Digital Ability and Livelihood Diversification in Rural China

Abstract

Livelihood diversification is an important strategy for rural households in developing countries, especially in China, which has the largest rural population in the world. In the current digital age, the existing literature lacks sufficient research on the influence of digital ability on livelihood diversification. Using survey data from 1914 rural households in China, this study evaluates the digital ability of rural households through item response theory. Additionally, the livelihood diversification of rural households is analyzed from two aspects: work type and industry. Finally, IV-Tobit models are set up to test the impact of digital ability on livelihood diversification. The results show that: (1) engaging in both agricultural production and employed work concurrently is the key strategy for rural households to diversify their livelihoods; (2) digital ability significantly promotes livelihood diversification, regarding both work type and industry; (3) digital ability’s capacity to diversify livelihood is particularly notable for low-income households, followed by the medium-income group, then the high-income group. These findings are meaningful for the sustainable improvement of rural households’ livelihoods.

1. Introduction

Livelihood diversification (LD) is an important process by which rural households expand and vary their activities to survive and improve their well-being [1]. First, LD can be seen as a strategy to diversify risk and reduce vulnerability to unpredictable crises such as natural disasters, extreme climate, market fluctuations, and diseases, which can help rural households increase their resilience and cope with external shocks [2,3,4]. Furthermore, diversifying livelihood has a significant impact on increasing food security and improving nutrition, as households can produce and consume a wider variety of crops and livestock [5,6]. In addition, as an important manifestation of LD, the transformation of rural households’ activities from those of traditional agriculture to non-agricultural activities can increase their income and well-being [7].

Existing research on LD mainly focuses on its strategies, effects and determinants. Studies exploring rural households’ livelihood strategies typically categorize them into two primary classifications: on-farm and off-farm strategies [7,8]. Regrettably, there exists a scarcity of additional breakdowns for industries. Numerous studies have demonstrated the substantial benefits that LD brings to the growth of rural households’ welfare and the advancement of rural development in underdeveloped areas [9,10,11]. Thus, the methods through which rural households diversify their livelihoods hold particular significance [7]. Existing studies have analyzed factors influencing LD from different aspects, including gender, age, and education of the household head [12]; family size [2], risk strategy [13], land possession [14], and asset possession [6,7] of the household. Studies have also found the effects of cash transfers [15], environmental regulation [16], and tourism [17] on LD.

In the present era of digital advancements, the influence of digital technology on LD has garnered increasing attention. For instance, research conducted among Maasai demonstrates that the utilization of mobile phones can effectively support pastoralists in expanding their range of livelihood activities [18]. Studies have also shown that the widespread ownership of mobile phones increases the possibility of rural households diversifying their livelihoods [19]. Similarly, studies carried out in China have revealed that the adoption of digital technology holds considerable potential for enhancing the LD of rural households, with notable variations observed across distinct groups [20]. However, little attention had been paid to examining the impact of digital ability on LD. First, since digital technology is not a panacea and relies heavily on users’ abilities to effectively leverage its potential benefits [21,22,23], the effect of digital ability deserves more attention in order to achieve digital inclusion and avoid digital divide [24]. Second, as digital technologies become increasingly popularized in rural China, the adoption of digital technologies is no longer a good indicator of digital technology utilization, rather, digital ability becomes a more critical indicator [25]. Third, digital ability forms the nucleus of the overall mechanism for achieving the sustainable development goals through digital technologies [26]. Therefore, the impact of digital ability on LD should be studied in depth.

To address this issue, we empirically tested the impact of digital ability on LD using rigorous econometric models. The data were obtained from a survey of 1914 rural households in eight provinces in China. Digital ability was measured by the item response theory (IRT) and LD was measured using the Simpson index. To analyze LD in detail, we examined both the diversity of the work type and industry in which rural households are engaged. IV-Tobit models were adopted for empirical estimation. For further research, robustness checks and group analysis were conducted.

