COVID and consumers’ food strategies in Madrid, Spain: Towards a resilient city-region food system approach Authors: José Luis Cruz, Mario González-Azcárate, Richard J Hewitt, Verónica Hernández-Jiménez, Isabel Bardají Published: 2024 ABSTRACT AbstractLockdowns, restrictions on people's movements and gatherings, and other COVID-19 containment measures have had economic, social, and environmental effects. The pandemic resulted in major disruption to food systems, highlighting serious weaknesses in the way such systems are currently configured. However, the pandemic also created important opportunities to develop transformative actions enabling the creation of more resilient and sustainable food systems. In this paper, we delve more deeply into the strategies developed by consumers, as key stakeholders in the food system, through an in-depth survey and statistical analysis of changes in consumers' habits during the pandemic lockdowns. We adopt a city-region food system (CRFS) approach to explore ways to enhance food system resilience to future crises.Data were collected through a telephone survey of a representative sample (600 consumers), conducted in the region of Madrid in May 2021. First, descriptive analyses allowed different consumer strategies to be identified. Differences in food preferences and behaviour changes between individuals in the sample were analysed by age, gender, income, and level of education using exponential family Principal Component Analysis (EPCA), logistic regression, and Multiple Correspondence Analysis (MCA). Lastly, Principal Component Analysis (PCA) was applied to classify the consumer strategies into four discrete profiles.The findings show significant differences in consumers' responses according to age, income, and level of education. Additionally, the results highlight increased consumption of local and organic products, and online shopping, as typical strategies employed by consumers to respond to the COVID-19 pandemic. In this context, alternative food networks were shown to play a key role in enhancing food security. Interestingly, no significant differences were found between urban and rural areas. Our discussion addresses this apparent homogeneity in consumers’ habits from a resilient CRFS approach.Keywords: Alternative food networks, Consumer profiles, Short food supply chain, Organic products, Urban consumption, Local products Introduction Large cities are especially vulnerable to unexpected and complex global risks and crises [1]. The global pandemic has placed great stress on urban areas due to restrictions on movement and their effects on food systems. Food resilience is key to ensuring food security [2,3]. The concept of resilience was defined by Walker et al. [4] as the capacity of a system to absorb disturbance and reorganize while changing, to retain essentially still the same function, structure, identity, and feedback. Food system resilience is the capacity over time of a food system and its elements at multiple levels to provide sufficient, appropriate and accessible food to all, in the face of various and even unforeseen disturbances [5]. Many researchers have attempted to define the qualities of food systems that make them generally resilient to disruptions [6]. The COVID-19 crisis focused attention on the importance of improving the way cities understand and respond to risks and crises [7,8]. Early efforts by the scientific community were focused on the direct impacts of the pandemic, for example, the factors affecting morbidity and mortality patterns [9], the effect of COVID control measures on urban mobility [10], air pollution [11] or the economy [12]. Indirect or more complex impacts of the pandemic, e.g. food systems, agriculture, and food security or rural-urban disparities and interactions have received far less attention [13]. Despite unprecedented demand, the food supply chain remained strong in most areas, since many supply chain actors, including farmers, producers, distributors, and retailers worked hard to keep shelves filled [14]. However, the extent to which the COVID-19 crisis caused consumers to change their habits remains unclear. Firstly, food behaviour is known to relate to sociodemographic and socioeconomic characteristics of individual populations (e.g. [15,16]), and while several studies have been conducted around consumer food behaviour change during the pandemic, few have examined how these changes varied from person to person. For example, Güney and Sangün [17] conducted a telephone survey of consumers in each of the major cities of the seven regions of Turkey, finding significant changes in food behaviour during the pandemic. However, the study did not shed light on how these changes differed depending on age, gender, income, or level of education. Hesham et al. [18] surveyed consumers in Saudi Arabia, revealing changes in food purchasing behaviours during the pandemic, but did not include age, gender, or income in their analysis of consumers’ profiles. One key aim of this research is to shed light on these important differences between consumers concerning food behaviour. Secondly, as the COVID-19 pandemic recedes into the past, there is a danger that key lessons will be overlooked. In the Madrid Region, the subject of our study, policymakers seek to reduce the potential risks to food systems from future crisis events by improving food system resilience [19]. This implies the development of shorter and more diverse food supply chains by strengthening food production from within the wider city region, for example, under a City Region Food System (CRFS) approach [8]. However, a strong empirical foundation for developing a CFRS approach in our region is lacking. Few studies combine surveys and analysis of food behaviour changes with a detailed discussion of strategies to increase resilience through better integration of food producers and consumers within city regions. The present study is therefore aimed to address these important research gaps. The overarching aim of our study is to provide an evidence base to improve and deepen the degree of alignment between food strategies developed by consumers during the pandemic and the goal of a resilient CFRS. To achieve this aim, we defined three specific objectives. The first objective (OR1) was to analyse changes in consumer habits because of the pandemic. The second objective (OR2) was to identify consumer profiles according to the strategies that they have implemented to cope with the pandemic, with special emphasis on the contribution of alternative food networks (AFNs). The last objective (OR3) was to analyse these changes from a CRFS perspective, building on existing discourse around food systems resilience in the Madrid Region. In this way, our study aims to contribute to the discussion on the vulnerabilities, strengths and resilience of urban-rural food systems. Information on consumers' strategies was collected by means of a telephone