Early childhood is the time when we first make sense of the physical world, forge our first social bonds and learn how to express and read basic human emotions. Normally, it is parents who lead children through these developmental firsts. Parental involvement in the cognitive and socio-emotional development of children is a critical factor in shaping their future well-being and success. Parenting is a fundamental component of child development that significantly influences a childs cognitive, emotional and social growth. As the primary care- givers, parents play a pivotal role in shaping their childrens future outcomes through their parenting practices and behaviors (Bornstein & Putnick, 2012). Effective parenting involves a range of practices that contribute to a childs development, including providing emotional supp-ort, setting appropriate boundaries and creating a nurturing environment for learning and growth (Bradley & Corwyn, 2002). The nature of these practices and their impacts on child development are areas of intense research interest, reflecting the critical role of parenting in the early years of life.

Children who experience responsive and supportive parenting are more likely to excel academically, exhibit better psychological health and develop stronger social skills (Conger et al., 2010). Parental engagement can manifest in numerous ways, from participating in educational activities to offering emotional reassurance during challenging times. These practices help children develop essential cognitive and socioemotional skills that serve as the foundation for future success (Shonkoff & Phillips, 2000). Conversely, inadequate caregiving can hinder optimal child development, particularly among parents who lack the necessary resources, knowledge, commitment or skills to foster their childrens growth and well-being (Bugental & Grusec, 2006). The lack of adequate parental caregiving can adversely affect childrens cognitive and socioemotional development. Research indicates that children who do not receive sufficient parental involvement are at a higher risk of developing be havioral problems, academic difficulties and emotional issues (Shonkoff & Phillips, 2000).

However, while the importance of effective parenting is well-established, there is significant variability in parenting practices across different cultural and socio- economic contexts. For example, in high-income countries, extensive research has explored diverse parenting styles and their effects on child development, revealing that both the quantity and quality of parental involvement are crucial for positive developmental outcomes (NICHD Early Child Care Research Network, 2004). In these contexts, the emphasis has often been on understanding how specific parenting behaviors, such as authoritative parenting, con-tribute to child development (Baumrind, 1991). In contrast, there is a notable gap in the research concerning parenting practices in low and middle income (LAMI) countries, where economic constraints, educational disparities and varying cultural norms can create different challenges and opportunities for parenting (Bornstein & Putnick, 2012). For instance, in many LAMI countries, parents may face significant economic hardships that affect their ability to engage in effective caregiving, which can in turn impact their childrens development (Bradley & Corwyn, 2002). The scarcity of nationally representative data from these regions limits our understanding of how diverse socio-economic and cultural factors influence parenting practices and consequently, child outcomes (Bornstein & Putnick, 2012).

Additionally, much of the existing literature has focused predominantly on maternal practices or has failed to distinguish between the parenting roles of mothers and fathers. Research shows that while mothers and fathers both contribute to their childrens development, they often do so in distinct and complementary ways (NICHD Early Child Care Research Network, 2004). This study seeks to identify and analyze the key determinants that influence parents involvement in both cognitive and socioemotional caregiving. By examining these factors, the research aims to provide insights that can inform policy interventions and programs designed to enhance child development outcomes in Bangladesh. Understanding the barriers and facilitators of effective caregiving will also help in crafting strategies that support parents in fulfilling their crucial role in their childrens lives.

Parental involvement is widely recognized as a critical factor in childrens cognitive and socio-emotional development. Numerous studies have demonstrated that both the quantity and quality of parental engagement significantly impact a childs academic achievement, psychological health and social skills (Henderson & Mapp, 2002). Effective parenting encompasses a range of activities including emotional support, academic encouragement and the establishment of a structured environment for learning (Parke & Clark-Stewart, 2011). 

The influence of socio-economic and cultural factors on parenting practices has been a significant area of research. Bornstein and Putnick, (2012) highlight that cultural norms and economic constraints shape how parenting practices are implemented across different societies. For example, in many low and middle-income countries, economic hardship can limit parents ability to engage in activities that promote cognitive and socioemotional development (Bornstein & Putnick, 2012). 

Socio-economic status (SES) is a critical determinant of parental caregiving. Families with limited financial resources often face numerous stressors that can negatively impact their ability to provide nurturing and supportive environments for their children (Bradley & Corwyn, 2002). Economic hardships can lead to parental depression and anxiety, reducing their emotional availability and responsiveness to their childrens needs (Conger et al., 2010).

Rahman et al. (2023) conducted a cross-sectional study using data from the Multiple Indicator Cluster Survey (MICS) 2019, which included 7,326 rural children aged 3 to 4 years in Bangladesh. Their findings revealed that children aged 4 years were more likely to be developmentally on-track compared to their younger counterparts. The study employed logistic regression analysis to examine the associations between home environment factors and early childhood development (ECD) outcomes. The impact of a childs gender on parental involvement has also been explored in various studies. 

Sultana et al. (2019) analyzed cross-sectional data from the 2014 Bangladesh Demographic Health Survey, focusing on acute respiratory infections (ARIs) in under-five children. Their study, which utilized logistic regression analysis, found that seeking care was significantly higher for female children (AOR = 2.19, 95% CI = 0.94, 5.12) compared to male children. This suggests that gender may play a role in parental caregiving behaviors, although further research is needed to explore this relationship specifically in cognitive and socioemotional caregiving contexts.

