Table 1 Household demands for water and electricity respond to heat

Point estimate (and standard error in parentheses) are expressed in log points	(1) Only HH FE, OLS	(2) with time, OLS	(3) with weather, OLS	(4) with PM_2.5, OLS	(5) Trend, not year, OLS	(6) PM_2.5 endogenous, 2SLS
Dependent variable: Log water use (m³ mo⁻¹)
Mean temperature over usage period (+1 °C)	0.0106***	0.0124***	0.0106***	0.0067***	0.0049***	0.0072***
Mean temperature over usage period (+1 °C)	(0.0004)	(0.0014)	(0.0015)	(0.0016)	(0.0016)	(0.0015)
Number of observations	2,001,389	2,001,389	1,998,127	1,998,127	1,998,127	1,998,127
Number of regressors (not counting HH FE)	1	23	27	28	26	28
Number of households	121,480	121,480	121,480	121,480	121,480	121,480
R² [first-stage F-statistic, excluded instr.]	0.830	0.830	0.830	0.830	0.830	[926,239]
Mean value of usage in sample (m³ mo⁻¹)	16.66	16.66	16.66	16.66	16.66	16.66
Dependent variable: Log electricity use (kWh mo⁻¹)
Mean temperature over usage period (+1 °C)	0.1026***	0.0987***	0.0828***	0.0751***	0.0738***	0.0746***
Mean temperature over usage period (+1 °C)	(0.0004)	(0.0009)	(0.0010)	(0.0010)	(0.0010)	(0.0010)
Number of observations	1,982,472	1,982,472	1,979,238	1,979,238	1,979,238	1,979,238
Number of regressors (not counting HH FE)	1	23	27	28	26	28
Number of households	120,099	120,099	120,099	120,099	120,099	120,099
R² [first-stage F-statistic, excluded instr.]	0.885	0.885	0.885	0.886	0.886	[930,901]
Mean value of usage in sample	405.91	405.91	405.86	405.86	405.86	405.86

The table reports estimates for 12 water or electricity use regressions. An observation is a household by usage period in the 2012–2015 utility usage microdata. The dependent variable is the natural logarithm (log) of water use (m³ mo⁻¹) in panel a and log electricity use (kWh mo⁻¹) in panel b. The key regressor is the average daily mean temperature (°C). This average is taken over the same days that are concurrent to each usage observation, and has empirical support 26.1–29.1 °C. All regressions include a full set of household fixed effects (FE). Relative to specification 1, specification 2 includes time controls (month FE, year FE, day type composition; see Methods). Relative to 2, specification 3 includes weather controls (average daily mean relative humidity, dew point depression, wind speed; proportion of days with some precipitation). Relative to 3, specification 4 controls for average daily mean PM_2.5. Relative to 4, specification 5 replaces year FE with a linear time trend. Relative to 4, specification 6 instruments for PM_2.5 with Southeast Asia fire activity variables and Singapore thermal gradients and wind direction. OLS regressions in columns 1–5, 2SLS regressions in column 6. Standard errors, in parentheses, are clustered by household. ***, **, * denote significance at the 0.01, 0.05, and 0.1 levels, respectively.

Search