on caregiver self-report, few studies have tried to quantify whether caregiver mental health symptoms were associated with their perception bias in caregiver self-assessments of child language development and their home language environment¹⁷.

Despite China’s status as an upper-middle income country, disparities remain between urban areas and the country’s vast rural population^18–20. About 40% of China’s population is rural, including 100 million children²¹. This population includes those residing in rural villages, as well as households from rural backgrounds in China’s cities. These “peri-urban” rural households include rural migrants (who have moved from their rural hometowns to urban areas for work) and urbanized farmers (whose rural farmland has been lost to urban expansion)^22–24. Although peri-urban households typically have slightly higher levels of income and education than those living in rural areas^25,26, both groups have significantly lower SES (in terms of both education and income) compared to established urban households in China’s cities^23,24,27,28. Past literature suggested high mental health issues among those populations⁹, yet, little is known about how caregiver mental health shapes early childhood language outcomes within these underrepresented populations.

Recent advances in the Language Environment Analysis (LENA) technology have enabled the recording and analysis of sounds in the child’s surrounding environment and those produced by the child, which allow for an objective measure of child language development and the verbal interactions that form the home language environment. Leveraging LENA data as benchmark objective measurements, the goals of the current study are to (a) investigate the association between caregiver mental health symptoms, child language development, and the home language environment; and (b) examine the association between those symptoms and perception bias in caregiver-reported assessments of child language development and their language home environment.

Methods

Study location

Data were collected from households in two distinct areas (a peri-urban district and two rural counties) in Sichuan Province, China. The peri-urban district is in the suburbs of Chengdu, the provincial capital, and has rapidly transitioned from an agricultural economy into one of the fast-developing areas in the province. Due to rapid urbanization, the district hosts a large population of (a) households with rural backgrounds, including rural migrants, that migrated from rural regions without access to urban public welfare benefits due to their rural household registration status, and (b) urbanized rural households, whose farmland in the area had been acquired by the government. In 2019, the average per capita disposable income of rural residents in this district was 30,405 RMB (4403 USD), much lower than the average per capita income in the Chengdu’s central urban districts (49,193 RMB or 7613 USD)²⁹. The second sampling location was in two counties in a rural, prefecture-level city in Sichuan. The average per capita income for rural residents in this prefecture was 14,670 RMB (2124 USD), less than half of the average per capita income of urban residents in Sichuan (36,153 RMB or 5235 USD)²¹.

Participants

The participants in the study are children aged 18 to 24 months and their primary caregivers, within the households with rural backgrounds. We focus on toddlers aged 18 to 24 months due to the rapid language development that occurs during this period: Around 18 months of age, many children experience a “vocabulary burst” and start to acquire new words at a much faster rate than before (Goldfield and Reznick 1990). By this age, they can also distinguish grammatically correct from incorrect placement of familiar function words (Santelmann and Jusczyk 1998). By 24 months, the speed and accuracy in spoken word understanding increase dramatically over the second year of life (Fernald et al. 1998). Therefore, any disparity in the home language environment at this stage is likely to have measurable impacts on the child’s language skills.

Sampling protocol

Due to the geographic and administrative differences between the rural and peri-urban communities, the research team separately sampled participants from the two study locations. To select the peri-urban sample, the research team followed a two-step protocol. First, the research team obtained a list of children between 18 and 24 months of age based on birth records from two local hospitals. Second, households were then randomly selected from the list and trained enumerators screened them for eligibility via phone calls. Households were invited to participate in a survey if their residency status met the legal criteria for rural migrants or urbanized farmers: A household was considered a rural migrant family if (a) the legal residence of either parent was in a rural area outside of Chengdu Municipality; and (b) one parent had lived in the sample district for at least 6 months.

Urbanized farmers are legally defined as people with rural residence status (hukou) whose farmland has been fully or partially expropriated by the government, resulting in each adult family member of the household owning less than 0.3 mu (approximately 200 square meters) of cultivated land on average³⁰. Accordingly, we defined a household as urbanized farmers if it met each of the following criteria: (a) The legal residence of either parent must be within the district; (b) All or part of the household’s farmland must have been expropriated by the government; And (c) any remaining cultivated land owned by the household must be less than 200 square meters per adult family member on average. Following this protocol, a total of 109 households were enrolled in the survey. Of these, 81 consented to the collection of home language environment data (34 rural migrant households, 47 urbanized rural households).

