Social practices, energy demand and time use data –methodological lessons from DEMAND, 17 October 2014

This workshop provided an opportunity to present, share and discuss some of the analysis techniques developed and data challenges encountered in DEMAND so far and to bring together others working on energy, mobility and Time Use data. Results and data challenges from these analyses were presented to provoke discussion on the main advantages and challenges of using Time Use data to investigate practices that underpin energy demand, the methods to be used to detect sequences of activities and how can we go from detecting practices to calculating energy demand.


MORNING CHAIR: Elizabeth Shove

10:10– 10:15               Opening up Demand

10:15 – 11:00              Who is doing what and what are we missing?

  • Introductions – who is doing what?
  • Roundtable discussion – what is missing?

11:00 – 12:30              Demand and the structure of everyday life

Four presentations in this session offered different perspectives on how Time Use data can be used to ‘look under the bonnet’ and detect traces of practices.

In ‘Measuring energy and time-use relationships’, Philipp Grunewald (Environmental Change Institute, Oxford) highlighted the need to understand household’s contribution to peak demand and presented a method of bringing together Time Use data with electricity use data. His research group have developed a gadget attached to a mobile phone and further connected to household electricity meters to collect 1 second resolution data. This is combined with event triggered time use diaries which send text message questions to participants when the energy data detects a change in usage, thus negating the need for continuous and burdensome diaries to be completed. In addition, memory jogger video cameras can be worn by householders. The devices also log participant movement which can, in turn, be used to determine co-location of householders. The data allows activities to be mapped on to energy use. The device is affordable and reliable enough to be used at scale. A future phase of the project will use the same devices to provide feedback and use the data to evaluate activity response and even further phases aspire to combine this gadget with measurement of energy service demands and other environmental factors in the home.

In ‘Defining practices as entities’, Mathieu Durand-Daubin (DEMAND, EDF R&D) outlined analysis of the French 2010 Time Use Survey (TUS) and the UK Multi-national TUS to look at the specific role of cooking in evening peak demand. Eating was characterised by various parameters in the datasets including the time it happens, the duration, where (in or outside the home), co-presence, secondary activities, satisfaction of the activity, preceding and following activities. The absence of cooking was also examined as well as its relation to trade-offs in time between different activities (more work or later working hours is associated with the absence of cooking) and competing for activities such as watching TV. Eating was found to structure the day and the analysis suggested it should be regarded as a social activity, particularly in France, due to the duration of the activity and the greater tendency to eat with guests. In the UK, cooking and eating was found to more likely be replaced by social activities. Interesting patterns were found depending on the type of dwelling with less cooking in flats and more eating out with friends the smaller the floor area.

In ‘What do people do at peak time?, Jacopo Torriti and Richard Hanna (DEMAND, Reading) focused on the interaction between household members and the fragmentation and synchronisation of activities over time and space. For this, the Trajectory Time Use dataset was used (a UK urban sample of three day diaries of 10 minute interval data including GPS traces). Even when controlling for employment, working women were found to spend the same amount of time in work as men but have more fragmented and less synchronized activity patterns, performing on average one more activity per day. Overall, synchronisation is higher in the morning than evening peaks, due to the smaller range of activities and more predictable sequence of activities which take place as respondents get ready to go to work under greater conditions of time squeeze. This is particularly true of men and respondents with children. Synchronisation matters because it generates peaks in energy demand and implies potential to manage social practices. However, the results suggest that the structuring effects linked to gender, work or the absence/ presence of children have implications for the potential to shift the timing of energy demand at the household level.

In ‘Using the National Travel Survey to explore frequency and rhythms of shopping’, Giulio Mattioli (DEMAND, Aberdeen) presented analysis of the UK National Travel Survey (a seven consecutive day, all household member travel diary) to identify genres of a specific car dependent practice – food shopping. The rhythm of food shopping is a challenging activity to study as it generally does not happen daily and yet is responsible for substantial amounts of travel and energy demand. The focus was on the frequency of activities and the identification of meaningful patterns of variation even among those households in the top quintile with respect to the total amount of food shopping they do (who are no the same people as the ‘top’ 20% of consumers overall). The challenge was to identify variables (frequency over a week, mode choice, percentage of distance accounted for by the longest trip, proportion of total household car miles and total travel time accounted for by food shopping, distance travelled) to be used in a cluster analysis. The analysis found that differentiation even among this top 20% of ‘polluters’ brought to light four distinct genres of shopping with different rhythms and, to some extent, socio-demographic characteristics of the shoppers. The question is whether this type of analysis identifies new targets or entry points for policy?