Our study makes the following contributions to the literature. First, we reflected on rural households’ utilization of digital technology from a new perspective of digital ability. Differing from the literature on technology adoption [18,20], we studied the impact of digital technology more accurately based on the background of high digital technology penetration in China. Second, we highlight the LD of rural households from the two new perspectives of work type and industry. Unlike the previous literature, which measures LD by the number of livelihood activities or the share of income sources [7,14], we divided the livelihood strategies into three types and seven industries, offering a more comprehensive and detailed understanding of rural households’ livelihood strategy choices. Third, we conducted further discussions from the perspective of income, which provides new ideas for realizing the sustainable development of rural households’ livelihoods.

2. Materials and Methods

2.1. Data

The data used in this study were obtained from a rural survey in eight provinces of China conducted by the Research Group of Rural Market and Information of the China Agricultural University. The provinces were selected from China’s three main regions with varying labor distributions and economic development. The survey was carried out from the end of 2021 to the beginning of 2022, reflecting the characteristics of rural households in 2021. To increase the reliability of the survey, a pre-survey was conducted in two provinces. Simultaneously, the survey was conducted through face-to-face interviews lasting approximately one hour per household. The investigators are senior undergraduates or postgraduates majoring in agricultural economics, recruited from across the country, and have received professional training.

The survey consists of two sets of questionnaires: household and village, covering a broad range of topics, including rural households’ information acquisition, family characteristics, employment status, income situation, and villages’ basic characteristics. After data cleaning, 1914 samples were used for the analysis of this study.

2.2. Methodology

The diversity of livelihoods can be categorized based on the research focus, such as sector (farm or non-farm), function (employed or self-employment), and location (on-farm or off-farm) [27]. This study focuses on the livelihood strategies of rural households and classifies LD according to the work type (LD1) and industry (LD2) by combining the factors of sector, function, and location used in previous studies. The work types include (1) agricultural production, (2) employed work, and (3) self-employment (self-employment refers to the state of not working for an employer but finding work for yourself or having your own business). The industries are grouped into seven categories (1) agriculture, forestry, animal husbandry, and fishery; (2) manufacturing; (3) construction; (4) mining, production and supply of electricity, heat, gas, and water, transport, storage, and post; (5) wholesale and retail trades, hotels and catering services; (6) services to households, repair, and other services; and (7) education, culture, sports, entertainment, health and social service, public management, social security, and social organization.

LD is generally measured by count data or a composite index, such as the number of livelihood strategies [7] or the proportion share of livelihood strategies [12,13]. Considering that the Simpson index accounts for the balance and evenness of livelihood strategies in addition to their number [20], we use it to calculate the LD of rural households:

$${LD}_i=1-{\sum }_{m=1}^n{P}_{i,m}^2$$

(1)

where LD_i denotes the livelihood diversification of household i; n refers to the number of work types (industries); m is a specific type (industry) of work; P_i,m represents the share of working time from type (industry) m of household i. According to the above formula, LD ranges from 0 to 1 − 1/n, and its value is positively correlated with the diversification of rural households’ livelihood.

The estimation strategy of digital ability is based on IRT. This method is used to determine latent traits that are difficult to directly measure, such as cognitive abilities, personality traits, and attitudes. In agricultural economics, IRT is used to measure the information acquisition ability [28] or digital ability [25,29] of rural households. In the IRT strategy, respondents are asked to answer a set of binary questions of varying difficulty and discrimination, and their probability of success on this set of questions is positively correlated with their ability. In our study, this set of questions is “whether to access a certain digital channel”, which is defined as 1 if yes and 0 if no. According to the previous literature [25,29], digital channels included WeChat, short video platforms, web pages, professional applications, internet forums, and government platforms. A two-parameter IRT model was established for estimation [30]:

$${\pi }_{ij}=\frac{\mathrm{e}\mathrm{x}\mathrm{p}\left[a_j\right({Abil}_i-b_j\left)\right]}{1+\mathrm{e}\mathrm{x}\mathrm{p}\left[a_j\right({Abil}_i-b_j\left)\right]}, {Abil}_i~N\left(\mathrm{0,1}\right)$$

(2)

where π_ij denotes the probability that household i has access to channel j; Abil_i refers to the digital ability of household i, and a larger value indicates stronger digital ability; a_j is the discrimination parameter of channel j, and a larger value can distinguish better between similar digital abilities; and b_j is the difficulty parameter of channel j, and a higher value means the channel is harder to access. In actual estimation, the discrimination and difficulty parameters of each channel are estimated first, followed by the ability parameters based on different channel combinations.