To select the rural sample, the research team followed a three-step protocol. First, all townships in the two sample counties were included in the study, except the townships where the county government was located, which are typically wealthier and more urbanized than townships in the rest of the county. This resulted in 11 townships. Second, we obtained a list of households with children aged between 18 and 24 months based on birth records from the township governments, which resulted in a total of 175 households across the 11 townships. Third, we excluded those that were not legally classified as rural residents and enrolled all remaining 109 eligible households in the study. Of these, 77 consented to the collection of home language environment data.

Of the 158 households that completed home language environment data collection, we excluded 21 in which the primary caregiver was not the respondent of the household survey. The final analytic sample includes 137 households: 40 urbanized rural, 27 rural migrant, and 70 rural households. A balance test found no significant differences in demographic information between households that consented to LENA data collection and those that did not [See Appendix Table 1].

Data collection

The research team collected data on the home language environment, caregiver mental health, child language development, and demographic characteristics of the household young children and their households. Trained enumerators collected data from all sample households in the summer and fall of 2020 following a standardized four-day protocol. On the first day, enumerators surveyed the primary caregiver of each sample child at their home or the hospital. After the survey, caregivers were instructed on how to record home language environment data with LENA. Caregivers then recorded their home language environment during the second and third days. On the fourth day, enumerators retrieved the recording devices and conducted an exit interview to ensure compliance with the data collection protocol.

Measures

Caregiver mental health

To measure caregiver mental health, enumerators administered the Depression, Anxiety, and Stress Scale-21 (DASS-21) to the primary caregiver of the child, identified by each household as the person primarily responsible for the child’s daily care. The DASS-21, a short-form version of the DASS-42³¹, is a self-report questionnaire with three subscales evaluating if the caregiver was at risk for the depression, anxiety, and stress, respectively (and the severity of the risk). Caregivers were asked to rank, on a scale of 0 to 3, how each of 21 statements (seven describing each mental health issue) applied to them in the past week. The score on each subscale is the sum of their ratings for the seven statements multiplied by two. Hence, the score for each subscale could range from 0 to 42, with higher scores indicating more severe symptoms. Following the DASS-21 scoring guidelines, participants were considered experiencing symptoms of mental health if they scored above 9 for depression, 8 for anxiety, and 14 for stress³¹. Notably, while the score cannot be used for clinical diagnosis, studies in multiple countries, including China, have demonstrated the validity of DASS-21 as a measurement of the severity of depression, anxiety, and stress symptoms^32,33.

Home language environment

The direct-assessment of children’s home language environment used the LENA system. LENA consists of a small recording device worn by the child and the corresponding software that records and analyzes all sounds in the child’s surrounding environment over the course of a day³⁴. Previous research has shown LENA to be a valid and reliable measure of language environments in American English³⁵, Spanish³⁶, Dutch³⁷, Korean³⁸, Vietnamese³⁹, and Mandarin-speaking populations^40,41.

Each sample household was given a LENA recording device, a specialized LENA shirt, and a LENA charger. Caregivers were instructed to record 16 h per day over two days that were representative of what their child typically experienced at home. The recording device was placed in the chest pocket of the LENA shirt which the child wore throughout the day and was removed only when the child bathed or slept at night. Caregivers were also instructed to charge the LENA device overnight between the first and second days of recording.

In this study, the home language environment is measured by two variables automatically generated by the LENA software: The first is the adult word count (AWC), defined as the number of words spoken to or near the child by adults in the recording. The second is conversational turn count (CTC), defined as the number of adult-child alternations in conversation in the recording.

In addition, the caregiver-report home language environment was measured using the four items in the Family Care Indicators (FCI) scale. The FCI is a validated survey instrument developed by UNICEF to evaluate the quality of the home environment and quantify the different types of parenting practices⁴². Caregivers reported the time they spent with their child in the previous day on interactive reading, storytelling, singing, and playing. These four parenting practices were selected because they are the most commonly reported measures of stimulating parenting practices for language development in the literature on rural China⁴. Factor analysis also was used to generate a standard index for stimulating parenting practices for language development.