Maternal education has consistently emerged as a significant predictor of parental involvement in caregiving. Sun et al. (2016) conducted a comprehensive study using data from the UNICEF 2005 Multiple Indicator Cluster Survey 3, which included 134,290 children aged 0-4 years from 28 low- and middle-income countries. Their analysis, which employed multilevel modeling, found that maternal education was a significant predictor of parents caregiving practices, with large effect sizes observed. The study emphasized the significance of a mothers education in improving parenting practices and child well-being.

Regional variations in parental involvement within Bangladesh have not been extensively studied. However, Haque et al. (2021) conducted a cross-sectional study across eight administrative districts and two city corporation areas in Bangladesh, focusing on parenting stress among caregivers of children with neurodevelopmental disorders. While not directly addressing cognitive and socio-emotional caregiving, this study highlights the potential for regional differences in parenting experiences and practices.

Hamadani et al. (2014) conducted a longitudinal study in Bangladesh, following 2,853 children from birth to 64 months. Using multiple regression analyses, they found that poverty had a significant impact on childrens cognitive development, with a mean cognitive deficit of 1.2 z scores of IQ by 64 months between the first and fifth wealth quintiles. The study revealed that parental education, pre- and postnatal growth and home stimulation mediated 86% of the effects of poverty on IQ, emphasizing the importance of socioeconomic factors in caregiving practices.

Bhattacharyya et al. (2023) conducted a cross-sectional study in urban slums of Bangladesh, focusing on fathers involvement in infant and young child feeding (IYCF) practices. Their study, which used multivariate logistic regression, found that fathers education and occupation were significantly associated with their involvement in IYCF practices. While this study focused on urban slums, it highlights the need for further research comparing urban and rural settings in terms of parental involvement in cognitive and socioemotional caregiving.

Bishwajit et al. (2017) conducted a study on male involvement in reproductive health in Bangladesh, using data from the Bangladesh Demographic and Health Survey. Their logistic regression analysis revealed that exposure to family planning information through newspapers (AOR = 1.952, 95% CI = 1.429-2.664) and television (AOR = 1.514, 95% CI = 1.298-1.886) was significantly associated with active male involvement. While this study focused on reproductive health, it suggests that media exposure may influence parental involvement in various aspects of caregiving (Barce LD., 2025).

The Multiple Indicator Cluster Survey (MICS) is an international household survey program that collects data on a wide range of indicators for monitoring the situation of children and women. We used publicly available data from the Bangladesh Multiple Indicator Cluster Survey (MICS) conducted in 2019. The MICS 2019 Bangladesh survey was the sixth round of the MICS in Bangladesh, and it was conducted by the Bangladesh Bureau of Statistics (BBS) with the technical and financial support of UNICEF and Other organizations, like UNFPA Bangladesh, SURCH, ISRT, and ICDDR,B played a significant role in providing crucial support for the survey.

Sample Design

The Bangladesh Multiple Indicator Cluster Survey (MICS) 2019 employed a robust sample design to ensure the collection of representative and reliable data at both the national and subnational levels. It encompasses various indicators like child health (vaccination, stunting), education (enrollment, literacy), womens empowerment (marriage age, employment), and access to basic necessities like sanitation and water. The data is disaggregated by factors like residence (urban/rural), division, and wealth quintile, enabling analysis for different population subgroups. The Bangladesh Bureau of Statistics (BBS) used the 2011 Bangladesh Population and Housing Census data as the foundation for selecting the sample.

The survey is based on a two-stage stratified cluster sampling design to ensure data collected is representative at the national level, for urban and rural areas, and for eight divisions and sixty-four districts. Primary Sampling Units (PSUs) were selected from urban and rural strata within each district using probability proportional to size (PPS). After a household listing was carried out within the selected enumeration areas, a systematic sample of 20 households was drawn in each sample PSUs. A total of 3,220 PSUs were selected from which 64,400 households were enumerated. Out of the 24,686 children under five years old listed in the household surveys, questionnaires were only completed for 23,099 children.

Limitations of the Study

Firstly, data on maternal and paternal engagement at home were primarily reported by the primary caregiver, usually the mothers. This arrangement could introduce bias into the data, as respondents might find it challenging to accurately report on their partners involvement. Additionally, mothers expectations of paternal parenting roles could influence how they rated actual paternal engagement; for instance, high expectations might lead to harsher ratings if perceived contributions fall short. Future studies would benefit from including multiple respondents who can provide firsthand information about their own involvement to mitigate such biases.

Secondly, the MICS 6 data offered insights into the types of activities that mothers and fathers participated in, but did not provide information on the actual frequency or quality of their engagement. Lastly, the MICS 6 datasets lack personal information on the fathers and are restricted to a limited range of variables, which constrains the scope of the analysis.

Multilevel Analysis and Model Fitting

In this study, 2-level modeling was used to deal with the hierarchical clustered nature of the MICS data, where children are nested within mothers. Our response variable, parental caregiving, has more than two categories (No involvement, moderate involvement, and high involvement,) and we want to estimate the effects of child-level and family-level factors on this outcome variable. For this purpose, we will fit a 2-level multinomial logistic regression, which includes both random effects (random intercepts only) and fixed effects models.

