Compared with large companies, SMEs are particularly vulnerable to the economic impacts of the pandemic predominantly because of their limited resources; they tend to have fewer assets, smaller cash reserves and lower levels of productivity (Saturwa et al. 2020). Moreover, as SMEs are diverse, the severity of the pandemic’s impact on them varies considerably depending on the economic sectors that they trade in (Sonobe et al. 2021). The SMEs that have been hardest hit by the COVID-19 are predominantly from retail, hospitality, food, entertainment and construction services (Albaz et al. 2020; Sonobe et al. 2021). Consequently, such SMEs have had less resilience and flexibility in dealing with the costs associated with economic shocks, making it more difficult for them to survive the COVID-19 pandemic and beyond compared with larger firms (UNCTAD 2020a; WTO 2020).

Cruz et al. (2020) reported that the COVID-19-induced shocks have adversely affected SMEs globally in terms of demand, supply and uncertainty. Demand has weakened across most industries because of weak consumer sentiment and job losses, as consumers defer their non-discretionary purchases and focus on buying essential items (Bouey 2020; Ernst & Young 2020). COVID-19 has reduced consumer spending on services like travel, entertainment, restaurants and demand from other firms, leading to a decline in exports because of disruptions in critical supply chains. In terms of supply-side shock, Cruz et al. (2020) found that COVID-19 has led to a decline in the availability of labour as businesses have reduced productivity or are under temporary closures, and employees’ lives are disrupted because of COVID-19 infections and/or travel restrictions. There has also been a decline in employee productivity as workers are less efficient in adapting to new working hours and modalities of work; there are reduced business trading hours. Consequently, businesses have suffered from capacity under-utilisation, and there are no clear indications as to when normalcy will return because of the ongoing uncertainty about finding a cure for the pandemic. Ernst and Young (2020) report that the limit on the movement of people and non-essential goods has disrupted critical supply chains, leading to a shortage of raw materials.

According to the World Economic Forum (WEF 2017:3), ‘the fourth industrial revolution is characterised by the convergence of breakthrough technologies – such as advanced robotics, AI, the IoT, virtual and augmented reality, wearables and additive manufacturing’. Hence, 4IR speaks to the essence of unexplored business opportunities through creative disruption (Munters & Marx 2017; Müller & Voigt 2018; WIPO Magazine 2019; Xu et al. 2018). A firm’s adoption of 4IR technology can facilitate effective implementation of cyber-physical integration into its supply chain (Lee 2020; Alqam & Saqib 2020; Lee 2021; Ivanov & Dolgui 2021). Through the use of 4IR technologies, enterprises across the globe can transition seamlessly from the traditional, static, linear supply chain designs towards smart, connected, intelligent, scalable, customisable and nimble supply networks, which, in turn, allow uninterrupted movement of products, services, information, and big data analytics for decision-making. Africa’s Pulse (2021) and BDO (2020) explains that enterprises can leverage cloud-based global positioning system (GPS) and Bluetooth Low Energy (BLE) asset tracking technologies for improving a business understanding of variable customer demand, visibility, responsiveness and resiliency throughout the supply chain.

It is observed, however, that in most developing countries, purchase, installation and subsequent running costs deter SMEs from readily adopting technology as nearly all ICT equipment is imported, let alone the relatively high rates of taxation, regulatory license fees and non-tariff barriers, exorbitant internet connectivity and telecommunications usage charges (Abualrob & Kang 2016; Saif-Ur-Rehman & Alam 2016; Mwai 2016). Rahayu & Day (2015), Walker et al. (2016) and Nugroho et al. (2017) find that whilst the progressive SME managers or owners are more inclined towards adopting technology with a view to spur on the business, technology adoption is less likely where SMEs find the IT infrastructure quite complex to optimise or merely use. Supply chain resilience is broadly concerned with the ability of supply chain to deal with instantaneous disruptive events (Soni et al. 2014; Yang & Xu 2015; Jain et al. 2017). Recently, business enterprises have become increasingly interested in enhancing their SCR (Kwak et al. 2018; Lima et al. 2018). The extant literature highlights the importance of SCR within a business enterprise. For instance, Alfarsi, Lemke and Yang (2019:3) posit that SCR enables businesses to ‘sustain and attain good financial performance, improve service quality and organizational reputation, which in turn, enhance overall business attractiveness’.