Child language development

The LENA-generated measure on child’s language development is the child vocalization count (CVC). CVC captures the frequency of child-produced utterances, including all sounds, words, and vocalizations that are not vegetative noises, cries, or coughs. Notably, researchers have proposed that children’s vocalizations can indicate the quality of the home language environment (Ramírez-Esparza et al. 2014; Werwach et al. 2021) ^67,70 because they are driven by the children’s interactions with other social entities (Moeller et al. 2007) ⁶⁹ and, as auditory feedback models suggest, hearing one’s own voice supports language development (Brainard & Doupe, 2000) ⁷¹. Nevertheless, we focus on CVC as an indicator of children’s language development in this study because it has been widely used as a measure of language development (Ramírez-Esparza et al. 2014) ⁶⁷ and has been shown to be highly predictive of language skills later in life (Wang et al. 2020; Werwach et al. 2021)^68,70.

The caregiver-report measure on child language skills is the Chinese version of the short-form MacArthur-Bates Communicative Development Inventory (CDI). The CDI measures expressive vocabulary of children aged 16 to 30 months⁴³, and the short-form CDI has been validated in Mandarin Chinese⁴⁴. The primary caregiver was given an inventory of 113 words and asked to identify the words that their child could say. For words the child could say, caregivers were asked to give an example of the context in which the child said the word. The number of words the child could say was then summed to reach a final CDI score which could range between 0 and 113, following CDI scoring guidelines⁴⁴.

Caregiver perception bias

Perception bias is measured as the discrepancy between the caregiver-report assessment and the LENA-generated measures of child language development and language environment. Since the LENA measures, CDI, and the parenting practices index are on different scales, direct comparisons between them are not feasible. To enable meaningful within-sample comparisons and account for increases in the raw scores with the child’s age, we internally standardize the scores by removing the effect of age (using the child’s age for child measures and caregivers’ age for caregiver measures)⁴⁵. We compute internal z-scores based on age-specific averages and standard deviations, which are estimated through non-parametric regression. This method reduces sensitivity to outliers and small sample sizes, producing smooth, standardized scores with a mean of zero across all age groups.

Then, we compute differences between the standardized scores to derive perception bias. Specifically, the perception bias in child language development is defined as the difference between the self-reported CDI score and the objectively collected CVC. The perception biases in child language environment are similarly defined as the difference between a parenting practices index computed from self-reported FCI items and AWC, as well as the difference between the parenting practices index and CTC. For the resulting bias scores, a positive score indicates an overestimation of the child’s language development or language environment relative to their LENA outcomes, while a negative score indicates underestimation.

Statistical analysis

We standardized the LENA measurements, including AWC, CTC, and CVC. To account for variation in recording start times across households and the skewness typical of count data, 16-h recordings were standardized to a 12-h observation window. Following common practice in LENA-based analyses⁴⁶, we aligned all five-minute audio segments to a shared reference time (6:20 a.m.) and modeled time-of-day patterns using Chebyshev polynomials, which provide stable polynomial approximations while minimizing boundary oscillations (Runge phenomenon)⁴⁷. Adaptive LASSO was used to select the optimal polynomial terms. Ordinary least squares (OLS) models were then estimated with five-minute AWC, CTC, and CVC as outcomes and the selected polynomial terms as predictors, and residuals were extracted to remove start-time-related variation. Residuals were added back to the model intercept to obtain adjusted five-minute values, which were subsequently summed across the valid segments within the 12-h window to generate standardized daily totals for each recording day.

To study the association between caregiver mental health symptoms, child language development, and the home language environment, we first compared the unadjusted distributions of standardized child environment and development outcomes by caregiver mental health status using kernel density plots, excluding the top and bottom 1% of values and without adjusting for demographic or sample strata information. This approach allows us to visualize the magnitude and direction of observed disparities before accounting for potential confounders. We also conducted univariable OLS regressions estimating unadjusted associations, with cluster-robust standard errors at the sample-source level (urbanized, migrant, and rural), to complement the visual analyses.

We then conducted multivariable analysis to account for demographic and sample strata information. Although we used multi-level sampling strategies in both rural and peri-urban areas, we did not apply hierarchical linear models (e.g., mixed-effects models) for two reasons. First, the intraclass correlation coefficients (ICCs) at the sample-source level were generally trivial (see Appendix Table 2). Prior studies suggest that hierarchical models are not warranted when ICCs are low, typically below 0.10^48,49. In our data, ICCs for all outcomes except AWC fall below this threshold. Second, our second-level sampling structures differ substantially across samples: hospitals were used in the peri-urban sample while townships were used in the rural sample, making harmonization challenging. Therefore, we opted not to estimate three-level mixed-effects models.