Multilevel models, also known by various names such as hierarchical linear models, nested data models, mixed models, random coefficient models, random-effects models, random parameter models, or split-plot designs, are statistical models that account for parameters varying at multiple levels. These models are especially suitable for research designs where data are organized hierarchically (i.e., nested data). Typically, the units of analysis are individuals (at a lower level) nested within contextual or aggregate units (at a higher level). While the lowest level often involves individual data, multilevel models can also handle repeated measurements of individuals. Therefore, they offer an alternative method for univariate or multivariate analysis of repeated measures. This concept is detailed in Gelman and Hills ”Data Analysis Using Regression and Multilevel/ Hierarchical Models” (2006).

A multilevel model can be viewed as a regression (linear or generalized linear) where the regression coefficients are assigned a probability model. This second-level model has its own parameters, known as hyperparameters, which are also estimated from the data. The two key components of a multilevel model are the varying coefficients and a model for those varying coefficients, which may include group-level predictors. While classical regression can sometimes handle varying coefficients using indicator variables, multilevel models stand out by explicitly modeling the variation between groups. To provide a preview of the notation, the regression equations for two multi-level models can be presented. For simplicity, we will consider one cluster-level (community-level) predictor x and one individual-level predictor.

Varying-intercept Model

First, the model is written in which the regressions have the same slope in each of the clusters and only the intercepts vary. The notation i for individual and j[i] for the cluster j containing individual i.

Here, xi and uj represent predictors at the individual and community levels, respectively, and i and ηj are independent error terms at each of the two levels. The model can be written in several other equivalent ways. The number of ”data points” J (here, cluster) in the higher-level regression is typically much less than n, the sample size of the lower-level model (for an individual in this study).

Varying-intercept, Varying-slope Model

More complicated is the model where intercepts and slopes both can vary by cluster,

Compared to the varying-intercept model, this has twice as many vectors of varying coefficients (α, β), twice as many vectors of second-level coefficients (a, b), and potentially correlated second-level errors η1, η2.

The 2-Level Model

Consider first a simple model for one cluster, relating y to x. We may write,

where standard interpretations can be given to the intercept (α), slope (β), and residual ϵ. We follow the normal convention of using Greek letters for the regression coefficients and place a circumflex over any coefficient (parameter) that is a sample estimate. To describe simultaneously the relationships for several clusters, we write, for cluster j,

This is now the formal model where j refers to the level 2 unit and i to the level 1 unit. As it stands, (2.4) is still essentially a single-level model, albeit describing a separate relationship for each cluster. In some situations, for example, where there are few clusters and interest centers on just those clusters in the sample, one may analyse (3.4) by fitting all the 2n+1 parameters, namely

assuming a common within-cluster residual variance and separate lines for each cluster. 

To make (3.4) into a genuine 2-level model, we let αj, βj become random variables. For consistency of notation, replace αj by βj0 and βj by β1j, and assume that

where u0j and u1j are random variables with parameters.

We can now write (*) in the form -

Estimation for Multilevel Model

Iterative Generalised Least Squares (IGLS) method is used to estimate the multilevel model. We consider the two-level variance components model

with m level 2 units and nj level 1 units in the j-th level 2 unit. When the residuals have normal distributions, it also yields maximum likelihood estimates. This estimation procedure is iterative. We would usually start from reasonable estimates of the fixed parameters. 

Single-level Multinomial Logistic Regression Model

Multinomial logistic regression is an extension of logistic regression that allows for more than two categories of the dependent or outcome variable. It is used when the dependent variable is nominal and has more than two levels. Lets assume the dependent variable Y has C categories (c = 1, 2, ..., C). The model will predict the probability of Y being in each category given a set of independent variables x. The probability of being in the cth outcome category P (Y = c) is πc (c = 1, 2, ..., C). For individual i, then, the probability of being in category c (c = 1, 2, ..., C − 1) versus the reference group (C) can be defined as follows,

The multinomial logistic regression model used to predict the odds of individual i being in outcome category c relative to outcome C (reference category) using the set of q predictors is given by,

Two-level Multinomial Logistic Regression Model

Heck et al. (2013) extended the single-level model to consider the nesting of individuals within groups. At Level 1, then, the multinomial logistic regression used to predict the odds of individual i in group j being in outcome category c relative to outcome C (reference category) using the set of q predictors is given by,

At Level 1, there is no separate residual variance term because the variance is dependent upon the mean. A more general model is summarized as,

 

At Level 2, we have the following model,

At Level 2, we can model one or more Level-1 intercepts or slopes as a function of a set of Level-2 predictors. At Level 2, we can model one or more Level-1 intercepts or slopes as a function of a set of Level-2 predictors (W) and corresponding variance terms (uqj).

Unconditional 2-level Model

We can start by estimating an unconditional (no predictors) model to examine the extent of variability of the nominal outcome across households. As we have developed, the unconditional model at Level 1 will have C − 1 estimates for individual i in group j as follows,

At Level 2, the combined set of models suggests that the intercepts vary between groups, and the general form is,

Estimation Procedure

When outcomes are categorical and therefore expected values result from probability distributions other than normal, nonlinear link functions are required. Model estimation requires an iterative computational procedure to estimate the parameters optimally. Maximum likelihood (ML) estimation methods are most often used for generalized linear models and also for multilevel models. ML determines the optimal population values for parameters in a model that maximize the probability or likelihood function - that is, the function that gives the probability of finding the observed sample data, given the current parameter estimates (Hox et al., 2017). Because the likelihood or probability can vary from 0 to 1, minimizing this discrepancy function amounts to maximizing the likelihood of the observed data.

