The Inverted U-Shaped Relationship Between Digital Literacy and Household Carbon Emissions: Empirical Evidence from China’s CFPS Microdata

Abstract

In the context of China’s dual-carbon agenda and the Digital China initiative, elucidating the role of digital literacy in shaping consumption-based household carbon emissions (HCE) is essential for advancing low-carbon urban living and supporting a broader green transition. Existing research has rarely examined, at the individual level, how digital capability shapes household consumption decisions and the structure of carbon emissions. Accordingly, this study draws on matched household-individual microdata from the China Family Panel Studies (CFPS). We employ a two-way fixed effects model, kernel density analysis, and qualitative comparative analysis. We test the nonlinear effect of digital literacy on household consumption-related carbon emissions and examine its heterogeneity. We also examined the mediating role of perceived environmental pressure, social trust and income level. The research results show that: (1) The net impact of digital literacy on carbon emissions related to household consumption shows an inverted U-shaped curve, rising first and then falling. When digital literacy is low, it mainly increases emissions by expanding consumption channels, reducing transaction costs and improving convenience. Once digital literacy exceeds a certain threshold, the mechanism will gradually turn to optimize the consumption structure, so as to support the low-carbon transformation of individuals. (2) The impact of digital literacy on HCE is structurally different in different types of consumption. In terms of transportation and communication expenditure, the emission reduction effect is the most significant, and with the improvement in digital literacy, this effect will become more and more obvious. For housing-related consumption, the turning point appeared the earliest. With the improvement in digital literacy, its effect will enter the emission reduction stage faster. (3) Digital literacy can reduce carbon emissions related to household consumption by enhancing residents’ perception of environmental pressure and strengthening social trust. However, it may also increase emissions by increasing residents’ incomes, because it will expand the scale of consumption, which will lead to an increase in carbon emissions related to household consumption. (4) The heterogeneity analysis shows that as digital literacy improves, carbon emissions increase more strongly among rural residents, people with low human capital, low-income households, and women. However, the turning-point threshold for emission reduction is relatively lower for women and rural residents. (5) Low-carbon transitions in household consumption are shaped by dynamic interactions among multiple factors, and multiple pathways can coexist. Digital literacy can work with environmental responsibility to endogenously promote low-carbon consumption behavior. It can also, under well-developed infrastructure, empower households and amplify the emission-reduction effects of technology.

1. Introduction

For a long time, climate governance has focused mainly on the production side. Many studies argue that the greenhouse effect driven by carbon emissions is primarily caused by production activities. This has often led to the neglect and underestimation of the ecological impacts of consumption-based carbon emissions [1]. In fact, global warming is closely linked to consumers’ consumption activities [2]. The United Nations Environment Programme’s Emissions Gap Report 2023 [3] indicates that emissions attributable to household consumption contribute roughly two-thirds of total global carbon emissions. In contrast to production-oriented emissions, consumption-based carbon emissions are largely driven by the decisions and daily practices of vast numbers of households and individuals. They are typically small at the individual level, widely dispersed, and highly heterogeneous. Moreover, as living standards and consumption levels rise, these emissions continue to grow [4]. Shifting consumption behaviors toward low-carbon patterns has the potential to cut emissions by approximately 40–70% over the coming three decades [5]. These points suggest that a low-carbon transition in household consumption will be a major new source of momentum for the broader green transformation of the economy and society. The resolution adopted at the Third Plenary Session of the 20th CPC Central Committee underscored the need to strengthen incentive mechanisms that encourage green consumption. The 2025 Government Work Report [6] further emphasized the need to “foster the development of environmentally sustainable, low-carbon patterns of production and everyday living” At present, China’s green consumption market is taking shape, but it still faces challenges such as the “easy to know, hard to do” gap among consumers and the problem that green products are “well received in principle but weak in actual sales.” How to turn green consumption from a socially promoted idea into a widely internalized and self-driven practice and how to build a multi-dimensional, coordinated, and incentive-compatible institutional system to drive a green and low-carbon transition across society have become core issues for advancing urban sustainability.

With the rising wave of a new round of scientific and technological innovation and industrial upgrading, the digital economy supported by next-generation network infrastructure, big data, artificial intelligence and platform-based business models has increasingly become the core driving force for China’s high-quality economic development [7]. Even if the digital economy continues to release consumption dividends, it also creates and strengthens low-carbon dividends, providing a new technical foundation for the transformation to green consumption and sustainable urban development [8]. By June 2025, the number of Internet users in China has increased to about 1.123 billion, and the national Internet penetration rate has reached about 80%. This shows that the digital economy is increasingly closely linked to the daily life of residents and the operation of urban governance. In this case, digital literacy has evolved into a series of basic abilities that enable individuals to acquire, understand, evaluate and generate information in a digital environment. It is increasingly regarded as the micro-foundation of consumer preferences, behavior patterns and environmental awareness [9]. This study defines digital literacy as a multi-dimensional capability system covering six dimensions: acquisition of digital devices, online communication and sharing, use of digital scenarios, cognition of digital formats, acquisition of digital information and mobile payment. Its indicators include the use of the Internet and smart devices, the frequency of online social interaction and online consumption, and the subjective evaluation of respondents on the importance of digital tools in work, learning and daily life. Stronger digital literacy enables residents to use digital tools and information platforms more efficiently, thus alleviating the long-standing information asymmetry in the green product and service market. Moreover, greater transparency of environmental information, coupled with online public supervision, can promote residents to internalize green social norms and strengthen low-carbon values [10].

Nevertheless, digital literacy does not affect the consumption carbon emissions of families in isolation. On the contrary, it is embedded in a specific regional socio-economic and policy environment. Uneven regional development will affect the coverage of digital infrastructure, the availability of green technologies, and the ability of households to pay. Access to clean energy directly restricts the decarbonization potential of household energy use. In addition, the strictness and design of local environmental policies will change the costs and benefits of behavioral change by changing incentives and constraints. These macro-level structural conditions can interact with digital literacy and jointly determine the way and effect of HCE. Thus, the analysis of the relationship between digital literacy and HCE should clearly take into account the differences caused by different external environments and constraints. In this study, digital literacy is defined as the core ability of residents to understand and apply digital technology. It aims to clarify how digital literacy can systematically affect HCE and urban emission reduction by enhancing environmental awareness and reshaping the consumption structure [11]. This is of important practical significance and policy guidance for improving urban governance capabilities and strengthening the policy framework to support green consumption.

Digital literacy is not a one-way tool to reduce carbon emissions related to consumption. On the contrary, it is like a typical double-edged sword, with both “green promotion effect” and “high carbon induction effect” [12]. High digital literacy improves the efficiency of information acquisition and the convenience of consumption. This may stimulate new consumer demand and expand digital consumption scenarios, thus increasing household energy use and carbon emissions. High digital literacy may enhance personal environmental awareness and willingness to buy green products. It can also promote the improvement in energy efficiency and the transition to low-carbon behavior [13,14]. This double impact means that the relationship between digital literacy and HCE is unlikely to be a simple linear relationship; on the contrary, it may present more complex and obviously nonlinear dynamic changes [15]. Against this background, how does digital literacy affect the overall level of carbon emissions related to household consumption? What is the mechanism that drives this relationship? Answering these questions is crucial for building an urban low-carbon governance system based on digital capabilities and promoting the inclusive green transformation of cities. These problems still need to be studied systematically. Under the guidance of China’s dual-carbon goal, this study starts from the micro-level family perspective, focuses on the mechanism by which digital literacy affects family carbon emissions, and explores whether this relationship follows the inverted U-shaped model of increasing first and then decreasing. This study aims to find a feasible path for family energy conservation and emission reduction, while ensuring the continuous improvement in living standards. This provides new ideas for promoting the dual-carbon strategy and an empirical basis for the design of China’s energy conservation and emission reduction policy.

2. Literature Review and Hypothesis

As the key ability of individuals to obtain, understand and use information in the digital environment, digital literacy has an obvious nonlinear impact on family consumption related to carbon emissions [16]. The relationship between digital literacy and consumption-based HCE is not linear, but shows an inverted U-shaped curve trend, that is, it rises at first and then begins to decline. This model can be explained from the perspective of the technological rebound effect. The theory points out that lower costs and greater convenience can stimulate the expansion of consumption in the early stages of technology or capacity improvement. This, in turn, increases energy use and carbon emissions. When digital literacy reaches a higher level, efficiency improvement and consumption structure optimization become more important, which in turn helps to curb carbon emissions. This process reflects how digital literacy shapes household consumption behavior and energy use patterns in different ways at different stages of development [17]. In the initial stage of the development of digital literacy, the improvement in family digital capabilities is mainly achieved by broadening consumption channels and reducing barriers to transactions and use [18]. Through algorithmic recommendation, e-commerce services and on-demand distribution, digital platforms can significantly increase the consumption demand of families in energy-intensive fields, such as electronic products, online entertainment and smart home devices [19,20]. In this process, although energy efficiency per unit of digital activity may be improved, the expansion of consumption and the stratification of energy use scenarios may lead to a small increase in HCE [21]. Therefore, at this stage, digital literacy is mainly manifested as a form of “consumption empowerment”. It prompts families to shift from traditional consumption patterns to more digital, more frequent and more diversified consumption patterns, and the corresponding carbon emissions will also increase [22].

However, once digital literacy reaches a key level, its development path will undergo a fundamental change. Families with high digital literacy no longer focus on expanding consumption scenarios, but have entered a green empowerment stage, which is characterized by deeper understanding and behavioral change. First of all, families can use technologies such as energy management systems and intelligent control devices more effectively to optimize their energy use structure. This enables dynamic monitoring and more accurate regulation of energy use. Secondly, high digital literacy enhances consumers’ ability to understand environmental issues and improves the efficiency of information acquisition. Therefore, they are more inclined to choose low-carbon products and services, participate in sharing economic activities, and reduce unnecessary digital consumption behavior. In addition, individuals with stronger digital literacy can use tools such as carbon footprint monitoring and green certification systems to guide emission reduction behavior in a more systematic and evidence-based manner. Therefore, once digital literacy reaches a high level, the impact of its increased emissions will gradually weaken, and eventually show a diminishing marginal effect [23,24]. This inverted U-shaped model highlights the dual role of digital literacy in the process of low-carbon transformation of the family. In the early stage, it played a role in promoting consumption expansion and rising energy demand. In the later stage, it has become a catalyst for improving energy efficiency and promoting more environmentally friendly consumption. Based on these, the following hypothesis is proposed:

Hypothesis 1:

Digital literacy and household consumption carbon emissions are non-linearly related in an “inverted U-shaped” manner.