According to OECD (2020), SCR equips organisations to be proactive and to quickly adapt their operations for transparency and business continuity in times of crises. Given the current uncertainty, complexity and vulnerability brought about by disruptions, a resilient supply chain ensures continuity of business operations and is a source of dynamic capability and competitive advantage (Alfarsi et al. 2019; Sadghiani, Torabi & Sahebjamnia 2015). Furthermore, yields of a 4IR-enabled resilient supply chain increased visibility and collaboration amongst all involved stakeholders (Ivanov & Dolgui 2021; Matt et al., 2016). Patsavellas, Kaur and Salonitis (2021:3), in turn, noted that such ‘supply chain collaboration leads to improved decision-making particularly in situations of contingency and production efficiency, has reduced bullwhip, overall inventory and administrative costs for the whole supply chain’. Their study further establishes that over time, retailers can maintain operations without interruptions and improve customer experience, whilst upstream suppliers can build loyalty and reduce costs through information-sharing across the supply chain network. Hence, leveraging 4IR technologies, SMEs can have a real-time sight of every process and part of the supply chain and quickly and proactively identify and rectify areas of potential risk before the occurrence or swiftly respond to disruptions. Therefore, based on the above discussion, the current study proposes the following hypotheses:

This exploratory study adopted a quantitative approach. Primary data were collected by a cross-sectional survey using an online questionnaire between 05 and 20 November 2021 using Google Forms. Respondents for the survey were selected through purposive sampling, a type of non-probability sampling method in which ‘particular respondents are deliberately selected in order to provide important information that cannot be obtained from other choices’ (Taherdoost 2016:7). These targeted respondents were either owners or managers across the 10 provinces in Zimbabwe whose business conformed to the Small and Medium Enterprise Development Corporation (SMEDCO 2010:3) definition of SMEs in Zimbabwe: ‘a business employing no more than 100 employees and generating a maximum annual revenue of USD830 000’. The use of online questionnaires proved cost-effective, allowed for ease of data collection, storage and visualisation and convenient, speedy and accurate feedback, including a wider reach (Bryman & Bell 2014; Nayak & Narayan 2019), allowing for compliance with the COVID-19 physical and social distancing health protocols.

The items used to operationalise the individual constructs were adapted from the literature (APEC 2021; Adam & Alarifi 2021; Hendijani & Saei 2020; Kamalahmadi & Parast 2016b; Thukral 2021; Christopher & Peck, 2004; FinScope, 2012). In addition to the binary (yes or no) option, a five-point Likert scale was used for the majority of the questions in sections 2-5, with responses ranging from 1 = ‘Strongly Disagree’, 4 = ‘Agree’ to 5 = ‘Strongly Agree’. Section 1 of the questionnaire focused on the respondents’ profile and the enterprise characteristics; Section 2 sought information on the COVID-19 pandemic and SME supply chains; Section 3 elicited response on SME 4IR adoption, whilst Section 4 dealt with SME SCR.

In order to enhance the validity of the survey items, the questionnaire was pre-tested on a few supply chain experts with areas of ambiguity addressed immediately. Afterwards, the corrected questionnaire was pilot-tested on 35 randomly selected respondents who complied with the study’s inclusion criteria before final use in the online survey. All the ethical requirements were adhered to. Informed consent was sought from all the potential respondents, and to encourage candidness, we provided assurance and reassurance of confidentiality of their responses. Purposive sampling led to a total of 318 participants responding to the online questionnaire. The responses were checked for completeness, plausibility and integrity before data analysis, with 249 questionnaires (78%) found usable for analysis.

