1 Introduction

Africa’s food market has experienced a significant amount of transformation with an increase in the share of purchased food from retailers. While the expansion of large food retail chains has been concentrated in urban areas, a significant amount to growth has been reported in rural areas. Various studies have reported the mixed effect that the supermarket revolution has had on household dietary quality [26]. This diet quality is mainly measured by the level of diversity of the diet and is measured by the level of inclusion of different food groups in one’s diet, which are beyond the carbohydrates, fats and sugars. Regardless of the information gaps, the penetration of supermarkets in rural has increased with innovative market penetration such as e-marketing making use of text messages to draw in new customers or ensure the loyalty of frequent customers to increase market integration of households. However, remains unclear whether e-marketing has had positive or negative effect on diet quality in the communities they serve. Hence, it is unclear if there be any reason for concern for the unregulated nature of e-marketing in food retailing or the need for policies guide the penetration of retail markets in the rural areas.

The introduction of mobile phones has accelerated the transformation in the rural food environment. Mobile phones have improved the wellbeing and livelihoods of poor households in distant rural areas through the various functions they perform. The positive influence has been reported in improved in integration in the both the supply of produce to the formal market as well as the demand of food. As explained by Sekabira and Qaim [20], mobile phones have enabled information dissemination, and this greatly decreased the costs associated with searching for information across different markets and has led to sellers and buyers receiving better prices. Sekabira and Nalunga [19] report that mobile phone usage has enabled the use of financial technology and facilitated securer and easier payment for food and access to remittances. Mobile phone ownership has also aided consumption smoothing as these assets can easily serve as a temporary store of wealth which can be converted into other forms of income-generating assets during times of economic prosperity or cash during times or hardship [12].

As the influence of mobile phone use on household diets has been multi-faceted, past studies have utilised a variety of approaches to measure their influence on food and nutrition security. Some have either measured the influence of this technology by evaluating the aggregate influence of cellphone ownership using the number of mobile phones owned [14, 22], mobile phone usage [19] and mobile coverage [3] as proxies. Others have measured the influence mobile phones in facilitating financial inclusion [12, 19]. Limited information is available on the effect the information dissemination on food offers using mobile phones has had on household food diversity. Hence, there remain many unanswered questions on the effects of the disaggregate functions of mobile phone. It is important that these individual effects be investigated to detail the potential of the individual welfare-enhancing services. Hence, this study makes a contribution to the existing literature by finding the disaggregated effect of retail information providing function on household diet diversity as it is enabled by mobile phone ownership. As retail shops’ messages are tailored to provide information on discount and pricing promotions, this study provides an account of how mobile phone ownership has been instrumentalization in the food industry. This is an important approach as it may reveal whether there is need to develop policy to safeguard the welfare of consumers. It also provides insights into the possible leveraging of asset endowment in driving the success of community development programs.

This study aims to evaluate the influence that mobile phone, particularly the influence of e-marketing using text messages on household food diversity. It uses the rural context of Moretle Municipality and also takes into account the role of endowment of other non-technological assets in facilitating improvements in dietary diversity. This approach has been underpinned by the entitlement theory and the sustainable livelihood framework [6], which provides that well-being they will experience in their lives is determined by their use of the other physical, natural, social, financial and human assets. The framework postulates that rural dwellers leverage these assts to improve their livelihoods. The more assets an individual or household owns, the less vulnerable they are to shocks and stresses, the more secure their livelihoods, and the better able they are to secure diet that diverse and meets their nutritional requirement [27]. This investigation key as it reveals key policy instruments that can be used by governments to nudge societies towards an improvement in diet quality [12]. The urgency of improving dietary quality is seen in the alarming prevalence in diet related diseases such as obesity and hypertension in South Africa [13, 29]. The need for this research is further emphasised by the significantly high proportion of the national budget is allocated to medication and treatment for non-communicable, diet-related and lifestyle diseases [25, 29].

