Invited paper presented at the 6th African

Conference of Agricultural Economists, September 23-26, 2019, Abuja, Nigeria

Copyright 2019 by [authors]. All rights reserved. Readers may make verbatim copies of this document for non-commercial purposes by any means, provided that this copyright notice appears on all such copies.

Modernization of African food retailing and (un)healthy food consumption: Insights from

Zambia

Makaiko G. Khonjea*and Matin Qaima

a Department of Agricultural Economics and Rural Development, Research Training Group

‘‘GlobalFood,’’ University of Goettingen, 37073, Goettingen, Germany

* Corresponding author. Tel.: +4915217034785.

E-mail address: [email protected] (M.Khonje) Modernization of African food retailing and (un)healthy food consumption: Insights from

Zambia

Abstract Food retail modernization continues to evolve rapidly; influencing consumer’s dietary choices and nutrition transition in developing countries. However, empirical evidence analyzing the linkage between retail modernization and people’s diets is still scanty. A few existing studies have focused exclusively on single modern retailers like supermarkets, and empirically they have neglected the fact that most people buy their food from different retail outlets. Using cross-sectional household survey data collected in urban Zambia, applying multivariate probit and tobit regressions, we analyze the associations between household socioeconomic status, the use of different types of modern and traditional retailers, and dietary patterns. Our results show that the use of different retail outlets is strongly associated with various household socioeconomic characteristics such as education and income. Several complementary and substitution correlations exist among retailers. We further find that the use of modern retailers is associated with a significantly higher consumption of ultra-processed foods at the expense of unprocessed foods. Interestingly, however, the same holds true for some of the traditional retailers. Hence, the findings suggest that simplistic statements about whether modern retailers improve or worsen household diets are not possible; more nuance is required. Keywords: Retail modernization; food consumption; dietary patterns; nutrition transition; multivariate probit/tobit regressions. 1. Introduction Food environments in developing countries have been evolving rapidly for the last few decades, partly as a response to urbanization, income growth, and changing consumer preferences (Andersson et al. 2015; Qaim 2017; Lu and Reardon 2018). However, consumer preferences and demand may also be influenced by structural changes in food supply chains. For example, the modernization of food retailing—with an ongoing shift from traditional markets to modern supermarkets, hypermarkets, and fast-food restaurants—has effects on the types of food offered, as well as on food variety, food prices, and shopping atmosphere, all of which may influence consumer choices (Asfaw 2008; Hawkes 2008; Reardon and Timmer 2014). Understanding the links between changing retail environments and food consumption patterns is important to promote food security and healthy diets. This is especially relevant in Africa, where poverty and undernutrition are still widespread but where overweight and obesity are also on the rise (IFPRI 2017). Available research suggests that the modernization of food retailing may make calories more affordable for urban consumers but—at the same time—may foster the nutrition transition towards more highly processed foods that are rich in fat and sugar, but contain low amounts of micronutrients for healthy nutrition (Gómez and Ricketts 2013; Umberger et al. 2015; Popkin and Reardon 2018). Recent studies with data from Africa show that the growth of supermarkets contributes to increased consumption of processed foods and a higher body mass index (BMI), also after controlling for living standard (Kimenju et al. 2015; Rischke et al. 2015; Demmler et al. 2018). However, the existing research has several shortcomings. First, many of the studies analyzing the effects of a modernizing retail sector compare food consumption and nutrition between users and non-users of supermarkets. While this approach provides interesting insights, it neglects the fact that most people do not buy all of their food in one retail outlet. Supermarket users may also use traditional retailers for some of the food purchases (Berger and van Helvoirt 2018; Lu and Reardon 2018). Conversely, supermarket non-users may frequently use fast-food restaurants or other types of modern retailers. Hence, focusing only on the differences between users and non-users of supermarkets may potentially provide an incomplete picture of the wider links between food retailing and dietary choices. Second, most available studies look at average effects of supermarkets on the entire sample of consumers without considering that different population segments may use different types of retailers to varying degrees. Third, available studies typically analyze diets and consumption patterns by differentiating between food groups but without properly accounting for the level of food processing. Standard food consumption surveys in developing countries often capture highly processed foods only in very aggregate form, because highly processed foods were traditionally not widely consumed in poorer population segments. However, this is changing with the modernization of the retail sector (Tschirley et al. 2015), implying that survey formats need to be adapted too. Here, we address these shortcomings with more detailed data and statistical analysis. Using primary survey data from urban areas in Zambia in Southern Africa and different econometric techniques, we analyze the associations between household socioeconomic status, the use of different types of modern and traditional retailers, and dietary patterns in terms of processing levels and food groups. To our knowledge, this is the first study that looks into such issues with detailed data from Africa. Zambia is an interesting empirical setting for this analysis, because it has recently experienced rapid growth of supermarkets, hypermarkets, and other modern retailers, especially in metropolitan areas (Tschirley et al. 2015). Moreover, Zambia is experiencing a triple burden of malnutrition, where undernutrition and micronutrient malnutrition coexist with rapidly rising overweight and obesity (Steyn and Mchiza 2014). Hence, our results may help to project how diets evolve with further changes in retail environments and what type of policy responses might be useful. We expect that the insights from Zambia can be useful also for other countries in Africa, where the modernization of the food retail sector is still at earlier stages. This paper is organized as follows. Section 2 provides a description on modernization of food retail environment in Zambia. Methodology—i.e., (i) data sources, (ii) measurement of key variables, and (iii) empirical strategy—and the results are presented in sections 3 and 4, respectively. While the last section (5) draws conclusion. 2. Modernization of food retail environment in Zambia For the past two decades, there has been a rapid growth in the retail modernization process as well as the economic growth of most countries in southern Africa; due to several factors such as technological innovations, globalization, foreign direct investments and regional integration (Tschirley et al. 2015; Ziba and Phiri, 2017). Currently, Zambia is one of the southern African countries with rapid growth in modern retailers (MR) such as superstores1, supermarkets and fast- food restaurants (FF) (Ziba and Phiri, 2017). For instance, between 1995 and 2018, the number of major shopping malls with MR rose from one to 25 (see Table A1 of the appendix) in Lusaka city. Consequently, a substantial share (42%) of the food consumed by urban households in Lusaka city is bought from MR—mostly from South Africa. The common feature of these MR is that they are mainly concentrated in the major shopping malls of main cities, urban towns and districts. On average, a shopping mall has at least a supermarket (e.g., Food Lover’s market, PicknPay, Shoprite and Spurs) or superstore (e.g., Choppies and Game stores), and three FF (e.g., Debonairs Pizza, Hungry Lion and KFC) in Lusaka, Zambia (see Tables 1 and A1 of the appendix for details). Most of these retail outlets operate from a shopping mall with a floor space of 100–200 m2 or more. Customers normally use self-service facilities with several modern cash tills (4–15) and limited credit facilities (Table 1). In addition to shopping malls, some of the FF such as Debonairs Pizza have several outlets in filling or gas stations. Furthermore, to maximize sales and profits, most MR offer occasional (i.e., month ends

1 We interchangeably use superstore and hypermarket as well as food retailer and food retail outlet in this paper. and the Christmas season) price discounts to customers by advertising in both electronic and print media (Table 1). [Insert Table 1] Unlike MR, most traditional retailers (TR) such as traditional markets, grocery stores, roadside markets and neighborhood kiosks operate from self-constructed shelters with limited floor space (e.g., 1–10 m2) and they hardly use cash tills (Table 1). However, customers may negotiate for price reductions as well as buying food items on credit. Furthermore, most TR also keep special traditional food items such as indigenous vegetables or fruits, wild mushrooms, bushmeat and caterpillars. These products—they are not sold in modern retailers—are nutritious and they provide a good source of vitamins, minerals and energy (Nyirenda et al. 2009). Perhaps, the other important feature for most TR is that they can repackage some food items such as sugar, cooking oil and flour into very small packets. This enables poor households to easily access these products especially if they are out of stock and as a result some consumers often use TR as opposed to modern retailers. While modern retailers keep a large variety of food and non-food products with relatively bigger package size and ultra-processed foods (UPFs), most TR have limited variety of such products. They rarely stock and sell frozen, canned and cooked foods, and beverages, possibly due to financial and infrastructure (e.g., refrigerators) limitations (Laska et al. 2018). However, most TR keep a fairly large variety of cereals, legumes, vegetables and other local foods. Nevertheless, typical for many developing countries, consumers in Zambia use both modern and traditional retailers simultaneously (Figure 1). For example, about 38% of the sampled households use two (both modern and traditional) retailers per week (Figure 1). [Insert Figure 1]

3. Methodology 3.1. Data sources The data used in our empirical analysis were collected through an expenditure survey of randomly selected households in Lusaka, the capital City of Zambia, between April and July, 2018; which was the harvest season in the region. We surveyed a total of 475 urban households from several compounds2 in Lusaka city through a two-stage stratified random sampling procedure. In the first stage, based on information from the Lusaka city council, key informants and a pre-survey visit conducted in November 2017, we selected 14 compounds with high concentration of major shopping malls (see Table A1 of the appendix) as primary sampling units. To ensure that households from diverse neighborhoods with different mean income levels and different access to modern retailers are represented in the sample, these compounds were stratified into three income categories: high (Avondole, Chalala, Kabulonga, Woodlands), medium (Chelston, Chilenje, Kabwata, PHI) and low (Chawama, Chazanga, Gardens, Kalingalinga, Kaunda Square, Ng'ombe)—see Figure A1 of the appendix for the distribution of the surveyed compounds. Finally, from each compound, we randomly sampled at least 33 households for the interviews. We interviewed the household head or another adult respondent responsible for most food purchase decisions. We used trained enumerators to conduct a face-to-face computer-based interview with the sampled households. To capture food consumption at household level, we conducted a 7-day food consumption recall using a structured questionnaire. We also captured a range of demographic and socioeconomic characteristics of the household members such as age, gender, education, ethnicity, religion status, income and employment, among others. Furthermore, we collected information on food shopping behavior, utilization of different food retail outlets and non-food expenditure.

