Declining Nipping of Alcohol and Increasing Sipping of Coffee
Description
Dataset Description The study uses a cross-national, multilevel dataset combining firm-level, city-level, and country-level data to examine whether rising coffee consumption substitutes for declining alcohol consumption, and how this relationship is moderated by national culture. The dataset covers 112 countries and nearly 50,000 coffee firms. At the firm level, the data include coffee-producing firms with variables such as firm size, firm age, listing status, subsidiary status, World Compliance (WOCO), technological sector (SIC-based), performance, inter-sector technological distance, and city cluster density. All firm-level variables are log-transformed and standardized and serve as Level-1 predictors in the multilevel models. Firms are nested within cities (Level 2), allowing the analysis to control for shared local conditions such as infrastructure, market size, and regional consumption environments through city-level random and fixed effects. At the country level (Level 3), the dataset includes: Alcohol consumption per capita (liters of pure alcohol per person aged 15+), from the World Health Organization. Coffee consumption per capita (kilograms per person per year), from the International Coffee Organization. Economic and demographic controls, including GDP per capita and the share of the Muslim population. National culture indicators, measured using Hofstede’s four dimensions: Power Distance (PDI), Individualism (IDV), Masculinity (MAS), and Uncertainty Avoidance (UAI). The main dependent variable is alcohol consumption per capita, while coffee consumption per capita is the key independent variable. Cultural dimensions enter the models both as direct predictors and as interaction terms with coffee consumption, allowing the study to test cultural moderation of the substitution effect. The dataset is designed for multilevel mixed-effects and fixed-effects estimation, enabling robust cross-level inference that links firm-level industrial dynamics to national consumption outcomes within distinct cultural and institutional contexts.
Files
Steps to reproduce
Steps to Reproduce the Analysis Data Collection Collect alcohol consumption per capita (liters of pure alcohol, population aged 15+) from the World Health Organization (WHO). Collect coffee consumption per capita (kg per person per year) from the International Coffee Organization (ICO). Obtain national culture scores (PDI, IDV, MAS, UAI) from Hofstede’s database. Collect country-level controls, including GDP per capita and Muslim population share, from World Bank or equivalent sources. Compile firm-level data on coffee firms (≈50,000 firms) including firm size, age, listing status, subsidiary status, compliance indicators, technological sector (SIC), performance, inter-sector distance, and city location. Data Cleaning and Harmonization Match firm-level data to city and country identifiers. Ensure all national-level variables are aligned to the same reference year. Exclude observations with missing core variables (coffee, alcohol, culture). Log-transform skewed firm-level variables. Standardize all variables on a 10-point scale. Dataset Construction Structure the data hierarchically: Level 1: Firms Level 2: Cities Level 3: Countries Aggregate coffee and alcohol consumption at the national level, while retaining firm-level predictors. Baseline Model Estimation Estimate a multilevel mixed-effects model with: Dependent variable: alcohol consumption per capita Main predictor: coffee consumption per capita Controls: firm-level, city-level, and country-level variables Random intercepts for cities and countries Cultural Moderation Tests Introduce interaction terms: Coffee × PDI Coffee × IDV Coffee × MAS Coffee × UAI Estimate models sequentially to test each moderator and then jointly. Fixed-Effects Robustness Re-estimate models with sector, city, and country fixed effects. Confirm coefficient stability in sign and significance. Endogeneity and Reverse Causality Regress coffee consumption on alcohol consumption. Use predicted values as alternative regressors. Re-estimate interaction models to confirm robustness. Diagnostics Check multicollinearity using VIF. Test non-linearity using quadratic specifications. Compare model fit across specifications. Result Validation Confirm that coffee consumption negatively predicts alcohol consumption. Verify that cultural interaction terms remain statistically significant and directionally consistent across models.
Institutions
- Liaoning University