Instead, we used OLS regression models to examine the associations between caregiver mental health, child language environment, and language development. The models controlled for demographic information with cluster-robust standard errors estimated by the sample strata. Demographic information including child age, sex, whether mother was the primary caregiver, primary caregiver’s age and education level, family assets, and the number of adults staying at home were collected through a questionnaire administered by trained enumerators to caregivers. To estimate family assets, caregivers reported whether they owned or had access to a series of household items like tap water, computer, internet, and a car. A family asset index was then calculated using polychoric principal component analysis⁵⁰. Given the small sample size, we used bootstrap resampling with replacement (1000 replications) to obtain more robust estimates⁵¹.

The same procedures were applied to examine associations between caregiver mental health symptoms and perception bias in caregiver-reported language measures. We first compared unadjusted distributions of perception bias by caregiver mental health status using kernel density plots and univariable OLS regressions. ICCs at the sample-source level were then evaluated; because all ICCs were below 0.10 (Appendix Table 2), we estimated ordinary least squares regression models rather than hierarchical linear models. Models adjusted for demographic characteristics and sampling strata, with inference based on 1,000 bootstrap replications, consistent with the previous analyses.

Results

Summary statistics

The demographic characteristics of the participants are presented in Part A, Table 1. The average age of the children was over 21 months, 58% of them were male, and 58% were primarily cared for by their mothers. The average age of the primary caregivers was approximately 37 years, and 50% of them (n = 74) had completed high school. Households from the rural counties make up 51% of the sample (n = 70), while another 29% were urbanized rural households (n = 40), and the rest 20% were rural migrant households (n = 27).

Table 1.

Descriptive information among the participants (N = 137).

Part A: Demographic characteristics
Child age (months)	21 (1.6)
Child is male (%)	80 (58%)
Mother is the primary caregiver (%)	80 (58%)
Primary caregiver age (years)	37 (13)
Primary caregiver graduated from high school (%)	74 (50%)
Family asset index	− 0.01 (1.3)
Number of adults at home (n)	2.8 (1.0)
Sample Strata
Urbanized rural households (%)	40 (29%)
Rural migrant households (%)	27 (20%)
Rural households (%)	70 (51%)
Part B: Caregiver mental health
Caregiver Mental Health Symptoms
Depressive symptoms (%)	17 (12%)
Anxiety symptoms (%)	21 (15%)
Stress symptoms (%)	11 (8%)
Part C: Child language development
Recorded Child Language Development
Child vocalizations count (CVC, n)	1923 (800)
Caregiver-report Child Language Development
CDI language score	45 (29)
Part D: Home language environment
Recorded Child Language Environment
Adult word count (AWC, n)	13,773 (5609)
Conversational turns count (CTC, n)	557 (286)
Caregiver-report Child Language Environment
Read books yesterday (mins)	5.4 (18)
Told story yesterday (mins)	4.8 (11)
Sang songs yesterday (mins)	4.9 (9.3)
Played games yesterday (mins)	28 (39)

Family asset index was calculated using polychromic principal component analysis based on whether households owned or had access to a number of household items such as tap water, computer, internet, and a car. Caregiver mental health was assessed using the Depression, Anxiety, and Stress Scale-21 (DASS-21). CDI refers to the Macarthur-Bates Communicative Development Inventory. CVC, AWC, and CTC were measured using the LENA technology for 16 h and then standardized into 12 h. Caregiver-report language environment was measured using Family Care Index (FCI).

Of the sample caregivers, 17 (12%) reported depressive symptoms, 21 (15%) reported anxiety symptoms, and 11 (8%) reported stress symptoms (Part B). The average CDI score was 45 words (out of 113), and the average CVC was 1923 vocalizations (Part C). The average AWC was 13,773 words and the mean CTC was 557 turns. On average, caregivers reported spending 5 min reading books, 5 min telling stories, 5 min singing songs, and 28 min playing with their child throughout the day prior to the survey (Part D).

