Socio-economic status, resilience, and vulnerability of households under COVID-19: Case of village-l

Socio-economic status

Socio-economic status (SES) of a household is an important factor in signifying the outcomes related to shocks and vulnerabilities. It can affect the results of an economic phenomenon in different ways. According to the American Psychological Association, the Socio-economic status (SES) is a person or group’s social class or status which is often calculated by combining education, income and occupation [10]. As stated by Adler & Snibble [11] households with higher SES have better access to health care, housing, knowledge and nutrition as compared to lower SES households. Several research studies have focused on the impacts of different shocks on households based on the status. Some have highlighted the impacts of economic shocks such as joblessness or income decline and their effects on the household’s status [12,13]. While others have argued on the health shocks effects on individuals health and in turn the productivity of the workers finally leading to decline in income level of the households [14]. There is still very few literature articles related to the impacts of shocks on the socio-economic status of different households and communities in the developing world.

Some researchers have highlighted the relationship between household status and shocks [15]. Other researchers argue that the effects of economic shocks on households are greater in households that are poor due to insufficient income to cushion against the external shock [16]. In terms of effects on communities and household, shocks at village level are usually not considered to affect the permanent income of the individual households. Others argued that household consumption declines due to negative shocks that impact the permanent income of the households [17,18].

Economists have formalized different form of models for the estimation of household response towards risks and shocks. Consumption response to income shock is one such model which focuses on use of combination of consumption, income or wealth in response to shocks [19,20]. Kruger and Perii [21] used consumption and wealth as household response towards the income shock. Nevertheless, the issue of which model fits the household response remains unclear. For this reason, this paper attempts to use the SES for the estimation purposes as it present a better proxy and a good indicator of the household’s economic situation.

Vulnerability and resilience

The concepts of vulnerability and resilience are found in various disciplines and subjects. A body of literature has emerged in the past few decades that have focused on different aspects of vulnerability. The notion of vulnerability has become an influential theoretical method for defining the condition of susceptibility to risk, inadequacy, and marginality of both physical and social institutions, and for influencing normative study of behavior to increase quality of life through risk mitigation [22]. The term vulnerability, according to the World Bank [23], is “the likelihood that a shock will result in a decline in well-being”, while some researcher have described vulnerability as a state of defenselessness and exposure to shocks [24]. This definition can be further explained that any external shock that can lead to the decline of the livelihood of households including income and other socio-economic factors, determines the vulnerability of the households.

Some studies focused on presenting evidence of vulnerability by examining the methods used by rural households in low-income countries to deal with HIV/AIDS [25]. A study by Ndirangu explores how farm households in rural Kenya react to shocks and the impact on quality of life of household risk management strategies. The research explores two primary causes of rural vulnerability: the weather and AIDS. The study shows strategies for coping with shock includes dependence on collaborative opportunities such as casual loans and gifts [6]. The impact of vulnerability in our case can be assessed and analyzed empirically through the addition of vulnerability of the households against COVID-19 in the model. We check the vulnerability and attributes of it relevant to poor households. By doing, useful understanding and insights can be obtained that can enable policy makers in formulating better policies to avoid the risk related to poor households in case of shock.

Similarly, the concept of resilience has gained much attention in recent years especially in terms of poor households [26]. Resilience refers to the ability of households to cope with basic necessities of livelihood in case of emergency or shock [27,28]. Resilience has been implemented by different scholars in diverse perspectives [29–31]. There is a broad consensus that resilience is not stagnant, but that it continues to evolve [29,32], that needs further analysis and examination.

Previous studies have focused on the resilience of household in case of natural disasters while others have focused on resilience of households to the socio-economic shocks [33–35]. Resilience of households has also been studied in terms of ecological and environmental effects [36,37]. Although different qualitative approaches have been undertaken in terms of analyzing the resilience but the studies based on quantitative resilience are scarce and limited [38,39]. The resilience of households taking into consideration the livelihood of the households is also an important factor. Several studies have focused on this component in terms of external shocks [40,41]. However, studies have focused on different aspects and factors linking resilience to household income and assets, yet there is a dire need to better understand the connection between these components in terms of rural households. This study will attempt to fill this gap and contribute to the empirical aspect of the literature.

