Author: Jonathan Russo
Chair: Joshua Abbott
Committee: Christa Brelsford, Kelli Larson

Predicting Participation in the Las Vegas Water Smart Landscaping (WSL)
Program

Abstract:
As arid cities’ water scarcity concerns grow, so does the importance of residential
water conservation. Understanding the drivers of participation in water conservation
programs can aid policymakers in designing programs that achieve conservation and
enrollment targets while achieving cost-effectiveness and distributional goals. In this
study I identify and analyze the characteristics that drive participation in the Southern
Nevada Water Authority’s Water Smart Landscaping rebate program – a program that
pays homeowners to replace their grass lawns with xeric landscaping – and how those
characteristics change over time as rebate values and water prices vary.

In order to determine what characteristics influence participation in this program
I gathered data from multiple sources. I use a panel dataset of household water
consumption that spans 12 years of approximately 300,000 homes. I merged this dataset
with home structural characteristics, geographical, and demographic context. I then use
these characteristics in a linear probability model, with school enrollment zone fixed
effects to determine their influence on a household’s probability of participation. School
zones are used to control for unobserved characteristics, such as demographics, which are

1

not at a household level. I then utilize these school zone fixed effects in a 2nd stage
regression to decompose these elements and analyze their effect on participation.
I find that a household’s water costs, as reflected in the marginal price faced in the
summer and the differential between summer and winter water bills, as well as yard size
are primary factors that influence participation. I also show that changes in rebate value
and water rates can affect different types of households. There is also evidence to
support that neighborhood characteristics affect a household’s likelihood of participating.

2

Predicting Participation in the Las Vegas Water Smart Landscaping (WSL)
Program
1. Introduction:

Water scarcity has long been a major concern for cities in arid regions. In the
face of limited water resources, these cities have two options to ensure water for their
communities. One way is to increase the supply by acquiring water rights. For instance,
Nevada has water-banking agreements with both Arizona and the Metropolitan District of
Southern California (Harrison, 2009). In these types of agreements one entity sells their
unused water to the other, normally with some drought restrictions to protect the seller.
However, these types of interstate agreements are rare, as are instate acquisitions of new
water sources. This is mainly because the transfer of water from the new source can
require extensive funding for conveyance infrastructure on top of the costs of locating the
new source and the transaction costs of negotiating an agreement.
A second, often preferred, option is to extend the current supply through water
conservation using price and non-price approaches. While there is a long-standing debate
as to which of these approaches yields the best outcome, there is still no consensus on
which is better at achieving water conservation. Many economists argue the most
efficient approach is to use prices to stimulate water conservation; however, raising
prices can have undesirable distributional consequences and be politically difficult to
achieve. Non-price policies, such as water-use restrictions, water conservation
technology adoption, and educational programs, are often the favored choice of water and
electric utilities, but may fall short in terms of their conservation goals and economic
cost-effectiveness. For instance, rebate programs for durable goods can have so-called

3

“additionality problems” in which many of the people who participate would have
replaced the good even in the absence of the rebate (Olmstead & Stavins, 2009; Bennear
et al., 2013). Paying households to undertake retrofits that they would have done anyway
obviously compromises the cost-effectiveness of such programs. Given these challenges,
what should an arid city do to promote water conservation amongst its residents?
This question has confronted water managers in the Las Vegas area for decades.
The city almost reached the limits of its water supply in the mid-1990s, which prompted
the creation of the Southern Nevada Water Authority (SNWA). The SNWA is now
considered by many to be a model for best water management practices. It has created a
plethora of programs targeted at water conservation utilizing both price and non-price
approaches. Their utilization of the wastewater management system has effectively
conserved indoor water use to the point where the majority of the efforts are now focused
on outdoor water use. The Water Smart Landscaping (WSL) program is one of the larger
and more notable programs that they have created aimed at outdoor water use. It is also
the focal point of this study.
The WSL program, informally known as “Cash for Grass,” is a voluntary program
that offers a rebate for the replacement of grass lawns for xeric landscaping. The purpose
of this study is to identify what observable structural, social and economic factors
affected household’s choices to self-select into the program. I then analyze these
characteristics to see which are the most influential on participation and affected
participation rates. This is done in order to get a better understanding of what types of
households are being motivated to enroll and how value changes in water price and the
rebate itself altered the uptake rate of the program and the type of household that was

