Population-scale dietary interests during the COVID-19 pandemic

The study investigates how dietary interests shifted during COVID-19-induced mobility restrictions in 2020. Given widespread non-pharmaceutical interventions (e.g., lockdowns, school and business closures, travel restrictions), the pandemic substantially changed daily life, raising concerns about health, well-being, and nutrition. Diet quality was suspected to worsen during confinement, with potential implications for cardiovascular risk, mental health, and vulnerable populations already affected by malnutrition, eating disorders, addictions, or obesity. While early evidence suggested increased cooking and baking, decreased eating out, and overall higher food interest, most prior work relied on surveys and lacked large-scale, passively collected data. The authors therefore pose the guiding question: How did dietary interests shift during COVID-19-related decreases in mobility in 2020, and how long did these shifts persist as mobility reverted toward normal? Leveraging global internet access and Google search behavior, the study aims to quantify population-scale changes in interest in specific foods and ways of accessing food across diverse countries.

Prior reports and studies during early COVID-19 confinement indicated increased interest in baking and cooking from scratch, reduced eating out, and altered health behaviors, including increased sedentary time and reduced physical activity. Survey-based research across various countries generally found more unhealthy meal patterns during lockdown, increased carbohydrate consumption (e.g., pizza, bread, cakes, pastries), more frequent snacking, and mixed findings on alcohol, with some studies noting decreased intake. Economic literature highlighted changing consumer needs, market readjustments, and supply chain issues affecting food access and security. Digital trace studies prior to and during COVID-19 demonstrated the utility of web search, purchase logs, online recipes, review platforms, social media, and mobility data for sensing population behaviors. However, comprehensive, global, passively collected measures of dietary interests across food categories remained limited, motivating the present study.

Data sources and scope: The authors used Google Trends to collect calibrated, weekly time series (via Google Trends Anchor Bank, G-TAB) for 1,432 specific food-related entities (e.g., bread, pizza) grouped into 28 categories, and 16 entities describing ways of accessing food (e.g., recipe, restaurant) grouped into four modes (prepared within household/third party × consumed at home/outside). Data span January 1, 2019 to December 31, 2020 for 18 countries selected for geographic diversity, large internet user bases, and varying lockdown severities: Brazil, Canada, Mexico, United States, France, Germany, Italy, Spain, United Kingdom, India, Indonesia, Japan, Egypt, Kenya, Nigeria, Australia, Sweden, and Denmark. Entity specification used language-independent Freebase identifiers to aggregate concept-related queries across languages. Calibration via G-TAB enabled aggregation and ratio comparisons across entities.

Food taxonomy construction: Food entities were sourced from Freebase types (food/dish/beverage/ingredient) and enriched with Wikidata properties (instance of/subclass of) to derive a 28-category taxonomy. Popular entities lacking Wikidata mapping were manually annotated to reach 95.7% coverage of global food search volume in 2019–2020. An epidemiologist specializing in nutrition refined the categorization.

Mobility data and changepoint detection: Country-level mobility data (Google COVID-19 Community Mobility Reports) provided daily percentage change in time spent in residential areas relative to a pre-pandemic baseline (early 2020). Weekly rolling averages were used to detect changepoints. The start of decreased mobility (first wave) and, where applicable, the second wave onset were defined as the first date when time at home increased by ≥10% versus baseline; the mobility increase (relaxation) was also identified. These changepoints were used as temporal cutoffs for modeling.

Modeling approach (quasi-experimental regression discontinuity design, RDD): To isolate the short-term effect of the abrupt mobility decrease on search interest while accounting for seasonality and trends, the authors fitted, per country and per entity/group, a quadratic RDD model on log-transformed calibrated interest volumes for weeks t ∈ [−10, 30] relative to the 2020 discontinuity, with a corresponding “fake” cutoff in 2019 to control for seasonal patterns. The model allowed different quadratic time trends before/after the cutoff and interacted year and post-cutoff indicators to estimate the discontinuity effect. Key quantities measured: (1) short-term effect (coefficient α, back-transformed via exp(α) to a multiplicative change); (2) time to return to normal (weeks post-cutoff until 2020 modeled interest is no longer significantly different from 2019 counterfactual using non-overlapping 95% CIs, capped at 30 weeks); and (3) long-term elevation (if not back to normal by 30 weeks, the relative elevation at week 30 vs. the same 2019 week). Correlations: Spearman rank correlations between seasonality-adjusted 2020 food interests (relative to 2019 corresponding weeks) and mobility were computed per country across 46 weeks of 2020 (Feb–Dec) for food categories and access modes. Associations between lockdown severity (peak increase in residential time; country-specific) and short-term effects were analyzed using Pearson correlations.

