Empirical Results
Analysis Results
Full results of the Spearman's rank correlation and OLS regression for all seven hypotheses. Results are shown both with and without IQR outlier removal. All findings should be interpreted as associative, not causal.
Overview
Spearman Correlation Coefficients — All Hypotheses
The chart below shows Spearman ρ for all hypotheses. Significance: p<0.001 · p<0.01 · p<0.05 · p<0.10 · ns = not significant.
Spearman's Correlation — Energy Prices
H0 and H1 are significant at p<0.001, indicating a moderate positive relationship between household electricity prices and data center concentration. H2 and H3 (non-household prices) are not statistically significant.
| Hypothesis | x (independent) | y (dependent) | ρ | p-value | n |
|---|---|---|---|---|---|
| H0 | DC_CAPITA | HH_E_P | 0.550 | 0.000613 | 31 |
| H1 | DC_SHARE | HH_E_P | 0.573 | 0.000397 | 29 |
| H2 | DC_CAPITA | NON-HH_E_P | 0.298 | 0.091579 | 29 |
| H3 | DC_SHARE | NON-HH_E_P | 0.299 | 0.091488 | 29 |
OLS Regression — H0 & H1
Adding GDP per capita as a control nearly doubled explanatory power for both models, suggesting economic development — not data centers alone — is a major driver of higher electricity prices.
Interpretation
H0 and H1 cannot be rejected — there is a statistically significant positive relationship between household electricity prices and data center concentration. However, when GDP per capita is controlled for, explanatory power increases substantially (R² from ~0.15 to ~0.30), indicating that more economically developed countries tend to have both more data centers and higher energy prices. Data centers are associated with higher prices, but are not necessarily the direct cause.
H2 and H3 yield no statistically significant findings — no clear relationship between non-household prices and data center concentration is observable.
Scatter Plot Explorer — Price Hypotheses
Select a hypothesis to visualise the raw data points and regression line. Red points are IQR outliers excluded from OLS regression.
Spearman's Correlation — ICT Sector
Results show high variability. H5 (ICT employment × DC per capita) is the strongest finding in the entire analysis: ρ = 0.703 at p < 0.001. H6b (ICT R&D × DC per capita) is also significant. H4 (ICT GVA) yields weak results.
| Hypothesis | x (independent) | y (dependent) | ρ | p-value | n |
|---|---|---|---|---|---|
| H4 | DC_SHARE | ICT_GVA | 0.206 | 0.267 | 31 |
| H5 | DC_CAPITA | ICT_EMP | 0.703 | 0.000020 | 29 |
| H6 | DC_SHARE | ICT_RnD | 0.192 | 0.318 | 29 |
| H4b | DC_CAPITA | ICT_GVA | 0.340 | 0.061 | 31 |
| H5b | DC_SHARE | ICT_EMP | 0.364 | 0.052 | 29 |
| H6b | DC_CAPITA | ICT_RnD | 0.559 | 0.00162 | 29 |
OLS Regression — H5 & H6b
For H5, adding GDP barely changes R² (+0.008), meaning DC per capita independently explains most of the ICT employment variation. For H6b, GDP adds more (+0.123), suggesting wealthier countries invest more in R&D regardless of data center density.
Scatter Plot Explorer — ICT Hypotheses
Select a hypothesis to visualise the raw data points and regression line. Red points are IQR outliers excluded from OLS regression.
Interpretation
H5 is the strongest finding of the entire analysis: countries with more data centers per capita also have a significantly higher ICT employment share (ρ = 0.703, p < 0.001). Data center infrastructure appears to act as a broader economic anchor for the ICT sector, even though individual facilities require few direct employees. The GDP control barely changes R² (0.447 → 0.455), confirming this relationship is largely independent of wealth.
H6b shows a significant positive correlation between DC density and ICT R&D spending (ρ = 0.559, p = 0.00162). However, adding GDP increases R² by +0.123, suggesting wealthier countries invest more in R&D regardless — so data center density and national wealth both play a role. H4 (GVA) shows only weak, borderline results.
Outlier Analysis
Key Outlier Countries
The IQR method flagged several countries as outliers due to extreme data center concentration. They were excluded from OLS regression but retained in Spearman correlation (which is rank-based and robust to extremes).
🇮🇪 Ireland
Data centers consume an estimated 18–22% of national electricity — the highest share in Europe. 95 of 127 data centers are in Dublin. Removed from OLS as a DC_SHARE outlier.
🇳🇱 Netherlands
Data centers account for approximately 5.2% of national energy demand, driven by the Amsterdam internet exchange hub. Outlier in DC_SHARE analysis.
🇩🇰 Denmark
Data centers account for approximately 4.5% of national demand — above the European median. Flagged by the IQR method in energy share analysis.
🇱🇺 Luxembourg & 🇱🇮 Liechtenstein
Very small populations produce extreme DC per capita values even with few facilities. Both removed from DC_CAPITA regression models.
Data Sources
Download the Data
All datasets and scripts used in this analysis are available below. The archive includes raw CSVs for energy prices, ICT variables, GDP per capita, and data center counts by country, as well as the cleaned datasets used in regression analysis.
Primary sources: Eurostat (nrg_pc_204, nama_10_pc, isoc datasets),
Data Center Map, prior studies on European data center energy shares.
All data covers 2014–2023 unless otherwise noted. Missing values filled via linear interpolation
or manual country-level research.