2 Literature review

Six asset classes are typically considered when discussing household dietary diversity. These are human, physical, financial, social, natural and technological assets. A review of past studies shows that most studies on household diet diversity have often grouped physical and natural assets and called them agricultural assets. Such an asset class includes farmland, farming implements, livestock, farm labour, forests, grazing land, irrigation equipment and water resources. Mixed results have been reported on the effect of agricultural resource ownership on household dietary diversity. Some studies [8,9,10, 23] have reported that ownership or access to natural resources is a major determinant of household diet diversity in rural areas. These findings suggest that encouraging subsistence farming households to produce various crop and animal species can be an effective strategy to improve dietary diversity [18, 22]. On the contrary, other studies fail to find a statistically significant effect of agricultural assets [21] while others could not find a positive relationship between own agricultural assets and diet diversity [16]. Mulenga et al. [12] report that agricultural asset ownership and use only has an impact on household diet diversity if asset endowment enables the production of healthier crops which were not possible or restricted before. Chegere and Stage [5] also explain that the contradictory results are because an unattainable degree of production diversification is often required to produce a significant move in household diet diversity.

Human assets can be defined as intangible resources that are provided by household members. These include intangible factors such as knowledge, motivation, health and skills [18]. Some of the most utilised proxies for this asset class have included the level of formal education of the household head, the number of years of schooling of the meal planner or the average educational attainment of all household members. Literature provides mixed findings on the association between human assets and household dietary diversity. Some studies [8, 11] have found a positive relationship between formal education and household diet diversity and explain that a higher education level improves knowledge about the benefits of dietary diversity, improves knowledge of nutrition and opens up other income-earning opportunities, and thus, it is likely to positively influence household diet diversity. Other studies such as those carried out by Gupta et al. [9] and Bandyopahdhays et al. [1] have found the opposite effect. These studies have found that the effect of improvements in human assets is often contravened by the lack of access to other assets such as land, water and finances. This is particularly true in cases where women empowerment investments do not yield the expected results. Similar results have been reported by studies that have used alternative measures of human capital. These have also reported an attenuating effect of age, sex, health and able-bodiedness on the positive effect of human capital improvement.

Recent studies have increasingly used contemporary human asset indicators such as membership in formal and informal groups, the relationship of trust, the bond between family members, the tradition of care for family members and access to wider institutions of society that people draw upon in pursuit of livelihoods [18, 24]. These may also be described as institutional assets. Bandyopadhyay et al. [1] measured human capital using intrahousehold gender relationships and found that healthier relations had a positive effect on household diet diversity. Sariyev et al. [18] found that altering household roles to include female participation in decision-making was associated with higher diversity both in terms of production and consumption.

Financial assets are arguably the most influential asset class affecting household diet diversity. This is because of the relatively greater ease of converting them to food. Financial assets include financial resources that are available to the household and include savings, credit or regular remittances from the government, relatives or friends [18, 24]. Pereira et al. [15] report that limited access to these assets is a major constraint in South Africa. The lack of access to financial resources also has a compounding effect as it also curtails access to all other resources (e.g. education). Researchers have taken different approaches to measure this asset class. While some such as Banerjee and Duflo [2] considered the per capita income, others such as Sinyolo et al. [22] have considered the amount of money that was dedicated to the purchase of food. The results on the influence of this asset class endowment have also been mixed with some studies showing an improvement in household diet diversity up to a determined threshold beyond which, increasing income was associated with declining diversity of diets.

A considerable number of studies have investigated the effect of technological assets on household diet diversity. This asset classification includes any physical goods that have increased productivity in food production or food acquisition process. Past studies have evaluated technology’s impact on-farm productivity and have measured it using proxies such as farmers’ access to irrigation, ownership of vehicles to transport produce to market and the number of kilometres with tarred roads. On the other hand, studies that have evaluated the effect of technology on improving food access have considered the availability of transport to reach the market or ownership of bicycles. Other studies have considered technology assets’ ability to increase food utilisation. These have considered variables such as ownership of a cooking stove or refrigerator. Both types of indicators show technological assets as conduits of better market integration. Most researchers prefer to aggregate technological assets and use the resultant variable as a measure of household wealth. Such aggregation has allowed for attribution of diet diversity improvement to the ownership of assets such as televisions. Most studies (see [18]) have captured technological assets as a composite variable that represents wealth. These wealth variables almost always find positive associations with household diet diversity. However, this may not be the case as diminishing returns occur between various asset ownership indicators and other measures of food security. This over-generalisation blurs the effect of the individual assets and keeps the number of options that government and stakeholders can use in community interventions that target improvements in household diet diversity.