2 Different locations or sections in the city are known as `Compound´.

3.2. Measurement of key variables Considering our interest in analyzing the associations of different food retail outlets with food consumption, we use two main outcome indicators in this study. First, we used food consumption data to compute food expenditure shares (%) by industrial processing levels. We focused on food consumption of three broad categories3: UPFs, primary processed foods and unprocessed foods (see Table A2 of the appendix). This disaggregation is important because it may help us to better understand which food outlet could be associated with nutrition transition. Moreover, poor dietary choices (e.g., consumption of UPFs) have been associated with increased risks of coronary heart diseases, stroke and diabetes (Steyn and Mchiza 2014). Lastly, we computed food quantities (kg/week) consumed by the household as a measure of dietary intake. While our study does not account for nutritional outcomes directly, we categorize the consumption data into several food groups; which are important in influencing consumer’s dietary choices and healthy eating. We focused on consumption of the following food groups: cereals and tubers, meat and fish, oils and fats, sugar and sugar-sweetened beverages, dairy products, eggs, fruits, legumes and vegetables. 3.3. Empirical strategy To explain utilization decisions and estimate associations among multiple food retailers, we employ a multivariate probit (MVP) model. Urban consumers can buy food from several food retailers simultaneously (see Figure 1), depending on several factors. Consequently, we theorize that the household utility is derived from the consumption of food items bought from both modern and traditional retailers (Lu and Reardon 2018). The MVP simultaneously models the complex relationship between a set of covariates and each of the food retailers, while allowing unobserved factors to be correlated (Green, 2012). Thus, our MVP model is estimated as: 1 > 0 = + , with = ∗ = 1,2, … 8 (1) 𝑖𝑖𝑖𝑖 ∗ ′ 0 𝑖𝑖𝑖𝑖 𝐹𝐹𝐹𝐹𝐹𝐹 where𝐹𝐹𝐹𝐹𝐹𝐹 𝑖𝑖𝑖𝑖 denotes𝑋𝑋𝑖𝑖𝑖𝑖𝛽𝛽𝑘𝑘 a 𝜀𝜀latent𝑖𝑖𝑖𝑖 variable𝐹𝐹𝐹𝐹𝐹𝐹𝑖𝑖𝑖𝑖 that� captures the expected𝑘𝑘 utility for household from utilization of∗ a particular outlet ( ). is expressed𝑜𝑜𝑜𝑜ℎ𝑒𝑒𝑒𝑒𝑒𝑒 as𝑒𝑒𝑒𝑒𝑒𝑒 a set of binary dependent variables ( ), 𝑖𝑖𝑖𝑖 each taking𝐹𝐹𝐹𝐹𝐹𝐹 the value of 1 if the household bought food from the outlet and zero otherwise. In𝑖𝑖 terms 𝑖𝑖𝑖𝑖 of retailers, we consider four modern𝑘𝑘 𝐹𝐹𝐹𝐹𝐹𝐹 types: superstores (SS), supermarkets (SM), convenience𝑘𝑘 stores (CS), fast-food restaurants (FF) and four traditional types: traditional market (TM), grocery store (GS), roadside market (RM) and neighborhood kiosk (NK) as dependent variables. To identify our covariates used in the regression analysis, we draw on the existing literature (e.g., Asfaw 2008; Rischke et al. 2015; Umberger et al. 2015; Demmler et al. 2018) on the retail modernization and nutrition transition. Thus, is a vector of observed covariates that represent control variables at household level such as gender, age, education, income, car ownership, 𝑖𝑖𝑖𝑖 ethnicity and religion status. represent vectors𝑋𝑋 of the parameters to be estimated, and is the error term. 𝑘𝑘 𝑖𝑖𝑖𝑖 The error terms in a MVP𝛽𝛽 model are distributed as multivariate normal, each with a 𝜀𝜀mean of zero, and variance-covariance matrix normalized to unity (Cappellari and Jenkins 2003). The matrix generated from the MVP represents correlation of the error terms for any two estimated equations. We use the generated correlation matrix to describe the interrelationship among food retail outlets. The MVP model in equation 1 is estimated using mvprobit command in Stata Statistical Software (STATA 15), and we present the empirical results in sections 4.2 and 4.3.

3 We adopted the classification of food groups or items by level of processing from Demmler et al. (2018), and the list of food items included in our analysis is presented in Table A2 of the appendix. We also hypothesize that the inequalities on food purchases from different food retail outlets could be associated with changes in food consumption patterns by the household. Therefore, to analyze these associations, we estimate the following basic econometric model4: = + + T + (2) where is an outcome for household . The outcomes of interest are the expenditure shares for 𝑖𝑖 0 1 𝑖𝑖 𝑖𝑖 𝑖𝑖 UPFs, primary processed𝑦𝑦 𝛼𝛼 foods𝛼𝛼 and𝑥𝑥 unprocessed𝛽𝛽 𝑢𝑢 foods. We also use food quantity (kg/week) for 𝑖𝑖 cereals𝑦𝑦 and tubers, meat and fish, oils𝑖𝑖 and fats, sugar and sugar-sweetened beverages, dairy products, eggs, fruits, legumes and vegetables as dependent variables. is the treatment variable of interest, and it represents the expenditure share for a particular food retail outlet—i.e., SS, SM, 𝑖𝑖 CS, FF, TM, GS, RM and NK—at household level. is a vector of our𝑇𝑇 exogenous covariates that represents household level characteristics as explained above. The estimable parameters from this 𝑖𝑖 model are , and , and is the random error term.𝑥𝑥 Since our food consumption data are left-censored at zero; e.g., fruits (74%), and right-censored 0 1 𝑖𝑖 at various 𝛼𝛼limits,𝛼𝛼 we 𝛽𝛽estimate𝑢𝑢 a doubly censored-tobit regression to address the issue of data censoring and corner solutions. To further account for the heterogeneity among food retailers, we also cluster standard errors at compound level and estimate our empirical models with disaggregated sample (e.g., poor and non-poor households). Additionally, we use an alternative outcome (e.g., food expenditure) in our empirical analysis for robustness checks. 4. Results and discussion 4.1. Descriptive statistics Table 2 presents summary statistics for our main outcome, treatment and explanatory variables used in the regression analysis. The results show that the average food expenditure is ZMW 112 per capita in a week, and it increases with increasing income—i.e., from lowest to highest income tercile (Table 2). However, the consumption patterns are mixed if we consider different food groups. For example, we observe that food consumption (quantity) increases with increasing income for all food groups (Table 2), cereals and tubers, meat and fish, dairy products, eggs and fruits (Table A3, Columns 2–4 of the appendix). In contrast, we find that the food expenditure shares decreases with increasing income for unprocessed foods unlike primary processed foods. [Insert Table 2] Most households buy their food from modern food retail outlets (MoFROs) such as supermarkets (73%) in the study area (Table 2, Column 1). However, the utilization of other MoFROs; especially SS (5%) and FF (2%) is largely low. Nevertheless, we generally find that the proportion of households using SM and SS increases with increasing income (Table 2, Columns 2–4). Interestingly, even poor (low income) households are largely utilizing these MoFROs. For example, 48% of the sampled households with low income used SS (Table 2, Column 2). This could be driven by two factors. First, the distribution of shopping malls with modern retailers in the city is very high—it has increased from one to 25 since 1995—compared with other cities in southern Africa. Second, retail giants like Shoprite target low income households within the city. Hence, they operate in these low income areas unlike other retailers like Food Lover’s market who target high income areas. Typical for many developing countries, we also observe that urban consumers in Zambia simultaneously buy food from both modern retailers and traditional retailers such as TM (73%), GS (45%), RM (36%) and NK (20%) (Table 2, Column 1). This finding is consistent with previous studies by Berger and van Helvoirt (2018), and Lu and Reardon (2018). This could be driven by several factors such as proximity, lower prices and unavailability of some food items in modern or traditional retailers.