survey of a representative sample of 600 consumers in the region of Madrid. The survey information was systematized and different strategies and consumer profiles were analysed. In the next section, we describe the background of the study and review relevant literature. Subsequently, we discuss our analysis methods (Section 3), present our results (Section 4), and discuss our findings in light of their implications for urban food systems resilience in crises in Spain and beyond. Background The impact of COVID-19 on urban food systems The COVID-19 pandemic presented a huge challenge for urban areas. The disruption it caused exposed serious weaknesses and inequalities in the global food system, leading to calls for transformative change [20]. Lockdowns, restrictions on the movement and gathering of people, and other containment measures have had economic, social, and environmental impacts. Sharifi and Reza Khavarian-Garmsir [21] reviewed early research on the impacts of COVID-19 on cities and found it to be mainly related to four major themes: (i) environmental quality, (ii) socioeconomic impacts, (iii) management and governance, and (iv) transportation and urban design. The strategic importance of the food sector was understood from the outset. To guarantee the food supply of the population was one of the main priorities of policymakers and with up to 70 % of the global food supply destined for urban consumption, urban areas were particularly vulnerable [22]. The quantity and density of the population and its dependence on the correct functioning of the entire food chain meant that no part of the supply chain (production, storage, distribution, sales) could be allowed to fail [23,24]. Some of the effects of the pandemic that were seen in the food system included: food price escalation, demand collapse due to lockdowns, decreased availability of some food products, shortage of food system inputs related to labour, and reduction or loss of income due to difficulties in commuting to work and unemployment of workers [3,25,26]. Food systems perspectives go beyond aspects of production, provision, and consumption, and recognize the interrelatedness of system components [27]. Urban food systems are complex and there are increasing concerns about their unsustainability, inequity, and lack of resilience [28]. Food systems in small towns were more resilient to shocks compared to those in larger urban areas, due to proximity to production areas and shorter supply chains [20,22]. The Food and Agriculture Organization of the United Nations (FAO) noted that “several initiatives to promote local production and short supply chains have emerged worldwide during the pandemic" [22]. For the FAO, a sustainable food system is “a food system that delivers food security and nutrition for all in such a way that the economic, social and environmental bases to generate food security and nutrition for future generations are not compromised” [29]. The potential for local food self-reliance in cities can play an important role in securing more sustainable food systems [30]. In this sense, the disruption to food systems from COVID-19 not only highlights the unsustainability of existing food systems but also the opportunities that can be captured by connecting local production and consumption. In this sense, COVID-19 provides opportunity for planners and policymakers, together with key local actors, to take transformative actions towards creating urban food systems that are more resilient and sustainable [5,21,24,[31], [32], [33]]. Resilience can be increased by developing a richer and more diversified food system, involving for example: the use of local sources of food [34]; agroecology [24,35], urban agriculture and gardening [5,34,[36], [37], [38], [39], [40], [41]], urban commons [42,43], and alternative food networks (AFN) [37,41]. AFNs can adapt to these kinds of disturbances in novel ways in a short time. Within the new regulatory landscape that emerged during the pandemic, the trust between consumers and producers has proved vital to the development of consistent responses [44]. Pulighe and Lupia [1], Bakalis et al. [4], Langemeyer [5], and Lal [36] supported the idea of fostering AFNs as a resilient measure. Carducci et al. [45] highlighted that effective, inclusive, resilient, and nutritious food systems must promote shorter and more diversified food value chains. Moreira and Fuster Morell [42] and Sardeshpande et al. [43] proposed the use of edible urban commons to alleviate the impacts of the pandemic and build more resilience in urban food systems. These approaches encompass collectively self-governed providers of resources and needs like agroecological consumer groups, community-supported agriculture (CSAs) schemes, community gardens, and other collective socio-economic structures for food provision. Consumers’ strategies for facing the challenges of the pandemic At the level of the individual consumer, the pandemic has been a source of concern for citizens all around the world. It has generated significant social and economic problems while aggravating existing ones [26]. From a food perspective, mobility constraints and lower incomes have affected access to food, purchasing habits, and perceptions of the food system [3]. At the same time, the pandemic had profound impacts on how people sourced, valued, and used food [46]. Concerns about food safety amid the global pandemic have drawn attention to the enormous infrastructure and workforce responsible for creating a safe and reliable food supply worldwide. In response to the pandemic, consumers developed various strategies, which have been highlighted in several studies (e.g. [[17], [18]]). Nonetheless, significant knowledge gaps remain, and the body of evidence on individual consumer behaviour in different parts of the world remains sparse. In this paper, we seek to improve understanding of consumer responses to the pandemic by examining in detail one of Raswami's [47] three main categories of social actors within the social subsystem of the city – individual food system users, representing households and businesses. Our research contributes to three strands of literature. Firstly, it amplifies and complements the literature about changes in consumer habits because of the pandemic. In addition, it analyses these changes from a CRFS perspective, building on the discussion of food systems on the CRFS resilience in Madrid Region. This approach is a contribution to the discussion on the vulnerabilities, strengths, and resilience of urban-rural food systems. Finally, consumer profiles are identified according to the strategies that they have been implemented to cope with the pandemic, with special emphasis on the contribution of AFNs. The City-region food system approach In large cities, achieving a transformation to a more sustainable and resilient food system implies the need for new spatial arrangements and land management approaches that