4

incentivized. It also gives the ability to understand the spatial footprint of the uptake of
the program across the city and how this uptake correlates with Census-level
sociodemographic variables. Understanding these factors illuminates the sorts of users
attracted to the WSL rebate program. It also gives insights into how the design of these
incentives can be potentially altered to improve program targeting.
To accomplish this I use a variety of data sources that capture aspects of water use
and the magnitude of the water bill from household level billing data, structural
characteristics of homes (i.e. bedrooms, pool), geographic context (i.e. proximity to
parks), and demographic characteristics (i.e. race, income). These characteristics are then
used in a linear probability model (LPM) to measure how they affect the probability of a
household enrolling in the program. Since demographic information isn’t available at a
household level I use elementary school enrollment zones “neighborhoods”, as a spatial
fixed effect. School zones are chosen instead of block group units from the Census, due
to the small amount of participants in WSL. By using the geographically larger school
zones I minimize the amount of neighborhood units that did not have a participant. It
also allows for a finer degree of econometric control over those unobserved features. I
then utilize these school zone fixed-effects in a second stage of the model to determine
how these elements affect a household’s probability of enrollment.
I also analyze the rebate’s effect and how this changed over time, rebate dummy
variables are utilized in conjunction with time variables. I use a flexible spline function
of time along with dummy variables for discrete changes in rebate prices and terms to
better understand how the relation between time and rebate value affected participation.
These changes of value over time also allow me to examine the differences in participants

5

who chose to enroll at different rebate values. To evaluate these changes, participants are
grouped into 4 cohorts based on the rebate value that was active at the time of their
enrollment. From there I use summary statistics as well, as well as Kruskal-Wallis tests,
to compare cohorts and identified characteristic changes that differentiate one cohort
from another.
Understanding the drivers of households’ participation in water-saving rebate
programs and how variation in the incentives alters both the rate of composition of
participants provides the knowledge-base for the creation of better policies. In an arid
region, effective water conservation measures are integral to sustainable water practices.
It is important to not only understand how efficient a policy is at conserving water, but
how that policy can be implemented to achieve higher rates of participation.

Figure 1 Illustrates the overall percentage of households that participated in the WSL program within
each elementary school zone by 2012

6

2. The Oasis Runs Dry: A Brief History of Water in Las Vegas:
The Las Vegas area has long been seen as an oasis in the desert. Even in the early
1900s its numerous natural springs and close proximity to two rivers provided an
attractive location for the railroad to create a train station. The station was the start of
Las Vegas and its growing community, but the population boom that was to come was
never expected. This lack of foresight is most evident in Nevada’s 300,000 acre-feet
allotment of water per year in the Colorado River Compact (CRC). At the time of the
agreement Nevada had a small population that was easily supported by its instate water
resources. It also, at the time, had no way to transport the water from the Colorado River
to Las Vegas. This led to Nevada accepting the smallest allotment of any of the Compact
states, since the region’s leaders never expected the population increase that was to come.
With the addition of the mega-casino in the early 1970s Las Vegas became one of
the fastest growing cities in the United States. Nevada small allotment under the CRC
became a liability as Las Vegas began closing in on the supply’s limit (Harrison, 2009).
Water Resources Management Incorporated, an outside consultancy agency, was
contracted to assess the water management issues of the area. In 1991, they released their
results, which concluded that the area would reach its water supply limits by the mid1990s. In order to deal with the water shortage five water utility companies and two water
waste management companies joined together to form one agency. In 1991 the Southern
Nevada Water Authority (SNWA) was established.
The SNWA initiated numerous water conservation programs and practices, none
more important than their utilization of the wastewater treatment system. They