In terms of measuring vulnerability and resilience, economists have argued on various economic factors including household income, status, capital, and access to social security nets [42–44]. In terms of understanding the response of household towards a certain external shock like the COVID-19 pandemic, resilience and vulnerability can be good measures. The impacts of COVID-19 on households are unprecedented and the responses are observed to vary from household to household due to the differences in the SES of the households. Similar, observations are reported in case of China. The assumption can be drawn that households with higher SES may have better coping strategy and resilience as compared to household with lower SES. In this paper, we analyze the difference in the households resilience based on SES using the survey data.

For empirical and quantitative analysis, some authors measured vulnerability based on its impacts on indicators of welfare of the households [45,46]. Others estimated vulnerability in relationship to the poverty dynamics of the households. In this paper, we follow the approach of Vulnerability as Expected Poverty (VEP). The VEP approach is defined as the probability that the expected Income expenditure of a household will fall into poverty in the future [47]. This approach was also adapted in terms of poor households in developing countries by other scholars [7,48,49]. In case of previous studies, the vulnerability was estimated in case of poverty status of the households, which stood above the absolute poverty line. These measures did not consider the effects in case of households below the poverty level, which may provide a deeper understanding, and vulnerability effects on poverty status of households. Chaudhuri [47] pointed out that welfare status of the households tend to decline where the households poverty status is below poverty line as compared to households above poverty line due to the increase in vulnerability and risk. Based on this notion, we build our model and apply it in case of household status both above and below the poverty line, which will provide a better, and in-depth measure of the vulnerability of different households as compared to previous ones.

Additionally, some past literatures have also focused on different coping strategies across multiple domains including access to credit, government support, and help from family and friends, use of savings, and other such strategies [50,51]. While, the coping strategies related to the SES of household especially poor households, have not been clearly observed [16]. Coping strategies towards shocks may differ and their impacts may vary based on the status of the households. In this paper, we observe a coping strategy specifically based on the significance of the policy for both poor and non-poor household status, which in this case are the loans provision on lower interest for business and its impacts on different households. There are very few research on the poor households susceptibility towards pandemic [52], there is still more to explore in terms of COVID-19 and its impacts on poor households and the response of the communities.

This paper attempts to provide the impacts of COVID-19 on household based on the SES of the households. Additionally, the paper further identifies the differences in the impacts of the outbreak taking into consideration the SES of the households by dividing the households into two groups (poor and non-poor households). Based on these differences, the resilience of the households towards a shock like COVID19 and the vulnerability of the households in face of coping with the outbreak are also analyzed and estimated for a deeper understanding of the differentiated effects of COVID19.

Sample and data collection

For the empirical estimations and predictions of our model, we use both primary and secondary data from different sources and methods. For the collection of primary data, an online survey was developed. The online questionnaire was shared using different social media platforms, which made it easier for data collection. The data was collected from village representatives (potential key respondents). The questionnaire was filled by different households with majority from the villages of Sichuan province (475 in Sichuan while 80 from other provinces). A total of 555 questionnaires were completely filled through the online channels over a period of two weeks. The province includes least, middle, and highly developed regions similar to mainland China. Due to measures taken by the Chinese government to contain the spread of the virus, including social-distancing and travel restrictions, the data was collected through administering an online questionnaire, which is also consistent with previous studies [53,54]. In case of Sichuan Province, there are about 183 counties that are different in terms of geo-political and socio-economic indicators [55]. For secondary data, online official websites were utilized. Health data and reports were collected from WHO website [56,57], while data on Chinese villages was collected from National Bureau of Statistics China website [58].

The data was collected through the snowball sampling technique due to its time and resources-saving advantages, as argued by various researchers [59,60]. The data was collected during mid of April 2020 for two weeks; at that time COVID-19 pandemic had already affected a large number of local people in terms of socio-economic and health effects of COVID-19.

