The determinants of household water consumption: A review and assessment framework for research and practice

Overview of paper search outcome

A general overview of the 231 general water consumption-related and 48 framework analysis scientific publications reviewed in this study (Fig. 3) shows that the number of papers published per year from the general water consumption-related set has been increasing, particularly since the early 2000s. Peaks of more than 10 papers per year in this category emerge since 2011, with a maximum peak of 34 papers recorded in 2018. This increasing trend in time can be attributed to the increasing development of smart metering studies, which have been increasingly allowing detailed household water demand/consumption and behavioral analysis^20,47. As a selected subset of the general water consumption-related papers set, the number of framework analysis papers has also increased in the last decade, compared to the ’80s and ’90s, constituting up to 5 papers per year. The final set of papers includes small-case studies comprising only a few units (11 individual households are considered as a minimum in⁴⁸), as well as large-scale studies comprising several thousands of households (e.g., more than 8000 individual households are considered in⁴⁹), or entire communities/towns⁵⁰.

Figure 4 shows the locations of the studies from the framework analysis set, with larger blue dots indicating more studies. The geographical distribution of the reviewed studies indicates that the interest in the determinants of water use is worldwide. Prominent interest emerges in particular areas, such as the US west coast, the east coast of Australia, and the Mediterranean area in Europe, perhaps reflecting the combination of areas more prone to drought and/or having the higher economic capacity to undertake water use related research.

Determinant representation by class

From the analysis of the 48 framework analysis papers, we identified a range of heterogeneous determinants and quantified different combinations of determinant classes, namely observable, latent, and external (see Determinant classification). Figure 5 shows an overview of the representation for the different classes of determinants over the 48 analyzed papers. Observable determinants were the most popular (47 total studies, i.e., 98% representation), with latent and external having lower representation of 52% and 56%, respectively. The values represented in the figure confirm our hypotheses that observable determinants are more common in literature than latent determinants, due to their availability in public databases, either at the household level or at coarser spatial sub-urban scales (e.g., census data collected at the block group-level, such as those used in⁴⁹). The slightly higher representation of external determinants, compared to latent determinants, is also as expected due to the widespread availability of weather records (e.g., temperature, rainfall) from national or international environmental agencies. While there is no full consensus in the literature on the effect of weather or price variables on water consumption^51,52,53, the high degree of representation of external determinants demonstrates that they are considered in more than half of the studies.

It is worth observing that multiple classes of determinants are simultaneously analyzed in most of the reviewed studies, with fewer than 20% analyzing observable variables alone. Further, almost every time external or latent variables are considered, they appear in combination with observable variables. Only one study specifically focused on analyzing the motivations for using and conserving water based on only latent determinants⁴², while no studies exclusively considered external variables. In contrast, nearly 30% of the studies included both observable and external variables, approximately 23% of the studies considered latent and external variables simultaneously, and 27% of the studies included all three types of determinant classes.

The high representation of observable determinants (Fig. 5) suggests that observable variables are widespread in the literature on modelling and forecasting of household water consumption. The prevalence of this class of determinants seems also to confirm the findings from previous studies, which demonstrated that meteorological variables have a greater influence on medium-term prediction and urban/suburban scales, but socio-demographics become more relevant when household-scale and short-term water demand models are developed^54,55.

Individual determinant representation, impact, and effort

To facilitate interpretation of the numerical values we obtained for the three determinant assessment criteria (i.e., representation, impact, and effort) we defined some regions of interest for each criterion based on thresholds (see the regions labelled as low/high/very high in Fig. 6). We selected the threshold values used to delimit the above regions of interest based on visual inspection of the empirical distribution of the representation, impact, and effort values. This simplification is carried out to facilitate the inference of general qualitative conclusions, while accounting for the low number of papers and, at the same time, high number of determinants. As a result, representation values above/below 30% are considered high/low. Impact values below 75% are considered low, values between 75% and 90% are considered high, and values above 90% are considered very high. Effort rate values above/below 8 are considered high/low.

From the resulting data visualized in Fig. 6, we can infer the following insights about determinant representation, impact, and effort. First, the determinants with the highest representation (top plot in Fig. 6) were household income (> 70%), family size (> 60%), and age (> 45%). As already suggested by the outcomes of class representation (Fig. 5), all the above determinants with high representation are observable. One exception is the awareness determinant, which is the only non-observable determinant we found with high representation. The majority of the other determinants had a representation rate of 10% to 30%.