Much of the discussion after these four presentations centred around the impact of different time metrics (frequency, episodes, fragmentation, synchronisation) to the activities or practices that can be investigated and our thinking about how these relate to each other. As the presentations highlighted, activities of less than one day frequency pose a clear challenge. In addition, the notion of ‘everyday’ practices were questioned given intra-person variability from day to day, season to season due to changes in weather or other non-routine factors. Complementing time use data with other datasets such as expenditure data or data from shopping loyalty cards was seen as part of the solution.


13:30 – 15:00              Detecting sequences

In recognition that life is not lived neat episodal ‘chunks’, the next three presentations presented attempts to develop and apply methods to detect meaningful sequences of activities from time use data.

In ‘VISUAL TimePAcTS’, Katerina Vrotsou (Dept. Science and Technology, Linköping University) gave an overview of software she has developed to explore and visualise sequences in activity diary data. The software is entitled ‘visualisation of time, place, activities, technologies and socialisation’ (Visual TimePAcTS). Sequences of activities are conceptualised as individual paths in an abstract activity space. To use her terminology, ‘projects’ akin to ‘practices’ are sets of activities that are performed together to achieve a certain goal and the software detects such sequences that are performed together often. This can take place ‘a-priori’ by starting with a user-defined activity and observing iteratively how this is interleaved with other activities. Once projects or practices are identified, the variation (or similarity – but how define?) in the performance of these can be examined and cluster analysis applied to various parameters (duration, start/end times, fragmentation, occurrences, size of the gaps between activities etc).

In ‘Detecting Mobility Intensive Practices, Giulio Mattioli used the TimePAcTS sequencing techniques with the 2000 British Time Use Study to characterise activities in terms of their mobility and car intensity. This analysis set out to identify which practices are responsible for a disproportionate amount of (car) travel and which practices might be inherently more difficult to switch away from the car. The approach was highlighted as theoretically and conceptually different from traditional transport studies which tend to focus on individuals, not activities. Time use data also provides a much finer-grained delineation of activities (265 activity codes as opposed to 23 in the NTS) and allows links to be made between in-home and out-of-home activities and sequencing of activities, rather than only ‘mobile’ episodes during the day. Taking data from the internationally harmonised Multi-national Time Use Survey also allows direct country comparisons not possible with disparate national travel surveys. The analysis involved recoding of the data to stitch together ostensibly contiguous main activities and to create indices to reflect the likelihood of activities being flanked by mobility and car-based mobility. Some surprising activities emerged as mobility intensive including the household disposal of waste and walking the dog! For the latter, when it is flanked by mobility other than the walking itself, there is a very high tendency to travel by car. As this is undertaken by relatively large numbers of people, it is responsible for non-trivial amounts of mobility related energy demand. Overall, activities with high car-based mobility intensity tend to require some kind of ‘cargo’ function including shopping, certain sports and activities requiring the transportation of children or other types of caring. The flanking analytical approach can give rise to distortions, however, if other activities are slotted between the travel and the activity that is being travelled to or if travellers chain different transport modes. In addition, the sequencing of more than two consecutive activities can be challenging as the sample size of these occurrences reduces as the number of activities in the sequence increases.