seek to reestablish a role for urban and peri-urban agriculture in local food production [1]. Jennings et al. [48] introduced the City Region Food System (CRFS) approach, which is increasingly recognized as a constructive entry point to building sustainable and resilient food systems. CRFS can be defined as: “the complex network of actors, processes, and relationships that has to do with food production, processing marketing, and consumption in a given geographical region which includes a more or less concentrated urban center and its surrounding peri-urban and rural hinterland” [49]. The geographical region for application of the CFRS approach can be established by taking into account different criteria, including perceptions of local policy actors and stakeholders, jurisdictional boundaries, or flows of produce and resources between the city and its surroundings [50]. The FAO and the international consortium Global Partnership on Sustainable Urban Agriculture and Food Systems (RUAF) promote the CRFS approach and have defined some specific aims which provide a useful basis for our own study, as follows: 1) understand the vulnerabilities and strengths of urban-rural food systems; 2) enhance communication and cooperation through multistakeholder, multiscale collaboration and system-centered planning to develop coherence across the rural-urban continuum; 3) coordinate action to help safeguard livelihoods and food and nutrition security [8]. Methods Data collection A Computer Assisted Telephone Interview (CATI) survey was conducted with inhabitants of the Region of Madrid aged 18 and over (n = 600; sampling error = ±4.1 % p = q = 50 %). The CATI was carried out between May 28th and June 2nd 2021. The number of surveys was divided proportionally between the capital city, the suburbs, and the rural area (one-third each), and quotas were set for gender and age (18–39; 40–59; >65 years old) in each of the three areas. This study was conducted under the principles of the Declaration of Helsinki, and informed consent was obtained from all respondents. The questionnaire was approved by the Ethical Committee of the Polytechnic University of Madrid (written report No. 2023-11-28). Four consumers tested the questionnaire to identify potential weaknesses or problems with the wording of questions and answer options. After some corrections, the survey was run. The CATI entailed a randomized phone selection composed of 47.5 % landlines and 52.5 % mobiles. The questionnaire was structured with closed questions based on a Likert scale (0–10) and dichotomous questions (yes or no). The interviews’ average length of time was 11 min. The questionnaire used in the study is provided in Appendix 1. The characteristics of the respondents are shown in Appendix 2. Data analysis The present paper aimed to identify consumer strategies and profiles to address the food challenges of the pandemic using a range of statistical analysis techniques. The analysis comprised six steps, as follows. 1)The first step was based on descriptive analyses focused on the changes in consumers' habits. The first step was based on descriptive analyses focused on the changes in consumers' habits. 2)Secondly, a Chi-square (χ2) test was used to identify a hypothesis based on the bibliography consulted. Secondly, a Chi-square (χ2) test was used to identify a hypothesis based on the bibliography consulted. 3)Thirdly, we applied an Exponential Family Principal Component analysis (EPCA) from the R package Logistic PCA to analyse the binary response questions (1–3). EPCA was employed initially to differentiate the responses to questions 1, 2, and 3 by gender, age (under 40, over 60) income (lowest, highest), and education level (lowest, university). We followed the approach described by Landgraf [51]. Thirdly, we applied an Exponential Family Principal Component analysis (EPCA) from the R package Logistic PCA to analyse the binary response questions (1–3). EPCA was employed initially to differentiate the responses to questions 1, 2, and 3 by gender, age (under 40, over 60) income (lowest, highest), and education level (lowest, university). We followed the approach described by Landgraf [51]. 4)Next, we used logistic regression modelling, also in R, to isolate specific patterns and relationships between the 7 predictor variables (gender, over_60, under_40, lowest_education, highest_education, lowest_income, highest_income) and the preference variables. To test the reliability of the regression models, we applied a cross-validation approach. Data were divided randomly in the ratio 80:20 between training and testing with each model being built on the training data and then used to predict the test data. To ensure that model outcomes were not due to chance, we used a 100-run Monte Carlo simulation to obtain aggregate model accuracy statistics (see e.g. [52]). Next, we used logistic regression modelling, also in R, to isolate specific patterns and relationships between the 7 predictor variables (gender, over_60, under_40, lowest_education, highest_education, lowest_income, highest_income) and the preference variables. To test the reliability of the regression models, we applied a cross-validation approach. Data were divided randomly in the ratio 80:20 between training and testing with each model being built on the training data and then used to predict the test data. To ensure that model outcomes were not due to chance, we used a 100-run Monte Carlo simulation to obtain aggregate model accuracy statistics (see e.g. [52]). 5)Next, we used a multiple correspondence analysis (MCA), to discover similarities and dissimilarities between respondents. In MCA, a 2-dimensional matrix is created from individual respondents in rows and dummy variables representing categories of the variables in columns. The distance between either row points or column points reflects the (dis)similarity between either individuals or variable categories. In this way, individual survey respondents that cluster together (rows) are interpreted as representing respondents with similar profiles, while variable categories that cluster together are considered more closely associated than those further away. Next, we used a multiple correspondence analysis (MCA), to discover similarities and dissimilarities between respondents. In MCA, a 2-dimensional matrix is created from individual respondents in rows and dummy variables representing categories of the variables in columns. The distance between either row points or column points reflects the (dis)similarity between either individuals or variable categories. In this way, individual survey respondents that cluster together (rows) are interpreted as representing respondents with similar profiles, while variable categories that cluster together are considered more closely associated than those further away. 6)Finally, we applied a Principal Component Analysis (PCA) to reduce the dimensionality of the datasets, while