In model estimation, given that LD is censored and data accumulate at 0, Tobit models [31] were established to explore the impact of digital ability on livelihood diversification:

$${LD}_i={\alpha }_i+{\beta }_i{Abil}_i+{\gamma }_i{C}_i+{\delta }_i{Prov}_i+{\epsilon }_i$$

(3)

where LD_i denotes the livelihood diversification of household i; Abil_i refers to the digital ability of household i; C_i represents the household characteristics that comprise Gen (gender), Age, Sage (age squared), Edu (school years), and Risk (risk attitude) of the household head; Popu (family size), Child (number of children under 16), Elder (number of elderly people over 65), Soci (social networks), Land (planting area), and Asset of the household; Prov_i is the province dummy variable.

Considering the potential reverse causality between digital ability and LD [9,18,20], we introduced an instrumental variable (IV)–village internet stability (VIS)–to address endogeneity. Intuitively, the exclusion restriction is expected to be valid because VIS depends on the distribution of network base stations and the frequency of the network, both of which are determined by the government and guided by technical principles. As a result, these factors introduce exogenous variations that impact households’ digital abilities, while having no direct influence on their livelihood choices. When considering the inclusion restriction, it becomes apparent that households’ access to digital channels serves as an indicator of their level of digital ability. This access is primarily facilitated through the internet, and a reliable and stable internet infrastructure plays a crucial role in enabling such access. Consequently, VIS exhibits a strong correlation with households’ digital abilities, further supporting the inclusion restriction. Thus, we adopted the IV-Tobit model to control for endogeneity and estimate it with a minimum χ² approach [32]. Stata 15.0 is used to estimate the above models.

2.3. Descriptive Statistics

Table 1. Variable definitions and descriptive statistics.

Variable	Definition	Mean	Std. Dev.
LD1	Livelihood diversification (work type, measured by Simpson index)	0.21	0.21
LD2	Livelihood diversification (work industry, measured by Simpson index)	0.32	0.25
Core explanatory variables
Abil	Digital ability (measured by item response theory)	0.00	0.80
Instrumental variable
VIS	Village internet stability (very unstable = 1, unstable = 2, average = 3, stable = 4, very stable = 5)	3.87	0.83
Control variables
Gen	Gender of household head (male = 1, female = 0)	0.83	0.38
Age	Age of household head (years)	49.62	11.56
Sage	Age squared of household head	2596.04	1159.88
Edu	School years of household head (years)	8.81	3.53
Risk	Risk attitude of household head (degree of agreement with the expression “when new things, new policies come out, I’d love to try them”, strongly disagree = 1, slightly disagree = 2, normal = 3, slightly agree = 4, strongly agree = 5)	3.44	1.09
Popu	Number of family members (pcs)	4.02	1.53
Child	Number of children under 16 in the family (pcs)	0.71	0.88
Elder	Number of elderly people over 65 in the family (pcs)	0.47	0.76
Soci	Social networks (family annual gift money expenditure, USD)	957.11	1128.17
Land	Planting land area (ha)	0.37	2.33
Asset	Family assets (cash, deposits, and financial products, 10,000 USD)	1.91	3.61
N	Number of observations	1914

lists the variable definitions and descriptive statistics. The average LD1 of rural households is 0.21, and LD2 is 0.32, indicating that many samples have a low livelihood diversity index. LD2, measured by work industry (a more specific indicator), is larger than LD1, measured by type, which aligns with common sense, and implies that LD2 can provide more detailed insights into the diversification of livelihood. The estimates of digital ability have a mean of 0.00 and range from −3 to 3, indicating that our prior assumption is reasonable [33].