For single-level categorical analyses, ML estimation using Newton-Raphson and Fisher scoring methods is applied, which are among the most efficient and widely used estimation methods for categorical outcomes. For multilevel models with categorical outcomes, model estimation becomes more difficult. Estimation typically relies on quasi-likelihood approaches, which approximate the nonlinear functions used by a nearly linear transformation. This complicates the solving of complex nonlinear mathematical equations representing the relationships among variables in the model (Hox et al., 2017).

Computing Predicted Probabilities

When there are more than two categories for the dependent variable, computing probabilities is a little more complicated than it is for a dichotomous outcome. For a dependent variable with C categories, we must calculate C−1 log odds equations. To calculate the probability for each category, we take each one of the C-1 log odds computed and then exponentiate it. Once the log odds are exponentiated, we simply divide each by the sum of the odds to obtain the probability for each category. For the C − 1 contrasts, beginning with c = 1 to C − 1, the computation of predicted probabilities is given by,

Where ηcij is the value of the linear component for specific values of the predictors (if there are any in the model). The probability of the reference category is

Since the reference category has the value of ηCij = 0 and exp (0) = 1; the log odds are therefore 0 [log(1) = 0].

Intra-cluster Correlation (ICC)

Intra-cluster correlation measures how similar the responses are that belong to the same cluster (Eldridge et al., 2009). For the sake of convenience, the variance of 

Logistic regression has been considered as 

The formula of intra-cluster correlation is, 

First of all, the nature of MICS data is hierarchical and clustered structure, which means children are nested within mothers and mothers are nested within house- holds. According to this scenario, children within a household can be more similar to one another than children from other households. We used two-level multinomial logistic regression model to test the fixed effects and random effects of child- level and family-level factors on maternal and paternal caregiving. We consider covariates to model fitting that are found to be significant in the bivariate analysis. Four different nested models were fitted. The models were null model (containing only the outcome variable), model 1 (a model fitted using child-level variables only), model 2 (fitted using family-level variables) and model 3 (fitted using both child-level and family-level variables). We presented the random effects and model statistics table for all the models. The intra-cluster correlation coefficient (ICC) was calculated for the measures of variation (random effects). The model with the lowest AIC value (model 2 for both dependent variables) was selected as our final model. We then presented the fixed effects table only for the final models. The adjusted Relative Risk Ratio (RRR) was reported and variables with p-value <0.05, in the multivariable analysis, were declared to be significant predictors of parental caregiving.

Results for Maternal Caregiving

Random Effects and Model Fit Statistics Comparison

For maternal caregiving, the AIC values for null model, model 1 (fitted using child- level variables only), model 2 (fitted using family-level variables) and model 3 (fit- ted using both child-level and family-level variables) are 209532.930, 196979.839, 172796.718 and 208945.722, respectively from the result of Table 1. It is seen that model 2 has lowest AIC value. It is well established that the model with lower AIC value is better than other model. So it is good to take family-level characteristics to figure out factors which are associated with maternal caregiving.

Table 1: Random intercept and model fit statistics comparison of multilevel multinomial logistic regression model.

Based on the results of the null model, the ICC value for mothers moderate involvement in caregiving compared to no involvement is 0.034. This indicates that approximately 3.4% of the total variance in caregiving in Bangladesh is attributable to the families in which the mothers live. Similarly, for mothers high involvement in caregiving versus no involvement, the ICC value is 0.102, which can be interpreted in the same manner. The ICC values estimated for model 1 are 0.061 for moderate involvement in caregiving activities and 0.140 for high involvement in caregiving activities, respectively which are higher than the null model. The estimated ICC value of 0.061 indicates that 6.1% of the total variance in caregiving in Bangladesh can be attributed to the families in which the mothers reside while examining child-level characteristics. The interpretation of the other ICC value is similar. The ICC values for Model 2 and Model 3 can be interpreted similarly, indicating that a comparable proportion of the total variance in caregiving is attributable to the families in which the mothers reside.

Fixed Effects Estimates for Selected Model (Model 2)

In this section, fixed effects are considered only for family-level factors. Table  2 and 3 show the result for this model. Only those variables that were found to be significant in the bivariate analysis are included and thus, the following discussion will focus on these family-level variables. From Table 2, it is found that mothers level of education appears to have a nuanced impact on their involvement in caregiving. Specifically, those with primary education exhibit a 1.147 times increased likelihood of being moderately involved in caregiving compared to mothers with pre-primary or no education, with this difference being statistically significant (p-value = 0.016). Furthermore, mothers with secondary education are 1.333 times more likely to engage moderately in care-giving compared to their counterparts with pre-primary or no education, a finding that is highly significant (p-value <0.001). In contrast, mothers with higher secondary education or above show a slightly decreased likelihood (0.967 times) of being moderately involved in caregiving relative to those with pre-primary or no education, though there is no statistical difference between mothers with higher secondary or above education and pre-primary or no education (p-value = 0.628).