preserving information about relationships between the variables as far as possible. PCA is a multivariate technique used to simplify the numerous and complex relations that can be found in a set of observed variables and it is typically employed in analysis of questionnaire surveys and structured interviews [53]. The statistical contrasts used to evaluate the goodness of the fit of the factorial models formulated were the mean of the KMO (Kaiser-Meyer-Olkin) measure and Bartlett's test of sphericity. PCA was used here to cluster the variables into four types of strategies to deal with the consequences of confinement. Finally, we applied a Principal Component Analysis (PCA) to reduce the dimensionality of the datasets, while preserving information about relationships between the variables as far as possible. PCA is a multivariate technique used to simplify the numerous and complex relations that can be found in a set of observed variables and it is typically employed in analysis of questionnaire surveys and structured interviews [53]. The statistical contrasts used to evaluate the goodness of the fit of the factorial models formulated were the mean of the KMO (Kaiser-Meyer-Olkin) measure and Bartlett's test of sphericity. PCA was used here to cluster the variables into four types of strategies to deal with the consequences of confinement. Depending on the analysis carried out, the number of valid responses that could be analysed was subject to variation. For example, bivariate grouping using Logistic PCA required valid responses only for the two variables considered (n = 451), while multiple correspondence analysis required valid responses for all variables analysed; hence n = 410. Care has been taken to state the number of observations in all cases. Case study area The central region of Spain, the subject of the present study, is characterized by a high concentration of population in the city of Madrid and its metropolitan belt (Fig. 1). The region of Madrid has almost 180 municipalities, with 6,779,888 inhabitants in an area of 802,000 ha. Of these, just over 209,000 ha. are used for crops. In addition, the proximity of rural areas of the region to such a huge market of about seven million people has led to the development of short-food supply chain (SFSCs) initiatives [54]. At the same time, the region is experiencing an ongoing process of urbanisation and de-agrariansation, in other words, agricultural land is disappearing. The main crops, in terms of surface and volume, are cereals, vineyards, and olive groves [55]. There are also important areas of fertile lowlands where agriculture is maintained. Like other parts of the world, the Madrid Region is characterized by a dominant globalized food system model that coexists with a growing demand for local, organic, and seasonal products. The region of Madrid, and its location in Spain and Europe. Figure: Florencia. A. Pera and Richard J Hewitt. Results Changes in consumer habits (OR1) Table 1 shows the purchasing habits of respondents regarding SFSCs and local and organic food. Consumers in the Madrid Region are characterized by their interest in local products and to a lesser extent by a more direct relationship with the farmer and the demand for organic products. Table 1Respondents' purchasing habits.Table 1How often do you buy …..NMeanStandard Deviation local products?5586.22.4 organic products?5954.42.6 directly to farmers?5983.22.50 = Never; 10 = AlwaysSource: Data collected by authors. Respondents' purchasing habits. 0 = Never; 10 = Always Source: Data collected by authors. Table 2 displays the results of the effects of the lockdown on the respondents. In the context of the pandemic, interest in buying food that is beneficial for the environment (65.3 %) and for health (75.8 %) has increased among consumers in the Madrid Region. One out of every four respondents has at some point been worried about running out of food at the market. 35.2 % have had problems finding food and 32.7 % have bought more food in the months of confinement. Table 2Effects of lockdown in the first months of the pandemic.Table 2Yes (%)No (%)Missing valuesMissing values (%)It has increased my interest in buying foods that are good for my health.75.824.200My interest in buying environmentally friendly food has increased.65.333.850.9I have bought more food in the shops in my neighborhood.49.250.800I have had trouble finding food.35.264.800My interest in buying from local or nearby food producers has increased.34.364.291.5I have bought more food in recent months.32.767.300I have been buying more from local or nearby food producers.31.260.5508.3My interest in living in rural areas has increased.29.070.810.2I have been worried at times about running out of food at the market.24.875.200I have bought more food than ever before through the Internet.24.575.500I have bought more food products from the Madrid Region.19.558.813021.7N = 600.Source: Data collected by authors. Effects of lockdown in the first months of the pandemic. N = 600. Source: Data collected by authors. Results of the χ2 test showed that concerns about access to food do not differ significantly between urban and rural areas. Changes in consumption habits do not differ significantly either. However, increased shopping in the neighborhood (χ2 = 9.68; sig. = 0.008) and internet shopping (χ2 = 13.02; sig. = 0.001) are more urban-related issues. The Exponential Family PCA model (Log SVD) for food adaptation strategies required 864 iterations to converge at a rank 4 solution, explaining 62.5 % of the deviance. This model demonstrated a clear separation in responses in terms of gender (Fig. 2A), age (Fig. 2B and C), education (Fig. 2D and G), and income (Fig. 2E and F). The strategies to deal with the challenges of the pandemic were different. A: Gender, n = 451, of which 211 are male and 240 are female. B: Under 40s, n = 162/451; C: Over 60s (1, red), n = 110/451; D: Lowest education, n = 32/451; E: Lowest income, n = 29/451; F: Highest income, n = 57/451; G: University education, n = 229/451. (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.) The over-60s group was less distinctive than the others were, and the low frequencies for lowest education, lowest income, and highest income mean that these patterns are less clear than for the other groups. These results indicated some clear differences between preferences by socio-economic group, age, and gender, around food responses to the COVID pandemic (Table 3). Table 3Statistically significant covariates from food adaptation strategies for different dependent variables (451 respondents) Significance codes: <0.05 ‘∗’, <0.01 ’∗∗’,<0.001 ‘∗∗∗’.Table 3Dependent variableIndependent variablescoefficientSEz-valuep-valueOdds ratioGender (male 0, female 1)Bought.more.food.recently−0.48∗0.22−2.170.030.62Bought.more.local.foods0.59∗0.292.070.041.80Bought.more.food.online0.52∗0.232.230.031.68Under_40 (no 0, yes 