Significance threshold NB: Only links above this
6.5 Value of data collection to reduce uncertainty about programme effectiveness
6.6.1 Practical implication of WDM design models
6.6.1.1 Behavioural dependencies model (Section 6.2)
Findings from the model in Section 6.2, which used the Theory of Planned Behaviour as a prior mode structure, demonstrates how Bns can be used to support analysis of household survey data. The use of an existing model added credibility and value to the findings because it allowed them to be considered within the context of historical discussions about the Theory of Planned Behaviour. Bns provide a number of techniques for analysing significance of data dependencies between drivers or constraints, and indicators to programme participation. The structural learning from responses to the questionnaire infers that uptake of household water conservation appliances in the context of Sofia conforms to the theory of planned behaviour.
Perceived behavioural control (pbc) among citizens in Sofia to adopt WSAs was identified as the chief constraint to programme participation. Financial reasons were the most commonly mentioned constraint to adopting water saving technology and financial reasons were also the single most commonly mentioned driver (subjective norm) for adopting water saving technology.
Water conservation behaviour in Sofia can be characterised as being subject to low volitional control particularly among low incomes and, therefore, introduction of WDM instruments to improve pbc are recommended. The model indicated that such instruments could increase coverage of WSAs by as much as 35%.
Use of the Theory of Planned Behaviour to structure the behavioural dependencies model demonstrates how making analogies between existing approaches in the
problem domain and the modelling approach used supports clarity of meaning of model outputs. This point is returned to in Section 6.6.2, below.
6.6.1.2 Programme participation forecasting model (Section 6.3)
The findings from the model in Section 6.3 support further application of Bns populated with household survey data for evaluating implementation conditions. The model used the ‘total market’ approach to identify neighbourhoods where coverage of WSAs is currently low and identified large variations in uptake potential between different areas. If outputs of the ‘total market’ model coincide with high water savings forecasts in the water savings forecasting model, then implementation effort can be focussed on these areas.
Looking forward, the models demonstrate how household surveys could be used to monitor ongoing programme participation rates and presented in Bns to communicate results to a wider policy audience.
Use of the total market approach to structure the programme participation forecasting model is a further example of how making analogies between existing approaches in the problem domain and the modelling approach used supports clarity of meaning of model outputs and is returned to in Section 6.6.2, below.
6.6.1.3 Single household water demand and water savings model (Section 6.4) The outputs from the models developed in Section 6.4 require further evaluation in terms of accuracy and precision. Measuring accuracy of precision of predictions in the WDM problem domain faces issues of repeatability, and these are discussed in detail in Chapter 8.
The findings from the models developed in Section 6.4 demonstrated how survey data, data from reports, and knowledge of experts can be combined using Bns to address data availability issues in forecasting water savings in household. The graphs in 0 also demonstrate how Bns can be used to evaluate household survey data sets in an iterative process to identifying missing data, and the flexibility of information types (i.e. survey data, data from reports, and knowledge of experts) that Bns support allow some issues of sampling to be addressed.
The ‘single household water demand and water savings model’ emphasizes the need for metering and the ability of Bns to combine metered data with expert knowledge
provides a potential solution where metering coverage is not 100%. It is possible to suggest that different levels of metering coverage have implications for how to go about developing the evidence-base and examples of how Bns might facilitate forecasting water savings under different metering scenarios are given below.
If the area under question is:
(a) fully-metered then metered demand data collected from participating households coupled with household survey data regarding demand variables can be used to develop the required data dependencies. The results can be compared to a control group of households with similar or identical characteristics to verify data dependencies. With careful sampling to achieve the required significance levels, this approach is likely to provide the best results.
(b) partially-metered it should be possible to included sufficient households in a pilot study to derive data dependencies between demand and water savings. Using household survey data, non-metered households with relevant demand variables (i.e.
indicators of high water saving potential) could be identified for inclusion in the programme. Advantages of this approach include reduction in implementation costs (i.e. household survey data is relatively inexpensive to collect compared to fitting meters) and it does not incur the cost of fitting data-loggers to monitor water savings in every household.
(c) unmetered the options available are either: (i) to use expert knowledge to populate the conditional probability tables for dependencies between water saving potential and demand variables and collect demand variable frequencies using household surveys or, (ii) to fit a sample of households with data loggers and collect household survey data for these household and learn data dependencies in this way.
The costs of installing data-loggers in a sufficient number of households to achieve statistical significance may make such a programme prohibitive. Finally, (iii) use a combination of (i) & (ii) (i.e. expert knowledge and data from loggers).
6.6.1.4 Model of indicators of high water saving potential (Section 6.5)
The model developed in Section 6.5 described the use of Value of Information (VOI) analysis in a test decision concerning data collection to direct implementation effort.
To understand how VOI analysis might be applied in practice to reduce implementation costs, consider the task of a plumber employed by a water company to retro-fit households with WSAs. A random selection process might involve one plumber installing ten low-flush WCs in a day with an average savings of say, 60
litres per household. Alternatively, by identifying relevant indicators of water saving potential and then selecting household based on these indicators prior to retro-fitting, the plumber could be more effective, increasing their rate of installing low-flush WCs to say, 15 households (e.g. due to closer proximity) and increase the average water savings to say 100 litres per day, thus significantly reducing the cost per m3saved.
Applicability of VOI analysis for market research have been discussed in former research by Lacava and Tull (1982) and Assmus (1977) who observed that “no other method has demonstrated an equally strong potential for analysing the returns from market research” (Assmus, 1977, p568)
Although Bayes’ theorem and VOI analysis appears to be well-suited and as a method for determining the value of new information, both Assmus (1977) and Lacava & Tull (1982) observed that it is seldom practiced. According to Lacava and Tull (1982, p383), reasons for this include data problems such as (i) difficulties of qualifying prospective gains and losses resulting from a decision and (ii) difficulties in assessing the probabilities required, and application problems including (i) unfamiliarity with how to calculate the expected value of information and (ii) the cost of using the method. If VOI analysis is to be applied to support WDM programme design the above issues need to be addressed to reduce their impacts on receptivity to this potentially useful method; this is an area for future research.