We sought permission from the relevant community leader of every village prior to administering the questionnaire. Besides, the consent of each village representative was obtained prior to partaking in the survey. Each respondent was allowed not to answer any question accordingly. The current research was approved by the Ethics Committee of Leshan Normal University. Prior to taking part in the online questionnaire, informed consent was obtained from each participant. The purpose and content of the study were first introduced to the potential participants before partaking in the survey. After assuring respondents of complete anonymity, confidentiality, and other ethical considerations, they filled the online questionnaire. There was no electronic record of the participant’s consent, but all respondents agreed on the purpose of the research and took part in the online survey voluntarily.

Measures

A questionnaire was developed to study and investigate the economic impacts of COVID-19 on households based on differences in the socio-economic status (SES). In this study we attempt to determine the household-level effects of the COVID-19 shock using different socio-economic factors. For this purpose, we asked several questions from participants such as SES level, household level (poor or non-poor), number of community groups in the village, per-capita annual disposable income (in normal times), main sources of income, industry involvement, agricultural supply level, cost of livestock breading (after COVID-19 outbreak), impact of COVID-19 on rural households to fall back into poverty, vulnerability, and resilience level of households, among others. For more details about the survey questions, seeS1 File.

In case of the measurement some researchers argue that although the consumption patterns and disposable income are currently applied for the evaluation of the welfare status and living standards of the households [61] but these are not the only proxy variables for representation of welfare and status of the households. The variables like welfare and living standards are somewhat hard to observe directly. Instead, different proxy variables are utilized for the indirect measurement of such values. Therefore, it is possible to estimate new proxy variables as suggested by Massari [61] which can best define the changes and effects on households. Using the Bank of Italy SHIW data, he included a new proxy variable “real assets” in addition to disposable income and consumption expenditure. Similarly, in this paper we assume the SES of the household may provide a better measure as a proxy representing the welfare of the households.

We develop a simple model based on step-wise binary logistic regression in order to estimate our variables. As we used an online survey (questionnaire) for the collection of the data, so some variables in our data are coded as binary variables while others in categorical form. Based on the design of the questionnaires, the responses and the data, it is convenient to use the logistic regression approach for better estimates, good model-fit and unbiased estimates [62,63].

The model

The model predicts the impact of different sources of income, industries, and policies under COVID-19 based on the SES in the villages of Sichuan province, China. As mentioned earlier, the households are divided into poor/low SES and non-poor/high SES household based on the survey data. The household level SES is based on village representative’s identification of the average SES level of the households in the villages (high SES or low SES village) and the reply of the respondent’s in the questionnaire. Based on this fact, we assume SES to be a better indicator for the indirect impacts of COVID-19 in case of different SES levels of the households.

We construct our model using SES as a proxy for the household’s main indicators of income, consumption, assets and social indicators based on the national statistics of China. SES contributes and expands the multiple indicator [61,64], by inclusion of social indicators such as education, members etc. with broader perspectives of the households status. As our dependent variable assumes binary values of 0 or 1, the binary logistic regression is adopted. In the literature works [62,63], a simple logistic model equation is given by:

Where,Yiis the linear function and the dependent variable,Xidenotes the independent regressors and β measures the coefficients of the regressors in the model.

Eq 2can also be written in the form of,

This represents a simple logistic linear regression equation for estimating binary. As the dependent variable in our model consists of binary values (0,1), it is better to treat the model as a linear equation [62]. As our dependent variable to be estimated is also binary coded, we can construct a binary logistic model, which is depicted as follows:

Where,Yiis the dependent variable representing SES of households ‘i’ whereasXi,ZiandNiare the independent regressors in our model.Xiwhich is divided into the four main sources of income in village ‘i’,Zirepresents the two main industries in village ‘i’, andNidenotes the two impact variables of Covid-19 on households in village ‘i’. All the independent variables are chosen based on the socio-economic features of the villages and in accordance with the national statistics of China. The variables are further explained inEq 4. Moreover,α0is the constant in the model whileβi,γiandδiare the relevant coefficient measures.μiis the overall error term in the model.