Table 2: Fixed effects estimates of the predictors of mothers moderate involvement in cognitive and socioemotional caregiving obtained from model 2.

Mothers from middle-class families are 0.954 times less likely to be moderately involved in caregiving compared to the poor mothers, though this difference is not statistically significant (p-value = 0.288). Similarly, mothers from rich households are 0.946 times less likely to engage moderately in caregiving compared to their counterparts from poor households, with this result also lacking statistical significance (p-value = 0.227). Children residing in rural areas are slightly more likely (RRR = 1.109) to get moderate caregiving from their mothers compared to those living in urban areas (reference category) and this result is statistically significant (p-value = 0.020).

Mothers in various divisions show differing probabilities of moderate involvement in caregiving compared to those in Barisal. Mothers residing in Sylhet (RRR = 0.454) have the most significant reduction in moderate involvement compared to the reference group (no involvement), with this result being highly significant (p- value = 0.000). Similarly, mothers in Chittagong (RRR = 0.825), Khulna (RRR = 0.675), Mymensingh (RRR = 0.834), Dhaka (0.545) and Rangpur (RRR = 0.779) are also less likely to be involved in caregiving activities, though to a lesser extent than Sylhet. In contrast, mothers in Rajshahi (RRR = 1.010) are somewhat more likely to be involved in caregiving compared to those in Barisal; however, this association is not statistically significant (p-value = 0.885).

Mothers with exposure to media are 1.091 times more likely to be involved moderately in caregiving activities in last three days compared to those with no exposure and this is statistically significant (p-value = 0.015). Mothers aged 20-24 years are substantially more inclined (3.072 times) to engage moderately in caregiving compared to their counterparts aged 15-19 years (reference category) and this association is highly significant (p-value <0.001). Similarly, mothers in all other age groups exhibit higher RRR values compared to young mothers (age group 15-19), with these differences being statistically significant (p-value = 0.000). However, mothers in age group 45-49 has lower likelihood compared to younger age groups and this is also statistically significant (p-value <0.001).

Table 3: Fixed effects estimates of the predictors of mothers high involvement in cognitive and socioemotional caregiving obtained from model 2.

From Table 3, it is found that mothers with higher levels of education are more likely to be involved in caregiving compared to mothers with pre-primary education or none. Specifically, those with primary education are 1.541 times more likely to be highly involved in caregiving compared to mothers with pre-primary or no education, with this result being statistically significant (p-value <0.001). Mothers with secondary education show an even greater likelihood, being 2.309 times more likely to exhibit high levels of caregiving involvement compared to their less educated counterparts, a difference that is also highly significant (p-value = 0.000). Furthermore, mothers with higher secondary education or above are 2.654 times more likely to engage highly in caregiving practices compared to those with pre-primary or no education and this association is similarly highly significant (p- value = 0.000). The table shows that mothers from wealthier families are more likely to be fully involved in caregiving activities relative to no involvement (reference category) compared to mothers from poorer families. Those from middle-class families are 1.048 times more likely to be highly involved in caregiving compared to mother from poor families, though this difference lacks statistical significance (p-value = 0.401). In contrast, mothers from rich households are 1.332 times more likely to demonstrate a high level of caregiving involvement compared to poor mothers and this association is statistically significant (p-value <0.001). 

Mothers residing in rural areas are 0.872 times less likely to be highly involved in caregiving compared to those living in urban areas, with this association being statistically significant (p-value = 0.006). Mothers in Chittagong are 0.902 times less likely to be highly involved in care- giving compared to those in Barisal, though this result is not statistically significant (p-value = 0.282). In contrast, mothers in Khulna are 1.789 times more likely to be highly engaged in caregiving, a difference that is highly significant (p-value <0.001). Similarly, mothers in Mymensingh, Dhaka, Rajshahi, Rangpur and Sylhet are more likely to be highly involved in caregiving activities compared to mothers with no involvement., with these findings also being statistically significant except for Dhaka division. Rangpur shows the highest likelihood, with mothers being 2.021 times more likely to engage highly in caregiving. Mothers who have media exposure are 1.215 times more likely to engage in caregiving activities in the last three days compared to those without media exposure. This association is statistically significant (p-value <0.001).

The analysis of maternal age reveals significant variations in caregiving involvement. Mothers aged 20-24 years exhibit a significantly higher likelihood (3.021 times) of being highly involved in caregiving compared to those aged 15-19 years and this association is highly significant (p-value = 0.000). Similarly, the RRR values range from 3.419 for the 25-29 age group to 3.632 for the 40-44 age group. In contrast, mothers aged 45-49 years are 1.439 times more likely to be highly involved in caregiving compared to those aged 15-19 years, but this difference is not statistically significant (p-value = 0.210).

Results for Paternal Caregiving

Random Effects and Model Fit Statistics Comparison

For paternal caregiving, the AIC values for null model, model 1 (fitted using child- level variables only), model 2 (fitted using family-level variables) and model 3 (fit- ted using both child-level and family-level variables) are 209532.930, 196979.839, 172796.718 and 208945.722, respectively from the result of Table 4. It is seen that model 2 has lowest AIC value. It is well established that the model with lower AIC value is better than other model. So it is good to take family-level characteristics to figure out factors which are associated with paternal caregiving.