7

successfully negotiated with the US Bureau of Reclamation to take advantage of the
return flow credit system that Arizona v California had set a precedent for previously
(Harrison, 2014). The return-flow credits allow treated wastewater to be returned to the
Colorado River system and credits the state’s allotment with the amount of water
returned. Consequently, almost all of Las Vegas’ indoor water use that enters the
treatment system is reclaimed. This outcome has two major benefits. First, it increases the
water supply because any water returned to Lake Mead does not count against Nevada’s
water allotment from the CRC. Second, since most indoor water is ultimately recycled it
incentivized the SNWA to focus their efforts on outdoor water conservation.
SNWA created a mixture of regulatory, educational, and incentive based
programs aimed at reducing outdoor water use. For instance, water pricing has been
employed using a tiered water price structure with increasing rates to promote lower
consumption. Wastewater enforcement has been combined with increased fines to deter
poor water management practices. The SNWA hosts numerous events and has created
multiple publications all targeted at educating the public on conservation efforts and
current drought plans. Rebates are offered for pool covers, smart irrigation clocks, and
rain sensors.
SNWA also utilized a block pricing approach for water (see Figure 3). From
1996 -2003 price tiers were set at: first 5 kgal, next 10 kgal, next 25 kgal, and over 40
kgal. Post-2003 the height and placement of tiers changed with the new tiers set with
increased pricing at: first 5 kgal ($.98 - $1.16), next 5 kgal ($1.42 - $2.08), next 10 kgal
($1.42 - $3.09), and over 20 kgal ($1.92 - $4.58). This lowered some of the volumetric
breakpoints from the previous rate structure and implemented significant increases in

8

price, mainly in the higher tiers. Prices were also raised in Feb. 2007 and once again in
May 2008, similarly to the increase in 2003 the largest raises are on the upper tiers.

Figure 2 Value of the rebate offered based on year. Blue represents the value and time frame of Cohort 1
or participants who enrolled at the value. Similarly, red represents cohort 2, green represents cohort 3,
and cohort 4 is orange.

The WSL program was initiated in 1996 as a pilot program offering a cash rebate
for the replacement of grass with xeric landscaping. The program was opened up to all
residents in 2000. While the rebates are offered to both commercial and residential
landowners, this study will focus only on the residential participants, specifically singlefamily households. The rebate varied over time between $.50 and $2.00 per square foot.
This study evaluates the program over a twelve-year period from 2000 to 2012 (Figure 2
above). During this time period, the only significant change to the program were the
length of time the landscape conversion needed to be kept and rebate price per square
foot. At the program’s inception, there was durability requirement. This changed in
2003, when conversions were required to be held for 5 years and again in 2004 when the
term was raised to 10 years or until the property was sold. In 2009 WSL conversions to
9

xeric landscaping were required to be maintained in perpetuity, regardless of the sale of
the home.
In order to enroll in the program participants must first file an application. The
yard is then assessed to confirm the land is eligible. Once approved a minimum
conversion of 50% of lawn must be replaced with either xeric landscaping or permeable
artificial turf. A final assessment of the home then occurs to verify the conversion has
taken place and meets all requirements before payment can be given. The time it takes to
complete this process is approximately 5 months on average (Brelsford & Abbott,
2017a). Currently there is a maximum conversion allowance of 5000 ft2 per year, with
no property receiving a rebate that exceeds $300,000 in a given year.
3. Literature Review/ Research Gap:
A number of studies have examined conservation programs that offer a rebate for
a durable good, such as and including the “Cash for Grass” WSL program in Las Vegas
(Brelsford & Abbott, 2017a; Sovocol et al., 2006; Bennear et al., 2013; Alcott, 2011).
From an economic perspective, much of this research has fallen into two major
categories. First, many papers examine the effectiveness of policies in achieving
reductions in conservation of a resource (Brelsford & Abbott, 2017a; Alcott, 2011; Datta
& Filippini, 2016; Datta & Gulati, 2014; Davis et al., 2014). Others examine the costeffectiveness of a policy or set of policies in achieving their conservation goals –
potentially comparing the relative economic efficiency of alternative approaches
(Bennear et al., 2013; Brelsford & Abbott, 2017b). The results of these studies, in many
instances, contribute to the debate of the efficiency of price versus non-price approaches.
Many studies show that utilizing price to signal scarcity creates a more efficient outcome