To further understand the impacts of COVID-19 on the counties and the resilience of the poor household against the outbreak, we add additional variables and modify the model. The new variables reduce the bias and increase the explanation effect of the regressors in the model. We fit our final model as follows:

Where,

yfarmi: Income from farmland in the household ‘i’

ybusni: Income from doing business in household ‘i’

yempi: Income from local employment in household ‘i’

ygovi: Income from Government Transfer payments in the household ‘i’

indagrii: Household member associated with the agricultural industry

indlivsti: Household member associated with the livestock industry

sagrii: Supply of agricultural products during the pandemic

clivestki: Cost of livestock feed during the pandemic

resi: Resilience of household ‘i’ against COVID-19

povi: Ability of household ‘i’ to counter falling to poverty during the pandemic

policyloani: Desired policy of loans for business on lower interest rate by household ‘i’

The new variableresirepresents the resilience of the household ‘i’ against COVID-19,Povishows the susceptibility of the household ‘i’ falling into poverty andpolicyloanidenotes the desired policy measures by the households in county ‘i’.εi,ρiandτiare the respective coefficients measures of the newly added regressors.

We estimated our model using step-wise binary logistic regression. As our independent variable is binary coded and independent variables are categorical, it is an appropriate method for estimation. The model was also tested for reliability and validity through tests for consistency and goodness of fit for the factors. The results show significant estimates and in line with our assumptions and research questions. The outcomes of the analysis are reported in the next part.

Sample characteristics and descriptive summary

In our sample, most respondents belong to high SES status level (74%). In terms of per capital annual disposable income during normal years, most participants reported less than RMB 13000 (61%) while only 12% of participants had income higher than RMB 20000. Further, in terms of main income sources, most survey participants were dependent on government transfer payment (89%), followed by business (66%), local employment or wage income (41%), and family farming (25%). Additionally, 59% reported that they are mainly involved in agricultural industry. When asked about supply of agricultural materials (in comparison to last year), 61% respondents reported that they are receiving adequate supply of materials. However, only 1% pointed out that there is severe shortage.

Additionally, 67% of participants said that due to COVID-19 outbreak, the cost of livestock breeding has increased. Similarly, when we asked survey participants to express the likelihood of the rural households to fall into poverty due to COVID-19 pandemic, 15% reported very high likelihood followed by high likelihood (46%). Likewise, when respondents were asked to express their resilience and ability to prevent returning to poverty independently without any external support, 46% reported that they possess strong resilience capability, while only 17% reported that they have no resilience capability.

Table 2shows the statistical summary of the dependent and independent variables. We collected the data through an online questionnaire. A total of 555 households were observed and surveyed. Most of the variables including the dependent variable ‘SESi’, are binary coded with values of 0 or 1. Some variables including ‘hhres’, ‘fallpov’ and ‘sagri’ are ordinal or ranked assuming values of 0, 1, 2 or 3.

Table 2. Descriptive statistics.

The mean value of the variables lies between 0 and 2 with minimum of 0.26 for the dependent variable ‘SES’. For ‘sagri’, the maximum mean value is 1.87. The standard errors of mean are low for all the variables and ‘ygov’ shows the lowest values of 0.013. The standard deviation for all the main variables is between 0 and 1. ‘hhres’ has the highest standard deviation of 0.933 followed by ‘fallpov’ 0.843. For the maximum and minimum values of the survey responses of the households, the codes range between 0 and 3. In case of the dependent variable, the code is in reverse order assuming 0 for positive response and 1 for negative response. The codes for independent variables present 0 for low level and 1, 2 or 3 for higher categories of responses. Overall, the data shows clear statistical values for the variables used in the estimation method.

Fig 2 shows the difference in the SES of the households in percentages and the response of the households in case of socio-economic changes. Household with better socio-economic status rely on income from business (68% of high SES households) and local employment (39% high SES households) while households having low SES are more dependent on transfer payments from the government (19% of low SES households) and farm output (93% of low SES households)

Fig 2. High and low SES households.

Table 3. Cronbach’s alpha test of reliability.

For testing of convergent validity, we use Factor Analysis for the four main components of all our factors, including 4 income factors in the first component, 2 industry factors as the second component, 2 cost factors as the third component, and the fourth component consists of 3 factors that have diverse loadings. The test is undertaken to check if the factors load are acceptable and distinct or not for each component. Table 4 shows the loading of the main components.

Table 4. Component matrix.

Table 4 clearly depicts that the factors load distinctly for the four groups and confirms the validity of convergence among the main factors in each of the groups. The average factors loading for all the groups are also estimated to be higher than 0.5 threshold level and most of the groups average above 0.7 which is a good determinant of validity and also suggests convergent validity.