Table 4: Random intercept and model fit statistics comparison of multilevel multinomial logistic regression model.

Based on the results of the null model, the ICC value for fathers moderate involvement in caregiving compared to no involvement is 0.070. This indicates that approximately 7% of the total variance in caregiving in Bangladesh is attributeable to the families in which the fathers live. Similarly, for fathers high involvement in caregiving versus no involvement, the ICC value is 0.191, which can be interpreted in the same manner.

The ICC values estimated for model 1 are 0.098 for moderate involvement in caregiving activities and 0.204 for high involvement in caregiving activities, respectively which are higher than the null model.  The estimated ICC value of 0.098 indicates that 9.8% of the total variance in caregiving in Bangladesh can be attributed to the families in which the fathers reside while examining child-level characteristics. The interpretation of the other ICC value is similar.

The ICC values for Model 2 and Model 3 can be interpreted similarly, indicating that a comparable proportion of the total variance in caregiving is attributable to the families in which the fathers reside.

Fixed Effects Estimates for Selected Model (Model 2)

In this section, fixed effects are considered only for family-level factors. Table 5 and 6 show the result for this model. As we have considered only those variables that were found to be significant in the bivariate analysis, the following discussion will be based on those variables.

The Table 5 of paternal involvement in caregiving in relation to maternal education levels reveals minimal and statistically insignificant effects. Fathers whose partners have attained primary education are only slightly more likely (RRR = 1.004) to be moderately involved in caregiving compared to those whose partners have pre-primary or no education; however, this effect is not statistically significant (p-value = 0.947). A similar trend is observed for secondary education, where fathers show a modest increase in caregiving involvement (RRR = 1.072), though this difference is also not significant (p-value = 0.254). Fathers with partners having higher secondary or above education show almost no change in likelihood compared to those with pre-primary or no education and this is not statistically significant (p-value = 0.979).

Table 5: Fixed effects estimates of the predictors of fathers moderate involvement in cognitive and socioemotional caregiving obtained from model 2.

The analysis of paternal involvement in caregiving across different income levels indicates a statistically significant decrease in involvement among fathers from wealthier households. Fathers from middle-income households are 0.869 times less likely (RRR = 0.869) to be moderately involved in caregiving compared to those from poor households and this difference is statistically significant (p-value = 0.004). Similarly, fathers from rich households are 0.873 times less likely (RRR = 0.873) to be moderately involved in caregiving compared to their counterparts from poor households, with this difference also being statistically significant (p- value = 0.005).

Fathers living in rural areas are 0.831 times less likely to be moderately involved in caregiving compared to those in urban areas. This result is statistically significant (p-value <0.001). This can be interpreted as a notable reduction in paternal caregiving involvement in rural settings. Fathers in Chittagong (RRR = 0.679), Khulna (RRR = 0.782), Mymensingh (RRR = 0.585), Dhaka (RRR = 0.554) and Sylhet (RRR = 0.638) are significantly less likely to be moderately involved in caregiving compared to those in Barisal, with Mymensingh and Dhaka showing the greatest reductions in involvement. In contrast, fathers in Rajshahi (RRR = 1.240) are 1.240 times more likely to be moderately involved compared to those in Barisal. No significant difference in caregiving involvement is observed between Rangpur and Barisal (RRR = 0.976; p-value = 0.729).

Fathers whose partners are exposed to media are 1.119 times more likely to be moderately involved in caregiving compared to those whose partners are not exposed. This association is statistically significant (p-value = 0.003). The analysis shows that fathers are more likely to be moderately involved in caregiving as their partners age increases, with the likelihood peaking among those whose partners are aged 35-39 years (RRR = 3.255) and 40-44 years (RRR = 3.340). Fathers with partners aged 20-24 years (RRR = 2.595) and 25-29 years (RRR = 2.787) also show a significant increase in involvement compared to those whose partners are of age group 15-19. Although the likelihood decreases for fathers whose partners are aged 45-49 years (RRR = 2.327), it remains substantially higher than the reference category. All the results are statistically significant.

Table 6: Fixed effects estimates of the predictors of fathers high involvement in cognitive and socioemotional caregiving obtained from model 2.

From Table 6, it is observed that fathers are more likely to be involved in care-giving who have partners with higher levels of education compared to those with pre-primary education or none. Fathers whose partners have a secondary education are significantly more likely (1.414 times) to be highly involved in caregiving compared to those whose partners have pre-primary or no education (p-value = 0.007). This likelihood further increases for fathers whose partners have higher secondary or above education, with a 1.912 times higher likelihood of being highly involved (RRR = 1.912; p-value <0.001). Although fathers with partners who have a primary level of education show a slight increase in caregiving involvement (RRR = 1.099), this association is not statistically significant (p-value = 0.478).

The table shows that fathers from wealthier families tend to be involved in caregiving activities relative to not involving in caregiving activities (reference category) compared to fathers from poor families. Fathers from rich families are 1.491 times more likely to be highly involved in caregiving compared to those from poor families and this difference is statistically significant (RRR = 1.491; p-value <0.001). However, fathers from middle-income families also show a higher likelihood of caregiving involvement (RRR = 1.128), though this effect is not statistically significant (p-value = 0.249).