In ‘Constraints and sequences over time, Ben Anderson (DEMAND, Southampton) was interested in change over time in order to learn something about the processes of change and where we might go next. He focused on laundry as something which is often identified as a potential candidate for time-shift. The Multi-national time use survey has usable data since the 1970s in roughly 10-15 year chunks and therefore has the potential to be used to examine change over time. However, challenges include the variability in the episode duration measured and inconsistencies in the coding. With the former issue, the only solution was to gross up to half-hour time slots as the lowest common multiple bringing obvious limitations for a study of an often time-fragmented activity such as laundry. For the latter, the definitions of laundry have changed between datasets, for example, sometimes including related activities such as sewing/mending. The analysis nevertheless revealed clear trends in the performance of laundry tasks with Sundays exhibiting the greatest change (increase). This reflects changes in the labour market participation of women over this time period and consequent time constraints affecting the way energy is consumed during the time spent at home. There is little change in what comes before or after laundry, however, but more fundamentally when people have the opportunity to do it. This begs the question as to whether shifting the timing of laundry will simply mean that another energy consuming activity will simply fill the gap. In conclusion, Ben emphasised that people are inherently flexible but they have constraints. Nevertheless, things change and therefore engineering the current system to meet the current state is likely to fail.

In the discussion, participants agreed that sequences are intrinsic to the concept of practices and operationalising practice theory requires being able to measure and detect how events are aligned and interdependent in this way. For instance, with respect to mobility, much policy rhetoric emphasises the need to shift travel mode for individual ‘journey purposes’ but has failed to understand activity patterns and the significance of the structure of everyday life on mobility. The purpose is to understand how activities fit into longer sequences in order to assess their flexibility. But the analysis of sequences is fraught with measurement challenges due to the underlying data structure which still forces episodic analysis and mis- or under-represents fragmented practices (is laundry 5 minutes or two hours and what is the role of the washing machine?), parallel activities whereby energy is used for different things in the same or different places at the same time and displaced energy demands due to the increasing disjuncture with battery powered appliances between where/when the energy is drawn and when/where it is used. Even where sequences ‘make sense’, the uncertainty in the analysis (accuracy and representativeness) compounds as more activities are strung together. The analysis presented arguably opens up more questions than it answers and struggles with the embeddedness of energy demand beyond the measurements of individuals, technologies and behaviour. Consequently, the question was posed as to whether the analyses of time use data to understand interleaving of activities and practices underlying energy demand are making very strong claims from weak data?

15:00 – 15:15              Tea and coffee

15:15 – 16:00              Calculating energy intensity

The two final presentations and discussion aimed to bring the focus on how we can relate time use activities to end use energy demands and implications for the energy system.

In ‘How people’s temporal behaviour and technology control drive energy service demands’, Mark Barrett (UCL Energy Institute) presented an important overview of the socio-technical energy system which illustrated how the whole chain of the household size, occupancy patterns, heating and appliance demand, building heat loss factors and how the energy is supplied including district heating and fuel synthesis is required to calculate total energy consumed. Indeed, the way energy is controlled and used in buildings is in part determined by how that energy is supplied (e.g. UK gas supply can be turned on and off easily, but in France electric heating is kept on more constantly including overnight reflecting nuclear and hydro which like a steady load at the ‘other end of the wire’) and thus future end use energy demand will be in part determined by supply. But, to go from current hourly demand load profiles to forecasting future demand requires large uncertainty in fundamental parameters such as the size of the future population (estimates for the UK range between 65 and 90 million), household formation.

In ‘Modelling and design strategies to predict activity dependent behaviours’, Darren Robinson (Environmental Physics and Design Group, Nottingham) explained how he uses time use data as input into stochastic simulation models that enable dynamic predictions of what people are doing in their homes through the course of an entire year on an hour by hour or sub-hourly basis. To provide more robust energy and design solutions and generate better load profiles for the sizing and control of energy systems, models need to account for a variety of types of behaviour including how firms and individuals invest in improvements to the performance of their buildings as well as activities, appliance use and comfort adaptations. He described Bernoulli (predicting the probability that a particular state is observed), discrete time Markow (modelling changes in state/ transitions) and continuous time random modelling (survival curves of activities) approaches. Combinations of datasets including time use data to predict if an appliance is owned and when it is switched on/off allow energy use to be predicted for particular household configurations. This is all integrated into multi-agent simulation environment to generate synthetic populations to observe patterns and randomness in behaviour. Future work will expand this to investigate water and heat usage and demand-side management.

  • Q&A/Discussion

16:00 – 17:00 What’s missing and where next?

  • Participants to discuss their ‘top 2’ issues
  • Discussion and wrap up
This entry was posted in Workshops and tagged , , . Bookmark the permalink.