To facilitate the analysis, we categorized rural households into two groups: high-ability (with ability values greater than 0) and low-ability (with ability values less than 0).

Table 2. Differences in the characteristics of high-ability and low-ability groups.

Variables	Low-Ability Households	High-Ability Households	Difference in Means
LD1	0.19	0.24	−0.05 ***
LD2	0.29	0.36	−0.06 ***
Abil	−0.59	0.79	−1.38 ***
VIS	3.64	4.18	−0.54 ***
Gen	0.84	0.80	0.04 **
Age	52.15	46.27	5.87 ***
Edu	7.77	10.19	−2.42 ***
Risk	3.26	3.69	−0.43 ***
Popu	3.76	4.35	−0.58 ***
Child	0.63	0.82	−0.19 ***
Elder	0.49	0.43	0.06 *
Soci	791.68	176.87	−385.19 ***
Land	0.24	0.55	−0.32 ***
Asset	1.40	2.59	−1.19 ***
N	1092	822

* p < 0.1, ** p < 0.05, *** p < 0.01.

shows the differences in characteristics between the high-ability group and the low-ability group. The comparison results show that the LD of the high-ability group is significantly greater than that of the low-ability group, both in terms of work type and industry, which preliminarily illustrates the role of digital ability in promoting LD. Naturally, the high-ability group tends to exhibit significantly higher levels of digital ability and internet stability compared to the low-ability group. From the perspective of household characteristics, the household heads in the high-ability group were younger, had higher educational levels, and had more positive risk attitudes. Simultaneously, high-ability households were more likely to have larger family sizes, raise more children, have wider social networks, and own more assets. These observations are consistent with a previous study [20].

3. Results and Discussion

3.1. Results of Livelihood Diversification

The details of the diversity of work types are shown in Figure 1. Employed work is the most basic livelihood strategy for rural households, followed by agricultural production, and finally self-employment. More than half of the rural households engaged in more than two types of work, mainly agricultural production and employed work, which is consistent with the reality of part-time farming in China [34]. The comparison between the high-ability group and the low-ability group shows that the livelihood of the former is more diversified, with 53% of the households engaged in two types of work and 12% of the households engaged in three types of work. The high-ability group has higher participation in the non-agricultural sector, especially in self-employment. Specifically, 7% of the high-ability households participated in both self-employment and agricultural production, 11% engaged in both self-employment and employed work, and 12% participated in all the three types, indicating that the high-ability group improved their LD through self-employment. These findings suggest that digital ability can promote rural households’ LD by increasing their participation in employed work and self-employment.

The number of work industries is reported in

Table 3. Number of work industries with different digital abilities.

Number of Industries	Low-Ability Households		High-Ability Households
	Frequency	Percent (%)	Frequency	Percent (%)
1	369	33.8	198	24.1
2	472	43.2	332	40.4
3	202	18.5	223	27.1
≥4	49	4.5	69	8.4
Total	1092	100.0	822	100.0

. Nearly half of the households were involved in two industries. The comparison between the high-ability group and the low-ability group shows that the average number of industries in the former is 2.2, which is higher than that of the low-ability group (1.9). Compared with the low-ability group, more households in the high-ability group were engaged in three or more industries, among which 27.1% of households engaged in three industries and 8.4% of households engaged in four or more industries. This suggests that digital ability is beneficial for rural households to diversify their livelihoods.

Table 4. Distribution of work industries with different digital abilities.