The dependent variable in the regression analysis was SME 4IR adoption, whilst the independent variables (regressors) were derived from the reviewed literature. Components of SME SCR were measured using a five-point Likert scale, with responses ranging from 1 = ‘Strongly Disagree’, 4 = ‘Agree’ to 5 = ‘Strongly Agree’.

The study employed logistics regression analysis, also known as the logit model and based on the cumulative distribution function, to analyse the determinants of 4IR technology adoption amongst Zimbabwean SMEs. Logistic regression analysis is used to determine the odds ratio in the presence of more than one explanatory variable and models the chance of an outcome based on individual characteristics (Sperandei 2014; O’Connell 2006; Fox 2015). The estimation method was selected because of its compatibility with the online survey data. Furthermore, the logistic model is favoured in adoption studies over linear regression, as it has well-organised and asymptotically reliable parameter estimates that are simple to compute (Ume et al. 2020), offers flexibility and the independent variables do not need to follow a normal distribution (Greene 2012).

As the adoption of 4IR technology (dependent variable) is dichotomous, the study sought to determine the likelihood of an SME owner or manager choosing to adopt 4IR technology. Accordingly, the dependent variable was coded 0 and 1. The SME owners or managers who adopted 4IR technologies were assigned a value of 1, whilst 0 was for the non-adopters. Consistent with Wondale, Molla and Tilahun (2016), Gujarati (1995) and Maddala (1992), the SME owner and manager distribution function for SME owner and manager 4IR technology adoption is as follows:

where

P_i is the odd technology adoption by SME owner or manager i ranging between 0 and 1, whilst 1– P_i is the likelihood of non-adoption. The observed response of the ith SME owner or manager is binary, with P_i = 1 for a 4IR technology adopter and P_i = 0 for a non-adopter,

e^–zi represents irrational numbers e to the power of z_i Whilst z_i is a function of the exploratory variables (regressors).

Following the lead of Ume et al. (2020), Equation (1) represents the ratio in favour of SME owner or manager’s 4IR technology adoption to the odds of non-adoption. Taking the natural log of both sides of Equation (1), it follows that:

Therefore, the above function can also be shown as follows:

where

X_i indicates the determinants of 4IR technology adoption by an SME owner or manager, β_i is the vector of parameter to be predicted using the maximum likelihood method and µ represents error term which is usually normally distributed with zero mean variance.

The study measured the reliability of the questionnaire constructs using the Cronbach alpha, wherein the alpha coefficient for the research items is 0.839, suggesting that the items have relatively high internal consistency in line with the literature (Bonett & Wright 2014; Taber 2018).

In order to assess whether data were normally distributed, two statistical tests of normality using SPSS were performed, namely Kolmogorov–Smirnov and Shapiro–Wilk. The results displayed in Table 3 reveal that significance values are greater than the alpha value of 0.05 (0.201 and 0.117 respectively), indicating that the data are normally distributed (Puan et al. 2019).

The study employed the exploratory factor analysis (EFA) technique as a variable-reduction tool for the survey data set to parsimoniously determine the essential control variables for use in the binary logit regression model. The EFA involved the correlation matrix, Kaiser-Meyer-Olkin (KMO) and Bartlett’s test, communalities and communalities total variance extractions.

The correlation matrix represents a simple rectangular array of numbers that provides the correlation coefficients between a single variable and every other variable in the study (Beckman & Quarles 1956). The Pearson correlations are computed to ascertain relationships between the independent and dependent variables (Seate, Pooe & Chinomona 2016). The correlation matrix for the cross-sectional survey data set of the Zimbabwean SMEs is shown in Table 4.

As shown in Table 4, the determinant of the correlation coefficients was 0.531, indicating the absence of computational difficulties. Therefore, consistent with Tripepi et al. (2008), it can be inferred that the model to be used for analysis is good. Thereafter, the KMO and Bartlett’s tests were employed to test for ample adequacy with the results displayed in Table 5.