Software: Analyses were conducted in Python 3.8 using standard scientific libraries (matplotlib, pandas, NumPy, scikit-learn, scipy, statsmodels). Data and code are publicly available.

• Global directional shifts: Across 129 countries with sufficient data, 2020 vs. 2019 showed a global increase in interest in recipes and a global decrease in restaurant interest.
• Overall surge magnitude vs. holidays: In the US, the peak surplus of food interest during the first wave matched that of Christmas and Thanksgiving 2019; cumulatively, the surplus of food interest in the first six months of 2020 was 9.0× that of Christmas week and 8.9× that of Thanksgiving week. Across countries with prominent Christmas seasonality, the peak surplus during the first wave averaged 1.9× the Christmas 2019 surplus, and the total surplus over the first six months averaged 18.8× the Christmas 2019 surplus.
• Access mode shifts: During the first wave, interest in food prepared within the household and consumed at home (recipes, cooking, baking, grocery stores, supermarkets) increased markedly, while interest in food prepared by third parties and consumed outside (restaurants, cafeterias, cafes, diners, food festivals) decreased. The peak surplus for at-home preparation/consumption was on average 1.7× the Christmas 2019 surplus; cumulative surplus in the first six months was 13.7× Christmas 2019. In several countries (US, Brazil, Japan, Denmark), interest in food prepared by third parties and consumed at home (delivery, take-out, drive-in) increased by over 30 pre-pandemic standard deviations at peak.
• Correlations with mobility: Interest in recipes was positively correlated with increased time at home in nearly all countries (Spearman ρ ranging from 0.36 in Egypt to 0.95 in Mexico; p<0.05 except Japan). Takeout interest was significantly positively correlated in 13/18 countries (ρ 0.34–0.82). Restaurant interest was significantly negatively correlated in all countries (ρ −0.40 in Kenya to −0.97 in Italy). Certain food categories showed strong positive correlations with staying at home, notably desserts (ρ 0.40–0.84) and pastries/bakery products (ρ 0.46–0.88) in most countries.
• Short-term causal-proximate effects (RDD estimates): In 15/18 countries, interest in at-home, within-household food preparation/consumption had significant short-term increases, ranging from +32.2% (Egypt) to +179.8% (India). In 14 countries, interest in out-of-home, third-party prepared food (restaurants, etc.) significantly decreased, from −32.1% (USA) to −81.7% (France). Interest in third-party prepared, at-home consumption (delivery/takeout) rose by >+100% in 8/18 countries.
• Dose-response with lockdown severity: The short-term increase in at-home preparation/consumption interest scaled positively with lockdown severity (Pearson R=0.713, p<0.001). Negative associations were observed for prepared-within-household/consumed outside (R=−0.702, p=0.004) and prepared-by-third-party/consumed outside (R=−0.56, p=0.016). Some outliers (UK, Kenya, Nigeria) potentially related to policies on outdoor congregation; excluding them strengthened negative associations.
• Food category specifics: Significant overall increases in total food interest were observed in all countries except Sweden and Japan (+24.6% in Denmark to +99.4% in Spain). The largest category increases were in calorie-dense, carbohydrate-based foods: pastries and bakery products, bread and flatbread, and pies, with effects >+200% in several cases (e.g., pastries/bakery in Spain, France, Canada, Egypt; bread/flatbread in Spain, France, Italy; pie in Spain). Smaller increases occurred for fresh produce (vegetables, fruit, salad, herbs), meats/fish, and alcoholic beverages, with country variation.
• Relative share analysis: Even after controlling for overall increased food interest, share of interest increased notably for pastries/bakery (share +>50% in 11/18 countries) and bread/flatbread (+>50% in 6/18 countries), while shares decreased for soft drinks, alcoholic drinks, and sandwiches—items often consumed in social, out-of-home contexts.
• Duration and persistence: Many shifts lasted months. Minimum duration for at-home preparation/consumption category was 6 weeks (Egypt); for specific categories, 9 weeks (e.g., wine/beer/liquor in France; pastries/bakery in Denmark; sandwiches in Sweden). Most interests returned to baseline within 30 weeks, but some remained elevated at week 30: third-party prepared/consumed at home (delivery) in Italy, Canada, USA, Australia, Denmark, Japan; at-home preparation/consumption (recipes) in Spain, UK, USA, Australia.
• Second wave (Oct–Dec 2020): Effects on dietary interests were much smaller than in the first wave; no significant increases in at-home preparation/consumption; notable exceptions were increased delivery interest in France and Italy; restaurant interest decreased in the UK, Italy, France but less than in the first wave.
• Alcohol-related interests: No population-level surges; in some countries, seasonality-adjusted interest in alcoholic drinks negatively correlated with time spent at home (e.g., cocktails: ρ −0.44 Italy; wine/beer/liquor: ρ −0.61 France), and relative share of alcohol interest decreased.

