Our main hypothesis is that a personalised behavioural feedback loop is more effective in inducing demand-reducing behaviour change than the consumption feedback loop, so the main independent variable is feedback type, in three categories: a control group receiving only household consumption feedback equivalent to that produced by Smart Meters, an experimental group receiving individual household consumption and behavioural feedback, and another experimental group receiving additionally social feedback on household consumption and behaviour. Social feedback will include, for example, normative comparisons with people in similar households. The control group will be recruited with the same offer and receive the treatment as the experimental groups. Thus the effects of repeated social interaction with the research team are controlled for, allowing us to rule out ``Hawthorne'' effects of social interaction per se. In order to simplify and standardise the feedback interface (a dedicated tablet computer, supported by a personalised website), we will select households that are broadband-equipped.

Our design uses 576 households divided into the three feedback types (192 in each category), as shown below. 

Beyond testing the main hypothesis with this design, we will evaluate the feedback loop's differential effectiveness in relation to the independent variables of household income (above and below UK median income) and household composition (single person; two or more adults; family of one or two parents and one or more children)-a 2x3 within-feedback-category design as shown below with 32 in each sub-cell. 

Household income and composition are expected to be a significant intervening variable in the impact of personalised behavioural feedback on energy demand. The size of the data set will also enable us to model, and differentiate, the interacting effects of social class, education and age; building type and existing energy-efficiency measures may also impact the effect of behavioural feedback. We will evaluate the effects of these additional variables in the Pilot Phase to inform recruitment and refine the study design, and will model these effects in the full data set.

The design will include households from demographics of significant social and environmental interest: 1) affluent/technically able, with the potential for high carbon savings, 2) low-income/fuel-poor, with the potential for significant social benefits as well as some carbon savings. The outcome will be a set of conclusions useful for policy setting, for commercial offerings targeted at different market segments and households, and for further studies.

Quantitative Measurement will determine:

1. Baseline and final energy consumption and user profiles for each household.
   
2. The impact of repeated visits, surveys and standard consumption feedback measures.
   
3. The impact of sensor-derived, personalised behavioural feedback during the course of the study.
   

These will enable quantitative assessment of:

1. Differential effects of social/normative and individual household feedback.
   
2. Differential impact of feedback due to household income, education, ages, composition, building type, and energy performance (regression).
   
3. The extent to which reductions in energy demand are sustained over time in different households.
   

Qualitative Measurement will examine why and how different forms of personalised behavioural feedback influence energy-related behavior, and explore the impact of household composition and dynamics. Six-monthly surveys (online and/or via the dedicated feedback tablet) over the course of 3 years will collect data on changes in attitudes to energy use and cost, carbon footprint of energy, the presence of the sensors, and responses to personalised feedback. When the sensors have been in place for 12 months, the attitudinal data will be used to identify a subset of around 50 households-selected for contrasting attitudes to feedback, and household composition-to participate in small scale evaluation of feedback presentations using semi-structured interviews. Household composition is important because responses to feedback will be affected by the dynamics of multi-person households. The qualitative dataset will provide insight into the complexity of interacting factors which modulate energy demand in households, and allow us progressively to optimise the design of feedback for different household types.