Caregiver mental health symptoms and child language development

The raw distributions of CVC were left-skewed for caregivers who reported depressive, anxiety or stress symptoms, compared to those who did not (Fig. 1), suggesting an unadjusted association between caregiver mental health symptoms and reduced CVC. Univariable regressions (Appendix Table 3, Part A) showed a significant association between depressive symptoms and CVC (= − 0.34, 95% CI [− 0.44, − 0.24], p < 0.01), whereas anxiety and stress symptoms were not significantly associated.

Fig. 1.

LENA child vocalizations count (CVC) by caregiver mental health symptoms. Note: Extreme 1% values on both sides are excluded.

After adjusting for demographic characteristics (Table 2, Part A), caregiver anxiety symptoms were significantly associated with lower CVC (= − 0.44, 95% CI [− 0.69, − 0.18], p < 0.01), and caregiver stress symptoms were significantly associated with both lower CVC (= − 0.75, 95% CI [− 1.27, − 0.24], p < 0.01) and lower CDI score (= − 0.57, 95% CI [− 1.00, − 0.15], p < 0.01). The regression results with the coefficients of covariates can be found in Appendix Table 4.

Table 2.

Regressions of caregiver mental health symptoms on child language development, language environment, and caregiver reporting bias (N = 137).

	Part A Language development		Part B Language environment			Part C Perception bias in language development	Part D Perception bias in language environment	Using CTC
	CVC	CDI	AWC	CTC	FCI	Using CVC	Using AWC
Depressive symptoms	− 0.38	− 0.12	− 0.24	− 0.51**	0.15	0.26+	0.39	0.66***
	(0.30)	(0.33)	(0.26)	(0.20)	(0.20)	(0.14)	(0.25)	(0.14)
Anxiety symptoms	− 0.44***	0.00	− 0.20	− 0.54***	0.10	0.44**	0.30	0.63
	(0.13)	(0.19)	(0.20)	(0.12)	(0.52)	(0.17)	(0.35)	(0.43)
Stress symptoms	− 0.75***	− 0.57***	0.03	− 0.62	− 0.36	0.18	− 0.39	0.26
	(0.26)	(0.22)	(0.33)	(0.39)	(0.24)	(0.17)	(0.24)	(0.29)

(1)⁺p < 0.10; **p < 0.05; ***p < 0.01. (2) Each row and column represent independent regressions. All regressions control for primary caregivers’ age, primary caregivers’ education, child age, child sex, the mother being the primary caregiver, number of adults in the residence and the family asset index. Regressions are bootstrapped 1,000 times, with cluster-robust standard errors estimated at the sample source level (urbanized, migrant, or rural). (3) Full regression tables with coefficients for all control variables are provided in Appendix Tables 3, 4 and 5. (4) CVC refers to LENA Child Vocalization Count, CDI refers to the MacArthur-Bates Communicative Development Inventory, AWC refers to LENA Adult Word Count, CTC refers to LENA Conversational Turn Count, and FCI refers to the Family Care Indicator.

Caregiver mental health symptoms and child language environment

The raw distributions of both AWC and CTC were also left-skewed among caregivers who reported depressive, anxiety, or stress symptoms, compared to those who did not (Fig. 2), indicating that, without adjustment, caregiver mental health symptoms were linked to lower AWC and CTC. Univariable regressions (Appendix Table 3, Part B) showed significant associations between CTC and both depressive symptoms (= − 0.43, 95% CI [− 0.66, − 0.21], p = 0.014) and anxiety symptoms (= − 0.42, 95% CI [− 0.88, − 0.04], p = 0.059), whereas no significant associations were observed for AWC.

Fig. 2.

LENA child language environment by caregiver mental health symptoms. Note: Extreme 1% values on both sides are excluded.

Controlling for demographic characteristics (Table 2, Part B), AWC was not significantly associated with any mental health symptoms. In contrast, lower CTC was significantly associated with caregiver depressive symptoms (= − 0.51, 95% CI [− 0.91, − 0.12], p = 0.011) and anxiety symptoms (= − 0.54, 95% CI [− 0.76, − 0.31], p < 0.01). The regression results with the coefficients of covariates can be found in Appendix Table 5.

Caregiver mental health symptoms and perception bias in child language development

The mean of the distributions of caregiver perception bias in child language development were to the right of zero for caregivers with depressive and anxiety symptoms (Fig. 3). In univariable regressions (Appendix Table 3, Part C), none of the associations were statistically significant.