10

than non-price approaches (Olmstead, 2010; Grafton & Ward, 2008; Mansur &
Olmstead, 2012). However, the cost of the reduction in consumption is often found to be
distributed unevenly and is paid mostly by low-income households. This leads to equity
issues and is one of the main reasons non-price policies are more common when it comes
water demand management. Another issue is raising rates is not politically palatable in
many areas. Low water prices are popular, regardless of drought conditions, which make
raising prices an uphill political battle (Olmstead & Stavins, 2009).
There is a substantial amount of research demonstrating that voluntary rebate
policies are inefficient relative to a direct pricing of water or energy consumption
(Bennear et al., 2013; Olmstead, 2010, Timmins, 2003). This is especially true when the
subsidies are given in return for replacing durable goods, such as retrofitting old
technology with new more efficient technology (Bennear et al., 2013; Davis et al., 2014).
The additionality of the rebates in these programs tends to be small. This is due to the
overall expected benefit of the rebate being minimized, since many participants would
have made the replacement with a smaller rebate and in some cases with no rebate at all
(Boomhower & Davis, 2014). There is also the concern of the “rebound effect”
occurring after the change has been made eliminating part of the expected benefits
(Gillingham et al., 2015). This happens when behavioral changes occur in response to
expected gains from using new technology that is more efficient.
Unlike many conservation programs that offer rebates for replacing durable
goods, the WSL program doesn’t suffer the same deficiencies. While irrigation systems
do require replacement at some point, this does not require a complete landscape change
and there are no specific time horizons to replace lawns. “Therefore, compared to many

11

other subsidies for the replacement of durable goods, there is little reason to suspect that
WSL-rewarded landscape renovations would have occurred in the absence of the
subsidy” (Brelsford & Abbott, 2017a). There is also little evidence to suggest any
significant “rebound effect” from the program (Brelsford & Abbott, 2017a). These
characteristics are uncommon in rebate programs and they contribute to the uniqueness of
the WSL program.
The overall size of the WSL program is yet another aspect that adds to its rarity
within conservation programs. This study consists of close to 300,000 households over a
span of twelve years with monthly observations for water use on each individual
residence. This creates a larger sample size over a longer period of time at a higher
resolution than most studies on comparable programs. At its inception, WSL program
was the first of its kind. Since its creation numerous other incentive based turf removal
programs have popped up in other cities located in arid regions across the southwest.
One place that the literature falls short is in studies that identify characteristics
that influence participation. Much of the existing literature in this area looks at programs
that involve different types of resources that are not directly comparable to water, such as
land conservation (Moon, 2013; Sorice et al., 2017; Drescher et al., 2017). There is a
scarce amount of research that examines this aspect of water and electricity rebate
programs. Understanding which factors affect the likelihood of participation and what
type of effect they have is a mostly overlooked facet in the research.
Of the literature that does exist, regardless of resource, a large portion of the
papers has their focal point on more subjective characteristics, such as trust and social
norms. While these characteristics are important and can help complete the picture of the

12

characteristic components that define a participant, little attention is paid to how
household and geographical attributes affect uptake of participation over time when
presented with different incentive amounts, which is the focus of this study.
The WSL program affords the opportunity to help fill in these gaps in
understanding not only how rebate values can elicit different rates of participation among
different type of households, but also its effects on similar households in a
characteristically different neighborhood. Due to its strength in additionality and noted
lack of a “rebound effect” understanding the differences between households that choose
to participate and those who do not can add valuable information into identifying
practices that are more effective in influencing higher enrollment. It can also help in
identifying what incentives attract specific types of householders, such as homes with
high outdoor water use. The results of these examinations can further validate the effects
of the WSL program. As more programs are created in the image of the WSL the more
important it is to address these aspects.
4. Data:
To identify significant variables and analyze participation in the WSL program, I
draw from multiple datasets. I use a SNWA panel dataset of 1999-2012 monthly parcel
level residential water consumption spanning 299,872 homes, of which 26,300 are
eventual participants in the WSL program. I matched each household’s records with the
structural characteristics of houses from the Clark County Tax Assessor’s records. Due
to unrealistic attributes found in the records, such as more bedrooms than rooms in a
home, 161 homes are excluded from the study, 22 of which are participants. This left
299,708 households in the study, with 26,278 (8.8%) of them being participants. The

