Determinants of behavioral change: combining insights from digital tools, surveys, and smart meters to understand prosumers' energy choices

Behavioral interventions like nudging are proposed as pow erful alternatives to price-based instruments in reducing en ergy consumption. Despite several promising examples, the evidence on nudging is mixed and in some cases simply un successful. Part of the knowledge gap is that research shows evidence on outcomes, but we still do not fully understand the
mechanisms. This paper explores the determinants of energy related behavior for 111 prosumer households in Germany
that were tracked over 1.5 years and treated with digital nudges in a field experiment from 2021 to 2023. Smart meter data on energy- and self-consumption are connected with five types of determinants: (i) intention as the psychological antecedent,
(ii) tool usage, i.e., whether, when, and how participants used the digital tools, (iii) techno-economic predictors, (iv) weather
conditions, (v) treatment-specific information. While the first two are directly linked to the consumption behavior and there fore the intervention, the other three are pre-determined, exog enous determinants. We exploit the unique condition of having different data types that cover both kinds of determinants. First, we analyse the determinants that are directly linked to the intervention in a descriptive analysis: the (i) intention (i) and engagement with the digital tool. We show that the tool
was used infrequently despite high stated intention, and that there is no systematic relationship between tool use and self reported intention. Second, we combine and compare all five kinds of determinants using regularization techniques (LAS SO regression) to understand what drives the outcome variables. The exploratory analysis reveals that pre-determined variables
are important predictors to both energy and self-consumption: heat pumps (in winter), photovoltaic size, and weather are con sistently among the effective predictors for changes relative to baseline consumption. By contrast, the most relevant behavio ral determinants differ across interventions and outcomes. Ad ditionally, these behavioral determinants can have unexpected
signs. In particular, we identify cases of a negative effect of app usage and intention, which we interpret in the context of the
intention-action gap. Overall, changes in consumption behavior are firstly driven by investments in energy technologies and their weather-dependent operation. Determinants directly linked to the behavioral inter vention have a secondary, partly counterintuitive impact on the consumption behavior. The paper helps to understand determi nants of behavioral mechanisms. This knowledge is critical for better intervention design and the effective use of behavioral approaches in policy portfolios.