1)Problem.finding.food1.00∗∗∗0.234.400.002.71Bought.more.local.shops0.48∗0.232.120.031.61Increased.interest.buy.healthy0.91∗∗0.332.750.012.47Over_60 (no 0, yes 1)Problem.finding.food−2.00∗∗∗0.38−5.240.000.14Bought.more.food.recently−1.49∗∗∗0.35−4.240.000.23Increased.interest.buy.green0.78∗0.342.300.022.19Lowest_educational_level (no 0, yes 1)Problem.finding.food−1.57∗0.64−2.440.0152.71University_educationProblem.finding.food0.53∗0.222.390.021.70Bought.more.food.recently0.49∗0.232.180.031.64Bought.more.Madrid.products−0.69∗0.29−2.360.020.50Increased.interest.live.rural−0.58∗0.23−2.510.010.56Increased.interest.buy.healthy−0.76∗∗0.29−2.660.0070.47Lowest_income (no 0, yes 1)Bought.more.food.online−1.49∗0.76−1.960.0490.23Highest_income (no 0, yes 1)Bought.more.Madrid.products−1.01∗0.52−1.960.0490.36Source: Data collected by authors. Statistically significant covariates from food adaptation strategies for different dependent variables (451 respondents) Significance codes: <0.05 ‘∗’, <0.01 ’∗∗’,<0.001 ‘∗∗∗’. Source: Data collected by authors. For the logistic regression analysis, Table 3 shows statistically significant covariates from food adaptation strategies for different dependent variables. Fig. 3 shows results of the Monte Carlo simulations for each regression model by dependent variable. Predictions based on education and income level models had too few positive class members to allow for reliable estimates of model accuracy and are not included in Fig. 3. Prediction accuracy results for the four dependent variables: Gender, Under_40, Over_60 and University. The figure shows accuracy results for a 100-run Monte Carlo simulation in which all questionnaire responses were used to predict the dependent variable in each case. Results suggested a possible (weak) correlation between certain food behaviours and gender, with women more likely to have bought more local food, bought more food online, and have become more interested in buying green products than men, but less likely to say that they had bought more food recently. However, gender effects are marginal, and might disappear entirely if correction for Type I errors were applied (e.g. Bonferroni correction). Under 40s were more likely to admit to having had problems finding food than over 40s, and this younger age group also bought more from local shops and was more interested in buying healthy food than older ages. Over 60s were significantly less likely to have problems finding food, or to have bought more food recently, but showed an increased interest in buying green. The lowest educational levels were significantly less likely to say that they had trouble finding food than other educational levels. University-educated respondents, conversely, were significantly more likely to admit to having trouble finding food and to have bought more food recently. They were significantly less likely to say that they had become more interested in buying healthy products, had bought more Madrid products, or had become more interested in living in a rural area than other respondents had. Lowest income respondents were significantly less likely to say that they had bought more food online than other respondents had, while highest income respondents were significantly less likely to say that they had bought more Madrid products. These results should be seen in terms of the reliability of the regression models (Fig. 3) at the 0.5 threshold. Prediction accuracy for dependent variables “Gender” (median = 0.56) and “University” (median = 0.62) were lower than results for age (median: under_40 = 0.67; median: over_60 = 0.74) which implies lower confidence in the results for gender and university than for age. These accuracy results are lower than those in some published studies (see, e.g. [56], who used sociodemographic variables to predict organic food purchasing, reporting model accuracy of 0.87), but few studies report aggregate scores from rigorous testing, as we have done here. Predictions of membership of the two age groups were more accurate, though the model for respondents over 60 was the best performing in this dataset. Notice, however, that the over 60s dataset had the lowest positive class membership on average, and hence was the most class-imbalanced, something that can affect the reliability of accuracy statistics. However, in general terms, these results suggest that food behaviour choices were stronger predictors of age than of gender or university attendance. These results are supported by an examination of the odds ratios (OR) of the regression model shown in Table 3. ORs express the ratios of the probability a respondent of a particular binary group (over 60, female, low-education, high income, etc.) will respond positively to a particular variable, over the probability that they will respond negatively. OR <1 indicates a decrease in the odds of an outcome in response to a variable, while OR>1 indicates an increase in the odds of an outcome in response to the variable. OR = 2 represents a 100 % increase, or a doubling of the odds of an outcome occurring. For example, in Table 3, the OR of problem.finding.food and Increased.interest.buy.healthy for respondents under 40, is 171 % and 147 % respectively, meaning that the odds of a person from this age group agreeing that they had a problem finding food or had become more interested in buying healthy food is more than double the odds the odds of them not agreeing. OR for over 60's and increased interest buy green were also more than double (119 %), as were ORs for people with the lowest educational level and problem.finding.food, though the direction of the coefficients was reversed. Likewise the low OR for the over 60s group with respect to the variables Problem.finding.food (OR 0.14 = 86 % decrease) and Bought.more.food.recently (OR 0.23 = 77 % decrease) indicate the low likelihood of these groups agreeing with these propositions. In short, analysis of the regression model ORs suggested large differences in food behaviour choices according to age, and that the lowest educational levels were much less likely to express concerns about finding food than respondents with other educational levels. - food choices during the pandemic: local, online, and organic products Local product selection was one of the main food choices during the pandemic (Table 1). Additionally, those who responded affirmatively to one aspect of local food choices (bought more local food, interested in local foods, bought more at local shops, bought more Madrid products) also responded affirmatively to the others. Localism in food choices was an important aspect of respondents’ preferences. Conversely, those who responded negatively to one aspect of local food choices also responded negatively to the others. There is thus a clear division between people who cared about local food and people who did not. Interestingly, these preferences were not differentiated by socioeconomic group. Fig. 4A shows that food preferences were closer to each other