13

dataset was then condensed down from monthly observations over twelve years to one
yearly observation, per household 1. This created a dataset with 3,215,017 observations.
Table 1: 1st Stage Regression Variables Summary Statistics
Definitions
Pervious Area
Outdoor porous area, such as grass or soil
(m2/1000)
Impervious Area
Outdoor and indoor non-porous area, such
as asphalt, not including the 1st floor area
of the home. (m2/1000)
Indoor Area
Total area of the home (m2/1000)
Pool
Dummy variable for presence of a pool
Bedrooms
# of bedrooms
Bath
# of bathrooms
Log Value
Log value of home. Value was calculated by
the C. C. Assessor’s Office
Residency
Years elapsed since last sale of home
Resident 2yr
Dummy variable that represents the 1st 2
years after the sale of the home
Summer Marginal Highest marginal price a resident is
charged per kgal during the summer
months (May-Sept.) in a given year.
Winter Marginal
Highest marginal price a resident is
charged per kgal during the winter months
(Oct.-April) in a given year.
Seasonal
The difference between the cost of the avg.
Difference
monthly summer and winter bills in a given
year
Consumption
Avg. monthly water use in kgal for a given
year
Park 1/4mi
Dummy variable that identifies if there is a
park with a water feature within a 1/4 mile
of the home
Rebate 2003
Dummy variable that identifies years
rebate value was $1
Rebate 2007
Dummy variable for the rebate value $2.00
Rebate 2008
Dummy variable for the rebate value $1.50
Vintage 1960
Vintage 1984
Vintage 1992
Vintage 1997

Dummy variable which represents homes
built before 1960
Dummy variable for homes built between
1961-1984
Dummy variable for homes built between
1985-1992
Dummy variable for homes built between
1993-1997

Mean
.397

Std. Dev.
.413

N = 3,215,017
Min/Max
[.001, 132.55]

.269

.128

[0, 9.65]

.183
.245
3.38
2.22
10.61

.079
.43
.809
.635
.607

[.025, 3.47]
[0, 1]
[0, 12]
[0, 14]
[5.3, 16. 43]

7.77
.19

7.61
.389

[0, 49]
[0, 1]

2.18

.747

[.847, 3.58]

2.07

.671

[.855, 3.65]

17.74

32.62

[-513.74,
5744]

14.48

14.83

[0, 3009.82]

.063

.243

[0, 1]

.305

.461

[0, 1]

.084
.424

.278
.494

[0, 1]
[0, 1]

.045

.208

[0, 1]

.24

.43

[0, 1]

.135

.341

[0, 1]

.174

(.379)

[0, 1]

1Billing records are related to the home and not household.

This is due in part to a lack of data on the
transfer of homes, which would allow for the identification of tenet changes in rental properties. In
this paper the terms “home” and “household” will be used synonymously even though there is a lack
of equivalence.

14

Vintage 2002
Vintage 2012
Year

Dummy variable for homes built between
1998-2002
Dummy variable for homes built after 2002
Elapsed time in years of the study 20002012, with 2000 = 1

.356

(.479)

[0, 1]

.19
7.415

(.392)
(3.64)

[0, 1]
[1, 13]

This data was then merged with spatial data including nearby parks and water
bodies. This was done under the assumption that public green space and/or bodies of
water could act as a substitutes or complements to the services provided by a home’s
private lawn. The mapping data for parks was collected from the Clark County
Government Center. Park water features data came from multiple local government
websites.2 Elementary school enrollment zones data came from the Clark County School
District. School zones are chosen as a proxy for “neighborhoods”, since these zones have
been found to share similar demographic characteristics (Clark, 1987; Richards, 2014).
The spatial characteristics are then linked to a household’s parcel using ArcView GIS
software. Variables are created for park distance from a household, parks with a water
feature (i.e. Pool, pond, etc.) distance from household, and the number of parks with and
without water features that fell within specific distances from a residence. Distance is
defined as the Euclidian distance from the residence to the park and does not necessarily
represent a true walking distance. Households are also matched with their elementary
school enrollment zone.