Industry	Low-Ability Households		High-Ability Households
	Frequency	Percent (%)	Frequency	Percent (%)
Primary and Secondary	254	35.6	84	13.6
Secondary	16	2.2	10	1.6
Tertiary	34	4.8	64	10.4
Primary and Tertiary	176	24.6	233	37.8
Secondary and Tertiary	99	13.9	96	15.6
Primary, Secondary, and Tertiary	135	18.9	129	20.9
Total	714	100.0	616	100.0

shows the industry choices of rural households in the process of expanding LD. Among the 1330 households with multiple livelihood industries, 1011 households were engaged in both agricultural and non-agricultural sectors, accounting for 76.0%, and only 319 households were completely engaged in non-agricultural sectors, accounting for 24.0%. This indicates that balancing agricultural production and non-agricultural work remains the most common approach for rural households to diversify their livelihoods. The comparison between the high-ability group and the low-ability group shows that a considerable part of the low-ability group was still concentrated in the primary industry or the secondary industry, accounting for 37.8%, which is much higher than that of the high-ability group (15.2%). Furthermore, 84.8% of the households in the high-ability group were involved in the tertiary industry when expanding their LD. This shows that digital abilities not only help rural households obtain more job opportunities, but also promote high-quality employment.

3.2. Results of Digital Ability

Table 5. Discrimination and difficulty of IRT models.

Channel	Discrimination		Difficulty
WeChat	1.934 ***	(0.162)	0.149 ***	(0.038)
Short video platforms	1.469 ***	(0.115)	0.810 ***	(0.058)
Web pages	2.391 ***	(0.222)	0.950 ***	(0.050)
Professional applications	1.464 ***	(0.118)	1.119 ***	(0.071)
Internet forums	2.674 ***	(0.390)	2.311 ***	(0.138)
Government platforms	1.688 ***	(0.137)	1.159 ***	(0.067)

*** p < 0.01; standard errors in parentheses.

reports the discrimination and difficulty parameters of the IRT model. The discrimination parameter indicates that internet forums play the most significant role in distinguishing digital ability, while professional applications play the least significant role. The comparison of difficulty parameters shows that internet forums are the most difficult channels, while WeChat is the easiest to access [25]. Due to the low immediacy and the high difficulty of operation of an internet forum, its difficulty and discrimination parameters are large. WeChat provides instant text, picture, voice, and video messages; these diversified information exchange methods provide rural households with a lower entry threshold.

Figure 2 intuitively shows the characteristics of the discrimination and difficulty of each channel. In Figure 2a, when the probability of accessing a channel is 0.5, the corresponding ability value is the difficulty of the channel [35]. The probability of accessing WeChat was higher than that of an internet forum at any ability level, indicating that WeChat is less difficult to access. In other words, a person would need a low ability level to access WeChat, but requires a higher ability level to access an internet forum. Furthermore, there exists a positive correlation between channel discrimination and the slope of the curve surrounding the difficulty value. A steeper curve indicates an enhanced ability to distinguish households with similar digital abilities [35]. The discrimination of channels can also be displayed more intuitively through Figure 2b, which illustrates the information provided by each channel in estimating digital ability. The height of the curve reflects the amount of information it provides in estimating the ability value, which is positively related to the discrimination of channels. It is obvious that internet forums provide the most information, followed by web pages and WeChat. The findings in the figures are consistent with the estimated coefficients.

Table 6. Predicted scores for digital ability.

Channel	Score	Channel	Score
None	−0.838	Internet forums only	0.154
Professional applications only	−0.224	Two channels	0.226–0.747
Short video platforms only	−0.223	Three channels	0.646–1.175
Government platforms only	−0.148	Four channels	1.077–1.501
WeChat only	−0.067	Five channels	1.612–1.903
Web pages only	0.073	Six channels	2.293

Since we cannot list all the channel combinations (49 in total), we list them by the number of channels.

presents the predicted scores of digital abilities under different channel combinations. In general, the higher the difficulty and discrimination of the channels that households could access, the greater their digital abilities. As the number of accessible channels increases, the corresponding score becomes larger, implying that rural households with access to a wider range of channels have stronger digital abilities [28].

3.3. Effect of Digital Ability on Livelihood Diversification

Table 7. Marginal effect of digital ability on livelihood diversification.

Variable	LD1 (Work Type)		LD2 (Work Industry)
	Tobit	IV-Tobit	Tobit	IV-Tobit
Abil	0.003	0.174 ***	0.013	0.244 ***
	(0.011)	(0.063)	(0.010)	(0.064)
Gen	−0.040 **	−0.023	−0.046 **	−0.024
	(0.020)	(0.022)	(0.019)	(0.023)
Age	0.011 **	0.012 **	0.012 **	0.013 **
	(0.006)	(0.006)	(0.005)	(0.006)
Sage	−0.000 *	−0.000	−0.000 **	−0.000
	(0.000)	(0.000)	(0.000)	(0.000)
Edu	0.005 *	−0.005	0.006 **	−0.007
	(0.003)	(0.004)	(0.003)	(0.004)
Risk	0.002	−0.012	−0.004	−0.021 **
	(0.007)	(0.009)	(0.007)	(0.009)
Popu	0.068 ***	0.054 ***	0.105 ***	0.086 ***
	(0.008)	(0.010)	(0.007)	(0.010)
Child	−0.029 **	−0.008	−0.069 ***	−0.042 ***
	(0.012)	(0.015)	(0.012)	(0.015)
Elder	−0.043 ***	−0.042 ***	−0.099 ***	−0.097 ***
	(0.013)	(0.013)	(0.012)	(0.013)
Soci	0.020 ***	0.010	0.034 ***	0.020 ***
	(0.007)	(0.007)	(0.007)	(0.008)
Land	0.188 ***	0.170 ***	0.031	0.007
	(0.037)	(0.026)	(0.030)	(0.027)
Asset	0.039 ***	0.009	0.022 *	−0.018
	(0.012)	(0.017)	(0.012)	(0.017)
Cons	−0.699 ***	−0.501 ***	−0.635 ***	−0.367 **
	(0.150)	(0.165)	(0.138)	(0.168)
Region	control	control	control	control
Wald test of exogeneity	–	8.56 ***	–	16.56 ***
F test in the first stage	–	47.46	–	47.46
VIS to Abil in the first stage	–	0.166 ***	–	0.166 ***
		(0.020)		(0.020)
N	1914	1914	1914	1914

* p < 0.1, ** p < 0.05, *** p < 0.01; standard errors in parentheses.

presents the impact of digital ability on LD. The second and third columns are the regression results of digital ability on work type diversity. The coefficient of ability in the Tobit model is positive, which is consistent with our expectations, but not significant, probably because of the existence of endogeneity. The IV-Tobit model shows that VIS is significantly positively correlated with ability in the first stage, and the F value is greater than 10, indicating that there is no weak instrumental variable problem. The result of the Wald test rejects the null hypothesis “ability is an exogenous variable” at the 1% level, indicating that the IV-Tobit model is more suitable than the Tobit model. The IV-Tobit model results indicate that digital ability promotes the diversity of households’ work types.

The fourth and fifth columns present the effect of digital ability on the diversification of work industry. Both the exogenous test of ability and the weak IV test of the VIS significantly reject the null hypothesis, which supports the IV-Tobit model. The coefficient of digital ability is significantly positive, indicating that with the improvement of households’ digital ability, the diversity of their work industries will increase. The comparison of the coefficients of digital ability in work types and industries shows that the latter is greater, indicating that digital ability plays a greater role in promoting industry diversification. Intuitively, it is easier for rural households to change industries than to change work types. This is also consistent with our previous analysis.

Digital abilities can contribute to rural households’ LD in several ways. First, rural laborers with strong digital abilities can broaden their job search channels, lower job search expenses [36], and access a wider range of job opportunities, thus expanding their range of livelihood options. Second, digital abilities help rural households acquire new skills, consequently enhancing their employability in various contexts [37]. Third, high digital abilities empower rural households by improving their market linkages and value chain integration. This provides opportunities for value-added processing and agritourism, contributing to the diversification of their livelihoods. Fourth, digital abilities enable rural households to access more financial support, facilitating their investments in different livelihood activities [38].

Regarding the control variables, age has a positive effect on LD, indicating that as rural laborers grow older, their experience increases, which facilitates the development of diverse livelihood strategies. Family size is positively associated with LD; larger households can take advantage of increasing returns to scale in housework, making it easier for family members to engage in multiple livelihood activities [39]. The negative correlation between caring for children and the elderly and LD indicates that taking care of family members will hinder labor participation. This demonstrates the necessity of digital abilities to improve LD from another perspective. High digital abilities can provide rural households with the opportunity to engage in remote work through digital platforms. This flexible form of work can improve labor participation, especially for those who have family members requiring care, allowing them to balance work and family commitments more effectively [37,40]. A social network promotes the diversity of work industry, and land has a positive effect on work type diversity.

3.4. Further Discussion

3.4.1. Robustness Check

To verify the robustness of the estimation results, we used two other methods for re-estimation. First, we defined the situation where the household engages in two or more livelihood strategies as multiple livelihoods, assigning a value of 1 and others as 0. Second, we re-estimated LD with the maximization index (The maximization index was calculated as: ${LD}_i=P_{i,max}/{\sum }_{m=1}^n{P}_{i,m}$, where LD_i denotes the livelihood diversification of household i; n is the number of work types (industries); m refers to a specific type (industry) of work; P_i,max is the working time of type (industry) m with the longest working time of household i; P_i,m is the working time of type (industry) m of household i. The maximization index is negatively correlated with diversity). The results of exogeneity and instrumental variable tests support our choice of models controlling for endogeneity, as shown in

Table 8. Robustness test of the effect of digital ability on livelihood diversification.

Variable	Multiple Livelihoods		Maximization Index
	Type	Industry	Type	Industry
Abil	0.486 *	1.088 ***	−0.147 ***	−0.209 ***
	(0.248)	(0.288)	(0.053)	(0.057)
Control variables	control	control	control	control
Region	control	control	control	control
Wald test of exogeneity	4.50 **	16.66 ***	8.70 ***	15.44 ***
F test in the first stage	47.46	47.46	47.46	47.46
VIS to Abil in the first stage	0.166 ***	0.166 ***	0.166 ***	0.166 ***
	(0.020)	(0.020)	(0.020)	(0.020)
N	1914	1914	1914	1914

* p < 0.1, ** p < 0.05, *** p < 0.01; standard errors in parentheses.

. The estimates for multiple livelihoods show that digital abilities promote the diversification of rural households’ work types and industries. Both the coefficient and significance of the work type are lower than those of the work industry, indicating that the promotion effect of digital abilities is more reflected in industry diversification. The results of the maximization index show that digital abilities are negatively correlated with the maximization index of the work type and work industry, indicating that digital abilities have a positive effect on the diversity of work types and industries. Moreover, the effect of work industry is larger than that of the work type. These results are consistent with the analysis above, confirming the robustness of the results in Table 7.

3.4.2. Relationship between Digital Ability, Livelihood Diversification, and Income

The ultimate goal of LD is to achieve stable income growth, so we continue to discuss the relationship between digital ability, LD, and household income. Figure 3 presents the households’ income under different digital abilities and livelihood strategies. In Figure 3a, the income of households tends to increase as the number of work types they engage in increases. This observation emphasizes the role of LD in enhancing household income. Having multiple sources of income and engaging in different types of work can contribute to higher earnings and overall financial well-being for rural households. When comparing households with the same types of work, it is observed that the high-ability group tends to have higher incomes than the low-ability group, which is consistent with an empirical study [29]. This observation highlights the role of digital ability in promoting the income of rural households. Figure 3b demonstrates a trend similar to Figure 3a, indicating that the income of rural households increases with both the expansion of work industries and the improvement of digital abilities. These findings further reaffirm the significance of enhancing digital abilities and promoting LD.

Furthermore, we divided rural households into three groups according to household income: low, medium, and high.

Table 9. Marginal effect of digital ability on livelihood diversification at different income levels.

