Preferences for Intrinsically Risky Attributes1 Zack Dorner2 , Daniel A. Brent3 & Anke Leroux2
LTU Choice Modelling Workshop 15 July 2016
1
Funded by the CRC for Water Sensitive Cities Monash University 3 Louisiana State University 2
Preferences for intrinsic attributes
Preference elicitation mostly focused on extrinsic attributes Preferences for intrinsic attributes matter for choice determination; especially if preferences depend on attitudes, e.g. towards others or towards risk Can we leverage information about attitudes to model preferences for unspecified intrinsic attributes? Implications for marginal utilities and model fit?
SP literature on risk and risk attitude
Studied extensively in the context of flood insurance (Botzen and Van den Bergh, 2012 a,b), wild fire protection (Bartzcak et al, 2015) and reducing health risks (Lusk and Coble, 2005; Cameron and De Shazo, 2013; Andersson et al., 2016) Outcome risk from environmental projects (Glenk and Colombo , 2013; Roberts et al., 2008; Wielgus et al. 2009; Rolfe and Windle, 2015) Risk attitudes and technology adoption (Qiu et al., 2014; Liu, 2013; Hidrue at al., 2011) We leverage information from a fully incentivized risk task to explain choices in a DCE over goods with unspecified intrinsic risky attributes
Our setting: choices over alternative sources of water Mix of mains supply in Australia (2009/10) (PC, 2011): Source Dams Groundwater Desalination Recycled Pipeline Total
Proportion of supply (%) 81.1 9.0 2.8 3.8 3.3 100.0
Drought related water shortages result in frequent restrictions to household water use and controversial public investments in water supply augmentations DCE to elicit preferences over alternative, additional sources of municipal water supply, allowing for water quality and cost.
Our hypothesis
Ex ante we hypothesize that there are two sources of intrinsic risk affecting the choices made by participants: Supply risk due to some sources being weather dependent Technology risk from new and unproven technologies
Note: These sources of risk were not mentioned in the information materials provided so as to avoid priming respondents
Intrinsic sources of risk
Source New dam Stormwater Pipeline Desalination Recycled Groundwater
Weather dependent? Yes Yes Yes No No No
New technology? No Yes No No Yes No
Intrinsic sources of risk
Source New dam Stormwater Pipeline Desalination Recycled Groundwater
Weather dependent? Yes Yes Yes No No No
New technology? No Yes No No Yes No
Risky attribute is instrinsic to the water source and cannot credibly be varied in a DCE setting
Data - survey
Professional door to door survey of 981 home owners in Manningham and Moonee Valley (Melbourne) and Fairfield and Warringah (Sydney) March to October, 2013 Random sample of home owners 167 people were randomly selected to complete an incentivised risk task first (based on Holt and Laury, 2002, to estimate coefficient of CRRA)
Presentation of incentivized risk elicitation task 10 choices over binary lotteries, one of which was selected at random for payment.
We introduced the choice task as follows:
When water shortages become more frequent, investments to increase urban water supply need to be made. There are a number of options in terms of water source and technology that a city can invest in. These options differ with respect to the quality of water provided and therefore their allowed use, as well as the cost of water provision. It is possible to install a third water pipe to your house, so that your tap water will not be contaminated with potentially lower quality water from the new source. You would NOT have to pay for the installation of the third pipe.
Discrete choice experiment
Theoretical framework Utility V from choosing a water source is given by: V = β j Xj + β q Xq + βc C Xj is dummy for source (relative to new dam) Xq is dummy for allowed use (relative to potable) C is cost
(1)
Theoretical framework Utility V from choosing a water source is given by: V = β j Xj + β q Xq + βc C
(1)
Xj is dummy for source (relative to new dam) Xq is dummy for allowed use (relative to potable) C is cost Adding risk: V = β j Xj + β q Xq + βc C + βr
Xr1−γi − 1 1 − γi
Xr = 1 when not risky Xr = 2 when risky γi is CRRA coefficient estimated by risk task
! (2)
Mixed Logit Results - Supply Risk
Fixed Coefficients Non-potable outdoor Non-potable indoor βr (supply risk) Random Coefficients Desalination
(1)
(2)
(3)
0.0265 (0.0470) −0.1452∗∗∗ (0.0514)
−0.0583 (0.1080) −0.2595∗∗ (0.1202) 0.2247 (0.2729)
0.0259 (0.0496) −0.1471∗∗∗ (0.0531) 0.7115∗ (0.3847)
−0.7724∗∗∗ −0.4354 −0.0546 (0.0879) (0.3400) (0.4014) Recycled −1.6845∗∗∗ −1.0638∗∗∗ −0.9622∗∗ (0.1109) (0.3551) (0.3995) Groundwater −2.5589∗∗∗ −1.9428∗∗∗ −1.8375∗∗∗ (0.1207) (0.3707) (0.4047) Stormwater −0.9977∗∗∗ −0.7071∗∗∗ −0.9998∗∗∗ (0.0788) (0.1558) (0.0845) Pipeline −2.2565∗∗∗ −1.6346∗∗∗ −2.2534∗∗∗ (0.0980) (0.1771) (0.0992) Cost+ −0.1118∗∗∗ −0.1543 −0.1086 (0.0425) (0.1107) (0.0927) AIC 23795.0 4129.5 23787.7 BIC 23893.8 4207.8 23893.6 Observations 8600 1370 8600 Individuals 860 137 860 Coef; (Std Err); *** p < 0.01, ** p < 0.05, * p < 0.1. Std Errs clustered at the respondent level; bootstrapped for Model 3. Allowed use relative to potable; source relative to new dam. + Triangular distribution. All others are normal.