The study addresses its central question by linking abrupt mobility decreases during COVID-19 to large, immediate, and category-specific shifts in dietary interests captured via Google searches. The strong temporal coincidence with mobility changepoints, significant RDD-estimated discontinuities, and dose-response relationships with lockdown severity support the interpretation that confinement drove increased interest in at-home food preparation and decreased interest in out-of-home consumption. Importantly, the interest increases were not uniform: calorie-dense, carbohydrate-rich foods (pastries, bread, pies) saw the most pronounced surges, while fresh produce and protein-rich meals did not rise proportionally. These patterns, sustained over weeks to months, suggest potential nutritional risks during confinement, particularly given reduced physical activity and possible stress- and boredom-related emotional eating. Conversely, greater interest in home cooking may offer public-health opportunities, since frequent away-from-home eating is associated with higher mortality risk; understanding how to steer increased home cooking toward healthier options could be beneficial. The findings broadly align with survey-based literature but add high-resolution, population-scale temporal insights free from self-report bias. They also inform market dynamics and food systems by documenting increased consumer interest in food delivery and baking staples, with potential implications for supply chains and food security. The lack of large increases in alcohol-related interest counters some early concerns, though subgroup risks may persist. Overall, the results highlight both risks (shift toward calorie-dense foods) and opportunities (sustained home-cooking interest) for nutrition policy and interventions during and beyond pandemic confinement.

This work provides a global, population-scale characterization of dietary interest shifts during COVID-19 mobility restrictions using calibrated Google search time series across 18 countries. Key contributions include: (1) documenting that the first-wave surge in overall food interest exceeded and outlasted typical holiday spikes; (2) demonstrating a pronounced shift toward at-home food preparation and away from out-of-home consumption, strongly associated with increased time at home and lockdown severity; (3) revealing disproportionate increases in interest for calorie-dense, carbohydrate-based foods (pastries, bread, pies); (4) quantifying durations and partial persistence of interest changes; and (5) showing diminished effects during the second wave. Future research should: (a) assess whether observed interest shifts translate into long-term consumption changes and health outcomes; (b) investigate heterogeneity across demographics and vulnerable subpopulations; (c) integrate other passively collected data (purchase logs, recipes, delivery transactions) for triangulation; (d) design and evaluate interventions that leverage increased home-cooking interest to promote healthier eating; and (e) compare COVID-19-induced stay-at-home effects with other events (extreme weather, pollution) to generalize insights for preparedness and public health.

• Search interest as proxy: Searches reflect interest, not necessarily consumption; behaviors may differ by country due to internet access, culture, or search habits. Nonetheless, prior work supports validity of digital nutrition proxies, and large interest shifts can influence consumption.
• Causal inference: While the RDD design leverages abrupt mobility shocks and controls for seasonality, the study does not claim strict causality; unobserved confounders may remain. Dose-response patterns bolster the interpretation but do not eliminate all alternative explanations.
• Coverage and representativeness: Only 18 countries were fully analyzed due to data collection constraints; results may not generalize globally or to subnational levels.
• Category construction and entity coverage: Taxonomy relies on Wikidata mappings and manual annotations; misclassification is possible. Despite 95.7% coverage of global food search volume, residual entities may be underrepresented.
• Multiple comparisons: Numerous correlations and effects were tested without adjustment for multiple comparisons, potentially increasing Type I error rates.
• Mobility measure: Residential time from Google mobility reports is an aggregate proxy and may not capture all aspects of confinement or policy heterogeneity within countries.
• Persistence assessment window: Duration analyses were capped at 30 weeks post-cutoff; longer-term persistence beyond 2020 could not be fully evaluated.