Second, the number of determinants with a high or very high impact (middle plot in Fig. 6) is much larger than the number of determinants with high representation. It must be noted that a high impact does not necessarily mean that a determinant was found to have a high influence on water consumption, but rather that it was found to have some influence on water consumption in many publications. Interestingly, some determinants from all classes achieve high or very high levels of impact. Observable determinants with very high impact include socio-demographic information (number of occupants), house characteristics (house age, value), and outdoor characteristics (garden size, and presence of rainwater tanks). While these latter attributes related to gardens ranked among those with the highest impact, garden composition was found to have one of the lowest impact rates across the analyzed studies. Also, the observable determinants related to the education level of occupants was found to have low impact. A latent variable that emerges as very important (GARD_C) is also related to garden characteristics, but, rather than representing any physical variable, it accounts for the psychological value given by occupants’ attitudes and habits towards gardening. Finally, all external variables were found to have high or very high impact, with rainfall and water price emerging as the two with impact above 90%.

The bottom plot of Fig. 6 shows that there was a wide variability in the effort rate for each individual determinant. Data on most of the observable determinants can be generally gathered with low effort, but some (e.g., appliance inventory and irrigation system) require house visits, and thus require high effort. In turn, all latent variables display an effort rate higher than 6, and three out of four are classified as high-effort. Conversely, data on all external determinants can be retrieved with low effort, as they are usually available from national agencies (weather data) and water utilities (water price). Obtaining information on higher effort determinants likely requires getting in contact with individual householders, via phone/online surveys, or house visits.

Overall, the results reported in Fig. 6 suggest that there are trade-offs between representation, impact, and effort. In the next section, we perform a joint analysis of the three criteria and their trade-off to infer the implications of the outcomes of this study for researchers and practitioners.

Trade-off analysis and implications for researchers and practitioners

Figure 7 shows the interaction between the representation, impact, and effort criteria. The distribution of blue and orange points in the figure demonstrates that there are different trade-offs among the three criteria. Each trade-off can have a different set of implications to derive recommendations for researchers and also practitioners. We identified the three groups of determinants marked with (A), (B), and (C) to illustrate the different needs of research and practice. Group A is characterized by high impact, high representation, and low effort. Determinants in this group include household family size, occupants’ age, and occupants’ income. This group of well-studied determinants with proven impact might be particularly interesting for practitioners aiming to gather knowledge on household water consumption with budget constraints. Group (B), which includes, among others, information on the household irrigation system, appliance efficiency, and occupant gender, is characterized by medium-to-high impact, but low representation, and a range of low to high effort. While this group might not be very appealing for practitioners due to low representation, researchers might be interested in focusing on these determinants to increase their representation and, thus, validate or contrast the limited findings on these determinants that appear in the literature. Finally, Group C refers to determinants with low representation and, compared to those in groups A and B, lower impact. As they also might require high gathering efforts, these determinants should be treated with caution until more research is performed to prove their potential impact on a larger sample of studies.

Accounting for similar trade-offs across the entire sample of determinants that we have identified from the review of the literature enables determinant-specific recommendations to be derived for practitioners and researchers. In the last step of this review and determinant classification effort we thus develop a trade-off analysis framework that considers different combinations of representation, effort, and impact to formulate such recommendations. In keeping with the goal of this study, our trade-off analysis aims at identifying groups of determinants that have proven cost-effective impact via extensive research and, thus, can be recommended for use in practice, compared with groups of determinants that require more research to address open questions related to representation, impact, and effort. The proposed trade-off analysis framework includes four main recommendation categories:

U

In this category, we include determinants characterized by high representation, high/very high impact, and low effort. We consider these determinants as determinants that practitioners can “definitely use” (U), as they have been extensively researched and have been shown to have an impact in most cases, while being affordable. For the same reasons, higher levels of research priority should be devoted to less explored determinants, while these can serve as references. The determinants included in box (A) in Fig. 7 belong to this group.