Some studies have ventured to measure the effect of information technology assets on diet diversity. Most studies have shown a positive association between household diet diversity and information technology assets and have used mobile phone technology as the main measurement proxy [23]. To be specific, Sinyolo et al. [22] measured mobile phone ownership, while Sekabira and Nalunga [19] considered mobile phone usage and Beuerman et al. [3] focused on mobile coverage. A few studies that are similar to one conducted by Sinyolo et al. [22] measured household technology endowment using four different technologies (television, radio, mobile phone and internet). Although the majority have measured mobile phone adoption, the contributions to literature have been varied. Some studies have looked at how the association between technology assets and household diet diversity is found when differences in gender [20] and food group classification [14] are taken into account.

3 Methodology

The study was conducted in the Moretele Local Municipality of the North West Province, South Africa. This is one of the five municipalities of the Province and was an ideal location for the study as it is predominantly rural and has a significantly high food insecurity rate. The Moretele Local municipality has a total population of 186 947 making up 52 063 households. It is classified as a rural settlement and is located 60 km to the north of Pretoria or Tshwane, the administrative capital city of South Africa. Due to the proximity to the economic hub of the country (Gauteng province), the municipality provides labour to the urban centres. Although the main agriculture (where small stock farming and vegetable production) is the main economic activity, residents’ ambitions tend to focus on moving to the city to obtain higher-paying jobs hence the agricultural enterprises have low productivity rates. An unemployment rate of 46.5% has been reported for the municipality and remittances play a key role in the food security of households. These remittances are spent in predominantly in formal supermarkets and some informal markets.

The Raosoft [17] calculator was used to calculate the study’s sample using the study area population (52,063), 95% confidence level and 6.89% margin of error. This calculation determined a sample size of 166. The sample size n and margin of error E are given by in Eqs. 1–3.

$$x = Z\left( {c/100} \right)^{{2}} r\left( {{1}00 - r} \right)$$
(1)
$$n = Nx/\left( {\left( {N - {1}} \right)E^{2} + x} \right)$$
(2)
$$E = {\text{Sqrt}}\left[ {(N - n)x/n(N - {1})} \right]$$
(3)

where N is the population size, r is the fraction of responses investigated, and Z(c/100) is the critical value for the confidence level c.

The determined 166 households were selected following a two-stage sampling technique. In the first stage, 6 villages were purposively selected and within each village, 28 households were randomly selected for the study. The villages that were within a comparable distance (5 km radius) from the formal markets were selected for the study. One-on-one interviews were carried out in which the household’s meal planner was questioned on the household’s diet and socio-economic circumstances. A structured questionnaire was administered to the sampled households. The questionnaire had three main sets of questions to full exploration of the study’s research questions. The first set included background information about household demographic characteristics while the second section captured information on food consumption. The last set of questions dwelt on a household’s asset possession. Consent was sought and obtained before the start of every interview. Other ethical guidelines such as assuring the study participants of their anonymity and informing them of their freedom to exit the study were observed. The study was granted ethical clearance certificate number NWU-00276-18-A9.

The socioeconomic characteristics considered in Section A of the questionnaire. These were classified into the five asst type considered in the the entitlement theory and the sustainable livelihood framework [6]. The variables measuring physical assets were infrastructure such as type of housing and road infrastructure. Natural assets considered were agricultural and include land size, number of livestock owned, and the availability water resources. Income, remittances, social grants were financial resources measured in the questionnaire. Human assets or capital was measured using the respondents’ educational level, household size, marital status, age, agency in household. Other socio-economic variable measured were age and gender.

To capture information in Section B of the questionnaire, the study used a 24-h recall period for nine food groups as prescribed by FAO [7]. The groups were as follows: cereals, pulses, tubers, vegetables, fruits, meat, fish, eggs, and milk products. An aggregate score household calculated for each household was taken as the indicator of the household dietary diversity. Each food group is assigned a score of 1 if consumed or 0 if not consumed. The household score ranges from 0–9 and it is equal to the total number of food groups consumed by the household. A Household Dietary Diversity Score (HDDS) of ≤ 3 is considered a low dietary diversity group, between four to six as medium and ≥ 7 as a high diversity score category. We have considered ≤ 3 as a low dietary score because as a general rule, consumption of four food groups over 24 h is considered good dietary diversity [7].

The HDDS was regressed against various household asset indicators using multinomial regression. The reduced model is shown in Eq. 4 below.

$$HDDS = {\text{ c }} + {\ss}_{{1}} X_{i} + {\ss}_{{2}} X_{2} +........ \, {\ss}_{{n}}\,X_{n}\,\mu_{i}$$
(4)

where Xi is a vector of all potentially time-varying socioeconomic characteristics of the household such, At is a vector of the asset endowment of the household, the ßs are the estimated coefficients and μi is a random error term.