4 We used a basic model for ( |T , ) directly because the counterfactual analysis with multiple treatments is more difficult to handle (Wooldridge 2010). 𝐸𝐸 𝑦𝑦𝑖𝑖 𝑖𝑖 𝑥𝑥𝑖𝑖 Column (1) of Table 2 further shows that the average household size is 4.5 and poor households have a bigger family size (4.9 vs 4.3) than non-poor households. We also observe that 53% of the sample is male-headed households. On average, the sampled households are roughly 44 years old with 12 schooling years. Furthermore, results show that 45% of the sampled households have an office job and the average household income is US$ 10,691 per year. Overall, we mainly observe that high income households have more education (schooling years) and income than low income households as expected. On transport mode, 28% of the sampled households own a private car and the ownership rate increases with increasing income. The main ethnic and religious groups are Bemba (29%) and Protestants (42%), respectively. 4.2. Factors explaining utilization of multiple food retail outlets (MFRO) Table 3 presents marginal5 effects from the multivariate probit (MVP) model on factors explaining utilization of MFRO (i.e., estimation of equation 1). The Wald test that all regression coefficients in each equation are simultaneously equal to zero is rejected ( (104) = 364; = 0.000), indicating that the model fits the data nicely. Hence, the MVP model2 is preferred over single‐equation probit models—i.e., the standard probit models with only one𝜒𝜒 dependent variable𝑝𝑝 . This suggests that our MVP model captures the utilization of MFRO better than the single-equation probit models. [Insert Table 3] Our marginal effects in Table 3 show that several factors such as gender, household size, education, office job, income, car ownership, ethnicity and religion status explain utilization of MFRO. We observe that male-headed households will likely use traditional retail outlets such as GS, RM and NK. In contrast, female-headed households will likely use SM. We also find that household size has a positive and significant influence on the use of CS, FF, GS and RM. Conversely, we note a significant negative association between household size and SM users. Perhaps, this is so because most households with a bigger family size are likely to be poor. Hence, with resource limitations they are unable to use SM regularly compared with other food retail outlets. Table 3 results show that education has mixed effects on utilization of MFRO. Generally, we find that households with more formal education are likely to use SM and FF as opposed to traditional retailers. We speculate that households who are more educated are likely to have more resources, and this could give them a comparative advantage to use modern retailers than traditional retailers. Another plausible explanation is that most shopping malls are usually constructed closer to high income areas of the city unlike low income areas. Unsurprisingly, most households who are less educated also stay in low income areas, and hence they may not use some of the modern retailers due to proximity limitations. We also find a positive association between office job and use of SM compared to traditional retailers such as RM and NK. Further, our results show that income and car ownership increases the probability of using SS and SM (Table 3, Columns 1–2). Unsurprisingly, households with more income will likely use these outlets, possibly because they have high purchasing power to buy cars—this helps them to easily access the outlets for food purchases. Moreover, occasionally non-poor households largely use modern retailers to ensure food safety, high quality products and accessing a large variety of food products (Gorton et al. 2011; Wertheim-Heck et al. 2015). In contrast, we find that income and car ownership are negatively associated with the use of GS (Table 3). Unsurprisingly, most poor households stay closer to traditional retailers, and therefore they may just walk to the informal market to buy food. Furthermore, some of the outlets (i.e., Food Lover’s market) offer relatively higher food prices compared to other SM retailers such as Shoprite. Therefore, most households with lower income levels may not patronize certain modern food outlets. Similarly, Figuié and