Fig. 3.

Bias in child language development by caregiver mental health symptoms. Note: Extreme 1% values on both sides are excluded.

After controlling for confounders (Table 2, Part C), there is a marginally significant association between caregiver depressive symptoms and a positive bias (= 0.26, 95% CI [− 0.02, 0.54], p = 0.064). Caregiver anxiety symptoms were also significantly associated with a positive bias (= 0.44, 95% CI [0.10, 0.77], p = 0.012). These findings suggest that caregivers experiencing depressive and anxiety symptoms were more likely to overestimate their children’s language development. The regression results with the coefficients of covariates can be found in Appendix Table 6, column 1–3.

Caregiver mental health symptoms and perception bias in child Language environment

The mean of perception bias in child language environment among caregivers were right or zeros for caregivers with depressive, anxiety, or symptoms, suggesting a tendency of overestimation linked to those symptoms (Fig. 4). The univariable regressions (Appendix Table 3, Part D) showed that caregiver stress symptoms were negatively associated with bias using AWC (= − 0.43, 95% CI [− 0.90, 0.04], p = 0.058) yet positively associated bias using CTC (= 0.19, 95% CI [− 0.06, 0.44], p = 0.084).

Fig. 4.

Bias in child language environment by Caregiver Mental Health Symptoms. Note: Extreme 1% values on both sides are excluded.

After adjusting for demographic characteristics (Part D, Table 2), depressive symptoms were significantly associated with a stronger positive bias using CTC (= 0.66, 95% CI [0.40, 0.93], p < 0.01), suggesting that caregivers with depressive symptoms were more likely to overestimate their home language environment. No significant associations were observed between anxiety or stress symptoms and perception bias in child home language environment. The regression results with the coefficients of covariates can be found in Appendix Table 6, column 4–9.

Discussion

Drawing on novel Language Environment Analysis (LENA) data collected from low-resource households with rural backgrounds in China, this study examines the associations between caregiver mental health, the home language environment, and children’s language development. The paper also investigates whether mental health symptoms are associated with the perception bias of caregivers in those outcomes.

As measured by LENA, we found that caregiver anxiety and stress symptoms were significantly associated with lower levels of child language development. Additionally, caregivers with depressive or anxiety symptoms engaged in fewer verbal interactions with their children. These findings align with prior studies using caregiver-report measures, which have shown that maternal mental health symptoms affect children’s language development in both high-income countries^52–54 and LMICs⁵⁵. Caregiver mental health symptoms are associated with reduced engagement in interactive parenting, including numerous studies in rural China^56,57. Our results also support recent studies from high-income countries using LENA, which suggests that caregiver depression and anxiety symptoms were associated with adverse home language environment^54,58.

Caregiver mental health symptoms were also associated with perception biases in their assessments of their children’s language development and the home language environment. Specifically, caregivers with depressive symptoms were more likely to overestimate their children’s language development. Depressive symptoms often manifest as persistent negative thoughts and cognitions⁵⁹, causing caregivers to adopt a more pessimistic view of their children’s language skills and undervalue the developmental milestones they reach. Additionally, the fatigue and low energy associated with depression may reduce caregivers’ awareness of their children’s day-to-day behaviors¹⁶ and lead them to overlook developmental achievements. These factors suggest a tendency toward underestimation—making the observed overestimation unexpected. These results highlight the need for future research to explore why caregiver depression may be linked overestimation of child language development.

Caregivers with depressive and anxiety symptoms also tended to overestimate how many verbal interactions they had with their children. While existing studies have documented a general tendency for caregivers to overreport verbal interactions⁶⁰, our result suggests that this bias could be more pronounced among those experiencing depressive or anxiety symptoms. One possible explanation is that verbal interactions may demand greater physical, cognitive, and emotional effort for individuals experiencing depression, making the interactions more salient to them and thus more likely to be overestimated⁶¹.

In addition, anxiety symptoms have been linked to heightened concerns on social evaluation and a greater susceptibility of social desirability bias⁶², suggesting that caregivers with anxiety symptoms may overreport their children’s language milestones to present the household in a more favorable light. Furthermore, anxiety symptoms are often associated with maintaining positive illusions that lead to overly optimistic assessments of oneself and the world⁶³. In this context, caregivers may exaggerate their children’s language skills to mitigate distress about their children’s development.