than to socioeconomic characteristics. In Fig. 4B, we see that respondents who reported increased interest in buying healthy food also often reported increased interest in buying environmentally friendly (green) options. Respondents who reported increased interest in these two types of food were also likely to report increased interest in living in a rural area. For online food purchasing (Fig. 4C) during the pandemic (but after lockdown had ended) respondents who bought more food online tended to be under the age of 40. Respondents aged 40–60 and those over 60 did not buy more food online. MCA Factor map showing: A: Increased preferences for local food choices during the COVID pandemic according to the socioeconomic characteristics of survey respondents (n = 410). B: Increases in online food purchasing during the COVID pandemic according to the socioeconomic characteristics of survey respondents (n = 410). C; increases in interest in living rurally or buying green or healthy food options during the COVID pandemic according to the socioeconomic characteristics of survey respondents (n = 410). (For interpretation of the references to colour in this figure legend, the reader is referred to the Web version of this article.) - Four types of consumer strategies to deal with the consequences of confinement on food (OR2) After clustering all these variables with a principal component analysis (PCA), it is possible to identify four types of strategies to deal with the consequences of confinement. These four components (eigenvalue higher than 1) explained 58.9 % of the total variance (Kaiser- Meyer- Olkin = 0.72; Sig. = 0.000). These four components (Table 4) showed four types of consumer profiles. (1)Local Consumers: these were consumers who were already buying local food, organic food, and food directly from the producer. Because of the pandemic, these habits have been reinforced. They buy more in their neighborhood shops. They buy more food produced in the Madrid Region and have increased their interest in buying from local or nearby producers. Moreover, they have increased their interest in buying food that is beneficial to the environment and their health. Local Consumers: these were consumers who were already buying local food, organic food, and food directly from the producer. Because of the pandemic, these habits have been reinforced. They buy more in their neighborhood shops. They buy more food produced in the Madrid Region and have increased their interest in buying from local or nearby producers. Moreover, they have increased their interest in buying food that is beneficial to the environment and their health. (2)Pandemic-alarmed consumers: these consumers have been worried about running out of food in the supermarket. They admit to having had problems accessing some food. They have bought more food than in previous months and have bought more online. They do not tend to buy local products directly from the producer. The lockdown has increased their interest in buying food that is good for their health. Pandemic-alarmed consumers: these consumers have been worried about running out of food in the supermarket. They admit to having had problems accessing some food. They have bought more food than in previous months and have bought more online. They do not tend to buy local products directly from the producer. The lockdown has increased their interest in buying food that is good for their health. (3)Local-online consumers: these consumers have an intermediate profile between types 1 and 2. These consumers buy local products but they do not buy local products as frequently as type 1. In addition, they show some concern about food supply but not as much as profile type 2. They have bought more food than before via the internet. In addition, they are not interested in buying food that is good for your health or more environmentally friendly food. Local-online consumers: these consumers have an intermediate profile between types 1 and 2. These consumers buy local products but they do not buy local products as frequently as type 1. In addition, they show some concern about food supply but not as much as profile type 2. They have bought more food than before via the internet. In addition, they are not interested in buying food that is good for your health or more environmentally friendly food. (4)Ecological consumers: they tend to buy ecological products and their interest in consuming environmentally friendly products has increased. Their consumption habits have not changed largely. Ecological consumers: they tend to buy ecological products and their interest in consuming environmentally friendly products has increased. Their consumption habits have not changed largely. Table 4Component Matrix of Principal Components Analysis (PCA) regarding questions about the effects of lockdown in the first months of the pandemic and purchasing habits (N = 451).Table 4Component1234Have you ever been worried about running out of food at the market?0.134.5640.3100.119Have you had any problems finding some foodstuffs?−0.049.6500.300−0.001Has your food shopping increased in recent months?0.159.4830.2270.040Have you bought more food produced in the Madrid Region?.6680.0610.082−.501Have you bought more from food producers in the area where you live?.6060.0210.126−.603Have you bought more food than before via the Internet?−0.0140.398.4150.268Are you more interested in buying environmentally friendly food?.5540.374−.5310.170Are you more interested in buying food that is good for your health?0.413.566−.521−0.059How often would you say you buy local products?.673−0.3450.2950.005How often do you buy protected designation of origin?.641−0.3040.0530.271How often do you buy organic products?.616−0.106−0.141.470How often do you buy directly from the producer?.602−0.1930.2900.241Source: Data collected by authors. Component Matrix of Principal Components Analysis (PCA) regarding questions about the effects of lockdown in the first months of the pandemic and purchasing habits (N = 451). Source: Data collected by authors. Discussion Towards improved knowledge of COVID-19 impacts on food habits Four main points of discussion emerge from the results presented in the previous section. Firstly, the descriptive results contribute to a better knowledge of COVID-19 impacts on food habits. Some groups of consumers in the Madrid Region already incorporated local, organic, and SFSC products into their shopping baskets before the pandemic. The results show that the pandemic has reinforced this habit, as the interest in consuming food products beneficial to health has increased (75.8 %) along with environmentally friendly food (65.3 %). These results complement previous research about the effect of the pandemic on food habits in other countries, such as the recent study by Marty et al. [57] that suggests that the consumption of “beneficial products” in France has increased by 35 %. The same study, in line with our