This data set contains estimates of the base rates of 550 food safety-relevant food handling practices in European households. The data are representative for the population of private households in the ten European countries in which the SafeConsume Household Survey was conducted (Denmark, France, Germany, Greece, Hungary, Norway, Portugal, Romania, Spain, UK).

Sampling design

In each of the ten EU and EEA countries where the survey was conducted (Denmark, France, Germany, Greece, Hungary, Norway, Portugal, Romania, Spain, UK), the population under study was defined as the private households in the country. Sampling was based on a stratified random design, with the NUTS2 statistical regions of Europe and the education level of the target respondent as stratum variables. The target sample size was 1000 households per country, with selection probability within each country proportional to stratum size.

Fieldwork

The fieldwork was conducted between December 2018 and April 2019 in ten EU and EEA countries (Denmark, France, Germany, Greece, Hungary, Norway, Portugal, Romania, Spain, United Kingdom). The target respondent in each household was the person with main or shared responsibility for food shopping in the household. The fieldwork was sub-contracted to a professional research provider (Dynata, formerly Research Now SSI). Complete responses were obtained from altogether 9996 households.

Weights

In addition to the SafeConsume Household Survey data, population data from Eurostat (2019) were used to calculate weights. These were calculated with NUTS2 region as the stratification variable and assigned an influence to each observation in each stratum that was proportional to how many households in the population stratum a household in the sample stratum represented. The weights were used in the estimation of all base rates included in the data set.

Transformations

All survey variables were normalised to the [0,1] range before the analysis. Responses to food frequency questions were transformed into the proportion of all meals consumed during a year where the meal contained the respective food item. Responses to questions with 11-point Juster probability scales as the response format were transformed into numerical probabilities. Responses to questions with time (hours, days, weeks) or temperature (C) as response formats were discretised using supervised binning. The thresholds best separating between the bins were chosen on the basis of five-fold cross-validated decision trees. The binned versions of these variables, and all other input variables with multiple categorical response options (either with a check-all-that-apply or forced-choice response format) were transformed into sets of binary features, with a value 1 assigned if the respective response option had been checked, 0 otherwise.

Treatment of missing values

In many cases, a missing value on a feature logically implies that the respective data point should have a value of zero. If, for example, a participant in the SafeConsume Household Survey had indicated that a particular food was not consumed in their household, the participant was not presented with any other questions related to that food, which automatically results in missing values on all features representing the responses to the skipped questions. However, zero consumption would also imply a zero probability that the respective food is consumed undercooked. In such cases, missing values were replaced with a value of 0.

505 Economics is on a mission to make academic economics accessible. We've developed the first monthly sub-national GDP data for EU and UK regions from January 2015 onwards.

Our GDP dataset uses luminosity as a proxy for GDP. The brighter a place, the more economic activity that place tends to have.

We produce the data using high-resolution night time satellite imagery and Artificial Intelligence.

This builds on our academic research at the London School of Economics, and we're producing the dataset in collaboration with the European Space Agency BIC UK.

We have published peer-reviewed academic articles on the usage of luminosity as an accurate proxy for GDP.

Key features:

• Granular: Data is provided at the following geographical units:
  • NUTS3 (e.g. German Districts/Kreise),
  • NUTS2 (e.g. Regions/Regierungsbezirke),
  • NUTS1 (e.g. States/Länder), and
  • NUTS0 (e.g. Germany) levels.
• Frequent: Data is provided every month from January 2015. This is more frequent than the annualised official datasets.
• Timely: Data is provided with a one month lag (i.e. the data for January 2021 was published at the end of February 2021). This is substantially quicker than the 18 month lag of official datasets.
• Accurate: Our dataset uses Deep Learning to maximise accuracy (RMSE 1.2%).

The dataset can be used by:

• Governments and policy makers - to monitor the performance of local economies, to measure the localised impact of policies, and to get a real-time indication of economic activity.
• Financial services - to get an indication of national-level GDP before official GDP statistics are released
• Engineering companies - to monitor and evaluate the localised impact of infrastructure projects
• Consultancies - to forecast the localised impact of specific projects, to retrospectively monitor and evaluate the localised impact of existing projects
• Economics firms - to create macro forecasts at the national and sub-national level, to assess the impact of policy interventions.
• Academia / Think Tanks - to conduct novel research at the local level. E.g. our dataset can be used to measure the impact of localised COVID-19 lockdowns.

We have created this dataset for all UK sub-national regions, 28 EU Countries and Switzerland.