In Section C of the questionnaire, human asset endowment was measured using the educational level of the meal planner and the size of the household. The former was measured using the highest level of education while the latter was measured using an adult equivalence index. Households’ financial asset endowment was measured using the aggregate household income and the number of remittances that the households received monthly. The agricultural asset endowment was measured using two indicators. The first was a dummy variable which took up the value of 1 when the household had access to a safe and private water source in the homestead and zero when the household used a public water source. The second was an index measuring the number of large and small stock, ploughs and tractors owned by the households. The information provision function enabled by technology asset endowment was measured using the number of text messages on food discounted offerings that were received from supermarkets on the meal planner’s mobile phone during the last 7-day period. The use a 7-day recall period was adopted from Sibhatu and Qaim [21] and Sinyolo et al. [22]. These messages are marketing tools used by supermarkets to advertise items that are normally discounted. They are meant to build customer loyalty as they are often coupled with loyalty rewards and are specific to consumers who choose to sign-up for such programs in supermarkets. This information for technological assets was solicited from and provided by the study respondents. An interaction term, mobile phone ownership and the number of messages received per month was used to measure the impact of improved marketing communication facilitated by mobile phone ownership. Institutional asset endowment was measured using the marital status of the meal planner. The study controlled for the number of meals eaten in households and the age of the meal planner.

4 Results and discussion

4.1 Descriptive statistics

The results in Table 1 show that the average age of a meal planner in the survey was 57. The majority (40%) of respondents were married, 28% were unmarried and the remaining 32% were either widowed or divorced. All survey respondents had received some form of formal education. Twenty-seven percent of the sample had ended at the primary school level, 65% ended at the secondary school level and 8% of the respondents had attended college or university. The results showed that 31.3% of households had an income below R5 670, 41.9% had an income falling between R5 670 and R15,000 while 26.8% had a total household income of above R15,000. An average family in the sample had an equivalence of 3 adults and the largest household had an equivalence of 7 adults. All households depended on the formal food markets as their main source of food, while approximately a quarter of the households subsidised this with subsistence farming outputs.

Table 1 Descriptive statistics
Full size table

Table 1 also showed that a few households (12.5%) had highly diverse diets, while 42.2% had moderately diverse diets. The majority of the households (45.3%) had low dietary diversity. A study by Sinyolo et al. [22] found similarly low levels of dietary diversity in South Africa. A study carried out in India by Geremew et al. [8] found about 17.5% of the sampled households were classified as having low dietary diversity, 61.2% as having medium dietary diversity and 21.3% as having high dietary diversity. In the India study, just 2.3% reported that they received one promotional text message, while approximately a fifth (20.3%) reported receiving two messages the previous month. Most respondents (53.1%) reported that they received messages from the local supermarkets three times the previous week, while about a quarter of the respondents (24.3%) of the sample received more than three promotional messages. Just over a tenth (10.7%) of the survey respondents indicated that they did not receive any retailers’ promotional messages, while 11.9% of the sample indicated that they received one or two messages during the week. Just over three-quarters of the respondents (76.2%) of the sample reported that they had 3 or more meals per day while the remaining 25.8% reported that had one or two meals.

4.2 Empirical results

Table 2 below provides a summary of the empirical results and attained from the regression analysis. It also shows the results of the validity and robustness tests. As shown in Table 2, the estimated model had an F-statistic p-value of 0.000 which was statistically significant at 99%. This indicates a very good measure of fit and implies that the joint variation in the explanatory variables included in the model explained the variation in the sampled households’ dietary diversity. The Ramsey test had a p-value of 0.7763 and this showed that the model did not suffer from any misspecification. A VIF mean value of 1.22 was acquired and this value shows that the model was free of multi-collinearity between the explanatory variables. Robust standard errors were used to control for heteroscedasticity. These results indicate the reliability of the estimated model for the sampled community.

Table 2 Influence of asset endowment household dietary diversity
Full size table

Table 2 shows that receiving promotional text messages on discounted food items on the meal planner’s phone was associated with higher household dietary diversity. This is because the messages provide timely market information and allow meal planners to easily access information on the different discounted market offerings. The effectiveness of text messages can also be attributed to the sense of gain that consumers normally obtain when purchasing goods at a discount. The study’s findings are consistent with those acquired by Sekabira and Nalunga [19], Sekabira and Qaim [20] and Parlasca et al. [14] which found similar positive associations between information technology and household dietary diversity.