findings, also reports a 49 % increase in the consumption of more environmentally friendly products [57]. Restaurants were initially closed and, later, opened with a capacity limit. Confinement has made it necessary to cook more at home and most consumers have opted for healthier and environmentally friendly products in their shopping baskets. In addition, interest in buying from local or nearby food producers has increased, but to a lesser extent. In the context of the pandemic, we also see that health concerns have led consumers to prioritize health over other issues when shopping for food. In our study, one in three consumers has bought more food, compared with >17 % in a study in Russia [58]. One in three consumers also reported buying more local products, a similar percentage to that found in two recent studies in France – 29 % in a study by Aday and Seckin Aday [14] and 37 % in a study by Philippe et al. [59]. The increase in sales is not known. Research on this topic in Italy found a higher proportion – between 50 % [60] and 81 % [61] local and zero-kilometre products were purchased. One in four consumers bought more food online, well below Italy (160 %, [62]), close to France (20 %, [14]), and above Russia (12 %, [58]). One in four consumers admitted to having been concerned about access to food, up from 18 % in Denmark or Germany, down from Slovenia [27]. Secondly, from a CRFS point of view, differences between urban and rural areas are of particular interest, since it might offer a way to infer vulnerabilities, strengths, and resilience of CRFS. As Dubbeling et al. [50] have noted a city region perspective can address broad challenges affecting both urban and rural areas and their inhabitants and build more resilient, fair, and sustainable food systems and territories. However, in our case, the results are quite homogeneous across the Madrid Region, with almost no statistically significant differences in either concerns or strategies developed to cope with the pandemic between urban, periurban, and rural areas. The region is highly urbanized, with a good communications network and an economy that is strongly weighted towards the tertiary sector. Although there are rural areas and agricultural activity, their food systems do not differ much from the more urbanized areas. Supply patterns in rural areas are very similar to those in urban areas. Good communications allow for easy mobility in the region and, therefore, the relationship with the globalised food system is very similar in both rural and urban areas. From the point of view of food insecurity, which might appear to be more of an urban issue, rural areas, being so de-agrarianized, do not seem to have differential access to food. Nor does there seem to be any difference in the purchase of local or proximity products. Governance, coordination, inter-municipal collaboration, and information sharing, as in González Azcárate et al. [63] and Fei et al. [64] is a challenge to support these processes. The similarity between consumer food habits and responses to the pandemic across rural, periurban, and urban areas in the region makes the CRFS perspective especially suitable. The good quality communications network between urban and rural areas in the region allows for greater efficiency in distribution and resource flows, and hence, potentially, a good opportunity to improve food system sustainability. Furthermore, while the Madrid Region has an important agricultural tradition, there is a lot of abandoned farmland. Under a CRFS approach, this land could be recovered and put back into productive use, contributing to rural employment and enhancing food security. Thirdly, these results indicated some clear differences between preferences by socio-economic group, age and gender, education, and income. In terms of food security, the under-40s were more likely to admit to having had problems finding food than over 40s. In regression models, food behaviour choices were better predictors of age than gender or educational level. This highlights a potential gap in the current food system. Fourthly, when analysing the different food strategies that have developed because of the lockdowns, very interesting nuances emerge. Those consumers who were already involved in local and organic consumption (Profile types 1, 3, or 4) were less concerned about supply, had no problems accessing food, and did not buy more food than before the pandemic. The results suggest that those consumers (Profile type 2) who recognized that they have been more concerned about access to food, have had problems, and have stockpiled food, are those who bought fewer local products, did not have a reference producer, and opted to buy via the internet. Internet food shopping in the Madrid Region was in the process of consolidation, but with the pandemic, it has experienced significant growth. According to Béné [26], it is important to pay attention to the access dimension of people's food security: "when individuals and households [do not] have adequate resources to obtain appropriate food". The lockdowns prevented attendance at work and closed businesses; as a result, many families suffered a reduction in their incomes. All these points reinforce the fact that the proximity of food systems and the relationship with local producers contribute to providing sufficient, appropriate, and accessible food, in the face of diverse and unforeseen disruptions. The use of short food supply chains (SFSCs) and the purchase of local and organic products demonstrates the existence of food purchasing networks that are alternative to conventional channels that increased consumers’ confidence and allowed them to respond better to the challenges posed by COVID-19. As Atalan-Helicke and Abiral [44] argue, AFNs have enabled better adaptation to the pandemic. The combination of global food supply networks and SFSCs leads to greater resilience since the two systems are vulnerable to disruption in different ways. Those consumers whose relationship to the food system contained more of the elements mentioned by Worstell [6] have been more resilient. The cases of consumers who have been concerned or who have had access problems show the need to reinforce those elements that contribute to the CRFS (OR3). For instance, public authorities like regional and local governments can contribute by playing an important role in reducing the discrepancy between consumers who already make use of SFSCs consumers and consumers who are interested in doing so but have not yet “made the leap” [62]. Lastly, our results point the way to some recommendations for regional food systems more broadly. In this sense, the CRFS approach is increasingly recognized as a constructive entry point to building sustainable and resilient food systems [[8], [48]]. CRFS can be defined as: “the complex network of actors, processes, and relationships that has to do with food production, processing marketing, and consumption