As the core human capital of the digital age, digital literacy not only directly affects the consumption behavior of families, but also indirectly promotes the transformation to low-carbon consumption through a variety of intermediary channels, including perceived environmental pressure, social trust and income level [25,26]. As shown in Figure 1, digital literacy can enhance residents’ perception of environmental pressure, thus providing intrinsic motivation for low-carbon consumption. Residents with high digital literacy can make more effective use of the Internet, social media and online scientific communication platforms. Therefore, they can obtain and understand complex information about environmental issues such as climate change, carbon footprint and resource shortage. This process is consistent with the risk perception theory [27], which believes that subjective risk awareness is the key driver of behavior. Digital literacy enhances families’ perception of the possibility and severity of the environmental consequences of consumption, thus increasing their willingness to mitigate these risks. Therefore, through the mechanism described in the “Planned Behavior Theory”, a stronger sense of risk and environmental responsibility can be translated into practical action [28]. Individuals have become more positive about low-carbon consumption. They also feel that there is a broader consensus on environmental protection. In addition, since they have the relevant information and skills, they are more confident that they can take effective action. These factors together promote the formation of a strong willingness to actively participate in low-carbon consumption. This encourages families to take the initiative to adjust the consumption structure. They reduce the use of high-carbon products and services and increase the choice of green, recycled and low-carbon goods. As a result, the consumption model has changed from expanding quantity to improving quality.

Secondly, the theory of social capital believes that trust, norms and social networks are key resources to support cooperation and collective action. Social trust can reduce transaction costs and uncertainty, and increase the possibility of team members taking consistent behavior [29]. Digital literacy can enhance social trust and help build a social network and market environment that supports low-carbon consumption, thus reducing the cost of transformation [30]. However, low-carbon consumption often faces challenges related to “green premium” and concerns about authenticity [31]. In this case, digital literacy serves as a “bridge of trust”. Family members can use digital skills to verify the authenticity of corporate environmental statements and identify “green bleaching” behaviors, so as to channel trust to companies that truly adopt sustainable practices. On the other hand, by participating in online communities and social media groups, individuals can share their low-carbon consumption experiences and recommend reliable brands to like-minded people, thus generating positive peer effects and community supervision. This platform-based social trust reduces the cost of information search and decision-making risks for families when choosing low-carbon products. It also creates a social atmosphere that makes low-carbon consumption an ideal choice. Therefore, low-carbon transformation has become a more sustainable and easier form of collective action [32].

Finally, the theory of human capital points out that human capital formed through the accumulation of education, skills and knowledge is a key determinant of personal productivity, employment opportunities and income [33]. As the economy becomes more digital, digital skills have become an important part of modern human capital. Digital literacy is able to enhance the competitiveness of individuals in the labor market and broaden their career paths, thus increasing household disposable income [34]. In addition, according to Maslow’s hierarchy of needs and its expansion in consumption research, with the improvement in income levels, consumer needs tend to shift from basic survival needs to development-oriented and enjoyment-oriented needs [35]. Consumers will then pay more attention to product quality, safety and long-term value. Therefore, on the one hand, revenue growth can alleviate the budget constraints caused by the “green premium” associated with low-carbon products. It enables families to undertake upfront investment in green commodities such as energy-saving appliances, new energy equipment and organic food [36,37]. On the other hand, it encourages consumer demand to shift from meeting basic needs to pursuing higher quality and more sustainable lifestyles, which further strengthens the preference for low-carbon, healthy and environmentally friendly products. Although income growth itself may increase emissions by expanding overall consumption, the combination of digital literacy and green awareness makes it more likely for high-income families to follow the benign path of “income growth-green consumption-emission reduction”. This helps to reconcile the relationship between economic improvement and low-carbon transformation. Based on these, the following assumptions are put forward:

Hypothesis 2

:Digital literacy indirectly affects households’ carbon emissions by enhancing residents’ understanding of environmental pressure, social trust, income level, etc.

The impact of digital literacy on HCE is not uniform. It has significant differences in different aspects such as urban–rural gap, human capital, income level, gender and age. Drawing on digital divide theory, the urban–rural structure produces pronounced disparities in digital infrastructure provision, skills development and training opportunities, and the breadth of real-world application contexts between cities and rural areas [38,39]. In urban areas, extensive network coverage and high device penetration rate provide a solid foundation for digital literacy. The impact of digital consumption is mainly reflected in structural optimization and energy efficiency management, and its marginal impact on carbon emission growth has been weakened [40,41]. In contrast, in rural areas, initiatives such as the “digital village” strategy have rapidly increased the Internet penetration rate [42]. Enhancements in residents’ digital literacy facilitate the adoption of emerging consumption modes, including e-commerce and on-demand delivery and logistics services. However, due to the weak green infrastructure, the limited promotion of energy-saving technology, and the relative lag of low-carbon awareness, the improvement in digital skills has not effectively brought about the improvement in energy efficiency. On the contrary, the expansion of consumption scenarios and the increase in energy demand have led to a rapid rise in carbon emissions, thus delaying the moment when emissions began to decline. Therefore, in rural areas, digital literacy in the “compensatory digitalization” stage has a stronger role in promoting the growth of carbon emissions. In addition, the theory of human capital points out that the accumulation of education, skills and knowledge shapes an individual’s cognitive ability and behavior [43].

Secondly, high human capital groups usually have stronger information processing capabilities and higher environmental awareness. Therefore, the improvement in digital literacy not only broadens the consumption scenario, but also quickly contributes to the formation of green consumption behaviors [44], such as the use of energy management tools or the choice of low-carbon products. In this way, their carbon emission growth is relatively small, and emission reduction measures can be implemented earlier. In contrast, the improvement in digital skills of low human capital groups mainly enhances the accessibility and convenience of consumption [45]. However, due to cognitive limitations and insufficient application of technology, it is difficult for these groups to effectively use digital tools to optimize energy efficiency. Therefore, they are more likely to fall into the cycle of “digital convenience-high-carbon consumption”, resulting in a more obvious linear growth trend in carbon emissions.

Third, the theory of income stratification points out that the uneven distribution of economic resources will lead to systematic differences in consumption capacity and structure [46]. High-income families have strong purchasing power and diversified consumption options. They benefit from higher digital literacy, which makes it easier for them to access high-carbon services such as smart homes and luxury travel. However, once a certain threshold is reached, they can obtain more resources to support green transformation [47], such as investing in energy-saving equipment and carbon footprint tracking, so as to establish a benign path of “expansion first and optimization”. Low-income families are bound by budgetary restrictions, and their consumption is mainly concentrated on basic needs. The improvement in digital literacy is mainly used to improve convenience and cost-effective consumption, which makes it difficult for them to support more in-depth green transformation. Therefore, their carbon growth inertia is stronger, and the potential for emission reduction adjustment is more limited.

Fourth, the gender dimension of heterogeneity can be explained by the theory of social roles, which emphasizes how gender roles affect individual consumption choices and preferences for using numbers [48]. Women are often responsible for daily shopping, child education and home management, which makes them more likely to use digital tools such as e-commerce and social networks to simplify the consumption process. With the improvement in digital literacy, their frequent and diversified consumption behavior has exacerbated the increase in carbon emissions. In contrast, men are more inclined to pay attention to energy efficiency tools and low-carbon technologies at an early stage. Therefore, the impact of digital literacy on the growth of male HCE is relatively weak, and the emission reduction adjustment has occurred earlier. Finally, age group differences are consistent with the life cycle theory, which shows that there will be significant differences in personal consumption needs, family responsibilities and technology adoption at different stages of life [49]. As the main decision-makers of family consumption, the level of digital literacy of middle-aged people has improved, which affects high-carbon areas such as family energy use, transportation and children’s education. Therefore, their carbon emission growth effect is the most significant, and the emission reduction point has not yet stabilized. They may be in the behavioral locking stage of “high digital capacity–high carbon consumption”. Although young people have also shown a significant increase in carbon emissions, they have begun to partially mitigate them through low-carbon methods such as the sharing economy and online second-hand trading. In contrast, minors and the elderly have less direct impact on digital skills. Their consumption patterns are more influenced by family members or traditional habits, which leads to a relatively small direct impact on household consumption carbon emissions [50]. The effect of digital literacy on consumption-based HCE is markedly heterogeneous across dimensions such as the urban–rural divide, human capital endowments, income, gender, and age. This reflects the fundamental structural differences in consumption behavior, technology use and environmental response between different groups in the process of digitalization. The following assumptions are put forward:

Hypothesis 3:

The impact of digital literacy on HCE varies in many aspects, including factors such as urban–rural disparities, human capital, income levels, gender and age.

Against the backdrop of the deep embedding of digital technologies in everyday household life, the mechanisms through which digital literacy affects HCE cannot be adequately captured by traditional linear models [51]. A substantial body of prior research treats digital literacy as an exogenous determinant and implicitly presumes a unidirectional linear association between digital literacy and carbon emissions, thus ignoring the systematic complexity and multi-dimensional coupling mechanism that shape family consumption behavior. From the perspective of complex system theory, HCE are generated by the nonlinear interaction between multiple factors, including digital literacy, environmental awareness, institutional environment and infrastructure conditions [52]. In particular, as digital technology increasingly affects consumption decision-making, the impact of digital literacy on HCE is not independent; on the contrary, it is systematically constrained by many factors such as environmental responsibility, access to clean energy, the strictness of environmental regulations, and the quality of digital infrastructure.

This influence reflects a structural dependence, which is characterized by the model of “conditional combination-behavior path-carbon emission result”. We must abandon the traditional linear model and adopt an analytical framework that can capture multi-dimensional coupling and asymmetric causality. Specifically, there are complex coupling mechanisms between digital literacy, environmental awareness, policy tools and infrastructure, which together shape the path of household consumption carbon emissions. Its influence mechanism can be explained from two levels. First of all, from the perspective of cognitive behavior, digital literacy enables families to obtain low-carbon information and identify green products, but its effect is largely affected by the regulatory factor of environmental responsibility. When the sense of environmental responsibility is high, digital literacy can effectively guide low-carbon consumption behavior, such as using intelligent platforms to compare energy efficiency or choose products with low carbon footprint. In contrast, when the sense of environmental responsibility is low, digital literacy may strengthen the preference for high-carbon consumption [53], such as frequently recommending the purchase of high-energy consumption or non-necessities through algorithms, thus increasing the carbon footprint. On the contrary, from the perspective of institutional opportunities, the severity of supervision and the breadth of digital infrastructure together constitute important external limiting factors. In an environment with developed digital infrastructure and strict environmental regulation, digital literacy can effectively encourage families to manage their carbon accounts and obtain emission reduction incentives. However, if there is not enough institutional support, even if the family has a high digital literacy, it may be difficult to turn it into meaningful emission reduction actions [54]. Therefore, the impact of digital literacy on consumption-based HCE shows a multi-dimensional, nonlinear and interdependent coupling effect. The effect will vary significantly depending on the differences in external conditions and internal cognitive configuration. We put forward the following assumptions:

Hypothesis 4:

The impact of digital literacy on consumption-based HCE cannot be fully reflected through linear models; on the contrary, it is determined by the interaction between multiple determinants of interaction.

Previous studies explored the relationship between digitalization, carbon emissions and sustainable development-oriented transformation from perspectives such as the expansion of the digital economy, the popularization of information technology and the evolution of green consumption models. These studies have established a relatively mature analytical framework and theoretical foundation. However, a large number of studies on digital technology and carbon emissions mostly adopt macro or medium-level research methods, which makes it difficult to observe individual differences in digital capabilities and determine how these differences are transformed into differences in carbon emissions based on consumption. In general, the existing literature has the following main limitations. First of all, many studies adopt a macro perspective, focusing on the expansion of the digital economy or the widespread popularization of digital technology; therefore, their evaluation of carbon emission results usually focuses on emissions in production links or overall emissions. In contrast, there are still limited studies on the use of micro-level household data to directly capture individual and family-level consumption behaviors and the carbon emissions they contain. This positioning method limits our understanding of the actual reaction of families in consumption decision-making. It also often ignores the significant differences in digital literacy and digital ability of different individuals. Therefore, in the digital environment, it is difficult to reveal the micro-level sources of differences in family consumption-related carbon emissions. Secondly, some studies based on the spread of digital technology and the increase in the convenience of consumption believe that higher digital literacy will significantly expand household consumption, thus increasing energy use and carbon emissions. Liu et al. (2024) used data at the family level to find that digital literacy usually increases carbon emissions related to household consumption by expanding consumption channels and strengthening online consumption behavior [15]. However, this research idea often ignores the fact that digital literacy is a dynamic human capital, and its operation mechanism may undergo structural changes at different stages of development. With a relatively low level of digital literacy, the rebound effect caused by consumer empowerment and technology may dominate, leading to an increase in HCE. At a higher level, efficiency improvement, stronger environmental awareness and optimization of consumption structure become more significant, and the direction of the impact on emissions may be reversed. Third, many studies regard digital literacy as an isolated factor and focus on its “average effect”. This makes it difficult to identify the asymmetric causal path generated under different combinations of conditions. Therefore, the conclusion is often only a binary expression, that is, digital literacy either increases emissions or reduces emissions, and cannot reveal the various mechanisms generated by carbon emissions related to household consumption under different conditions.

In view of these gaps, this study expands and improves the previous research in three aspects. At the theoretical level, we have integrated the insights of risk perception theory, human capital theory and Maslow’s hierarchy of needs theory to build a coherent framework, which shows that the relationship between digital literacy and consumption-based HCE may be nonlinear and may present an inverted U-shaped pattern. This helps to solve the problem of insufficient attention to stage dependence effects in previous studies. Secondly, at the mechanism level, we examined multiple channels through which digital literacy affects carbon emissions related to household consumption—perceived environmental pressure, social trust and income. This transcends the limitations of single-channel interpretation. Finally, at the methodological level, we have introduced a configuration analysis framework to identify the coupling effect of digital literacy with multiple dimensions such as motivation, opportunity and ability. This method reveals the multiple ways and asymmetrical causal patterns behind the carbon emissions related to household consumption. Therefore, it provides more systematic and detailed empirical evidence for understanding the low-carbon transformation of families in the context of accelerated digitalization.

3. Research Design

3.1. Benchmark Model Setup

We use a two-way fixed-effect panel regression model to quantify the impact of digital literacy on consumption-based HCE, which is shown in Formula (1):

$$\ln HC{E}_{ikmt}={\beta }_0+{\beta }_1{D}_{it}+{\beta }_2{D}_{it}^2+{\beta }_3{X}_{it}+{\beta }_{4}fa{m}_{kt}+{\beta }_{5}pro{v}_{mt}+{\mu }_m+{\chi }_t+{\epsilon }_{it}$$

(1)

HCE_ikmt represents the carbon emissions of household k in provincial administrative region m for individual i at time t, X_it denotes the characteristic variables of individual i at time t, fam_kt represents the characteristic variables of household k at time t, and prov_mt refers to the characteristic variables of the provincial administrative region m at time t. µ_m and χ_t are the fixed effects for urban individuals and time, respectively, controlling for time-invariant characteristics of the urban context and common time trends across all cities. ε_it is the random disturbance term.

3.2. Fuzzy-Set Qualitative Comparative Analysis

The fsQCA is a case-based analysis method based on set theory and Boolean algebra. Its main advantage is that it can reveal a variety of different configuration causal paths, which can lead to the same results together. In this way, it transcends the traditional regression analysis, which usually only focuses on the net effect of a single variable [55]. It is particularly suitable for the analysis of complex social outcomes, such as HCE, which are shaped by factors such as technology, economy, socio-psychology and institutions. This study follows the standard QCA procedure. First of all, based on the theoretical framework, we define the consumption level related to HCE as a result variable. Secondly, we have selected seven conditional variables: digital literacy, cultural literacy, environmental responsibility, public environmental concerns, environmental regulations, digital infrastructure and clean fuel choices. Third, we calibrate continuous variables and convert them into fuzzy set affiliation scores from 0 to 1. Based on the collection of these calibrations, we have built a true value table. Based on the consistency and coverage standards, we then determined the typical configurations that are sufficient to cope with high carbon emissions and low carbon emissions. Through qualitative comparative analysis, this study aims to answer three questions: what combination of conditions together contributes to the high-carbon consumption pattern of families? What are the diverse and effective ways to achieve low-carbon consumption? Finally, does digital literacy play a central or auxiliary role in different paths? This research method helps to reveal the systematic and complex causal mechanism behind the family carbon emission behavior. It also provides a solid empirical basis for designing coordinated and targeted intervention strategies.

3.3. Entropy Weight Method

First, based on the CFPS questionnaire items related to digital literacy across survey waves, we construct an individual-level composite digital literacy index using the entropy-weighting method. First, indicator direction alignment and missing-data handling. Positive indicators, where larger values indicate stronger digital capability, are retained as they are. If a small number of indicators have the opposite meaning, we first transform them so that all indicators have the same direction. Binary (“yes or no”) items are coded as 0 or 1. Likert-scale items and frequency items keep their original ordered coding. For duration and expenditure variables, we apply moderate winsorization to reduce the influence of outliers before standardization. Second, Dimensionless transformation. To remove scale differences, we apply min–max normalization to each indicator according to Equation (2), mapping it linearly to the [0, 1] interval:

$$z_{ij}={\displaystyle \frac{x_{ij}-\min (x_j)}{\max (x_j)-\min (x_j)}},j=1,2,\dots m$$

(2)

In Equation (2), x_ij is the original value of indicator j for individual i, and z_ij is the normalized value.

Third, entropy calculation and objective weight determination. Equation (3) shows the information entropy of indicator j.

$$e_j=-{\displaystyle \frac{1}{\ln (n)}}{\displaystyle \sum _{i=1}^n{p}_{ij}\ln (p_{ij})}$$

(3)

Based on this, we compute the variation (diversity) coefficient and then normalize it, as shown in Equation (4), to obtain the entropy weight w_j:

$$w_j={\displaystyle \frac{1-e_j}{{\displaystyle {\sum }_{j=1}^m(1-e_j)}}}$$

(4)

The economic meaning of entropy weights is as follows: the greater the dispersion of an indicator in the sample, the more information it contains, and the higher its weight in the composite index. Finally, we construct the composite index. The individual digital literacy index is defined in Equation (5):

$$D{L}_i={\displaystyle \sum _{i=1}^m{w}_j{z}_{ij}}$$

(5)

3.4. Data Sources and Processing

The data based on this empirical analysis comes from the CFPS conducted by the Chinese Social Science Survey Center of Peking University (https://www.isss.pku.edu.cn/cfps/ (accessed on 9 October 2025)). The CFPS baseline survey began in 2010, with follow-up surveys every two years. Seven rounds of surveys have been completed to date. The sample of CFPS covers 25 provinces and involves a total of 16,000 families, which makes it a highly representative and large-scale micro-level data set in China. The CFPS survey covers many topics, including family income, expenditure, decision-making, employment, and details of each family member. These data form the basis of this study. In addition, data on regional control variables was also obtained from China Regional Economic Database (https://olap.epsnet.com.cn/#/datas_home?cubeId=773 (accessed on 9 October 2025)).

3.4.1. Dependent Variable

Household energy use can be decomposed into two components: direct and indirect consumption. Direct consumption denotes the energy households utilize in routine activities. Electricity for lighting and appliances, as well as fuels used for heating and cooking. Indirect consumption refers to the energy embodied in the production and distribution processes of the goods and services that households purchase, thereby capturing the energy content embedded in their overall consumption basket. Accordingly, consumption-based household carbon emissions comprise both direct and indirect emissions, and the relevant emission factor can be expressed as the sum of the direct- and indirect-HCE factors. Household direct energy use mainly includes fuel and various public utility products, such as coal, kerosene, gasoline, liquefied petroleum gas, natural gas, urban gas, thermal energy and electricity. Heat and electricity do not directly generate carbon emissions but produce indirect emissions during their production and transportation. Therefore, this study measures the direct energy consumption of households through five energy sources (coal, kerosene, liquefied petroleum gas, natural gas and urban gas), but it is not included in the scope of consideration due to the serious lack of gasoline data. In this study, greenhouse gas emissions from household consumption are measured in terms of carbon dioxide (CO₂). Equation (6) presents the method for calculating the emission factor of household direct consumption-related carbon emissions:

$$E_d={\displaystyle \sum _{i=1}^5{\alpha }_i{d}_i}(i=1,2,\dots 5)$$

(6)

In this context, E_d represents the direct HCE (kg), d_i is the energy produced by the i-th type of direct energy consumed by the household (kJ), and α_i denotes the carbon emission factor associated with each type of energy (kg/kJ). Since direct energy consumption falls under residential consumption, this study includes direct energy carbon emissions as part of residential consumption emissions. Therefore, Equation (7) gives the household direct consumption-related carbon emission coefficient as follows:

$${\zeta }_d=E_d/H$$

(7)

In the equation, ζ_d represents the direct carbon emission factor for household consumption (kg/CNY), and H represents household residential consumption expenditure (CNY). It should be noted that, due to data limitations in CFPS, we cannot obtain complete information on the physical quantities or expenditures of all types of direct energy use. Therefore, H is not a direct measure of energy spending. Instead, it serves as a scale variable that is highly related to household energy demand. It is used to monetize and normalize E_d, which improves comparability in direct carbon emissions across households and reduces measurement bias caused by differences in housing conditions and household size.