Variable	Type			Industry
	Low	Medium	High	Low	Medium	High
Abil	0.331 **	0.217 *	0.184 **	0.407 **	0.246 *	0.193 **
	(0.165)	(0.130)	(0.088)	(0.174)	(0.130)	(0.085)
Control variables	control	control	control	control	control	control
Region	control	control	control	control	control	control
Wald test of exogeneity	4.49 **	4.19 **	4.85 **	6.07 **	6.20 **	6.14 **
F test in the first stage	14.66	11.22	12.18	14.66	11.22	12.18
VIS to Abil in the first stage	0.126 ***	0.135 ***	0.216 ***	0.126 ***	0.135 ***	0.216 ***
	(0.027)	(0.034)	(0.044)	(0.027)	(0.034)	(0.044)
N	638	638	638	638	638	638

* p < 0.1, ** p < 0.05, *** p < 0.01; standard errors in parentheses.

shows the impact of digital ability on LD at different income levels. The first-stage regression results show that there is no weak instrumental variable problem, and the exogenous test results reject the null hypothesis of exogenous digital ability, which supports our choice of IV-Tobit models. The results indicate that digital ability has a positive impact on LD, with the greatest impact on low-income households, followed by medium-income households, and finally high-income households. Low-income households often face a capital constraint that limits their participation in various production and operation activities [9], leading to low LD. As a result, digital ability will have a greater marginal effect on enhancing livelihood diversity for this group [9,20]. The difference in the promotion effect of digital ability on rural households with varying incomes highlights its important role in narrowing income gaps and fostering common prosperity.

4. Conclusions

This study used digital ability to reflect rural households’ utilization of digital technology and analyzed their LD from two new perspectives: work type and industry. We found that the main livelihood strategy of rural households is employed work, showing that the non-agricultural sector has become an important source of livelihood for rural households and highlighting the effectiveness of China’s urbanization. The most important method through which rural households can diversify their livelihoods is by balancing agricultural production and employed work. As digital abilities continue to advance, self-employment assumes a growing significance in fostering livelihood diversity, and rural households’ involvement in the tertiary sector has emerged as an important means to expand their range of livelihood activities.

Recognizing the pivotal role non-agricultural employment plays in enhancing the LD of rural households, it is essential to prioritize policies aimed at fostering non-agricultural employment within rural communities. These policies encompass initiatives like providing skill training to rural households and bolstering social security measures. However, it is vital to acknowledge that excessive rural labor force migration could hinder the achievement of rural revitalization. Encouraging rural households to participate in local non-agricultural work not only diversifies their livelihoods and improves their welfare, but also spurs on the advancement of the rural economy. Consequently, appropriate attention should be given to policies that support the growth of diverse industries in rural regions, fortify the development of rural infrastructure, facilitate market access for rural enterprises, and extend credit support alongside preferential incentives.

The estimation results show that digital ability has a significant positive impact on LD. The promotion effect is more obvious in the low-income group, followed by the medium-income group, and finally the high-income group. Simultaneously, the relationship between digital ability, LD, and household income shows that both digital ability and LD have a positive relationship with household income, emphasizing the importance of improving digital ability and LD.

Given the beneficial influence of digital ability on advancing rural households’ LD, prioritizing strategies to enhance rural households’ digital ability is imperative. The government can invest in digital technology training programs to provide rural households with basic digital education, such as smartphone operation and internet navigation. These training programs can be implemented through avenues like agricultural technology extension stations and rural libraries. Additionally, offering online training courses would be an effective approach, enabling rural households to engage in learning at their own convenience, fitting their personal schedules and locations. Simultaneously, the construction of digital information platforms and the improvement of digital infrastructure should also be taken into consideration. To enhance digital inclusion, it is crucial that the measures mentioned above prioritize the needs and challenges faced by low-income households.

The promotion of digital ability in the context of LD not only boosts rural income but may also reduce inequality among different income groups, providing a new focus for increasing rural households’ income, alleviating income gaps, and promoting sustainable rural development. However, this study has some limitations. First, cross-sectional data pose challenges in observing the dynamic changes of rural households’ livelihood strategies. Second, constrained by the existing data, the current assessment of internet stability is evenly coded, which fails to accurately capture variations in its “quality”. Conducting more comprehensive long-term follow-up research would be more meaningful.