As shown in Table 5, a KMO measure of 0.741 therefore suggests that this study’s sample was adequate, permitting EFA, in line with Kaiser’s (1974) recommendation of values between 0.7 and 0.8. The table also displays a Bartlett’s test of sphericity’s p-value of 0.00 at the 5% significance level, indicating that the EFA procedure was valid and allowed for further analysis.

Communality, also referred to as h² describes the common variance of extracted factors ranging between 0 and 1. The closer the values are to 1, extracted factors account for a greater amount of the variance of a particular individual item (Grimm & Yarnold 1995). Table 6 presents the study’s communalities that were selected for further analysis in the study. Communality values of these variables following extraction were greater than 0.5.

Table 7 shows the components extracted through the principal component analysis procedure by their eigenvalues.

Not all the control variables recommended by the literature and availed in the survey data set were retained for data analysis. As shown in Table 7, only three components whose eigenvalues were greater than 1.00 were selected through variable combination. The first component accounts for the greatest proportion of variance, and hence, had the highest eigenvalues. The next component explained as much of the remaining variance as it could, all up to the final component. The first three components cumulatively accounted for 76.5% of the total variance. Component 1 explained 35.27% of the total variance, Component 2 accounted for 21.40% and Component 3 constituted 19.83%, whilst the remaining three components represented 23.5%. Thereafter, we proceeded to estimate the logit model, as shown in Table 8.

The Omnibus tests of model coefficients is a goodness-of-fit test, which ‘gives an overall indication of how well the model performs over the one with none of the predictors entered into the model’ (Palliant 2010). As shown in Table 8, the Omnibus tests of model coefficients display a p-value of 0.000, and therefore is statistically significant at the 95% confidence level. Thus, the Omnibus tests of model coefficients’ chi-square statistic shows that all variables considered in the model were significant determinants of 4IR technology adoption amongst Zimbabwean SMEs. From the model summary in Table 8, the Nagelkerke R² statistic indicates that the model consisting of SME age, personal innovativeness, owner or manager ICT literacy, owner or manager age, costs, security and ICT as independent variables represents 83.9% of the total amount of variance in the SMEs’ 4IR technology adoption decision. The Nagelkerke R² values provide an indication of the amount of variation in the dependent variable explained by the model (Palliant 2010). The Nagelkerke R² value suggests a good model fit, as only 16.1% of the variance in SME 4IR adoption is accounted for by other variables, which are not in the model.

At a 95% confidence level, the Hosmer–Lemeshow test is not statistically significant (p = 0.702). However, with a p-value greater than the 0.05 significance level, the logit regression model is an adequate fit to the data (Hosmer & Lemeshow 2000). The classification of the false positive and false negative error rates shown in Table 8 shows that the decision rule predicts a decision to adopt 4IR technology 112 times, and the prediction was wrong only five times. Therefore, there is a false positive of 4.5% (5/112). The decision rule predicted the non-adoption of mobile money 142 times, and that prediction was correct, making a false negative of 0% (0/142), with an overall model correct classification of 91.7%.

The Wald test in the variables in the equation section of Table 8 shows that at both the 5% and 10% significance levels, SME age, owner or manager age and gender are statistically significant in determining 4IR adoption amongst Zimbabwean SMEs. The dominant variables include personal innovativeness, owner or manager ICT literacy, education, costs, security and ICT ease of use. The coefficients of these variables are all statistically significant at the 95% confidence level, implying that they are significant predictors of 4IR technology adoption amongst Zimbabwean SMEs. Thus, the fitted logit model equation for the study is as follows:

The relationship between the SME 4IR adoption and SCR was tested using a chi-square test, and the results are displayed in Table 9.

As shown in Table 9, the chi-square statistic reports a p-value of 0.000, which is less than the 5% significance level. Consequently, the study therefore rejects the null hypothesis. Evidence from the sample indicates that there is a statistically significant association (n = 58; p = 0.000) between SME owner or manager’s adoption and SCR.