Indirect HCE refer to the embodied energy use and carbon emissions induced by household purchases of goods and services. In terms of accounting, this study adopts a final-demand attribution approach based on the input–output framework. We attribute sectoral energy use and carbon emissions to the share of output driven by households’ final consumption demand. Because sectors such as food, clothing, and transportation and communication serve not only households but also non-household entities (e.g., restaurants, hotels, schools, hospitals, and government agencies), using total sectoral energy use as a proxy for household indirect energy use without distinction would create a logical bias. To avoid this problem, the first step in our accounting procedure is to map household consumption expenditures by category to their main corresponding production and service sectors. Following prior studies [56], we systematically establish the correspondence between consumption categories and industrial sectors (

Table 1. Household consumption categories and corresponding industries.

Consumption Items	Industrial Sectors
Food	Agro-processing, Food Manufacturing, Beverage Manufacturing, Tobacco
Clothes	Textile Industry, Textile Apparel, Footwear and Hat Manufacturing, Leather, Fur, Feathers (Down), and Related Products
Residence	Construction Industry, Electricity and Heat Production and Supply, Gas Production and Supply, Water Production and Supply
Living goods and services	Wood Processing and Wood, Bamboo, Rattan, Palm, and Grass Products Manufacturing, Furniture Manufacturing, Electrical Machinery and Equipment Manufacturing, Metal Products Manufacturing, Plastic Products Manufacturing
Transportation and communications	Transportation Equipment Manufacturing, Communication Equipment, Computer and Other Electronic Equipment Manufacturing, Transportation, Storage, and Postal Services
Education, culture, and entertainment	Paper and Paper Products Manufacturing, Manufacture of Educational, Cultural, and Sports Goods, Printing, Recording Media Reproduction, Instrument and Meter, and Cultural and Office Machinery Manufacturing
Medical and healthcare	Pharmaceutical Manufacturing
Other supplies and services	Wholesale and Retail Trade, Accommodation and Catering, Handicrafts and Other Manufacturing Industries

). Based on this mapping relationship, we calculate the specific energy consumption intensity and carbon emission coefficients of each category, and then use these data to estimate the indirect HCE in each consumption category. The calculation is provided in Equation (8):

$${\zeta }_m={\displaystyle \frac{{\displaystyle \sum {\displaystyle \sum a_n{w}_{mkp}{J}_p{\lambda }_p}}}{{\displaystyle \sum a_{mn}}}}$$

(8)

In the equation, ζ_m represents the indirect HCE factor for the m-th category of household consumption (kg/CNY). a_m is the total added value of industries within the m-th consumption category. a_mn is the added value of the n-th industry within the m-th consumption category. w_mkp denotes the amount of standard coal consumed for the p-th type of energy by the k-th industry in the m-th consumption category. J_p is the calorific value of 1 kg of standard coal for each type of energy. λ_p is the carbon emission factor for each type of energy.

The total carbon emission coefficient for housing consumption equals the sum of the direct and indirect emission coefficients. For all other consumption categories, the total emission coefficient equals the indirect emission coefficient. Key parameters and auxiliary statistics—including the emission coefficient of specific fuels, household energy use coefficients, total household expenditure, incremental consumption of fossil fuels, industrial added value and related indicators—are all from the Intergovernmental Panel on Climate Change (IPCC), the China energy statistical yearbook and the provincial statistical yearbooks.

As is shown in Figure 2, we obtain the HCE coefficients for the eight major consumption categories using the above approach. From 2014 to 2022, the emission factors across consumption categories display clear time-varying patterns. For example, the emission factor for housing consumption is still much higher than that of other categories, but it shows a sustained downward trend overall. This pattern suggests a cleaner electricity mix and improved energy efficiency. In contrast, categories such as transportation, communications and food show cyclical fluctuations. This is consistent with the changes in energy prices, travel methods and supply chains in the real world. The time evolution of emission coefficients in each category is not completely synchronized. This shows that different consumption areas follow different adjustment paths in terms of unit emission intensity, which are driven by technological progress, changes in supply structure, and changes in consumption patterns. In general, these results show that our measurement methods can reasonably capture changes in the energy structure.

3.4.2. Core Independent Variable

Digital literacy refers to a series of abilities and qualities required by citizens in study, work and daily life in a digital society. It covers the ability to acquire, create, use and evaluate digital information, as well as the ability to interact, share, innovate, ensure safety, and comply with ethical norms. According to the definition of digital literacy, this study builds a measurement system with six main dimensions. First of all, “obtaining digital devices” reflects the basic conditions for using the Internet and digital terminals. Secondly, “digital communication and sharing” reflects the ability to interact and communicate with online social tools. Third, the “application of digital scenarios” measures the breadth of the application of digital technology in learning, work and daily life. Fourth, “cognition of digital formats” reflects the extent to which individuals understand online information formats and their importance. Fifth, “digital information acquisition” captures the ability to obtain and screen information through the internet. Sixth, “mobile digital payment” reflects the level of application in online consumption and digital transactions. The specific CFPS items and dimensions used to measure digital literacy in each survey wave are reported in

Table 2. Digital Literacy Indicator System.

Dimension	Questions	Year					Indicator Attribute	Value Range
		2014	2016	2018	2020	2022
Digital Device Access	Internet usage	√					Positive	0 or 1
	Mobile phone usage	√	√	√			Positive	0 or 1
	Using mobile devices for internet access		√	√	√	√	Positive	0 or 1
	Using computer for internet access		√	√	√	√	Positive	0 or 1
	Daily internet usage time on mobile devices (min)				√	√	Positive	0–1440
	Daily internet usage time on computers (min)				√	√	Positive	0–1440
	Weekly internet usage time (h)	√	√	√			Positive	0–168
Digital Communication and Sharing	Importance of the internet for social interaction	√	√	√	√	√	Positive	1–5
	Frequency of using the internet for socializing	√	√	√			Positive	0–6
	Do you use WeChat?				√	√	Positive	0 or 1
	Sharing frequency on social media?				√	√	Positive	0–6
	Whether to send or receive emails?	√	√	√			Positive	0 or 1
	How many days per week do you log in to your email?	√	√	√			Positive	0–7
Digital Competence in Contexts	Does the job require the use of computers	√	√	√	√	√	Positive	0 or 1
	The importance of the Internet to work	√	√	√	√	√	Positive	1–5
	How often do you work on the Internet	√	√	√			Positive	0–6
	Whether to use the Internet for learning				√	√	Positive	0 or 1
	The Importance of the Internet for Learning	√	√	√	√	√	Positive	1–5
	Frequency of learning using the Internet	√	√	√			Positive	0–6
	The Importance of Internet to Daily Life				√	√	Positive	1–5
	The Importance of the Internet to Entertainment	√	√	√	√	√	Positive	1–5
	Frequency of using Internet entertainment activities	√	√	√			Positive	0–6
	Do you play online games					√	Positive	0 or 1
	Do you want to watch short videos				√		Positive	0 or 1
Digital Political Awareness	Frequency of accessing political information online (per week)	√	√	√	√		Positive	0–7
	Have you ever posted political opinions on websites	√	√	√	√		Positive	0 or 1
Digital Information Access	Importance of the internet as an information channel	√	√	√	√		Positive	1–5
	Importance of SMS as an information channel	√	√	√	√		Positive	1–5
Digital Mobile Payments	Online shopping				√	√	Positive	0 or 1
	Importance of the internet for business activities	√	√	√			Positive	1–5
	Frequency of using the internet for business activities	√	√	√			Positive	0–6
	Expenditure on online shopping (thousand CNY)	√	√	√			Positive	0–1000

Note: √ represents that the problem exists in the corresponding observation year.

. Variables with values of 0 or 1 are binary indicators. Variables ranging from 0–6 or 0–7 are frequency measures. Variables ranging from 1–5 are Likert-scale items. The remaining indicators are continuous variables.

3.4.3. Control Variables

Drawing on the research of existing scholars, this paper extracts characteristic variables from three levels: individual characteristics, family characteristics, and regional socio-economic environment, to control for other factors influencing household carbon emissions. At the individual level, we control for respondents’ education. Based on the highest degree attained, we assign values as follows: illiterate or semi-illiterate = 0; primary school = 6; junior high school = 9; senior high school, secondary technical school, vocational school, or technical school = 12; junior college = 15; bachelor’s degree = 16; and master’s degree = 19. Gender is coded as a dummy variable (female = 0, male = 1). Age is a continuous variable measured in years. Marital status is also coded as a dummy variable: unmarried, divorced, or widowed = 0; married or cohabiting = 1. Health status and cognitive ability are measured as ordered variables on 1–5 and 1–7 scales, respectively, where higher values indicate better conditions. Employment status is coded as a dummy variable: employed = 1; unemployed, not working, or out of the labor force = 0. At the household level, the control variables include: human resource investment, measured as the log of household spending on education and training; social capital, measured as the log of household spending on gifts and social exchanges; and household economic status, measured as the log of total household income. Household size is measured by the number of household members. At the regional level, we control for the type of residence area (urban = 1, rural = 0). Marketization is measured following prior studies [57], where we project the marketization level based on the average growth rate of the historical marketization index. Industrial structure is measured at the provincial level as the share of tertiary industry value added in provincial GDP. Educational attainment is measured as the average number of students enrolled in higher education institutions per 100,000 people at the provincial level. Economic development is measured as the log of provincial GDP per capita.

3.4.4. Mechanism Variables

CFPS classifies household expenditure according to the consumption classification standards of the National Bureau of Statistics. On this basis, we roughly divide family expenditure into two categories: basic consumption and developmental consumption. Basic consumption covers expenditure on food, clothing and housing, as we cover the expenditure needed to meet basic living needs. Developmental consumption includes expenditure on daily necessities, health care, transportation and communication, cultural and entertainment activities and other related items. This type of consumption is aimed at improving living standards and pursuing higher-quality life goals. We reflect the escalation of consumption through the proportion of developmental expenditure in total household consumption. Perceived environmental pressure reflects an individual’s subjective assessment of how severe local environmental problems are. This variable is measured by respondents’ rating of the question “How serious are environmental problems in China?” As shown in

Table 3. Variable definition.

Variable		Variable Description
Dependent Variable	HCE	Calculated as the sum of direct and indirect HCE from household consumption
Core Independent Variable	Digital Literacy	Individual digital literacy score derived through factor analysis with weighted components
Control Variables	Educational Attainment	Coded as 0 for illiterate/semi-literate, 6 for primary school, 9 for junior high school, 12 for senior high school/vocational school/technical school, 15 for junior college, 16 for bachelor’s degree, 19 for master’s degree
	Gender	Female = 0, Male = 1
	Age	Respondent’s age
	Marital Status	Coded as 0 for unmarried, divorced, or widowed; 1 for married or cohabiting
	Health Status	Self-reported health status, ranging from 1 (lowest) to 5 (highest)
	Cognitive Ability	Self-reported cognitive level, ranging from 1 (lowest) to 7 (highest)
	Employment Status	Employed = 1, unemployed, job-seeking, or outside the labor market = 0
	Urban and Rural Categories	Urban = 1, Rural = 0
	Human Capital Accumulation	Log of household expenditure on education and training (CNY)
	Social Capital Accumulation	Log of household expenditure on social gifts (CNY)
	Household Economic Status	Log of total household net income (CNY)
	Household Size	Number of household members
	Marketization Level	Predicted following Li et al. (2020) [57] based on the average growth rate of historical marketization indices
	Industrial Structure	Share of tertiary industry value added in provincial GDP
	Education Level	Number of higher-education students per 100,000 population at the provincial level
	Economic Development Level	The logarithm of per capita regional GDP at the provincial level
Mechanism variables	Environmental Pressure Awareness	How serious do you think the environmental problem is in China? 0 means it is not serious, and 10 means it is very serious.
	Social Trust	The degree of trust in local government officials, 0 means very distrusting, and 10 means very trusting.
	Income	The total family income divided by the number of family population is taken by the logarithm.

, it ranges from 0 to 10, where 0 means “not serious” and 10 means “very serious.” Higher values indicate stronger environmental concern. Social trust refers specifically to an individual’s trust in local government officials. It is measured by respondents’ rating of their trust in local officials, also on a 0–10 scale. A value of 0 indicates “complete distrust,” and 10 indicates “complete trust.” Higher scores represent higher trust in local government officials.

3.4.5. Descriptive Statistics

Table 4. Descriptive Statistics of Variable Data.

Variables	Observations	Mean	Std. Dev	Min	Max
HCE	57,453	9.0015	0.9626	4.3292	14.0669
Digital Literacy	57,453	0.2545	0.1808	0.0001	0.9719
Educational Attainment	57,453	9.6971	4.3475	0.0000	22.0000
Gender	57,453	0.5196	0.4996	0.0000	1.0000
Age	57,453	34.4867	13.4456	9.0000	96.0000
Marital Status	57,453	0.6735	0.4689	0.0000	1.0000
Health Status	57,453	3.3564	1.0773	1.0000	5.0000
Cognitive Ability	57,453	5.6407	1.2857	1.0000	7.0000
Employment Status	57,453	0.7090	0.4542	0.0000	1.0000
Urban and Rural Categories	57,453	0.5767	0.4941	0.0000	1.0000
Human Capital Accumulation	57,453	5.3848	4.2627	0.0000	13.1224
Social Capital Accumulation	57,453	7.4470	2.2596	0.0000	12.4292
Household Economic Status	57,453	11.1983	0.9537	0.0000	16.2481
Household Size	57,453	4.2765	1.9510	1.0000	17.0000
Marketization Level	57,453	8.0837	1.8414	0.6470	11.0150
Industrial Structure	57,453	2515.9320	579.9745	1220.0280	5428.8300
Education Level	57,453	43.9499	7.7964	35.4000	77.9000
Economic Development Level	57,453	10.7244	0.3986	10.1824	11.5639
Environmental Pressure Awareness	57,453	7.0215	2.1934	0.0000	10.0000
Social Trust	57,453	5.0688	2.5463	0.0000	10.0000
Income	57,453	9.8567	0.9987	0.0000	10.0000

shows that, in terms of individual characteristics, the average age of the residents in the sample is about 34 years old, indicating that most people are within the working age range, have high consumption activity, and are more highly adopted of digital technology. The sample was relatively balanced in terms of gender distribution, and 51.9% of the respondents were male. Most people are married, which reflects high family stability and provides a stable living environment for their consumption behavior and energy needs. The average number of years of education of the residents in the sample is 9.7 years, which is equivalent to the level from junior high school to high school.

These models show that although the level of education in the sample is generally similar, there is still room for improvement. In contrast, the average health status and cognitive ability are relatively high, indicating that the respondents are in good physical condition and cognitive ability, these factors provide a solid foundation for making wise consumption choices and adopting digital technology. In terms of employment, about 71% of people have jobs, reflecting the high employment rate. This shows that most families have a stable labor income, which provides a solid economic foundation for their consumption behavior. The sample also shows a slightly higher proportion of urban residents, although the gap between urban and rural areas still exists, which may lead to systematic differences in household consumption patterns, energy use and access to digital technology. The average family size is about four people, reflecting the typical structure of Chinese families, that is, the nuclear family plus some collateral members. This shows that the consumption needs of most families cover basic living, education, health care and developmental expenditure, providing a diverse background for analyzing the carbon emission structure associated with consumption. The average of human capital and social capital reserves is 3.5848 and 7.4470, respectively, indicating that families have a medium level of education and social capital resources. These factors affect their efficiency in obtaining information, consumption preferences and environmental behavior to a certain extent.

Figure 3 shows the nuclear density estimation results of residents’ digital literacy from 2014 to 2022. The distribution shows a dynamic pattern of “continuous rightward movement—gradual convergence—periodic differentiation”, highlighting the remarkable progress and structural changes in China’s digitalization efforts in recent years. At first, the curve continued to move to the right every year, indicating that residents’ digital literacy is continuously improving. In 2014, the distribution was concentrated in a lower range. However, from 2016 to 2022, the peak gradually approached zero, indicating that residents’ ability to use mobile Internet, smart devices and various online services is increasing. Over the years, the curve has become more concentrated, the peak has increased, and the tail has become narrower. This shows that the gap in digital capacity between different groups has narrowed. The overall distribution of digital literacy in society has changed from a “multi-level” to a “centralized” model, reflecting the improvement in digital inclusiveness driven by expanded digital infrastructure and a wider range of digital public services. It is worth noting that the curve in 2020 shows an obvious “double peak” structure, with local peaks in the low literacy rate and medium literacy rate intervals. This temporary differentiation is mainly driven by the sudden surge in digital use during the COVID-19 pandemic. Some families have rapidly improved their digital skills through online work, education and contactless consumption, while some elderly and rural groups are difficult to adapt due to equipment and skill limitations. Therefore, there is a temporary gap in digital capacity. However, since 2022, this gap has narrowed significantly, and the curve in the latest year has returned to a single peak distribution, and the peaks are more concentrated. This shows that with the improvement in digital service systems and the widespread adoption of intelligent devices, the digital divide that has widened during the epidemic has been greatly alleviated.

Figure 4 shows the estimated nuclear density of carbon emissions related to household consumption by category. In general, the density curves of different categories have obvious differences in shape and position. In terms of distribution location, the curves of food, housing, transportation and communication shift to the right, and the peaks are concentrated in the high emission range. This shows that these three categories constitute the main sources of HCE and are closely related to daily energy use and high-frequency consumption behavior. In contrast, the emission distribution of clothing, daily necessities and services is single-peak and relatively concentrated, which means that the difference in emissions between households in this category is smaller. In particular, the peak of clothing is obviously within the low emission range. This shows that clothing-related consumption contributes less to total household emissions, and the difference between different families is small. In contrast, the curves of education, culture and entertainment and health care show relatively flat peaks or multiple peaks, which are more distributed in the medium and high emission ranges. This suggests that, shaped by household income, consumption preferences, and regional differences in public service provision, service-related consumption emissions are highly heterogeneous across households. The curve for other goods and services is shifted to the left but has a long right tail. This indicates that the category can generate relatively high emissions for a small number of households, while its overall contribution remains relatively low in the full sample.

4. Empirical Results

4.1. Benchmark Regression

As described in

Table 5. The impact of digital literacy on HCE: OLS regression.

Dependent Variable	HCE
	(1)	(2)	(3)	(4)
Digital Literacy	1.3344 ***	1.3343 ***	0.8351 ***	0.8004 ***
	(0.0717)	(0.0750)	(0.0706)	(0.0711)
Digital Literacy2	−1.1193 ***	−1.1076 ***	−0.7249 ***	−0.7251 ***
	(0.1078)	(0.1090)	(0.1013)	(0.1013)
Education Level		0.0080 ***	0.0038 ***	0.0054***
		(0.0011)	(0.0011)	(0.0011)
Gender		0.0028	0.0046	0.0033
		(0.0079)	(0.0073)	(0.0081)
Age		−0.0009 **	0.0006 *	0.0005 ***
		(0.0004)	(0.0004)	(0.0004)
Marital Status		0.2753 ***	0.1354 ***	0.1341 ***
		(0.0108)	(0.0105)	(0.1113)
Health Status		−0.0020	−0.0101 ***	−0.0085 ***
		(0.0038)	(0.0035)	(0.0037)
Intellectual Level		0.0075 **	−0.0014	−0.0001
		(0.0032)	(0.0030)	(0.0029)
Employment Status		−0.0684 ***	−0.0551 ***	−0.0534 ***
		(0.0095)	(0.0089)	(0.0093)
Urban–Rural Status		0.2240 ***	0.1793 ***	0.1732 ***
		(0.0087)	(0.0085)	(0.0091)
Human Capital Accumulation			0.0213 ***	0.0219 ***
			(0.0009)	(0.0010)
Social Capital Accumulation			0.0369 ***	0.0347 ***
			(0.0018)	(0.0018)
Household Economic Status			0.3063 ***	0.2921 ***
			(0.0084)	(0.0085)
Household Size			0.0211 ***	0.0214 ***
			(0.0023)	(0.0025)
Marketization Level				0.2098 ***
				(0.0809)
Industry Structure				−0.0376 ***
				(0.0105)
Education Level				−0.0002 ***
				(0.0001)
Economic Development Level				0.5148
				(0.0147)
Year FE	Yes	Yes	Yes	Yes
Province FE	Yes	Yes	Yes	Yes
R2	0.0847	0.0985	0.2128	0.2637
Observations	57,453	57,453	57,453	57,453

Notes: *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively. Standard errors are shown in parentheses. Individual characteristics include education level, gender, age, marital status, household registration, political affiliation, health status, and intellectual level. Household characteristics include human capital accumulation, social capital accumulation, household economic status, and household size. Regional characteristics include marketization level, industry structure, education level, and economic development level.