5 We also estimated standard coefficients from the MVP model and the results are presented in Table A4 of the appendix. Moustier (2009) and Berger and van Helvoirt (2018) found that poor households buy very little from SM due to several constraints—e.g., higher prices and proximity limitations—in Vietnam and Kenya, respectively. Hence, they rely on traditional retailers to ensure food accessibility, credit opportunities and low prices. Surprisingly, we observe a positive association between income and use of NK. Perhaps, this is so because most non-poor households are using this outlet (see Table 2). On ethnicity and religion status, we observe that consumers who are Tongas and Catholics, respectively, will unlikely use SM as opposed to SS (Table 3). We also find that consumers who are Chewa are likely to use GS. Overall, the possible explanation to these mixed results is that urban consumers are largely heterogeneous in terms of religion, ethnicity (culture and beliefs) as well as affinity for certain food products. Hence, they will likely purchase different food items from MFRO. 4.3. Interrelationships among multiple food retail outlets (MFRO) To understand the interrelationships among MFRO, we also report correlation matrix results from the MVP in Table 4. Based on the likelihood ratio test results ( (28) = 85; = 0.000) in Table 4, we reject the null hypothesis of equal correlation coefficients2 among MFRO. We find several significant positive and negative correlations among MFRO; implying𝜒𝜒 that the𝑝𝑝 associations can be interpreted as complementary or substitutes, respectively. [Insert Table 4] Several strong and significant complementarities exist among the following retailers: e.g., (i) SS and CS, (ii) CS and RM, and (iii–iv) GS with RM or NK (Table 4). Overall, this is not surprising, because most of these retailers buy and sell fairly different (uncompetitive) food products. Hence, consumers will visit other food outlets, if a particular food item is unavailable in the outlet. For instance, most consumers will likely buy indigenous vegetables and legumes from selected traditional retailers (TR) compared to modern retailers like CS. Similarly, a few previous studies (e.g., Figuié and Moustier 2009; Berger and van Helvoirt 2018) have found that consumers use both modern and traditional retailers to access certain food products. Conversely, we find that there are a few substitution associations among retailers: e.g., (i) SM and GS, and (ii) TM and RM (Table 4). Perhaps, this is so because these outlets buy and sell fairly homogenous (competitive) food products such as bread, milk, flour, sugar, beverages, vegetables and fruits. Hence, TR are facing increased competition from modern retailers (Hovhannisyan et al. 2018; Stewart and Dong 2018). Similarly, within the traditional retail sector, there is competition among TR (e.g., TM and RM), and this is consistent with findings by Suryadarma et al. (2010) in Indonesia. 4.4. Effects of multiple food retail outlets (MFRO) on food consumption Our investigation on the associations (i.e., estimation of equation 2) between the use of MFRO and food consumption patterns follows two separate steps. First, we estimated the heterogeneous effects of retail outlets on food consumption by categorizing processing levels into (i) UPFs, (ii) primary processed foods and (iii) unprocessed foods. Empirical results are presented in Table 5. Second, we further focus on consumption of specific food groups under two arbitrary broad categories: processed and unprocessed foods. We classify the consumption of cereals and tubers, meat or meat products (i.e., sausages) and fish, oils and fats, sugar and sugar-sweetened beverages, and dairy products as processed foods. While the consumption of eggs, fruits, legumes and vegetables is considered as unprocessed foods. We estimated a separate model for each food group and our regression results are presented in Table 6. 4.4.1. Food consumption by industrial processing levels Panel A of Tables 5 and 6 presents results with supermarket expenditure share only as a treatment outcome of interest. We find that supermarket purchases are associated with an increase in consumption of UPFs at the expense of unprocessed foods (Panel A, Table 5). These findings are consistent with previous studies by Asfaw (2008) and Rischke et al. (2015), where they found that supermarket purchases increase the consumption of UPFs at the expense of unprocessed foods in Guatemala and Kenya, respectively. [Insert Table 5] We also accounted for the heterogeneous effects of both modern and traditional retailers on food consumption by processing levels, and the results are presented in Panel B of Table 5. We find that a greater reliance on modern retailers (MR) is associated with an increase in the consumption rate of UPFs by roughly 0.15 to 0.61 percentage points (Table 5, Column 1). Probably, this is so because most MR largely stock and sell fast-food and UPFs with higher fat, caloric, sugar and salt content (Popkin 2014; Qaim 2017; Popkin and Reardon 2018). Hence, the rise in consumption of unhealthy foods e.g., UPFs from MR may contribute to rising rates of overweight or obesity (Kimenju et al., 2015; Demmler et al. 2018). Interestingly, however, we also find that a few traditional retailers (TR) such as GS and NK are associated with higher consumption rates of the UPFs by 0.22 and 0.27 percentage points, respectively (Table 5, Column 1). This finding is probably driven by improvements in food supply chain systems (innovations, infrastructure and financial resources). Hence, even TR are able to buy and sell UPFs such as chips, sugar-sweetened beverages and ready- to-eat dishes. Therefore, these findings imply that the effects of retail modernization on the consumption of UPFs are suddenly not so clear anymore but needs additional nuanced interpretation. Turning to unprocessed foods, the results show that a one-unit-percentage-point increase in the expenditure share for SM and GS or NK is associated with a decrease in the consumption of unprocessed foods by approximately 0.14 and 0.17 percentage points, respectively (Table 5, Column 3). In contrast, we find that an increase in the expenditure share for RM is associated with significantly higher consumption rate for the unprocessed foods by 0.12 percentage points; which is consistent with our results in Columns 6 and 9 of Table 6. This finding is also consistent with previous studies by Suryadarma et al. (2010) and Gorton et al. (2011), where they found that consumers mainly purchase unprocessed foods such as assorted fresh vegetables from traditional retailers compared to modern retailers (MR). To further check the heterogeneous effects of using MFRO on consumption of UPFs, we disaggregated the sample into poor and non-poor households, and the results are presented in Table A5 of the appendix. Interestingly, we find that an increase in the expenditure share in modern retailers e.g., CS is associated with higher consumption rate for the UPFs by roughly 0.27 percentage points for the poor households. This finding suggests that even urban poor households may be at higher risk of major public health problems such as overweight or obesity, diabetes and cardiovascular diseases due to the rise in consumption of UPFs (Demmler et al., 2018). 4.4.2. Food consumption by key food groups We largely find that purchases from modern retailers (MR) are significantly associated with an increase in consumption for cereals and tubers, and meat and fish (Table 6, Columns 1–2). Conversely, we find that most of the UPFs such as oils and fats, beverages and dairy products are sourced from both modern and traditional retailers (Panel B of Table 6, Columns 3–5). Interestingly, these results could have two important contrasting implications on dietary changes and nutrition transition. First, the consumption of dairy products could promote healthy diets and reduces major public health concerns such as obesity. Second, the rise in consumption of UPFs from MR—in some cases even traditional retailers (TR)—may contribute to unhealthy diets and increasing the risk of major public health problems to consumers. Nevertheless, our results are consistent with previous studies (e.g., Asfaw, 2008; Reardon and Timmer, 2014; Popkin and Reardon, 2018), as they also observed that the retail modernization could have mixed effects on dietary changes and nutrition transition. [Insert Table 6] Panel B of Table 6 further shows that TR play a crucial role for consumers to source most unprocessed foods. For example, in Columns 6 and 9 of Table 6, we find that purchases from TR such as NK and TM increases the consumption of eggs and vegetables, respectively. Similarly, Suryadarma et al. (2010) and Gorton et al. (2011) found that fresh vegetables are mainly sourced from TR in Indonesia and Thailand, respectively. Hence, the retail modernization process should be more inclusive to TR to promote healthy eating. Traditional retail owners could be encouraged to supply both indigenous and exotic vegetables to modern retailers, possibly through contractual arrangements. Regardless of the outcome used (see also Table A7 of the appendix), we find that purchases from both modern and traditional retailers are associated with lower consumption rates of fruits (Table 6, Column 7). A plausible explanation is that some households economically may not value the importance of buying fruits at the expense of other food groups such as cereals, meat and legumes. Further, this could be explained by the unavailability and higher prices of fruits in selected food retailers. As noted by Mergenthaler et al. (2009) and Berger and van Helvoirt (2018), fresh vegetables and fruits were more expensive in modern retailers (e.g., SM) than traditional retailers in Vietnam and Kenya, respectively. Therefore, fruit purchases and consumption could improve in the study area and beyond (the trend is similar in most sub-Saharan African countries) if food retailers adopt healthy checkouts—selling low-cost, pre-packaged, convenient, pairs of fruits and vegetables—as part of public health marketing strategy (Payne and Niculescu, 2018). Apart from healthy checkouts, creating consumer awareness on the importance of fruits in healthy eating could be a feasible intervention. 5. Conclusion Food retail modernization has been evolving rapidly for the past century in developing countries including Zambia. The retail modernization may have significantly contributed to dietary diversity as well as nutrition transition. This could be associated with several health problems such as obesity, diabetes and cardiovascular diseases. However, empirical evidence analyzing the linkage between food retail modernization and people’s diets is still scanty. A few existing studies have focused exclusively on single modern retailers like supermarkets, and empirically they have neglected the fact that most people especially in Africa buy their food from different—i.e., both modern and traditional—retail outlets simultaneously. Using household survey data, we analyze the associations between household socioeconomic status, the use of different types of modern and traditional retailers, and dietary patterns. Our results show that the use of different retail outlets is strongly associated with various household socioeconomic characteristics. For instance, richer and better educated households are more likely to use supermarkets than poorer and less educated households. People with an office job are also more likely to use supermarkets. Larger households are more likely to use grocery stores than smaller households. However, the results are not uniform across all modern and all traditional retailers, showing that differentiating by type of retailer is indeed important. Several complementary and substitution correlations exist among food retailers. In terms of dietary patterns, regardless of the poverty status (poor vs non-poor) of the household, we find that the use of modern retailers is associated with a significantly higher consumption of UPFs at the expense of unprocessed foods. Interestingly, however, the same holds true for some of the traditional retailers, especially grocery stores and neighborhood kiosks. Hence, the findings suggest that simplistic statements about whether modern retailers improve or worsen household diets are not possible; more nuance is required. Our findings also suggest that the retail modernization process should be more inclusive to traditional food outlets to promote healthy eating; traditional retail owners could be encouraged to supply both indigenous and exotic vegetables to modern retailers, possibly through contractual arrangements. Furthermore, since we largely find that the consumption of fruits is very low; its consumption could improve if food retailers adopt healthy checkouts as a public health marketing strategy (Payne and Niculescu, 2018). Apart from healthy checkouts, creating consumer awareness on the importance of fruits in healthy eating could be a possible policy intervention. While our study has contributed to a better understanding on associations between multiple food retail outlets and food consumption patterns for urban households in an African context, a few issues could be interesting for future research. First, we use cross-sectional data and we could not account for potential endogeneity in our empirical strategy due to challenges in identifying valid instruments. Lastly, we captured few observations for some retailers such as fast-food restaurants. Further, we conducted this study in Lusaka city of Zambia, which may be different with other cities in southern Africa. Therefore, future research should focus on causality using regional panel data with a larger sample size for certain food retail outlets. References Andersson CIM, Chege CGK, Rao EJO & Qaim M, 2015. Following up on smallholder farmers and supermarkets in Kenya. American Journal of Agricultural Economics 97(4): 1247–66. Asfaw A, 2008. Does supermarket purchase affect the dietary practices of households? Some empirical evidence from Guatemala. Development Policy Review 26 (2): 227–43 Berger M & van Helvoirt B, 2018. Ensuring food secure cities – Retail modernization and policy implications in Nairobi, Kenya. Food Policy 79: 12–22. Cappellari L & Jenkins SP, 2003. Multivariate probit regression using simulated maximum likelihood. The Stata Journal 3(3): 278–94. Demmler KM, Ecker O & Qaim M, 2018. Supermarket shopping and nutritional outcomes: A panel data analysis for urban Kenya. World Development 102: 292–303. Figuié M & Moustier P, 2009. Market appeal in an emerging economy: Supermarkets and poor consumers in Vietnam. Food Policy 34: 210–17. Gómez MI & Ricketts KD, 2013. Food value chain transformations in developing countries: Selected hypotheses on nutritional implications. Food Policy 42: 139–50. Gorton M, Sauer J & Supatpongkul P, 2011. Wet markets, supermarkets and the “big middle” for food retailing in developing countries: Evidence from Thailand. World Development 39(9): 1624–37. Greene WH, 2012. Econometric analysis. Seventh edition. Prentice Hall, Boston, MA. Hawkes C, 2008. Dietary implications of supermarket development: A global perspective. Development Policy Review 26(6): 657–92. Hovhannisyan V, Cho C & Bozic M, 2018. The relationship between price and retail concentration: Evidence from the US food industry. European Review of Agricultural Economics 46(2): 319– 45. IFPRI, 2017. Global nutrition report 2017. Washington, DC: International Food Policy Research Institute. Kimenju SC, Rischke R, Klasen S & Qaim M, 2015. Do supermarkets contribute to the obesity pandemic in developing countries? Public Health Nutrition 18: 3224–33. Laska MN, Sindberg LS, Ayala GX, D’Angelo H, Horton LA, Ribisl KM, Kharmats A, Olson C & Gittelsohn J, 2018. Agreements between small food store retailers and their suppliers: Incentivizing unhealthy foods and beverages in four urban settings. Food Policy 79: 324–30. Lu L & Reardon T, 2018. An economic model of the evolution of food retail and supply chains from traditional shops to supermarkets to e-commerce. American Journal of Agricultural Economics 100(5): 1320–35. Mergenthaler M, Weinberger K & Qaim M, 2009. The food system transformation in developing countries: A disaggregate demand analysis for fruits and vegetables in Vietnam. Food Policy 34: 426–36. Nyirenda DB, Musukwa M, Habulembe R & Shindano J, 2009. Zambia food composition tables. Fourth edition. Lusaka, Zambia. Payne C & Niculescu M, 2018. Can healthy checkout end-caps improve targeted fruit and vegetable purchases? Evidence from grocery and SNAP participant purchases. Food Policy 79: 318–23. Popkin BM, 2014. Nutrition, agriculture and the global food system in low and middle income countries. Food Policy 47: 91–6. Popkin BM & Reardon T, 2018. Obesity and the food system transformation in Latin America. Obesity Reviews 19: 1028–64. Qaim M, 2017. Globalisation of agrifood systems and sustainable nutrition. Proceeding of the Nutzrition Society 76(1): 12–21. Reardon T & Timmer CP, 2014. Five inter-linked transformations in the Asian Agrifood economy: Food security implications. Global Food Security 3(2): 108–117. Rischke R, Kimenju SC, Klasen S, & Qaim M, 2015. Supermarkets and food consumption patterns: The case of small towns in Kenya. Food Policy 52: 9–21. Stewart H & Dong D, 2018. How strong is the demand for food through direct-to-consumer outlets? Food Policy 79: 35–43. Steyn NP & Mchiza ZJ, 2014. Obesity and the nutrition transition in Sub-Saharan Africa. Annals of the New York Academy of Sciences 1311(1): 88–101. Suryadarma D, Akhmadi AP, Budiyati S, Rosfadhila M & Suryahadi A, 2010. Traditional food traders in developing countries and competition from supermarkets: Evidence from Indonesia. Food Policy 35(1): 79–86. Tschirley D, Reardon T, Dolislager M & Snyder J, 2015. The rise of a middle class in East and Southern Africa: Implications for food system transformation. Journal of International Development 27: 628–46. Umberger WJ, He X, Minot N & Toiba H, 2015. Examining the relationship between the use of supermarket and over-nutrition in Indonesia. American Journal of Agricultural Economics 97(2): 510–25. Wertheim-Heck SCO, Vellema S & Spaargaren G, 2015. Food safety and urban food markets in Vietnam: The need for flexible and customized retail modernization policies. Food Policy 54: 95–106. Wooldridge JM, 2010. Econometric analysis of cross section and panel data. Second edition. The MIT Press, Cambridge, Massachusetts. Ziba F & Phiri M, 2017. The expansion of regional supermarket chains: Implications for local suppliers in Zambia. WIDER Working Paper 2017/58. Table 1: Key features of MFRO in Lusaka city, Zambia Modern retail outlets Traditional retail outlets Characteristic Superstore Supermarket Convenience Fast-food Grocery Traditional Roadside Neighborhood store restaurant store market market kiosk Business place Bigger shopping Bigger shopping Smaller Bigger shopping Very small Formal and Self-constructed Self-constructed mall mall shopping mall mall or Filling shopping permanent counter counter station mall counter Floor space (m2) >200 100–200 <100 10–30 10–70 1–10 1–5† 1–5†