IR-UC

In this category, we classify those determinants characterized by low representation, high/very high impact, and low effort. Given their promising, but not extensively proven, impact, and overall affordability, further research on these determinants should be prioritized to increase their representation (IR). We consider these determinants as determinants that practitioners can “use with caution” (UC), as they have not been extensively researched, but at the same time might have high impact at low-cost. The determinants included in box (B) in Fig. 7 and classified as low effort (blue color) belong to this group.

LE/IR-UC

In this category, we include determinants characterized by generally low representation, high/very high impact, and high effort. Similarly to the previous category, we believe that practitioners can use these determinants “with caution” (UC), as they have not been extensively researched and require high effort for data collection, but at the same time might have high/very high impact. Given their promising, but not extensively proven, impact, and high cost, further research on these determinants should be prioritized, primarily to lower the effort (LE) needed to collect them and, thus, facilitate their consideration in more studies (increased representation – IR). The determinants included in box (B) in Fig. 7 and classified as high effort (orange color) belong to this group.

IR/LE/AI-NP

In this category, we include determinants characterized by low representation, low impact, and mainly high effort. Given the limited knowledge on these determinants, we suggest that these determinants are “not prioritized” (NP) for use by practitioners unless further research demonstrates that the effort required to collect these determinants is worth the benefit of considering them. Further research should then aim at increasing their representation (IR), lowering the effort needed to obtain data on these determinants (LE), and further assessing their impact (AI) to acquire better knowledge on their actual value. The determinants included in box (C) in Fig. 7 belong to this group.

Summary information on the above categories is reported in Table 5. Based on the proposed trade-off analysis framework and the threshold values defined in Fig. 6, we associated each of the different determinants identified in the framework analysis papers with a level of recommendation (see Fig. 8). Some relevant insights for researchers and practitioners emerge. First, only observable determinants are classified as “U”. At present, there are some socio-demographic determinants (i.e., number of occupants, income level, and occupant age) that can be reliably used by practitioners in most cases to model household water consumption and can be easily and affordably retrieved.

Second, all external variables (i.e., average rainfall, temperature, and water price) are classified as IR-UC. Consequently, they have a proven impact, but have been used sporadically in connection with household water consumption (while they have been used more often at larger, urban scales), thus results might be case-specific and further research is needed to assess their impact on a larger number of studies.

Third, a mix of observable and latent external variables deserves further research to lower effort (e.g., by improving technology/data gathering practices or identifying lower-effort proxies for the same type of information) and increase representation. These variables are either observable determinants, the collection of which requires significant effort and house visits/calls to occupants (e.g., to build an inventory of appliance efficiency or storing information on irrigation systems), or latent variables the impact of which is still not proven due to low representation. The increasing availability of high-resolution metering and behavioral studies fostered by smart metering development is likely to contribute more knowledge on these determinants and more complete guidelines for use by practitioners in the coming years^7,17,20.

Fourth, we would like to stress that the recommendation “Do not favor adoption until further research” for the determinants classified as IR/LE/AI-NP does not mean that they should not be considered in future applications or no research should be done on them. Conversely, we recognize that many existing studies are based on limited data or data with coarser spatio-temporal resolutions, thus conclusive statements on the impact of such determinants would require further validation. Since large uncertainty about their impact remains, more studies are actually needed to increase the representation of these determinants and increase the statistical significance and generality of their impact assessment. Joint research that also includes other determinants with higher levels of representation could be beneficial to discover more information on the determinants in this group and better understand whether practitioners should eventually include one/more of these determinants in their analysis. Further research could be also developed to assess the degree to which these determinants are correlated with others, and hence redundant, and to which extent these and other determinants can relate to particular characteristics of water consumption (e.g., demand peaks, end use components).

Finally, some of the determinants that we recommend to use with caution (UC) in practice, and that should be prioritized for research, might become determinants to definitely use (U) in the future. Two limitations currently prevent us to recommend “definitely use” for these determinants, i.e., generally low representation and high effort for data collection. Low representation indicates that the determinant has not been well-studied in the literature. Hence it might not be generalizable to a wide range of locations. To address the disadvantages of low representation, the following is recommended for practitioners:

• Check the literature and if there are studies with similar context (location/climate/application) to the practitioners’ required application, and the impact of the determinant is high, then the determinant could be considered for use.