, columns (1) to (4), in all models, the linear term coefficient of digital literacy is significantly positive, while the quadratic term coefficient is significantly negative. Overall, these estimates provide consistent evidence of the inverted U-shaped relationship between digital literacy and consumption-based HCE. The emission-increasing effect weakens at the margin as digital literacy rises. Therefore, Hypothesis H1 receives initial support. Under the specification that includes individual-, household-, and regional-level controls, the turning point of digital literacy is located at the 79.61st percentile of its distribution. Under the specification that includes individual-, household-, and regional-level controls, the turning point of digital literacy is approximately 0.552. Only about 6.78% of residents have digital literacy levels above this point, meaning that digital literacy reduces HCE only for a small share of the population.

To identify heterogeneous effects across consumption domains, we divide household consumption expenditure into three categories—basic, enjoyment-oriented, and development-oriented—and estimate separate regressions. As described in

Table 6. The impact of digital literacy on HCE from different types of consumption.

Type	Basic HCE			Enjoyable HCE			Development HCE
	Food	Clothes	Residence	Transportation and Communications	Education, Culture, and Entertainment	Medical and Healthcare	Living Goods and Services	Other Supplies and Services
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Digital literacy	0.2182 ***	0.3907 ***	0.0503 ***	−1.3425 **	0.0286 ***	−0.3448 ***	0.1351 ***	0.0747 ***
	(0.2847)	(0.3305)	(0.3346)	(0.5513)	(0.3965)	(0.3525)	(0.2338)	(0.0732)
Digital literacy2	−0.1615 ***	−0.3919 **	−0.0931 ***	1.3392 ***	−0.0392 ***	0.3019 *	−0.1092 ***	−0.0606 ***
	(0.2559)	(0.2969)	(0.3007)	(0.4954)	(0.3563)	(0.3168)	(0.2101)	(0.0657)
Control variables	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Province FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
R2	0.0273	0.0088	0.0402	0.0061	0.0241	0.0214	0.0152	0.0112
Observations	57,453	57,453	57,453	57,453	57,453	57,453	57,453	57,453

Notes: *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.

, Enjoyment-oriented consumption mainly refers to spending that enhances psychological satisfaction and living quality, such as education, culture and entertainment, and high-end personal services. Development-oriented consumption mainly refers to spending that builds human capital, safeguards health, and expands social connections, such as healthcare and transportation and communication.

In terms of emissions related to basic consumption, digital literacy has a positive impact on the HCE of food, clothing and housing, while the square term has a significant negative impact. The estimates indicate an approximately inverted U-shaped nonlinear relationship between digital literacy and carbon emissions associated with basic consumption. Among these categories, the positive coefficients for food and clothing are relatively larger, while the effect size for housing is smaller. However, the nonlinear pattern remains statistically significant. For enjoyment-oriented consumption-related carbon emissions, the results differ clearly across categories. For transportation and communication, the linear coefficient on digital literacy is significantly negative, whereas the squared term is significantly positive. The results for healthcare also indicate that as digital literacy increases, its effect on emissions becomes more suppressive at higher levels. In contrast, in the fields of education, culture and entertainment, linear terms are positive and square terms are negative, showing an obvious inverted U-shaped pattern. For development-oriented HCE, there is a significant inverted U-type relationship between digital literacy and the HCE of daily necessities and services, as well as other goods and services.

4.2. Robustness Test

We further adopted the instrumental variable strategy to alleviate potential endogenous problems. Specifically, take the individual’s cognitive ability as a tool variable to measure digital literacy. Following existing studies, we construct the cognitive ability measure using respondents’ word test scores and math test scores from the CFPS questionnaire [58]. This choice has clear theoretical and empirical support for both the relevance and the exogeneity conditions. First, in terms of relevance, cognitive ability is a foundational capacity for individuals to understand, learn, and use digital technologies. Higher cognitive ability helps individuals master digital tools more quickly, comprehend information, and engage in more complex digital scenarios. It is therefore closely related to digital literacy. Second, regarding exogeneity, cognitive ability reflects a relatively stable individual trait that is largely formed early in life. It is unlikely to directly affect a household’s current consumption behavior or HCE in the short run. Therefore, cognitive ability is expected to satisfy the exclusion restriction for a valid instrument.

As reported in

Table 7. Results of robustness test.

Type	2SLS	First Stage		OLS
	HCE	Digital Literacy	Digital Literacy^2	HCE
	(1)	(2)	(3)	(4)	(5)	(6)
Digital literacy	4.8745 ***			0.7906 ***	0.7728 ***	0.7484 ***
	(1.3972)			(0.1040)	(0.0856)	(0.1183)
Digital literacy2	−9.8652 ***			−0.6004 ***	−0.7485 ***	−0.7305 ***
	(2.2958)			(0.1620)	(0.1389)	(0.2015)
Cognitive		0.0086 ***	−0.0094 ***
		(0.0016)	(0.0011)
Cognitive2		0.0031 ***	0.0024 ***
		(0.0003)	(0.0002)
Control variables	YES	YES	YES	Yes	Yes	Yes
Year FE	YES	YES	YES	Yes	Yes	Yes
Province FE	YES	YES	YES	Yes	Yes	Yes
K-P F-stat	149.9080
Observations	57,453	57,453	57,453	57,453	57,453	57,453

Notes: *** indicate significance at the 1% levels.

, the first-stage results in Columns (2) and (3) show that cognitive ability and its squared term significantly explain digital literacy and its squared term. The estimated signs are consistent with theoretical expectations. The Kleibergen–Paap F statistic is 149.908, which is well above the conventional threshold for weak instruments. This rules out weak-instrument concerns and indicates strong relevance of the chosen instrument. In addition, the estimation results in Column (1) show that after the correction of endogenous factors, the linear terms of digital literacy are still significantly positive, while the quadratic terms are still significantly negative. The inverted U-type relationship obtained from this is strengthened, and its absolute coefficient value is larger than the value obtained in the benchmark model. This implies that ignoring endogeneity may underestimate the true effect of digital literacy on HCE. The instrumental variable estimates are consistent with the baseline and lagged regressions in both direction and functional form. They confirmed the robustness of the nonlinear path of digital literacy affecting HCE, which shows that our conclusions are not mainly influenced by endogenous factors.

4.3. Heterogeneity Analysis

To further clarify subgroup-level differences in how digital literacy influences HCE, we conducted a heterogeneous analysis from five dimensions (urban residence and rural residence, human capital, income, gender and age). The analysis results are shown in

Table 8. The impact of different groups’ digital literacy on HCE.

Type	HCE
	(1)		(2)		(3)
	Urban	Rural	High Human Capital	Low Human Capital	High Income	Low Income
Digital literacy	0.7124 ***	0.9124 ***	0.5944 ***	0.9290 ***	0.7760 ***	0.8360 ***
	(0.0905)	(0.1172)	(0.1131)	(0.1016)	(0.0990)	(0.1040)
Digital literacy2	−0.5689 ***	−0.8098 ***	−0.4427 ***	−0.8428 ***	−0.6974 ***	−0.8629 ***
	(0.1233)	(0.1874)	(0.1461)	(0.1761)	(0.1325)	(0.1678)
Control variables	Yes	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes	Yes
Province FE	Yes	Yes	Yes	Yes	Yes	Yes
R2	0.2046	0.1719	0.2201	0.1874	0.1860	0.1354
Observations	33,136	24,317	23,639	33,814	29,227	28,226
	HCE
Digital Literacy	(3)		(4)
	Male	Female	Minor	Youth	Middle-Aged	Elderly
Digital literacy2	0.6568 ***	1.0643 ***	0.6277 **	0.8041 ***	1.3000 ***	0.6662 *
	(0.0981)	(0.0967)	(0.2611)	(0.0866)	(0.1761)	(0.3534)
Control variables	−0.4817 ***	−1.0473 ***	−0.4990	−0.6789 ***	−1.4078 ***	−0.7786
	(0.1389)	(0.1476)	(0.4656)	(0.1185)	(0.2833)	(0.7056)
Year FE	Yes	Yes	Yes	Yes	Yes	Yes
Province FE	Yes	Yes	Yes	Yes	Yes	Yes
R2	Yes	Yes	Yes	Yes	Yes	Yes
Observations	0.2109	0.2169	5028	39,847	9719	2859

Notes: *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.

. In general, digital literacy has a positive influence on HCE of all subgroups, but the magnitude of the impact varies significantly. This shows that the effect of improving digital skills is structurally constrained by resource endowment, spending capacity and lifestyle. The following analysis systematically explains these results based on grouping standards and intra-group differences.

Overall, digital literacy is significantly and positively associated with HCE across all subgroups, although the estimated effect sizes differ markedly. These results indicate that the influence of rising digital literacy is structurally conditioned by households’ resource endowments, consumption capacity, and prevailing lifestyles. The following part will systematically discuss the results based on grouping standards and intra-group differences. In the dimension of human capital, groups are defined according to the level of education and skill level. Specifically, the sample is divided into high and low human capital groups using the mean value of the percentile-based indicator: individuals above the average are classified as high human capital, whereas those at or below the average are classified as low human capital. Within the high-human-capital subgroup, the estimated coefficient on the linear digital literacy term is notably smaller than the corresponding estimate for the low-human-capital subgroup. Although the quadratic term remains significantly negative, its magnitude is relatively modest. The turning point (−0.6713) is also substantially lower than that of the low human capital group (−0.5513), indicating that high human capital households reach the emission-reducing stage earlier. In the income dimension, groups are defined by household per capita income. The sample is split into high- and low-income groups using the mean value of the percentile-based income measure: households above the average are classified as high income, whereas those at or below the average are classified as low income. For high-income households, the coefficient on the linear term of digital literacy is slightly smaller than that for low-income households, while the negative moderating effect of the quadratic term is stronger. Accordingly, the emission-reducing turning point for high-income households (−0.5564) occurs slightly earlier than that for low-income households (−0.4845). In the gender dimension, groups are defined by the respondent’s biological sex. The effect of digital literacy on HCE is significantly stronger for females than for males: the linear term has a larger coefficient, and the quadratic term shows a more pronounced negative effect. Accordingly, the turning point for females (0.5082) is also markedly lower than that for males (0.6818). In the age dimension, groups are defined by life-cycle stage: minors are those under 18; young adults are aged 18–35; middle-aged adults are aged 36–59; and older adults are those aged 60 and above. Differences across life-cycle stages are pronounced. Digital literacy shows a significant positive effect for both young and middle-aged groups, with the middle-aged group having the lowest turning point (0.2707), followed by the young group (0.4221). In contrast, the coefficients for minors and older adults are generally smaller, and the quadratic term is not statistically significant for either group. The corresponding turning points are 0.6277 for minors and 0.6662 for older adults.

4.4. Mechanism Analysis

This study examines several theoretical paths, paying special attention to its potential nonlinear effects. First of all, cognitive effect: digital literacy improves the ability to obtain and process environmental information, but its impact may not be linear, and there may be a decrease in marginal benefits or other complex situations. Secondly, social effect: digital literacy promotes social interaction and information dissemination, and enhances residents’ trust in the government and social institutions. Third, income effect: digital literacy can affect family income by improving human capital and employment opportunities, thus promoting the expansion of consumption. The results of columns (1) to (3) in

Table 9. The Mechanism Effect of Digital Literacy on HCE.

Explained Variable	Cognition of Environmental Stress	Social Trust	Income	HCE
	(1)	(2)	(3)	(4)	(5)	(6)
Digital literacy	2.7682 ***	1.4757 ***	0.0285 *	0.8492 ***	0.8122 ***	0.4881 ***
	(0.1675)	(0.1859)	(0.0148)	(0.0693)	(0.0694)	(0.0998)
Digital literacy2	−2.8392 ***	−1.1722 ***	0.0312	−0.8215 ***	−0.7338 ***	−0.6185 ***
	(0.2367)	(0.2599)	(0.0195)	(0.0976)	(0.0981)	(0.1315)
Cognition of environmental stress				−0.0047 ***
				(0.0017)
social trust					−0.0076 ***
					(0.0015)
Income						0.1927 ***
						(0.0393)
Control variables	Yes	Yes	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes	Yes	Yes
Province FE	Yes	Yes	Yes	Yes	Yes	Yes
R2	0.1057	0.1038	0.9641	0.2511	0.2641	0.2231
Observations	57,453	57,453	57,453	57,453	57,453	57,453

Notes: * and *** indicate significance at the 10% and 1% levels, respectively.

show that the linear terms of digital literacy have a significant positive impact on environmental awareness, social trust and income level. However, with the improvement in digital literacy, this positive impact will gradually weaken. Secondly, in terms of the communication path, the coefficients of environmental awareness and social trust are significantly negative in the carbon emission equation, which shows that both are effective emission reduction mechanisms. Digital literacy enhances residents’ environmental awareness through information dissemination, thus establishing the path of “cognition-behavior”.

Simultaneously, increased social trust promotes the spread of low-carbon knowledge within communities and imposes soft constraints on high-carbon behaviors through informal norms. However, the relatively small coefficients for these two pathways indicate that, without strong external incentives and accessible low-carbon infrastructure, relying solely on cognitive improvement or social norms is insufficient to overcome the “lock-in effect” of high-carbon lifestyles. Finally, the “consumption expansion” driven by the income effect of digital literacy largely counterbalances the emission reduction benefits from the cognitive and social dimensions. This helps explain why, in the early stages of digitalization, HCE tend to rise overall.

5. Further Analysis: Configurational Pathways of the Low-Carbon Transition in Household Consumption

5.1. Variable Description and Calibration

The MOA theory (Motivation–Opportunity–Ability), states that behavior and outcomes depend on motivation, ability, and external opportunities. It has been widely applied in research on consumer behavior, green consumption, and technology adoption. The theory shows that behavior is shaped by multiple factors, not determined by a single reason [59]. In the household consumption study, the MOA model explains the interaction between will, ability and opportunity, and provides a clear framework for low-carbon behavior. It has become an important method in the study of green consumption and sustainability. Unlike the TPB model, which focuses on internal psychology, the MOA framework considers both internal factors (such as intentions and values) and external conditions (such as technology, public resources and institutions) [60]. Therefore, it is very compatible with China’s dual-carbon strategy and its multi-level policy system. More importantly, in the digital age, green consumption depends not only on environmental intentions and values, but also on factors such as digital literacy, the ability to access information and digital infrastructure. Therefore, the MOA theory provides a clear framework for explaining how digital literacy supports the low-carbon transformation of household consumption.

First of all, motivation represents intrinsic values and intentions, which are the core elements of low-HCE. In this study, motivation is measured by the sense of environmental responsibility and the choice of clean fuel. The sense of environmental responsibility reflects the understanding of resource constraints, climate risks and sustainable development goals, which constitutes the moral foundation of green behavior. On the other hand, the use of clean fuel shows a practical preference for a low-carbon lifestyle. However, without supportive social and institutional conditions, green consumption is still difficult to achieve even if there is a will. Secondly, with the rapid development of urban digitalization in China and the acceleration of green transformation, the opportunity dimension is particularly important. This study takes digital infrastructure, environmental regulations and public environmental awareness as opportunity variables. Digital infrastructure affects the ways to obtain green products and low-carbon services, and supports green consumption. Environmental regulations guide green behavior through rules and incentives. The public’s environmental awareness reflects the influence of social norms and media, and acts as an external driver. Together, these factors define the opportunity structure for green consumption. Finally, ability is the mechanism that converts intention into action. In this study, ability includes digital literacy and cultural literacy. Digital literacy enables access to low-carbon information, use of digital payment for green incentives, understanding of energy-saving products, and operation of smart devices. It is a basic requirement for a green lifestyle in the digital age. Cultural literacy, shaped by education and values, increases awareness of sustainability and strengthens support for green living. It also supports long-term practice and transfer across generations. Together, these abilities form a “cognition-skill-action” chain that moves green ideas into daily behavior.

Overall, the MOA theory shows strong practical value and good contextual fit in this study. Under China’s goals of carbon neutrality and digital development, household green consumption reflects not only greater environmental awareness, but also better digital infrastructure, greener markets, and higher overall capability. In the QCA analysis, data calibration assigns set-membership scores to cases. This study uses the direct calibration method to convert variables into fuzzy sets. Thresholds of 0.95, 0.50, and 0.05 are used for full membership, the crossover point, and full non-membership, respectively. The final membership scores range from 0 to 1 (

Table 10. Variable selection and calibration.

Dimension	Variable	Thresholds for the Calibration Process
		Full Membership	Cross-Over Point	Full Non-Membership
Outcome Variable	HCE	9.4085	8.9504	8.6846
Motivation	Environmental Responsibility Awareness	7.7885	7.0529	5.9989
	Clean Fuel Choice	4.3937	3.4762	2.4820
Opportunity	Digital Infrastructure	0.3798	0.2300	0.0997
	Environmental Regulation	0.0074	0.0015	0.0002
	Public Environmental Attention	655.6962	275.7514	85.1367
Capability	Digital Literacy	0.2366	0.0197	−0.1038
	Cultural Literacy	0.2685	0.0170	−0.2073

).

5.2. Single-Factor Necessity Analysis

Before the configurational analysis, each condition is tested for necessity. A consistency score above 0.90 is commonly used to identify a necessary condition [61]. As shown in

Table 11. The necessity analysis of the QCA method.

Antecedent Variables	HCE		~HCE
	Consistency	Coverage	Consistency	Coverage
Environmental Responsibility Awareness	0.652	0.694	0.607	0.621
~Environmental Responsibility Awareness	0.645	0.631	0.701	0.659
Clean Fuel Choice	0.705	0.673	0.675	0.619
~Clean Fuel Choice	0.601	0.658	0.644	0.677
Digital Infrastructure	0.665	0.695	0.586	0.588
~Digital Infrastructure	0.605	0.603	0.696	0.666
Environmental Regulation	0.557	0.636	0.638	0.699
~Environmental Regulation	0.737	0.679	0.668	0.591
Environmental Attention	0.744	0.797	0.568	0.584
~Environmental Attention	0.612	0.596	0.802	0.750
Digital Literacy	0.695	0.714	0.597	0.589
~Digital Literacy	0.600	0.608	0.710	0.691
Cultural Literacy	0.715	0.721	0.637	0.616
~Cultural Literacy	0.619	0.640	0.711	0.706

Note: “~” represents the logical operation “not”.

, all consistency values are below 0.90. This means that no single condition is necessary for either high or low HCE. These results indicate that household emissions are shaped by multiple factors, in line with the MOA framework that behavior results from the joint effect of several conditions.

5.3. Configuration Analysis

This study uses a QCA method to move beyond the limits of conventional regression models that focus on the net effect of a single variable. We examine multiple configurations of conditions that lead to high-carbon and low-carbon household consumption-related emissions. The analysis includes seven conditions: environmental responsibility, clean fuel choice, digital infrastructure, environmental regulation, public environmental concern, digital literacy, and cultural literacy. As shown in

Table 12. Configurations that produce high.

Antecedent Variables	High HCE		Low HCE
	A1	A2	NA1	NA2	NA3
Environmental Responsibility Awareness	⨂	⨂	●	⨂	▲
Clean Fuel Choice	⨂	⨂		▲	⨂
Digital Infrastructure	☐	⨂		⨂	●
Environmental Regulation	●	●			⨂
Environmental Attention	⨂	☐	●	●	●
Digital Literacy	☐		●	●
Cultural Literacy		⨂	▲	●	●
Consistency	0.957	0.958	0.958	0.935	0.955
PRI	0.834	0.787	0.905	0.828	0.863
Coverage	0.266	0.240	0.395	0.332	0.281
Unique coverage	0.039	0.013	0.055	0.019	0.037
Summary consistency	0.954		0.926
Aggregate coverage	0.279		0.515

Note: ● and ⨂, respectively, indicate the presence or absence of core conditions; ▲ and ☐, respectively, indicate the presence or absence of edge conditions. This form can clearly indicate the relative importance of each condition in the configuration.

, both high-carbon emissions (A1–A2) and low-carbon emissions (NA1–NA3) can be explained by multiple, functionally equivalent pathways. There are significant differences in conditional combination patterns under these paths, which highlights the systematic complexity and asymmetry of carbon emission behavior drivers. The consistency of the solutions in each group of configurations exceeds the recommended threshold of 0.9, indicating that it has strong explanatory power.