Table 5. Hosmer and Lemeshow test.

According to Hosmer [69], the value of significance has to be above the threshold of 0.05% in order to determine a good fit. The null hypothesis assumes that the data is a good fit for the model. In our case, Table 5 clearly shows that the data fits our model, and our model has the goodness of fit as the significance of Hosmer-Lemeshow test estimates are above 0.05, which means we are unable to reject the model and our data fits the model. In addition, for the overall significance of the model, the Omnibus test of the Model coefficients also proved that coefficients of the independent regressors used in the model are significant. Table 6 presents the Omnibus test outcomes of the final estimations of the model.

Table 6. Omnibus tests of model coefficients.

Main findings and discussions

Table 7 shows the stepwise binary logistic regression trends for our model based on the survey of households in different villages of Sichuan Province.

Table 7. Binary logistic regression output.

As our dependent variable ‘SES’ assumes binary values (0, 1), therefore, binary logistic regression analysis is adopted for the estimation of the coefficients in our model. The binary logit regression provides better estimations and is recommended in case the dependent variable consists of binary values. The use of SES as the representation of the households proved to be a better indicator of showing the impacts of shock on different households’ and their response to an external shock (COVID-19) as it takes into account the main economic factors relevant to households. All the independent variables are regressed and estimated for the effects on the dependent variables.

The table also shows the coefficients ‘B’ and the exponential form of the coefficients ‘Exp (B)’ of all the independent variables in the model. Exp (B) interprets the odds of the effects of the independent variables in case of the binary dependent variable.

Table 8. Binary logistic regression (categorical variables).

The estimations in the table show different categorical estimates for our main variables in the model. The results are similar to the non-categorical estimations with a few changes suggesting robust outcomes in our main model. The income effect and industrial effects of the variables show similar signs as the previous model. Similarly, the Nagelkerke R Square value is also high at 70.1%, which shows a higher predictive power of the variables. As in the case of Binary logistic regression, the Nagelkerke R Square values loosely define the effect of independent variables, therefore, we also calculated the correlation value of the probabilities of our independent variables and the dependent variable using the Pearson Correlation test. The value of the Pearson Correlation test is 61%, which is also high and shows a good choice of our independent variables in explaining the dependent variable.

The resilience has an overall significant impact in determining the odds of non-poor households to be more resilient as compared to poor households. The first category representing resilience in the table suggests that there is no resilience. Based on the estimation value of 1.941 it can be interpreted, as the odds of households with no resilience are about 1.9 times more likely to be poor and low SES households. Based on exp(B) value of 6.965 it can also be explained that a household with no resilience is approx. 6.96% more likely to be poor, which is highly significant. On the contrary, the resilience coefficients for 2nd and 3rd category we obtain the odds ratios of 0.166 and 0.018. Thus, the odds of a household to be resilient or highly resilient are less likely to be poor households.

Similar trends are represented in terms of countering poverty. The signs are negative showing less likeliness towards poor households, and exponential coefficients for the three categories assume values of 0.455, 0.121, and 0.020, respectively. All the category estimations were negative, predicting less likeliness of being poor or low SES households. The estimates were significant except for the first category. The results suggested that the odds of households countering poverty strongly or very strongly are less likely to be a poor household. It can be explained that a household with low SES is more vulnerable to the impacts of COVID-19 and may fall back below the poverty line, while the high SES households show more resilience and less vulnerability towards the pandemic. The results further highlight that the number of new poor may increase in the future if the pandemic continues as these households are less resilient and less likely to counter falling back into poverty.

For further explanation of the model, we estimated the predicted probabilities of the variables against the dependent variable. Fig 3 shows the predicted probabilities of the independent variables against the values of dependent variables. Although some values overlap, the majority of the distribution provides a good prediction of the model. It predicts that households with low SES are likely to fall on the positive side while a household with high SES over the negative side. The same outcomes are also presented in Fig 4.

Fig 3. Predicted probabilities and SES.

Piling of the data responses and predicted probabilities on both sides of the figure explains a good outcome related to the variables and the model.