Office building daily energy use

What it is good for

• Recognise a U-shaped relationship in a scatter plot and why correlation near zero does not mean no relationship.
• Compare linear, quadratic and degree-day (piecewise linear) regression models.
• Use residual plots and residuals over time to diagnose misfit and autocorrelation.
• Include a working-day indicator and interpret it.

Methods: scatter plot, polynomial regression, piecewise regression, dummy variables, residual diagnostics, Durbin–Watson statistic, autocorrelation.

Data quality notes

• energy_kwh is null on 5 days (meter outage).
• The relationship with temperature is nonlinear (U-shaped and asymmetric: cooling costs more than heating).
• Residuals are autocorrelated because weather and building behaviour persist from day to day.

Provenance and licence

Temperature = 21 + 12·seasonal cosine (coolest mid-January, hottest late July) + AR(1) weather noise (φ = 0.7, sd 2.2 °C). Energy (kWh) = base load (2 300 on working days, 1 100 on Friday/Saturday) + 55 × heating degrees below 16 °C + 95 × cooling degrees above 22 °C (both halved on non-working days) + AR(1) noise (φ = 0.5, sd 90). Five days of meter outage have no energy reading.

Synthetic data · CC0-1.0 (synthetic data generated by Statistics Quest) · generated deterministically (seed 20241027). It does not describe real people or organisations; any resemblance is coincidental.