In energy monitoring and targeting, it is very often the case that expected consumption E is related to a driving factor D by a simple straight-line formula:
E = c + m.D
where c and m are constants: c being the intercept on the vertical axis and m being the gradient of the line on the scatter diagram of consumption against driving factor. The intercept represents the fixed base-load consumption (per week or whatever interval) while the gradient tells us about the sensitivity of consumption to changes in the driving factor.
The gradient value is expressed in units energy per unit of driving factor: for example kWh per tonne of product or kWh per degree day. These numbers have some physical significance: if you analysed a glass-melting process and the gradient was 355 kWh per tonne, that actually represents how much energy is embedded in a tonne of molten glass, disregarding any standing loss from the furnace. Another example: air compressors should yield a gradient of around 0.12 kWh per cubic metre of air, again ignoring any fixed overhead consumption. The gradients of regression lines rather useful for benchmarking different installations against each other. They are inherently much more comparable than overall specific energy ratios, which include fixed consumptions that could be very different from one installation to another, notably just because they may be different sizes, but also because there may be other non-throughput-related ancillary loads on the same metered circuit.
Unfortunately you cannot compare the kWh-per-degree-day gradients of buildings, because their different sizes, ventilation rates and insulation levels all affect how much fuel they need for particular weather conditions. However, it is worth bearing in mind that knowing a building’s parameters it is possible to calculate its theoretical heat rate—how much heat would flow out for a given temperature difference inside to out, commonly measured in watts per kelvin (W/K). If you multiply the W/K value by 0.024 you get a value in kWh per degree day – which is what you would expect the gradient of its regression line to be.