Modern cash tills 4–15 4–10 <4 <4 No cash tills No cash tills No cash tills No cash tills

Service type Self-service Self-service Self-service Pressing order Over the Direct contact Direct contact Direct contact counter with customer with customer with customer service† Credit facility No No No No Possible Possible Possible Possible

Promotions via media Often Often Often Often Very rare No No No

Price discounts Occasional Occasional (e.g., Occasional (e.g., Occasional Very rare No No No (e.g., month month ends) month ends) (e.g., month ends) ends) Price negotiations No No No No No Often Often Often

Product range† Large variety of Large variety of Limited variety Only fast food Limited Fairly large Fairly large Fairly large food and non- food and non- of food and non- products and variety of variety of variety of fruits variety of food products food products food products beverages food legumes, and vegetables legumes, cereals products cereals and and vegetables vegetables Large variety of Large variety of Limited variety Limited variety fruits and few fruits and few of fruits and of vegetables vegetables vegetables vegetables Frozen, canned Frozen, canned Limited variety Occasional and cooked food and cooked food of frozen, cooked food canned and e.g., chips cooked food Packaging size† Small to very Small to very Small to very Small to very Small to very Small to very Small to very Small to very large large large large large small small small Repacking No No No No No Often Often Often

Key actors Game stores, Shoprite, Numerous Hungry Lion, Numerous Soweto, Numerous Numerous (examples) Cheers, PicknPay, Food Debonairs Pizza Compound Choppies Lover’s, Spurs , KFC, KEG markets Note: Source: Author’s compilation through observations, personal interviews with key actors and † denote features adopted from Berger and van Helvoirt (2018), and Rischke et al. (2015).

13

Table 2: Descriptive statistics for key variables by income category Full By income tercile Variable Unit sample Lowest Middle Highest

(1) (2) (3) (4) Outcomes Household income US$/year 10691.40 1855.67 7548.14 22920.93 (12163.16) (1036.68) (2134.58) (14347.06) Food expenditure ZMW†/capita/week 112.46 96.32 115.61 125.69 (62.98) (65.99) (59.37) (60.18) Food quantity Kg/capita/week 6.52 5.89 6.59 7.10 (3.27 ) (3.15 ) (3.28 ) (3.30 ) Treatments: Food retail outlets Superstore 1=Food purchases from the outlet; 0 0.05 0.01 0.04 0.12 otherwise (0.23) (0.08) (0.19) (0.33) Supermarket 1=Food purchases from the outlet; 0 0.73 0.48 0.78 0.92 otherwise (0.45) (0.50) (0.42) (0.27) Convenience store 1=Food purchases from the outlet; 0 0.12 0.12 0.09 0.16 otherwise (0.33) (0.33) (0.29) (0.37) Fast-food restaurant 1=Food purchases from the outlet; 0 0.02 0.01 0.01 0.04 otherwise (0.14) (0.11) (0.11) (0.19) Grocery store 1=Food purchases from the outlet; 0 0.45 0.64 0.43 0.28 otherwise (0.50) (0.48) (0.50) (0.45) Traditional market 1=Food purchases from the outlet; 0 0.73 0.70 0.74 0.74 otherwise (0.45) (0.46) (0.44) (0.44) Roadside market 1=Food purchases from the outlet; 0 0.36 0.54 0.33 0.20 otherwise (0.48) (0.50) (0.47) (0.40) Neighborhood kiosk 1=Food purchases from the outlet; 0 0.20 0.17 0.20 0.23 otherwise (0.40) (0.38) (0.40) (0.42) Covariates Male 1=If the household head is male; 0 0.53 0.46 0.49 0.65 otherwise (0.50) (0.50) (0.50) (0.48) Household size Number of household members 4.52 4.53 4.47 4.56 (1.63) (1.79) (1.66) (1.43) Age of household head Years 43.83 45.13 41.98 44.40 (12.86) (13.67) (12.68) (12.02) Adolescent 1=If age ranges 11 to 18 years; 0 0.47 0.50 0.49 0.43 otherwise (0.50) (0.50) (0.50) (0.50) Child 1=If age is less than 11 years; 0 0.59 0.71 0.53 0.54 otherwise (0.49) (0.45) (0.50) (0.50) Education of household Schooling years 12.03 9.48 11.88 14.77 head (3.93) (3.62) (3.46) (2.71) Office job 1=If a household member has office 0.45 0.10 0.51 0.74 job; 0 otherwise (0.50) (0.30) (0.50) (0.44) Car ownership 1=if the household owns a car; 0 0.28 0.03 0.21 0.60 otherwise (0.45) (0.16) (0.41) (0.49) Bemba (used as a 1=If ethnicity is Bemba; 0 0.29 0.28 0.24 0.36 reference group) otherwise (0.45) (0.45) (0.43) (0.48) Tonga 1=If ethnicity is Tonga; 0 otherwise 0.19 0.15 0.21 0.21 (0.39) (0.36) (0.41) (0.41) Protestant (used as a 1=If religion is Protestant; 0 0.42 0.42 0.46 0.38 reference group) otherwise (0.49) (0.49) (0.50) (0.49) Catholic 1=If religion is Catholic; 0 0.26 0.31 0.19 0.29 otherwise (0.44) (0.47) (0.39) (0.45) Observations 475 159 160 156 Note: † denotes Zambia Kwacha currency unit. The average exchange rate was ZMW 9.87 to a US dollar (US$) between April and July, 2018. Standard deviations are reported in parentheses.

14

Table 3: Factors explaining utilization of MFRO: Marginal effects from MVP model

Modern retail outlets Traditional retail outlets Convenience Fast-food Grocery Traditional Roadside Neighborhood Superstore Supermarket store restaurant store market market kiosk (1) (2) (3) (4) (5) (6) (7) (8) Male 0.007 –0.088** 0.022 –0.008 0.105** 0.009 0.168*** 0.091** (dummy =1 if true) (0.023) (0.035) (0.032) (0.015) (0.044) (0.041) (0.043) (0.038) Household size –0.004 –0.031** 0.019* 0.009* 0.054*** 0.017 0.044*** –0.001 (number) (0.009) (0.012) (0.011) (0.005) (0.016) (0.016) (0.015) (0.026) Age of household head 0.000 0.000 0.001 –0.001 –0.002 –0.001 –0.002 0.001 (years) (0.001) (0.001) (0.001) (0.001) (0.002) (0.002) (0.002) (0.002) Adolescent† 0.014 0.053 –0.003 0.010 –0.017 0.050 –0.060 0.043 (dummy =1 if true) (0.022) (0.037) (0.034) (0.015) (0.046) (0.045) (0.045) (0.040) Children† –0.002 –0.019 0.015 0.011 0.030 0.061 0.009 0.016 (dummy =1 if true) (0.022) (0.040) (0.034) (0.016) (0.048) (0.046) (0.046) (0.042) Education of household –0.002 0.025*** –0.007 0.006** –0.021*** 0.000 –0.028*** –0.012* head (schooling years) (0.005) (0.006) (0.005) (0.003) (0.007) (0.007) (0.007) (0.006) Office job 0.004 0.109** –0.015 –0.033* –0.072 0.123** –0.125** –0.100** (dummy =1 if true) (0.027) (0.043) (0.038) (0.018) (0.052) (0.053) (0.051) (0.046) Household income‡ 0.031** 0.063*** 0.027 0.011 –0.045** 0.015 –0.043** 0.072*** (ZMW/year) (0.015) (0.017) (0.018) (0.009) (0.022) (0.022) (0.021) (0.022) Car ownership 0.054** 0.157*** 0.086** 0.010 –0.124** –0.113** 0.008 –0.012 (dummy =1 if true) (0.024) (0.058) (0.038) (0.017) (0.056) (0.054) (0.054) (0.048) Chewa –0.035 –0.011 –0.008 –0.176 0.107* –0.024 –0.098 0.007 (dummy =1 if true) (0.047) (0.050) (0.048) (6.286) (0.064) (0.063) (0.062) (0.055) Tonga 0.058** –0.118** 0.067 0.008 0.005 0.005 –0.057 –0.008 (dummy =1 if true) (0.024) (0.048) (0.041) (0.017) (0.058) (0.060) (0.056) (0.050) Catholic 0.039* –0.089** 0.052 0.020 0.078 0.036 –0.041 0.067 (dummy =1 if true) (0.023) (0.039) (0.033) (0.016) (0.049) (0.047) (0.047) (0.041) Seventh Day Adventist –0.017 0.031 –0.059 0.001 0.010 0.083 –0.049 –0.007 (dummy =1 if true) (0.028) (0.053) (0.049) (0.014) (0.060) (0.063) (0.059) (0.058) Note: Number of observations=475. Log pseudo likelihood=–1460, and Wald χ2 (104) =364***. ‡ represents a variable expressed as a logarithm. † The age category for children and adolescent is less than 11 years and 11–18 years, respectively. Bemba and Protestant are used as the reference group for ethnicity and religion status, respectively. Standard errors are reported in parentheses. *, **, and *** represent 10%, 5%, and 1% significance levels, respectively.