2

Websites used: www.clarkcountynv.gov, www.lasvegasnevada.gov, www.summerlink.com,
www.cityofnorthlasvegas.com, www.cityofhenderson.com

15

Table 2: 2nd Stage Regression Variables
Definition
Poverty Income
% of people with an income <
$25,000
Low Income
% of people with an income
between $25,001-$35,000
Low Mid Income % of people with an income
between $35,001-$45,000
Middle Income
% of people with an income
between $45,001-$100,000
High Mid
% of people with an income
Income
between $100,001-$150,000
High Income
% of people with an income <
$150,000
HH Size 1&2
% of households whose size is 2
or less people
HH Size 3&4
% of households whose size is 3
or 4 people
HH Size 5&up
% of households whose size is 5
or more people
No HS Diploma
% of people who have a 12th
grade education or less, without a
high school diploma or equivalent
HS Diploma
% of people who have a H.S.
diploma or its equivalent.
Some College
% of people who have some
college or an associates degree
Bachelor Degree % of people with an Bachelors
Degree
Graduate
% of people with a Graduate or
Degree
Professional Degree
White
% of people who identify their
race as White & non-Hispanic
Black
% of people who identify their
race as Black or African American
Asian
% of people who identify their
race as Asian
Other Race
% of people who identify their
race as some other race or
combination of races.
Hispanic
% of people who identify as
Hispanic regardless of race.
Owner
% of households who own their
home
Renter
% of households who rent their
home
Under 18
% of people who are < 18
Age 18to29
% of people who are between the
ages of 18 to 29

Mean
.226

Std. Dev.
(.13)

N = 135
Min/Max
[.064, .645]

.123

(.037)

[.045, .208]

.28

(.048)

[.14, .422]

.241

(.086)

[.054, .428]

.084

(.051)

[.003, .234]

.047

(.041)

[.002, .27]

.567

(.094)

[.282, .778]

.303

(.056)

[.14, .433]

.13

(.056)

[.048, .351]

.204

(.121)

[.047, .639]

.301

(.049)

[.164, .417]

.32

(.067)

[.118, .444]

.116

(.057)

[.018, .268]

.058

(.035)

[.004, .159]

.761

(.16)

[.091, .961]

.099

(.144)

[0, .834]

.046

(.03)

[0, .174]

.093

(.056)

[0, .336]

.148

(.106)

[.026, .571]

.654

(.21)

[.07, .964]

.346

(.21)

[.036, .93]

.255
.163

(.049)
(.042)

[.152, .405]
[.047, .291]

Age 30to44

% of people who are between the
ages of 30 to 44

.317

(.033)

[.206, .422]

Age 45to64

% of people who are between the
ages of 35 to 54

.163

(.037)

[.078, .263]

16

Over 65

% of people who are over 65

.102

(.044)

[.037, .377]

*All % rates are in relation to elementary school enrollment zones.

The final datasets I utilize are demographic characteristics from the U.S Census
Bureau. I utilized the 2000 Census surveys for block level data on income, education,
race, age, home ownership, and household size. Income was composed into six groups
for simplicity: poverty level (>$25,000), low income ($25,000-$35,000), low middle
income ($35,000-$45,000), middle income ($45,000-$100,000), high middle income
($100,000-$150,000), and high income (over $150,000). Household size was also
grouped: 1-2 people, 3-4 people, and more than 5 people. Educational data consists of:
no High School (HS) diploma or Equivalent, HS diploma or equivalent, some college or
Associates Degree, Bachelor degree, graduate or professional degree. Age was separated
into: under 18, 18 to 29, 30 to 44, 45 to 64, and over 65. Race & ethnicity was broken
down into Non-Hispanic White, Black, or Asian, Hispanic of any race, and other.
Ownership has two categories to identify residents who own their home and those who
rent. Utilizing this block level data, weighted averages for these characteristics are
created for the school zones. The averages are based on block demographics within the
enrollment zones. Each household was then matched with their corresponding school
zone and the demographic attributes.
5. Estimation Models & Methodology:
The nature of this study brings about some methodological concerns that needed
to be addressed. The first concern is dealing with the binary outcome variable. A linear
probability model (LPM) was used for models 1 - 4 (Cameron & Trivedi, 2010). While
logit or probit models were also considered, there are several reasons that made the LPM
17

the more appropriate choice. First, our overriding concern is to provide easily
interpretable estimates of the average marginal effects of changes in the exogenous
variables of the model on changes in the probability of WSL participation. The LPM has
robust properties in this regard, despite the potential for it to predict outside of the (0,1)
interval for individual probabilities in the presence of continuous covariates. Second, due
to the large size of the dataset using a logit model with a large number of spatial fixed
effects led to convergence challenges on a laptop computer. Third, and most importantly,
the LPM leads to a natural interpretation for the elementary school zones spatial fixed
effects included in the models. