But why? The need for a causal understanding of changes in energy use
Phil Grunewald
Abstract
Change is at the heart of the energy transition. Technologies, buildings and people - all change. But why? Some changes happen naturally, other changes are the result of deliberate or random interventions. Some result from centrally devised policies, others are personal choices. For effective and deliberate change to be achieved, a sound understanding of what causes the change can be helpful in developing the right mechanisms, and to avoid wasting time and effort on ineffective ones. The importance of ‘understanding the causes of change’ seems self-evident, but the practical implementation for studies that provide evidence of causes (rather than mere correlation) are challenging. Ambition is required. Even after the event, it can be difficult to attribute effects to causes. Was a demand reduction the result of price, policy or products? We present findings from ten years of deliberate research design intended to get a handle on ‘causes’ in changing energy demand. Inspired by the work of Judea Pearl (Pearl and Mackenzie 2018), we present an iterative three step process to understand causation. 1) Causal model 2) Observation and 3) Intervention. The causal model creates a framework for the research design and helps to systematically hypothesise about causal pathways and confounding variables. It informs what experimental design is required and what variables need to be observed. In some cases deliberate and controlled intervention are needed to test and revise the hypothesis. Examples of the successful implementation are presented for behavioural interventions (demand response), technology interventions (heat pumps) and market interventions (price elasticity). This approach is currently being scaled up as part of the UK Energy Demand Observatory and Laboratory (EDOL), which establishes statistically robust control groups to act as counterfactuals for interventions, which are tested in matched laboratories. Only with such longitudinal data at sufficient scale can causes be understood.