15

Table 4: Correlation coefficient matrix from multivariate probit estimates for MFRO Modern retail outlets Traditional retail outlets

SS SM CS FF GS TM RM NK (1) (2) (3) (4) (5) (6) (7) (8)

Superstore 1.000 (SS)

Supermarket 0.161 1.000 (SM) (0.122)

Convenience 0.252** 0.149 1.000 store (CS) (0.114) (0.099) Fast-food –0.088 –0.047 0.198 1.000 restaurant (FF) (0.236) (0.220) (0.205)

Grocery store –0.098 –0.304*** 0.009 0.388*** 1.000 (GS) (0.108) (0.073) (0.090) (0.122)

Traditional 0.074 –0.164* 0.064 –0.046 0.022 1.000 market (TM) (0.108) (0.084) (0.091) (0.127) (0.080)

Roadside 0.060 –0.040 0.163* 0.285** 0.249*** –0.282*** 1.000 market (RM) (0.105) (0.086) (0.091) (0.124) (0.076) (0.081)

Neighborhood –0.003 –0.145* –0.086 0.137 0.222*** –0.124 –0.026 1.000 kiosk (NK) (0.117) (0.086) (0.096) (0.114) (0.081) (0.086) (0.083)

Note: The likelihood ratio test of equal correlation coefficients is rejected; χ2 (28) =85***. Standard errors are reported in parentheses. *, **, and *** represent 10%, 5%, and 1% significance levels, respectively.

16

Table 5: Associations between MFRO and food expenditure share by processing levels: OLS estimation results By industrial processing levels (Expenditure share (%)) Expenditure share (%) Ultra-processed foods Primary processed foods Unprocessed foods (1) (2) (3) Panel A: Single food retail outlet Supermarket only 0.051** 0.043* –0.094*** (0.022) (0.021) (0.027) Other covariates Yes Yes Yes R-squared 0.035 0.122 0.146 Panel B: MFRO

Superstore 0.146* –0.018 –0.128 (0.071) (0.095) (0.091)

Supermarket 0.196*** –0.053 –0.143* (0.052) (0.075) (0.075)

Convenience store 0.293*** –0.267** –0.026 (0.091) (0.110) (0.097) Fast-food restaurant 0.611*** –0.671*** 0.060 (0.109) (0.091) (0.168)

Grocery store 0.217*** –0.043 –0.174** (0.055) (0.070) (0.066)

Traditional market 0.063 –0.122* 0.058 (0.044) (0.063) (0.070) Roadside market 0.041 –0.164** 0.122* (0.054) (0.061) (0.063) Neighborhood kiosk 0.274*** –0.101 –0.173*

(0.079) (0.093) (0.098) Other covariates Yes Yes Yes R-squared 0.116 0.149 0.256 Observations 475 475 475

Note: OLS=ordinary least squares. All estimations include the following covariates: age, education, household size, log of household income, a dummy (1, 0) variable for male, ethnic groups—Chewa, and Tonga; Bemba is the reference group, and religion status—Catholic and SDA (Seventh Day Adventist); Protestant is the reference group. Robust standard errors—clustered at compound level—are reported in parentheses. *, **, and *** represent 10%, 5%, and 1% significance levels, respectively.

17

Table 6: Associations between MFRO and food consumption for selected food groups: Tobit estimation results Food quantity (kg/week) Cereals Meat Oils Sugar Dairy Eggs Fruits Legumes Vegetables Expenditure share (%) and and Fish and Fats and Beverages products Tubers (1) (2) (3) (4) (5) (6) (7) (8) (9) Panel A: Single food retail outlet Supermarket only –0.003 0.015*** –0.003 –0.010*** 0.014* –0.002 –0.005 –0.001 –0.001 (0.006) (0.006) (0.002) (0.004) (0.007) (0.001) (0.004) (0.005) (0.009) Other covariates Yes Yes Yes Yes Yes Yes Yes Yes Yes

Pseudo-R-squared 0.060 0.073 0.068 0.024 0.081 0.053 0.017 0.027 0.011 Panel B: MFRO Superstore 0.025 0.043* 0.009 0.040*** 0.053* 0.007 –0.009 –0.009 0.013 (0.031) (0.023) (0.005) (0.008) (0.029) (0.006) (0.018) (0.019) (0.020) Supermarket 0.011 0.030* 0.005 0.015* 0.055*** 0.005 –0.031** 0.003 0.027 (0.018) (0.016) (0.004) (0.008) (0.020) (0.003) (0.015) (0.012) (0.021) Convenience store 0.058** 0.022 0.014** 0.020 0.014 0.002 –0.039* –0.007 0.012 (0.025) (0.015) (0.006) (0.013) (0.050) (0.005) (0.023) (0.011) (0.019) Fast-food restaurant –0.100*** 0.110* 0.105** 0.132** (0.037) (0.062) (0.049) (0.055) Grocery store 0.013 0.026 0.005 0.028*** 0.063** 0.008** –0.030* –0.003 0.016 (0.016) (0.016) (0.004) (0.007) (0.029) (0.004) (0.017) (0.013) (0.023) Traditional market 0.011 0.015 0.011*** 0.024*** 0.023 0.004 –0.033** 0.016 0.058*** (0.018) (0.015) (0.004) (0.008) (0.022) (0.003) (0.015) (0.013) (0.015) Roadside market 0.010 0.007 0.005 0.010 0.038** 0.006 –0.038** 0.012 0.038** (0.019) (0.016) (0.004) (0.007) (0.015) (0.004) (0.018) (0.013) (0.016) Neighborhood kiosk 0.030 –0.010 0.007 0.027* 0.057** 0.017*** –0.014 –0.013 –0.008 (0.027) (0.019) (0.005) (0.015) (0.027) (0.004) (0.017) (0.015) (0.025) Other covariates Yes Yes Yes Yes Yes Yes Yes Yes Yes Pseudo-R-squared 0.065 0.086 0.085 0.035 0.101 0.074 0.044 0.045 0.028 Observations 475 475 475 475 475 475 475 475 475 Note: All estimations include the following covariates: age, education, household size, log of household income, a dummy (1, 0) variable for male, ethnic groups—Chewa, and Tonga; Bemba is the reference group, and religion status—Catholic and SDA (Seventh Day Adventist); Protestant is the reference group. Robust standard errors—clustered at compound level—are reported in parentheses. *, **, and *** represent 10%, 5%, and 1% significance levels, respectively.

18

70

60 Modern retailers Traditional retailers Both

50

40

30

20 Percent households (%) Percent of

10

00 0 1 2 3 4 5 6 7 Number of food retail outlets used per week

Figure 1: Utilization of different food retail outlets by urban households Source: Survey data (2018)

Appendix

Figure A1: Map of the study area: Lusaka urban, Zambia Source: Survey data (2018)

Table A1: The distribution of the shopping malls, key modern food retail outlets and its surrounding compounds in Lusaka city, Zambia Modern food retail outlets: Supermarket or No. Major shopping mall Location or compound name superstore, (fast-food restaurant) Roma, University of Zambia 1 Arcades Spurs (UNZA), 2 Cairo Central Lusaka, Shoprite, (Food Fayre, Hungry Lion, Machachos) 3 Chawama Chawama†, John Haward, Kuku Spur 4 Chazanga Shoprite Chazanga†, SOS† Shoprite Choppies, Shoprite, (Debonairs Pizza‡, MM 5 Chilenje Shoprite Chalala†, Chilenje†, Woodlands† Chickens, Naaz) 6 Choppies Complex Kabulonga†, Sundel, Zamtel flats Choppies Chawama†, John Howard, Jon-Lengi, Game Stores, Shoprite, (Chicken Inn, Galito’s, 7 Cosmopolitan Makeni, Misisi Hungry Lion, Mochachos, Pizza Hut) Cross Road, Kabulonga†, Nyumba 8 Cross Roads Spurs, (Gigibonta, Major Meat) Yanga, Sundel Chibolya, Jon-Lengi, Kabwata†, Spurs, (Big Bite, Debonairs Pizza‡, Down Town 9 Down Town Kamwala, Misisi Foods) Childley, Kalingalinga†, Kalundu, Food Lover's, PicknPay, (Fishaways, Gigibonta, 10 East Park Ng’ombe†, Roma, UNZA GoatnChips, Hungry Lion, KEG, Pizza Hut)