Fathers living in rural areas are 0.825 times less likely to be highly involved in caregiving activities compared to those in urban areas, which is statistically significant (p-value = 0.035). This can be interpreted as a notable reduction in paternal caregiving involvement in rural settings.

Fathers high involvement in caregiving activities in the previous three days relative to no involvement in caregiving shows disparity in terms of divisions. Fathers in Khulna, Dhaka, Rangpur and Sylhet are significantly more likely to be highly involved in caregiving compared to those in Barisal, with the highest involvement observed in Dhaka (RRR = 5.961) and Rangpur (RRR = 6.643), both showing highly significant differences (p-value = 0.000). While fathers in Rajshahi are 1.648 times more likely to be highly involved (RRR = 1.648; p-value = 0.022) and those in Sylhet are 1.687 times more likely (RRR = 1.687; p-value = 0.032), these results are also statistically significant. In contrast, fathers in Chittagong and Mymensingh show a slight increase in involvement compared to Barisal, but these differences are not statistically significant.

Fathers are 0.984 times less likely to be highly involved in caregiving practices for those children whose mothers are exposed to media compared to those who are not exposed but this is not statistically significant since the p-value is >0.05. Mothers media exposure and non-exposure has no significant difference for fathers moderate engagement in caregiving activities during the previous three days relative to no involvement.

Fathers high engagement in caregiving activities have significant difference across mothers different age groups (p-value <0.05). The analysis indicates that fathers are significantly more likely to be highly involved in caregiving when their partners are older than 15-19 years. Fathers with partners aged 20-24 years (RRR = 2.779), 25-29 years (RRR = 2.887), 30-34 years (RRR = 2.977) and 35-39 years (RRR = 2.806) are nearly three times more likely to be highly engaged in caregiving activities. Although the likelihood of high involvement slightly decreases for fathers with partners aged 40-44 years (RRR = 2.436) and 45-49 years (RRR = 2.399), they still exhibit a significantly higher likelihood of involvement compared to the youngest age group.

Discussion of the Findings on Maternal Caregiving

According to multilevel multinomial logistic regression model it is noticed that higher levels of maternal education are strongly associated with greater maternal involvement in caregiving. Mothers with secondary or higher education are particularly more likely to participate in caregiving practices compared to those with no or pre-primary education. Numerous studies have consistently identified maternal education as key to quality maternal parenting (Walker et al., 2011).

According to the analysis of wealth index it is found that economic status positively influences maternal caregiving involvement. For mothers with moderate engagement in caregiving activities, involvement tends to decrease with increasing wealth. Conversely, mothers with high engagement in caregiving activities are more likely to participate in multiple caregiving activities if they come from wealthier households, compared to those from poorer households. This finding is consistent with the findings of the research conducted by Cuartas et al. (2020).

Mothers living in rural areas exhibit greater involvement in caregiving activities when it comes to moderate engagement in caregiving practices. In contrast, mothers living in urban areas display higher levels of involvement in caregiving concerning high engagement. Armistead et al. (2002) got similar result in their study. Armistead et al. (2002) reported findings in their study that were similar to the results observed in this research. This suggests that geographical location within rural or urban settings substantially impacts the likelihood of maternal caregiving involvement.

Regional variations significantly impact maternal involvement in caregiving activities. Mothers in most divisions are less likely to engage moderately in caregiving activities compared to those in Barisal. However, mothers in Rajshahi show a higher likelihood of engaging in caregiving activities compared to the reference category. In terms of high involvement, mothers across all divisions, with the exception of Chittagong, are more likely to participate highly in caregiving activities.

Mothers who have media exposure are more inclined to participate actively in caregiving activities, suggesting that access to information and increased awareness positively influences caregiving practices. From this study it is observed that maternal age is a strong predictor, with mothers aged between 20-44 years being the most involved in caregiving activities. Younger mothers (15-19 years) and older mothers (45-49 years) show lower involvement, with older mothers particularly less inclined to participate in more caregiving activities.

Discussion of the Findings on Paternal Caregiving

Mothers education level shows a nuanced impact on paternal caregiving. While higher levels of maternal education are associated with increased paternal caregiving, the statistical significance varies across different levels of education. Higher maternal education generally correlates with more paternal involvement in caregiving activities, but this relationship is only statistically significant for the highest education levels (e.g., secondary and higher secondary education) for fathers high involvement in caregiving activities.

Economic factors also play a significant role in paternal caregiving. Fathers from wealthier households are less likely to be moderately involved in caregiving compared to those from poorer households. This inverse relationship suggests that as household wealth increases, the direct involvement of fathers in caregiving activities decreases, which might be associated with the availability of alternative caregiving resources or shifts in caregiving responsibilities. On the other hand, wealthier fathers, especially those in rich households, are more likely to be highly involved in caregiving due to better access to resources, reduced financial stress, greater work flexibility and higher awareness of child development.

The geographical context significantly affects paternal caregiving practices. Urban fathers are more likely to be involved in caregiving due to better access to resources and services, higher income levels, more flexible work arrangements and greater emphasis on gender equality in urban settings. These factors make it easier for them to balance work and family responsibilities.