A1 and A2 have an obvious common feature: in both paths, environmental regulation is a core factor of the status quo, while the elements of motivation and ability are completely missing. This shows that there is a practical dilemma. Under the pressure of external regulation, if internal motivation and key enabling conditions are generally absent, the household consumption model may remain in a high-carbon state. More specifically, when mandatory external policy constraints fail to activate residents’ internal pro-environmental intentions, and are not supported by clean energy options or accessible digital tools, policy measures may have limited effectiveness on their own. Households may be unable to translate regulatory pressure into real low-carbon consumption because alternatives are inconvenient or because awareness is limited.

The core mechanism behind the low-carbon pathways is “public environmental concern,” which acts as a shared hub (NA1–NA3). The realization of low-carbon results depends on a combination of digital literacy, cultural literacy and digital infrastructure. This shows that at the family level, low-carbon transformation depends more on the external social atmosphere and public discussion that people can perceive, learn and imitate. When public concern for the environment becomes a lasting and obvious social norm, families are more likely to regard low-carbon choices as reasonable and necessary choices in daily life. This can greatly reduce the path dependence on high-carbon consumption. Low-HCE paths can be summarized into three types. This path emphasizes that digital literacy enables families to identify low-carbon information more effectively, compare product energy efficiency, and obtain green consumption and service scenarios. In this way, external consensus is transformed into actionable consumption decision-making. Secondly, it is the path of “ability compensation and behavior change”. Even with the lack of digital infrastructure and weak environmental awareness, the low-carbon goal can still be achieved by closely combining the public’s environmental awareness, digital literacy and cultural literacy. This path shows that when external technical opportunities are limited, the accumulated capacity can play a compensatory role. They encourage families to make more cautious consumption decisions and choose cleaner ways to use energy on a cleaner basis, so as to achieve low-carbon goals. Third, it is the path of “infrastructure support and cultural internalization”. Even if environmental supervision is weak and there is no choice of clean fuels, the low-carbon goal can still be achieved through the close combination of digital infrastructure, public environmental awareness and cultural literacy. This path shows that low-carbon consumption does not necessarily depend on strict regulation or a single form of energy alternative. When the digital infrastructure is fully developed, green information is easy to obtain, and cultural literacy supports long-term value consistency and self-restraint, families can still form relatively stable low-carbon consumption habits.

6. Discussion

This study systematically evaluates how digital literacy affects the HCE from the micro level, focusing on the mechanism behind it, nonlinear dynamics and multiple ways of influence. Through such research, it provides a new empirical basis for understanding the transformation of low-HCE of families in an increasingly digital environment. Compared with existing research, our findings expand the relevant literature from several important aspects.

First of all, we found that there is a statistically significant inverted U-type nonlinear relationship between digital literacy and HCE. This result helps to explain the contradictory evidence in the literature about whether digitalization will increase or reduce HCE. On the one hand, existing research believes that the advancement of digital technology and the improvement in digital skills may increase energy consumption and emissions by expanding consumption, strengthening online shopping and increasing the demand for on-demand services. On the other hand, they also emphasize the possibility of achieving emission reductions by improving energy efficiency, enhancing information transparency and guiding green consumption more strongly. Our research results show that these two effects are not mutually exclusive. On the contrary, they play a leading role in different stages of the development of digital literacy. At a lower level, the effect of consumption expansion and technological rebound is more significant. Once digital literacy exceeds a threshold, efficiency improvement, consumption structure upgrading and the internalization of low-carbon behavior will become more and more important, thus reducing HCE related to household consumption. This result addresses a key limitation of previous studies, that is, these studies rely on linear models, making it difficult to capture the long-term and dynamic effects of digitalization.

Secondly, from the perspective of consumption structure, the impact of digital literacy is obviously different in different consumption categories. Spending on transportation and communication is most sensitive to changes in digital literacy. With the improvement in digital literacy, this category has shown strong emission reduction potential. This model reflects the role of online substitution, information-driven travel and efficiency improvement in this field. In contrast, the turning point of housing-related emissions appeared the earliest. This shows that household energy use in the housing sector is moving faster from a digitally driven expansion phase to an adjustment phase dominated by energy efficiency improvements and finer energy management. This finding shows that the low-carbon transformation in the field of household consumption is not carried out in various fields at the same time, but in a clear order in different sectors, and its transformation paths are also different.

Third, our research results confirm that digital literacy does not affect the relationship between family carbon emissions and consumption through a single channel. On the contrary, it will function simultaneously through multiple mechanisms, including perceived environmental pressure, social trust and income levels. On the one hand, digital literacy enhances the intrinsic motivation of low-carbon consumption by improving information acquisition and risk awareness. On the other hand, as a form of human capital, it can increase employment opportunities and income, thus enhancing purchasing power, and may exert upward pressure on carbon emissions. The coexistence of these mechanisms further explains why the overall effect of digital literacy is non-linear. This also shows that relying on a single intermediary channel is not enough to fully understand the impact of digital literacy on the environment.

Finally, our analysis shows that there is no single sufficient condition for the transformation of household consumption to low-carbon. On the contrary, this transformation is determined by the dynamic interaction between multiple factors, including digital literacy, environmental responsibility, public environmental awareness, digital infrastructure and institutional environment. In different contexts, digital literacy can work together with a stronger sense of environmental responsibility and public attention, so as to cultivate low-carbon consumption values. In addition, when the infrastructure and institutional conditions are supportive, it can also strengthen the family and amplify the emission reduction effect of technology. This finding goes beyond the category of “average effect” that traditional regression analysis focuses on. It highlights the diversity of family decarbonization paths and its dependence on environmental conditions.

7. Conclusions and Policy Implications

7.1. Conclusions

Using CFPS microdata from, this study combines a two-way fixed effect model with robustness testing, mechanism testing, heterogeneity analysis and fsQCA to systematically determine how digital literacy affects HCE. The research results show that:

First, there is a robust inverted U-shaped nonlinear relationship between digital literacy and HCE. At early stages of digital literacy improvement, wider use of digital technologies lowers transaction and information search costs. It also expands online shopping and digital service scenarios, and increases demand for logistics delivery and related services. These changes raise HCE. When digital literacy reaches a higher level, households become better at acquiring and processing information and applying digital tools. They are more able to use these tools for energy-use management, identifying green products, and improving consumption decisions. As a result, the consumption pattern shifts from “scale expansion” to “efficiency improvement,” and HCE decline.

Second, the effect of digital literacy on HCE differs structurally across consumption types. The strongest emission-reducing effect is observed in transportation and communication, followed by healthcare, and both effects become stronger as digital literacy increases. Both basic and development-oriented consumption display an overall inverted U-shaped pattern. However, the turning point appears earliest for housing-related consumption. As digital literacy improves, its effect shifts more quickly from an emission-increasing phase driven by consumption expansion to an emission-reducing phase dominated by energy-efficiency gains and energy-saving management.

Third, there are systematic differences in the effect of digital literacy among different groups. With the improvement in digital literacy, the emissions of rural residents, those with low human capital, low-income families and women have increased more. This may be because digital technology has stimulated more demand for carbon-intensive basic commodities that may be concentrated in these groups. However, the turning point threshold for emission reduction is relatively low for women and rural residents. This shows that once these groups acquire a certain amount of digital capabilities, they can transform digital skills into emission reduction practices at an earlier stage, so as to achieve low-carbon transformation faster.

Fourth, digital literacy affects the HCE through the dual transmission mechanism. On the one hand, digital literacy can improve residents’ perception of environmental pressure and enhance social trust. This prompts families to adopt low-carbon consumption models and energy-saving technologies, thus reducing HCE. Higher digital literacy can also increase the income of residents, which may expand the scale of consumption and accelerate the upgrading of consumption. In particular, it will increase the demand for energy-intensive goods and services, thus increasing the HCE.

Fifth, the low-carbon transformation in household consumption reflects the dynamic interaction between many factors, and multiple paths can coexist. For example, the path of “consensus-driven and digital support”, the path of “capacity compensation and behavioral transformation”, and the path of “infrastructure support and cultural internalization”. Digital literacy can be combined with environmental responsibility and public environmental concerns to promote the value transformation to low-carbon consumption endonously. Under the condition of perfect infrastructure, it can also provide support for families and enhance the effectiveness of technology in emission reduction.

7.2. Policy Implications

This study puts forward the following suggestions. First of all, policy efforts should take the improvement in digital literacy as a key entry point, while avoiding the belief that digitalization will automatically reduce emissions. The impact of digital literacy on HCE is obviously phased and non-linear, which shows that digitalization itself does not necessarily bring emission reduction. With a low level of digital literacy, digital technology is more likely to increase carbon emissions by reducing consumption barriers and expanding consumption. Therefore, policies to promote the popularization of digital skills should also include low-carbon guidance.

Secondly, policymakers should implement targeted policy programs that reflect the differences in emission reduction potentials in different consumption areas. In the process of digitalization, the response to emission reductions in areas such as transportation and communication and housing consumption is more rapid or more significant, indicating that the implementation of policies in these areas is expected to yield higher returns. On the one hand, by expanding intelligent transportation systems, promoting online alternative services, and strengthening information-based travel management, the digital emission reduction potential in the fields of transportation and communication can be further released. On the other hand, in the field of housing, policies should accelerate the popularization of smart home equipment, energy use monitoring and refined energy management systems. This helps families shift from “digital consumption expansion” to “digital energy efficiency management” as soon as possible. In contrast, stronger price signals and clearer institutional constraints are needed for consumption areas with weak or delayed emission reduction responses. This can prevent digital convenience from being transformed into continuous high-carbon consumption habits.

Thirdly, a low-carbon transition in household consumption does not rely on technology or policy alone. It is the result of joint effects from cognition, capability, and institutional conditions. Household mitigation policies should not be limited to subsidizing energy-efficient products or promoting single technologies. Instead, they should coordinate “soft constraints” with “hard conditions.” For example, policymakers can integrate low-carbon information disclosure, public environmental participation, and community-level digital governance into the household mitigation system. This can reduce both the cognitive and the implementation costs of low-carbon behavior.

Fourth, low-carbon transitions in household consumption can follow multiple coexisting pathways. In some regions or groups, the transition relies more on internal drivers such as values and a sense of responsibility. In areas with stronger digital infrastructure and more mature institutions, capability empowerment and technological substitution play a more decisive role. Policies should adopt differentiated strategies by group, region, and scenario to improve the sustainability and inclusiveness of low-carbon transitions at the household level.