CRRA Imputation
Marginal utility of choosing desalination over new dam, by γ ˆ¯ − V ˆ¯ =βˆ¯ + βˆ V r D ND D
=βˆ¯D − βˆr
11−γi − 1 1 − γi
21−γi − 1 1 − γi
− βˆr
21−γi − 1 1 − γi
(3a) (3b)
Mixed Logit Results - Technology Risk
Fixed Coefficients Non-potable outdoor Non-potable indoor βr (new technology risk) Random Coefficients Desalination
(1)
(4)
(5)
0.0265 (0.0470) −0.1452∗∗∗ (0.0514)
−0.0587 (0.1080) −0.2601∗∗ (0.1201) −0.1209 (0.2759)
0.0259 (0.0481) −0.1455∗∗∗ (0.0498) −0.3891 (0.4581)
−0.7724∗∗∗ −0.6748∗∗∗ −0.7746∗∗∗ (0.0879) (0.1783) (0.1021) ∗∗∗ ∗∗∗ Recycled −1.6845 −1.1771 −1.2903∗∗∗ (0.1109) (0.3528) (0.4823) Groundwater −2.5589∗∗∗ −2.1664∗∗∗ −2.5616∗∗∗ (0.1207) (0.2538) (0.1331) Stormwater −0.9977∗∗∗ −0.5778∗ −0.6053 (0.0788) (0.3283) (0.4747) Pipeline −2.2565∗∗∗ −1.6342∗∗∗ −2.2554∗∗∗ (0.0980) (0.1774) (0.1074) Cost −0.1118∗∗∗ −0.1528 −0.1138 (0.0425) (0.1112) (0.0912) AIC 23795.0 4130.0 23794.2 BIC 23893.8 4208.3 23900.1 Observations 8600 1370 8600 Individuals 860 137 860 Coef; (Std Err); *** p < 0.01, ** p < 0.05, * p < 0.1. Std Errs clustered at the respondent level; bootstrapped for Model 5. Allowed use relative to potable; source relative to new dam. + Triangular distribution. All others are normal.
Conclusion
We incorporate intrinsically risky attributes in a choice model over alternative water sources, where water quality and costs vary exogenously Leverage information on risk attitudes, elicited in an incentivised risk task Non-potable indoor water is disliked relative to other allowed uses Supply risk (weather dependence) is an intrinsic attribute that matters to individuals, depending on their level of risk aversion In contrast, technology risk does not significantly affect choice behaviour
CRRA Imputation Coefficient (Std. Error) Constant -0.4976 (0.7710) Flood risk perception -0.0916 (0.1109) Own flood insurance -0.0225 (0.2435) Don’t know flood insurance -0.3901 (0.2498) Flood insurance*Flood risk percep 0.3412∗∗ (0.1503) Age -0.0089 (0.0061) Female 0.0810 (0.1840) Education (yrs) 0.0591 (0.0452) Middle income -0.2087 (0.2376) High income -0.1150 (0.3547) Fairfield 0.6196∗∗ (0.2870) Moonee Valley 0.3039 (0.2558) Manningham 0.6336∗∗ (0.2644) σ 0.9600∗∗∗ (0.0729) Pseudo R-squared 0.0798 P-value 0.0047 N 124 ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1. Middle and high income are dummies relative to low income. Dummies for Fairfield, Moonee Valley and Manningham are relative to Warringah. Return to Results
