Household Water Consumption in an Arid City: Affluence, Affordance, and Attitudes

Abstract

Reducing consumption in affluent urban households is perhaps the most important driver of future natural resource conservation. This article examines how water consumption in individual households is affected by income and determines whether household amenities or attitudes toward community and the environment mediate the effect of income on residential water use, net of other factors. We matched household social surveys, property characteristics, and climate variables with 24 months of individually metered water usage records for single-family houses in Phoenix, AZ. Household income had a positive, significant effect on consumption that was mediated by house size. Irrigable lot size and landscape type also had significant effects on consumption, although attitudes did not. In order to promote environmentally sustainable behavior we must develop better models of the social organization of consumption and encourage affluent households to be more attuned to the water affordances of their lifestyles.

Keywords:

• affluence
• conservation behavior
• environmental attitudes and behavior
• equity and sustainability
• household consumption
• sustainable consumption
• water resources

This research was supported by grants from the National Science Foundation's Biocomplexity in the Environment program (SES 0216281) and the Central Arizona–Phoenix Long-Term Ecological Research project (DEB 9714833 and DEB 0423704). We thank Lela Prashad and Shapard Wolf for preparing the water usage data used in this study and Darren Ruddell for research assistance. This article has benefited from conversations with our colleagues, Kelli Larson, Patricia Gober, Marco Janssen, and Chris Martin, as well as the comments of anonymous reviewers. We also thank the Phoenix Water Services Department for its cooperation.

Notes

Note. N = 205 single-family households.

Note. Significance indicated by ∗p < .05, ∗∗p < .01, ∗∗∗p < .001; two-tailed tests; t-statistics in parentheses. Reference category for Backyard = mostly lawn; Years lived in Phoenix = less than 5 years; Household size = one occupant.

In addition, during the 24-month time span of water records from January 2001 through December 2002, 31 residents moved into their homes. Models using only water records from months in which current survey respondents occupied their homes were not substantially different from models using water records from all months.

Values were imputed for 49 cases with missing responses on the income question. Differences in median household income among the eight neighborhoods were significant (F = 11.68, p < .001). Therefore, we assumed respondents’ earning profiles were much like their neighbors’. Median incomes for each separate neighborhood were calculated using valid survey responses cross-classified into a four-cell table that specified two demographic variables: whether respondents were married and whether they had a high school education or less versus more than a high school education. The resulting median income values for each cell were assigned to missing cases matched by neighborhood, marital status, and education.

Future studies could explore a variety of house characteristics that may be correlated with indoor water consumption. The Maricopa County Assessor's database is most useful for property evaluation and assessment and therefore is limited in relevant information. Number of bathrooms is the only other such variable included in the database. However, its theoretical utility for characterizing water-intensive lifestyles is less obvious than house size, and house size is also the stronger predictor of water use in our analysis.

The bivariate correlations among independent variables in the analyses range from near 0 to 0.76, and all but 6 in the matrix are below 0.40. For example, house size (log) and irrigable lot size (log) are correlated with a coefficient of −0.03. Phoenix is not a densely settled city and many smaller older houses are on properties nearly as large as those of newer houses on the urban fringe. Many newer homes, however, have a larger footprint, leaving less yard for cultivation. Variables that are significantly correlated (e.g., house size (log) and income (log) = 0.76, meaning they share about half their variance) do not introduce multicollinearity problems into our analysis. Using OLS regression, we calculated the variance inflation factors (VIFs) [1/(1 – R²)] for both of these variables in our final model with all predictors (Table 2, model 5). VIFs for these variables are less than half the often-cited threshold value of 10 (O'Brien 2007; Neter et al. 1989).