Regional differences reveal significant variability in paternal caregiving practices. Fathers in Barisal, Rajshahi and Rangpur are more likely to be moderately involved in caregiving compared to those in other divisions. Regarding high involvement, significant regional variation exists, with the highest levels observed in Dhaka and Rangpur. Barisal is used as the reference point for comparison, revealing that many other regions exhibit significantly greater levels of involvement. These regional disparities underscore the impact of local cultural, economic and social norms on paternal caregiving behaviors.

Mothers media exposure does not show a consistent and statistically significant impact on paternal caregiving across all analyses. While there are some indications of increased paternal involvement associated with media exposure, these results are not always statistically significant, suggesting that media exposure alone may not be a strong determinant of paternal caregiving behavior.

Mothers age is also a significant factor for fathers involvement in cognitive and socioemotional activities with children. Fathers are more likely to be moderately involved when their partners are aged 20-44 years, with the highest likelihood observed in the 35-44 years age range. Additionally, fathers are more likely to be highly involved if their partners are aged 20-39 years, with peak involvement in the 30-34 years age group. While the likelihood of high involvement slightly decreases for older age groups, it remains significantly higher compared to the reference category.

Comparative Insights on Maternal and Paternal Caregiving

The analysis of factors associated with both maternal and paternal engagement in cognitive and socio-emotional caregiving reveals distinct patterns and influences. Higher maternal education consistently increases maternal engagement in caregiving. This is likely due to the greater awareness and understanding of the importance of early childhood development among more educated mothers. Similarly, fathers are more likely to be involved in caregiving when the mother is better educated, particularly at the secondary level or higher. This suggests that maternal education not only influences the mothers own caregiving practices but also encourages greater paternal participation.

Economic status positively impacts maternal engagement, with wealthier mothers more likely to engage in caregiving activities. This could be due to the availability of resources and the ability to afford time for caregiving. Fathers from wealthier households are also more involved in caregiving. However, the impact of economic status on paternal engagement is more pronounced compared to maternal engagement, suggesting that financial stability might be a more critical factor for fathers.

Urban mothers are generally more involved in caregiving than rural mothers, likely due to better access to resources and information about child development. The same pattern is observed for fathers, with urban fathers showing higher engagement than their rural counterparts. This highlights the broader urban-rural divide in caregiving practices, driven by differences in lifestyle, access to education and cultural norms.

Regional variations in maternal engagement are significant, with certain regions displaying higher levels of caregiving than others, likely influenced by local cultural norms, economic conditions and access to resources. Paternal engagement also varies significantly across regions, with fathers in regions like Dhaka and Rangpur being the most involved. The regional patterns of engagement are similar for both mothers and fathers, indicating that regional culture and socioeconomic conditions impact both parents.

Younger mothers, particularly those in their 20s and 30s, are more involved in caregiving. This could be due to a combination of physical energy, cultural expectations and the timing of childbearing. Fathers tend to be more involved when their partners are in the 20-39 age range. This suggests that the age and life stage of the mother significantly influence paternal engagement, potentially due to the dynamics of family life and responsibilities at different stages.

Media exposure plays a role in increasing maternal engagement by providing mothers with information and resources on child-rearing practices. Notably, mothers media exposure does not substantially affect fathers involvement, suggesting that direct exposure to media and information may have a greater influence on mothers than on fathers in the context of caregiving.

Both maternal and paternal engagement in caregiving are influenced by maternal education, economic status and place of residence. This highlights the interconnected nature of parental roles, where factors that empower or constrain one parent can have ripple effects on the other. However, paternal engagement appears to be more sensitive to economic status and regional differences, while maternal engagement is more directly influenced by maternal age and media exposure. This suggests that while both parents are influenced by broader socio-economic and cultural factors, the pathways through which these influences operate may differ.

The family, particularly the parents, often provides a childs first learning experiences. As a result, the caregiving practices they adopt have a significant impact and warrant closer examination, especially in families vulnerable to various environmental risk factors. This study explores how specific child- and family-level factors are associated with maternal and paternal engagement in cognitive and socio-emotional caregiving for young children in Bangladesh. These findings help us better understand and improve parental caregiving practices in LAMI countries like Bangladesh. Educational opportunities and support should be provided for mothers to pursue higher education. Also, it is important for well-educated mothers to live in a society that is supportive of quality parenting. Regional programs should be developed that cater to the specific cultural and social needs of different areas. These programs should be informed by regional data and tailored to meet the diverse needs of different communities. It is important to create specialized media literacy programs for mothers that emphasize how to use media resources effectively for parenting and accessing educational content. Additionally, there should be a focus on promoting the creation of high-quality, accessible parenting content through various media channels. Targeted support services should be provided for older mothers to help them balance caregiving with other responsibilities. Additionally, resources should be allocated to fund parenting education programs specifically designed for younger mothers, focusing on enhancing caregiving skills and providing support. Policies and practices should be promoted that encourage both parents, regardless of age, to share caregiving responsibilities equally.

Both the authors contributed in conceptualizing and designing the study. M.R.D.: contributed in data analysis and report writing. M.A.U.: supervised the research and contributed in preparing article from the study.

First and foremost, the authors are grateful to Almighty Allah. The authors are also thankful to anonymous reviewers and editors for their helpful comments and suggestions.

The author declares no conflict of interest.