11 Embassy Chawama†, Jon-Lengi, Makeni, Misisi Embassy, Spurs, (Papas, Piatto, Zorbas) 12 Garden City Avondole†, Chelston† Food Lover's, PicknPay, (Bushman, Foodano,) Melissa, PicknPay, (Debonairs Pizza‡, KFC, 13 Kabulonga and Melissa Kabulonga† Nando’s, Subway) Central Lusaka, Chilulu†, Evelyn Food Lover's, PicknPay, (Chicken Inn, Hungry 14 Levy Junction Home College, Gardens†, Nippa, Lion, KFC, Pizza Inn, Wimpy) North Mead, Roads Park, Thorn Park Food Lover's, PicknPay, (Debonairs Pizza‡, 15 Makeni Chawama†, Jon-Lengi, Makeni, Misisi KFC, Nando’s) 16 Mama Betty Foxydale Ngo’mbe†, Roma Spur, (Debonairs Pizza‡, Gigibonta) Shoprite, Game Stores, (Bread Café, Debonairs Central Lusaka, Chilulu†, Gardens†, Pizza‡, Galito’s, Hungry Lion, Mugg and Bean, 17 Manda Hill Longacres, Olympia, Roads Park My Asia, Nando’s, Pizza Inn, Steers, Subway, Vasila) 18 Matero Matero Shoprite, (Hungry Lion) 19 Novara Great North§ Chazanga†, SOS† PicknPay, (GoatnChips, Hungry Lion ) 20 PHI Kaunda square†, PHI†, Mtendere PicknPay, (Debonairs Pizza‡, King-Pie) 21 SOS Spurs Chazanga†, SOS† Spur Avondole†, Chelston†, Ibex, Salama Shoprite, (Chicken Inn, Debonairs Pizza‡, 22 Twin Palm Park Hungry Lion) 23 Waterfalls Avondole†, Chelston† Shoprite, (Gigibonta, Hungry Lion) PicknPay, (Creamy, Debonairs Pizza‡, Galito’s, 24 Woodlands Chilenje†, Kabulonga†, Woodlands† Nachies, O. Hagans, Pizza Inn) 25 Zappa§ Chawama† (Debonairs Pizza‡)

Note: § denotes shopping malls that were still under construction, but some food retail outlets had started doing business at the time (April to July, 2018) of the survey. We excluded shopping malls (e.g., Airport and Munali) which were under construction at the time of the survey and many other minor shopping malls. ‡ Most of Debonairs Pizza outlets are also located in some filling or gas stations (e.g., Chelston, Kabwata, Kapiri, Oryx and Woodlands) in Zambia. † indicates the surveyed compound. Key fast-food restaurants are presented in parentheses. Source: Author’s compilation during the survey period.

Table A2: Food groups/items by level of processing

Processing level Food group Food items

Cereals and Tubers Maize (dry/green), cassava, irish potatoes, sweet potatoes, yams

Eggs and Milk Eggs, fresh whole milk Apples, avocadoes, banana (ripe/boiled), guavas, mangoes, Fruits pawpaw, pineapples, pumpkin, oranges/tangerines, sugar plum, watermelons Unprocessed foods Beans (fresh/dry), cowpea (fresh/dry), groundnuts (fresh/dry), Legumes pigeonpea (fresh/dry), soybean, velvet beans Bean leaves, blackjack, cabbage, carrot, cassava leaves, cowpea leaves, cucumber, eggplant, garlic, greengram, lettuce, Vegetables mushroom (cultivated/wild), okra, onions, pepper, pumpkin leaves, rape/mustard/chinese, tomatoes Bottled/clear beer, bottled water, roasted cashew/macadamia Drinks and Snacks nuts Beef, bush/game meat, chicken, ducks, turkey, goat meat, sheep Primary processed foods Meat and Fish meat, pork, fish (fresh/frozen/dried) Cereals Rice, millet, oats, sorghum Bread and Pasta Bread, buns, pasta, instant noodles

Cereals and Tubers Maize flour, cornflakes, porridge mix, wheat flour, cassava flour

Dairy products Cheese, milk, yoghurt

Oils and Fats Butter/margarine, coconut oil, cooking oil/fat, groundnut flour Ultra-processed foods Meat and Fish Sausage (beef/chicken/pork), soya meat, tinned fish Frozen fruits/vegetables, mandazi, mixed fruits/salads, pizza, Miscellaneous samosa, tinned vegetables soft drinks (e.g., coco-cola, fanta, sprite), ginger, fruit juice, Sweet drinks, wine, jam, tomato sauce, salt, sugar, biscuits/cookies, cake, Sweeteners and Snacks chips, chocolate, crisps, puffed salted corn chips, popcorn, salted nuts

Note: The classification of food items by level of processing is adopted from Demmler et al. (2018).

Table A3: Descriptive statistics for key outcome variables by income category By income tercile Full sample Lowest Middle Highest (1) (2) (3) (4) Food quantity (kg/week) 26.92 23.94 26.59 30.29 (10.96) (11.33) (10.06) (10.61) Cereals and Tubers 11.88 11.23 11.45 12.97 (5.20) (5.48) (4.56) (5.38) Meat and Fish 4.81 3.38 4.85 6.24 (3.45) (2.86) (3.24) (3.64) Oils and Fats (liters/week) 0.69 0.65 0.72 0.70 (0.60) (0.58) (0.60) (0.62) Sugar and Beverages 1.68 1.28 1.65 2.13 (2.59) (1.99) (2.31) (3.26) Dairy products 0.61 0.25 0.48 1.11 (1.27) (0.65) (1.01) (1.74) Eggs 0.44 0.28 0.34 0.69 (0.77) (0.64) (0.67) (0.92) Fruits 0.28 0.22 0.26 0.37 (0.82) (0.73) (0.83) (0.89) Legumes 1.22 1.27 1.34 1.03 (1.59) (1.55) (1.83) (1.34) Vegetables 4.22 4.36 4.57 3.70 (3.74) (3.78) (3.87) (3.52) Processing levels (Expenditure share (%)) Ultra-processed foods 0.35 0.36 0.35 0.35 (0.14) (0.14) (0.14) (0.12) Primary processed foods 0.40 0.35 0.41 0.45 (0.17) (0.18) (0.17) (0.15) Unprocessed foods 0.25 0.29 0.25 0.20 (0.14) (0.16) (0.13) (0.12) Food expenditure (ZMW†/week) 456.02 369.62 465.00 534.88 (202.05) (171.36) (179.61) (218.67) Cereals and Tubers 106.41 87.37 108.25 123.94 (57.02) (49.60) (55.01) (60.40) Meat and Fish 172.84 126.04 178.54 214.69 (116.61) (100.16) (107.59) (124.26) Oils and Fats (liters/week) 9.82 9.14 10.28 10.05 (9.47) (8.65) (9.09) (10.61) Sugar and Beverages 33.86 27.55 32.23 41.96 (50.67) (42.59) (43.95) (62.54) Dairy products and Eggs 23.53 14.45 18.54 37.90 (33.24) (18.70) (25.32) (45.25) Fruits 7.88 7.23 6.75 9.70 (20.64) (20.94) (17.37) (23.25) Legumes 30.15 30.84 32.36 27.16 (43.99) (43.52) (49.17) (38.63) Vegetables 59.63 57.99 64.98 55.82 (44.19) (39.59) (47.19) (45.21) Observations 475 159 160 156

Note: † denotes Zambia Kwacha currency unit. The average exchange rate was ZMW 9.87 to a US dollar between April and July, 2018. Standard deviations are reported in parentheses.

Table A4: Factors explaining utilization of multiple food retail outlets: Multivariate probit model results Modern food retail outlets Traditional food retail outlets Convenience Fast-food Grocery Traditional Roadside Neighborhood Superstore Supermarket store restaurant store market market kiosk (1) (2) (3) (4) (5) (6) (7) (8) Male 0.073 –0.391** 0.115 –0.175 0.317** 0.029 0.543*** 0.347** (dummy =1 if true) (0.236) (0.157) (0.166) (0.334) (0.132) (0.133) (0.139) (0.146) Household size –0.043 –0.137** 0.100* 0.215* 0.163*** 0.054 0.143*** –0.002 (number) (0.091) (0.055) (0.058) (0.112) (0.048) (0.050) (0.049) (0.052) Age of household head 0.003 0.000 0.008 –0.017 –0.005 –0.002 –0.007 0.003 (years) (0.009) (0.006) (0.006) (0.014) (0.005) (0.005) (0.005) (0.006) Adolescent† 0.156 0.237 –0.015 0.234 –0.051 0.158 –0.192 0.162 (dummy =1 if true) (0.242) (0.167) (0.172) (0.343) (0.138) (0.141) (0.144) (0.151) Children† –0.021 –0.083 0.078 0.245 0.089 0.192 0.029 0.061 (dummy =1 if true) (0.233) (0.173) (0.176) (0.363) (0.142) (0.146) (0.147) (0.157) Education of household –0.017 0.113*** –0.037 0.148** –0.063*** 0.001 –0.089*** –0.046* head (schooling years) (0.045) (0.025) (0.028) (0.072) (0.022) (0.023) (0.023) (0.025) Office job 0.041 0.481** –0.078 –0.757* –0.217 0.387** –0.404** –0.381** (dummy =1 if true) (0.273) (0.190) (0.200) (0.411) (0.158) (0.165) (0.164) (0.177) Household income‡ 0.334** 0.278*** 0.144 0.254 –0.134** 0.048 –0.140** 0.274*** (ZMW/year) (0.165) (0.076) (0.095) (0.215) (0.065) (0.071) (0.069) (0.083) Car ownership 0.596** 0.695*** 0.459** 0.235 –0.376** –0.354** 0.026 –0.045 (dummy =1 if true) (0.269) (0.258) (0.205) (0.389) (0.169) (0.170) (0.175) (0.182) Chewa –0.388 –0.047 –0.043 –4.005 0.324* –0.074 –0.317 0.027 (dummy =1 if true) (0.516) (0.218) (0.256) (142.148) (0.192) (0.195) (0.201) (0.212) Tonga 0.640** –0.524** 0.357 0.189 0.015 0.015 –0.183 –0.031 (dummy =1 if true) (0.268) (0.215) (0.221) (0.393) (0.176) (0.181) (0.180) (0.194) Catholic 0.422* –0.395** 0.277 0.461 0.236 0.114 –0.132 0.256 (dummy =1 if true) (0.248) (0.172) (0.175) (0.364) (0.148) (0.149) (0.152) (0.158) Seventh Day Adventist –0.190 0.136 –0.313 0.031 0.031 0.260 –0.159 –0.025 (dummy =1 if true) (0.317) (0.234) (0.260) (0.439) (0.184) (0.198) (0.190) (0.208) Constant –5.766*** –2.945*** –3.467*** –7.293*** 1.465** –0.436 1.939*** –3.657*** (1.782) (0.835) (1.044) (2.474) (0.722) (0.763) (0.749) (0.876) Note: Number of observations=475. Log pseudo likelihood=–1460, and Wald χ2 (104) =364***. ‡ represents a variable expressed as a logarithm. † The age category for children and adolescent is less than 11 years and 11–18 years, respectively. Bemba and Protestant are used as the reference group for ethnicity and religion status, respectively. Standard errors are reported in parentheses. *, **, and *** represent 10%, 5%, and 1% significance levels, respectively.