• Continue to monitor the literature, to see if new studies appear using that determinant.

The other limitation, high effort, means that in the reviewed past studies it has been costly for practitioners to collect some of the required determinants. With the advent and widespread use of new technologies, the effort required to collect some of the required high-effort determinants may be substantially reduced. Lowering the effort related to some high-impact, yet also high-effort, determinants (see, e.g., those indicated with orange color in Group B in Fig. 7) would have a two-fold benefit. The direct reduction of the costs required to collect information on those determinants will also enable wider consideration of these determinants in a larger number of studies, thus increasing their representation. As technology enhances the capture of such determinants, there is an opportunity to revisit past studies/datasets and increase the representation of these determinants, which might then transition to determinant group A in in Fig. 7. To address the current limitations and disadvantages of high effort, the following is recommended for practitioners:

• Monitor the use of emerging technologies that provide an opportunity to lower the cost required to collect the determinant. For example, there is an opportunity for analysis of high resolution satellite maps/photos to provide automated estimates of observable determinants such as garden size (GSZE) over large number of households, which would lower the cost substantially⁵⁶. Similarly, latent determinants such as water consumption awareness (AWARE_C) could be based on the uptake of user-friendly smart metering and phone apps on water consumption if they were widely available^17,57.

• Evaluate overall costs vs benefits based on preliminary experiments on small sample data (to evaluate benefits while avoiding high costs), and consider the use of lower cost proxy data for the “high effort” determinant.

These recommendations provide some guidance for practitioners to handle determinants classified as “use with caution”.

Limitations and future research

This work provides evidence and a quantitative framework for the analysis of household water consumption determinants, yet several limitations and questions remain for further research. First, alternative formulations of determinant representations, impact, and effort could lead to different results. This also stands for the subjective thresholds we adopted to distinguish between high and low representation, impact, and effort. Such thresholds and criteria formulation could be changed based on needs and subjective judgement.

Second, in this review we focused on the analysis of individual determinants of household water consumption. However, some determinants could be correlated, present redundant information, or be accounted for in alternative ways to build models for forecasting water demand (e.g., rainfall amount versus rainfall occurrence⁵³). Input feature engineering, variable redundancy, and data accuracy can substantially affect the performance of water demand models. Future studies focused on comparative analysis of alternative determinant formulations and inter-links/dependencies among different determinants can help define non-redundant determinant sets to train models of water demand and recommendations for variable pre-processing.

Third, the findings of this study are consistent with previous review papers that identified both observable and latent variables as the most important with respect to domestic water consumption⁵⁸. Yet, other meta-analyses and review studies found partly contradictory results. Differently from our study²⁶, found that the most important determinants of water use behaviour are related to individual opportunities and motivations, gender, income, and education level. In turn⁵⁹, formulated a model that accounted for a wide range of variables including demographics, dwelling characteristics, household composition, conservation intention, trust, perceptions, habits, and perceived behavioral control. It must be noted that the above studies do not consider household water consumption per se as we do here, but relate potential determinants of water consumption also to individual consumption or behavior changes (i.e., changes in water consumption over time). Future, potentially contrasting, studies could then expand the scope of this work and relax the exclusion criteria we adopted here to achieve more inclusive comparative analyses that investigate the effect of different determinants in relation to quantified intervals of total household water consumption, and other heterogeneous aspects of domestic water demand, including statistics on end use components (e.g., flow rate, duration, or frequency of individual appliances)⁶⁰ and temporal changes of water consumption levels due to external stressors such as droughts, or demand management interventions^39,49, for example.

Fourth, the set of framework analysis papers includes case studies primarily located in the United States, Australia, and Europe (see Fig. 4). Geographical coverage is thus skewed. There is a need for more studies from other geographical regions (including countries with low-income economies) in order to obtain a more balanced picture and consolidate/expand the results obtained so far.

Finally, recent works have highlighted that urban and household water demands have been modelled at different spatial and temporal resolutions⁴⁷. The choice of the temporal and spatial resolution of interest is determined both by data availability and the specific modelling and management purpose. Multi-scale studies combining different levels of spatial and temporal aggregation of water demands and potential determinants would further advance our analysis and contextualize specific recommendations for data collection and processing at the different spatial and temporal scales of interest.

Source: Resources - nature.com