The study’s results show that being married is associated with higher levels of household dietary diversity. This implies that there was a positive relationship between household dietary diversity and the institutional asset endowment of the meal planner. This outcome can be explained by the fact that marriage often provides meal planners access to a larger number of resources (e.g. financial assets) that can be used to acquire more diverse food types [1]. On the other hand, the study showed that a negative relationship existed between household dietary diversity and larger household size. As explained by Beyene and Muche [4], larger family sizes are not often associated with higher household dietary diversity because their resources become shared by a larger number of individuals so the households tend to focus on procuring food items from the basic food groups (e.g. carbohydrate-high foods) and neglect other food items rich in other nutrient groups. This is especially true for poor communities whose farming produce does not cater for their diverse nutrition requirements and heavily depend on remittances such as the one used in this study. In such communities, an increase in farming family size (human assets) may offset the increase in the number of mouths to be fed with an increase in free labour.

The results presented in Table 2 show that financial assets (household income and remittances) have a positive effect on household dietary diversity. This is an expected result as the majority of the households in this study relied on the market for the supply of food. The study’s findings are very similar to those of Sinyolo et al. [23] who found that social grant recipients increased their dietary diversity when they had high financial resources. These results show that an increase in the employment of household members would improve the quality of their diet and this could be a possible area of government intervention for development.

The two indicators of agricultural asset endowment had a positive association with higher household dietary diversity. The ownership of livestock had a positive statistically significant influence on dietary diversity in households. This is most probably because households did not produce many crops but instead kept livestock which provided milk, meat and eggs to supplement purchased food. The livestock, especially the small stock are normally sold when finances were low to ensure consumption smoothing. These results are similar to those acquired by Mulenga et al. [12] and Zanello et al. [28] which showed positive associations between agricultural asset endowments and improvements in food security.

Contrary to expectations, the results showed that an increase in the number of meals consumed in a household had a negative influence on dietary diversity. This could be indicative of the low education on the importance of diverse diets, the community’s preference for high-calorie foods or the composition of the market offerings (this includes pricing and shelf space) in the supermarkets in their vicinity. This finding highlights the unattenuated effect of the spread of supermarkets’ footprint on consumer welfare and emphasizes the need for intentional efforts to encourage consumers to make healthy consumption decisions. Other studies have attributed the growing trend in poor diets despite an increase in food retail availability to the large volumes of nutritionally poor foods provided by food retailers.

5 Study limitations

We acknowledge that there are certain limitations to our study. First, more information can be acquired by doing a comparative study between survey respondents receiving text messages about retail market offerings and those that do not. This analysis would require a sampling method that is different from the one used in our study; hence we recommend that future research venture into carrying out such an investigation. Second, the study design did not accommodate an exploration between the general diversity in the offering of food retailers with household dietary diversity. This can be explored in future research.

6 Conclusion

Key developments of the twenty-first century such as the spread of supermarkets and technological advancements have filtered through to influence the dietary diversity of households in rural communities. As discussed in the current study, despite the higher number of meals that are made possible from such developments, households’ dietary diversity still remains limited. As the study’s results showed that endowments in technological, agricultural, financial, institutional and technological assets were found to counter this trend, then opportunities through policies and/or developmental programs should be developed to make use such assets to improve household dietary patterns. Community development programs can use retailing messaging platforms to inform households of healthier food offerings and discounted prices. This could assist in increasing household dietary diversity. However, further research is required on the frequency, the timing of messages, and the length and content of the messages as these aspects were not investigated in the current study. Nationwide data would also be required to ascertain the measure of influence such an intervention would have at this scale.

Additionally, other interventions would be necessary to enable the effectiveness of the communication using mobile phone text messages. These are: an increase in employment and an improvement in dietary education. An improvement in the former would increase market access while the latter would ensure prudent use of the use of the gained financial resources. It would also be necessary to have institutional enablers other than marriage that would assist female meal planners in accessing other enabling assets. Further research into the nature of the necessary institutions would be required for this to become practical. As shown, promoting agricultural production could also improve dietary diversity in rural communities but as shown, investigations into the enterprises that assist in achieving this specific goal would be necessary.