in a given geographical region which includes a more or less concentrated urban center and its surrounding peri-urban and rural hinterland” [49]. Limitations and future research challenges The research has some limitations, which should be highlighted here. First, the number of consumers interviewed (600) might be regarded as low, at least when compared with large-scale questionnaire surveys. However, telephone surveys are more time-consuming than questionnaires, and this number is comparable with other recently published studies [17,18,[65], [66], [67]]. Secondly, the focus of this study on the Madrid Region could be considered a weakness. Future research might explore other regions or attempt a national-scale study. Complementary research should look beyond the administrative boundaries of the Madrid Region, since the food system of the Madrid Region covers a larger area in terms of ecological footprint. Finally, further studies would be of interest to assess the extent to which the new practices adopted by our informants have continued after the pandemic. Additionally, further qualitative research would be useful to explain the reasons for the differences in preferences and strategies adopted by the different sociodemographic and socioeconomic groups in addressing these kinds of challenges. Conclusions COVID-19 changed the food consumption habits of consumers during the pandemic. It is not clear which, if any, of these changes will persist over time. However, the pandemic offers a great opportunity to rethink our food systems, as consumers have become more aware of the importance of food because of their experiences during the crisis. Those consumers who have focused more on proximity consumption have been more resilient than those without access to such networks. Buying local products, using Alternative Food Networks (AFNs), and consuming organic products have given consumers peace of mind. Those who have been most concerned by the pandemic have been those who were more strongly dependent on the agro-industrial food supply chain and who had no links with local producers. These results offer support to efforts to strengthen short food chains through an agroecological and territorial perspective, known as the city region food systems (CRFS) approach. CRFS enables the identification of flows of resources, food, and funding, among other things, to foster innovation and make the food system more resilient and sustainable. Our research contributes to three strands of literature. Firstly, it amplifies and complements the literature about changes in consumer habits resulting from the COVID-19 pandemic. In addition, consumer profiles have been identified according to the strategies that they have been implemented to cope with the pandemic, with special emphasis on the contribution of AFNs. Changes in consumers’ behaviour choices have been linked to particular sociodemographic and socioeconomic profiles through multivariate statistical analyses. Finally, these changes have been explored from a city region food systems (CRFS) perspective, building on existing discourse around food systems resilience in the Madrid Region. The research contributes to the broader societal discussion on the vulnerabilities, strengths and resilience of urban-rural food systems. CRediT authorship contribution statement José Luis Cruz: Writing – review & editing, Writing – original draft, Software, Resources, Project administration, Methodology, Investigation, Funding acquisition, Formal analysis, Conceptualization. Mario González-Azcárate: Writing – review & editing, Investigation, Formal analysis. Richard J Hewitt: Writing – review & editing, Writing – original draft, Visualization, Validation, Supervision, Software, Methodology, Investigation, Formal analysis. Verónica Hernández-Jiménez: Writing – review & editing, Investigation. Isabel Bardají: Writing – review & editing, Supervision. Data availability statement The data associated with this study has not been deposited into a publicly available repository because the data that has been used is confidential. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Acknowledgments This work was funded by the European Union EAFRD, Ministerio de Agricultura, Pesca y Alimentación (MAPA) and Comunidad de Madrid (through IMIDRA), grant PDR18-MadridKmRegion under the Madrid Rural Development Programme PDR-CM 2014-2020. The database used in this study was provided by IMIDRA. MGA is a recipient of a PhD fellowship awarded by IMIDRA (BDNS code 420371). Richard J Hewitt gratefully acknowledges support provided by a Ramón y Cajal Research Fellowship award (IMOSET project) from the State Research Agency (AEI) of the Spanish Ministry of Science and Innovation (MCIN) 10.13039/501100011033 through the “ESF Investing in your future” funding framework. Contributor Information José Luis Cruz, Email: joseluis.cruz@madrid.org. Mario González-Azcárate, Email: mario.gonzalez@madrid.org. Richard J Hewitt, Email: richard.hewitt@cchs.csic.es. Verónica Hernández-Jiménez, Email: veronica.hernandez.jimenez@upm.es. Isabel Bardají, Email: isabel.bardaji@madrid.org. Questions During these months of the COVID-19 pandemic, … YesNoDK/NA Have you ever been worried about running out of food at the market? Have you had any problems finding food? Have you bought more food in recent months? Have you bought more food in your neighborhood shops? Have you bought more food produced in the Community of Madrid? Have you increased your interest in buying from local or nearby food producers? Have you bought more from local or nearby food producers? Have you bought more food online than ever before? Have you increased your interest in buying environmentally friendly food? Have you increased your interest in buying foods that are good for your health? If 1 never 10 usually, how often do you buy … ? 1–10DK/NALocal products or products from their territoryDirectly from the producerprotected designation of origin productsOrganic products Profile of the respondents who took part in the survey for this paper Zone (of 3)Frequency%Valid %Cumulative %ValidUrban30350.550.550.5Metropolitan25642.742.793.2Rural416.86.8100.0Total600100.0100.0Sex of intervieweeFrequency%Valid %Cumulative %ValidMen28347.247.247.2Women31752.852.8100.0Total600100.0100.0Age quotaFrequency%Valid %Cumulative %Valid18–39 yrs20133.533.533.540–59 yrs23338.838.872.3over 6516627.727.7100.0Total600100.0100.0What is your highest level of completed studies?Frequency%Valid %Cumulative %ValidNo studies30.50.50.5Basic Education386.36.36.8Secondary Education569.39.316.2Final secondary educationl/Baccalaureate19332.232.248.3University28647.747.796.0PhD244.04.0100.0Total600100.0100.0Note:Data collected by authors. Note:Data collected by authors. References Associated Data Data Availability Statement The data associated with this study has not been deposited into a publicly available repository because the data that has been used is confidential.