Table A5: Associations between multiple food retail outlets and expenditure share (%) for ultra-processed and unprocessed foods by poverty status: OLS estimation results By poverty status† Poor households Non-poor households Expenditure share for Ultra-processed Unprocessed Ultra-processed Unprocessed food retail outlet (%) foods foods foods foods (1) (2) (3) (4)

Panel A: Single food retail outlet Supermarket only 0.031 –0.058 0.057** –0.111*** (0.050) (0.087) (0.026) (0.025) Other covariates Yes Yes Yes Yes R-squared 0.166 0.106 0.029 0.125

Panel B: Multiple food retail outlets

Superstore 0.165** –0.118 (0.060) (0.087)

Supermarket 0.035 –0.165 0.231*** –0.144 (0.128) (0.205) (0.047) (0.095)

Convenience store 0.274* 0.114 0.329** –0.076 (0.130) (0.187) (0.111) (0.125) Fast-food restaurant –0.627 –1.055 0.679*** 0.071 (0.853) (1.003) (0.081) (0.219)

Grocery store 0.009 –0.165 0.269*** –0.225* (0.118) (0.163) (0.052) (0.106)

Traditional market –0.029 –0.022 0.064 0.106 (0.098) (0.151) (0.048) (0.084) Roadside market –0.073 0.020 0.056 0.158* (0.075) (0.147) (0.067) (0.084) Neighborhood kiosk 0.040 –0.375** 0.358*** –0.067

(0.147) (0.165) (0.053) (0.123) Other covariates Yes Yes Yes Yes R-squared 0.205 0.241 0.144 0.282

Observations 126 126 349 349

Note: OLS=ordinary least squares. Superstore was dropped for the poor household model due to collinearity. †We used $1.90/person/day as a poverty line to classify the sample into poor and non–poor groups. All estimations include the following covariates: age, education, household size, log of household income, a dummy (1, 0) variable for male, ethnic groups—Chewa, and Tonga; Bemba is the reference group, and religion status— Catholic and SDA (Seventh Day Adventist); Protestant is the reference group. Robust standard errors—clustered at compound level—are reported in parentheses. *, **, and *** represent 10%, 5%, and 1% significance levels, respectively.

Robustness checks We also estimated our empirical models on processing levels and specific food groups using an alternative outcome e.g., food consumption expenditure (ZMW/week) as a means of robustness checks and the results are presented in Tables A6 and A7 of this appendix, respectively. After dropping households with zero expenditure, coefficients in Table A6 for expenditure models by processing levels remain similar with our main results in Table 5 of the paper. We also find that an increase in the expenditure for both modern and traditional retailers is associated with higher consumption rate of ultra-processed foods at the expense of unprocessed foods. However, the expenditure results are slightly lower than the expenditure share results. Despite using an alternative outcome, we generally find that our main results in Table 6 of the paper are similar for most of the food groups in Table A7; a noticeable exception is for sugar and beverages. This confirms that urban consumers source different food items from multiple food retail outlets. Nevertheless, we observe that coefficients of interest from the food quantity model (Table 6) are consistently smaller than those in the expenditure model (Table A7). This is possibly due to observed factors such as food prices, which are not accounted for in the food quantity model. Overall, since we are using different outcomes, the exact scale of the coefficients of interest in Tables A6 and A7 varies with our main results in Tables 5 and 6 of the paper as expected. Hence, some caution is necessary when interpreting the exact scale of our empirical results.

Table A6: Associations between multiple food retail outlets and food expenditure by processing levels: OLS estimation results By industrial processing levels Expenditure share for Ultra-processed foods‡ Primary processed Unprocessed foods‡ food retail outlet (%) (ZMW/week) foods‡ (ZMW/week) (ZMW/week) (1) (2) (3) Panel A: Single food retail outlet Supermarket only 0.002 0.002 –0.004* (0.001) (0.001) (0.002) Other covariates Yes Yes Yes R-squared 0.145 0.220 0.086 Panel B: Multiple food retail outlets

Superstore 0.012*** 0.007* 0.006 (0.004) (0.004) (0.006)

Supermarket 0.009** 0.003 0.000 (0.003) (0.003) (0.006)

Convenience store 0.014*** –0.003 0.009 (0.005) (0.003) (0.006) Fast-food restaurant 0.041*** 0.000 0.029*** (0.006) (0.012) (0.006)

Grocery store 0.009** 0.001 –0.002 (0.003) (0.003) (0.006)

Traditional market 0.006* 0.002 0.011** (0.003) (0.003) (0.005) Roadside market 0.003 –0.002 0.010** (0.003) (0.003) (0.004) Neighborhood kiosk 0.010** 0.000 –0.004

(0.004) (0.003) (0.006) Other covariates Yes Yes Yes R-squared 0.199 0.232 0.200 Observations 475 469 471

Note: OLS=ordinary least squares. ‡ represents a variable expressed as a logarithm. All estimations include the following covariates: age, education, household size, log of household income, a dummy (1, 0) variable for male, ethnic groups—Chewa, and Tonga; Bemba is the reference group, and religion status—Catholic and SDA (Seventh Day Adventist); Protestant is the reference group. Robust standard errors—clustered at compound level—are reported in parentheses. *, **, and *** represent 10%, 5%, and 1% significance levels, respectively.

Table A7: Associations between multiple food retail outlets and food consumption expenditure for selected food groups: Tobit estimation results Food expenditure (ZMW/week) Expenditure share for Cereals Meat Oils Sugar and Dairy food retail outlet (%) and Fruits and Fish and Fats Beverages products and Eggs Tubers (1) (2) (3) (4) (5) (6) Panel A: Single food retail outlet Supermarket only 0.095 0.445*** –0.017 –0.086 –0.059 –0.315* (0.069) (0.167) (0.032) (0.054) (0.053) (0.140) Other covariates Yes Yes Yes Yes Yes Yes Pseudo-R-squared 0.035 0.028 0.018 0.010 0.018 0.009 Panel B: Multiple food retail outlets Superstore 0.863** 1.115* 0.237*** 0.109 0.895** 0.248 (0.358) (0.664) (0.066) (0.133) (0.366) (0.813) Supermarket 0.233 1.093*** 0.159** –0.011 0.284* –0.866** (0.177) (0.381) (0.069) (0.149) (0.152) (0.330) Convenience store 0.952** 0.436 0.255*** 0.024 0.384** –1.084* (0.409) (0.356) (0.077) (0.220) (0.173) (0.564) Fast-food restaurant 0.866 4.527** 0.355 1.518 (1.317) (1.789) (0.369) (1.231) Grocery store 0.189 0.879** 0.140** 0.074 0.348* –0.740* (0.180) (0.400) (0.065) (0.123) (0.209) (0.342) Traditional market 0.008 0.722* 0.220*** 0.054 0.203 –0.711* (0.141) (0.379) (0.062) (0.130) (0.150) (0.294) Roadside market –0.143 0.508 0.110* –0.111 0.329** –0.878** (0.168) (0.383) (0.063) (0.129) (0.164) (0.301) Neighborhood kiosk 0.229 0.146 0.219** 0.282 0.670*** –0.108 (0.221) (0.477) (0.091) (0.262) (0.177) (0.445) Other covariates Yes Yes Yes Yes Yes Yes Pseudo-R-squared 0.044 0.031 0.027 0.013 0.026 0.024 Observations 475 475 475 475 475 475

Note: All estimations include the following covariates: age, education, household size, log of household income, a dummy (1, 0) variable for male, ethnic groups—Chewa, and Tonga; Bemba is the reference group, and religion status—Catholic and SDA (Seventh Day Adventist); Protestant is the reference group. Robust standard errors—clustered at compound level—are reported in parentheses. *, **, and *** represent 10%, 5%, and 1% significance levels, respectively.