The perception bias associated with caregiver mental health symptoms are concerning, as most assessments of child development and parenting practices rely on caregiver self-reports¹⁷. In LMICs, where caregiver mental health symptoms are common, such bias may compromise the accuracy of these measures. Relying on caregiver reports in surveys may introduce bias in estimation of intervention impacts on child development if caregiver mental health changes differentially between treatment and control groups over time. Similarly, using caregiver-reported assessments to screen for developmental delays may fail to identify vulnerable children, particularly those whose caregivers experience depression or anxiety symptoms.

These findings highlight the need for caution when using and interpreting self-report tools in early childhood development research. Whenever possible, researchers should employ direct-assessment tools to measure child outcomes. When self-reports are unavoidable, studies should account for caregiver mental health status during both data collection and analysis periods. Additionally, developing benchmarks that adjust for caregiver mental health symptoms may improve the accuracy of caregiver-reported assessments.

We acknowledge several limitations of this study. First, the construction of perception bias relies on the assumption that non-parametric regressions fully account for the effects of child age. This may not always hold, since residual age-related variation could still influence the standardized scores and introduce bias into our estimates. To mitigate potential confounding, we included child age (in months) as a covariate in our analyses. Second, the measurement of perception bias is based on two specific instruments, which limit the generalizability of our findings to other self-report instruments. Third, although our sample size of 137 is the largest of LENA-based studies worldwide to date, it remains small relative to studies using other assessment approaches. Future research should validate and expand our findings using larger, representative samples to enhance statistical power.

Despite these limitations, the study makes several noteworthy contributions. To the best of our knowledge, this is the first study to examine associations between caregiver mental health and LENA-generated measures of language development and the home language environment in LMICs. It is also among the first to quantify the perception bias in caregiver self-report assessments of early development and to examine the association between this bias and their mental health issues. Despite the exploratory nature of our analysis, our findings provide initial evidence that depressive and anxiety symptoms can distort caregiver assessment of their children’s language skills and the home language environment, calling for greater caution when administering self-report assessments in ECD research.

Conclusion

This study provides novel evidence that caregiver mental health symptoms, particularly depression, anxiety and stress, are significantly associated with both diminished home language environments and child language development. Those symptoms are also associated with perception biases in caregiver-reported assessments of children’s language development. These results underscore the importance of incorporating caregiver mental health considerations into both the design and interpretation of assessments in early development studies. When direct assessments are not feasible, researchers should adjust for potential reporting biases and consider developing correction benchmarks to enhance the accuracy of caregiver-reported data.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

AWC

Adult word count

CDI

MacArthur–Bates Communicative Development Inventory

CTC

Conversational turn count

CVC

Child vocalization count

DASS-21

Depression, Anxiety, and Stress Scale-21

FCI

Family care indicators

LENA

Language environment analysis (LENA) technology

LMICs

Low- and middle-income countries

OLS

Ordinary least squares

Author contributions

**QJ: ** Conceptualization, Investigation, Methodology, Software, Formal Analysis, Writing- Original draft preparation, Writing- Reviewing and Editing. **YQ: ** Writing- Reviewing and Editing, Methodology, Formal Analysis. **TF** : Investigation. **MD: ** Investigation **. HZ: ** Writing- Reviewing and Editing. **EZ: ** Writing- Original draft preparation. **S-ED: ** Writing- Reviewing and Editing. **YM: ** Conceptualization, Data curation. **SR: ** Conceptualization, Writing- Reviewing and Editing.

Funding

This study was supported by private gifts from individual donors. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Data availability

Data and analytic code necessary to reproduce the analyses presented in this paper are available from the first author upon reasonable request. The analyses presented here were not preregistered.

Declarations

Ethics approval and consent to participate

The study was approved by the Stanford University Institutional Review Board (Protocol ID 49552). The study was conducted in accordance with the Declaration of Helsinki. Informed consent was obtained from the legal guardians or parents prior to data collection. Enumerators introduced the study, allowed time for questions, and used a standardized script in both Mandarin and the Sichuan dialect to obtain consent. Caregivers were made aware that their recordings would be collected and reviewed for research purposes. The caregivers were compensated for their participation in the study with books and educational toys.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.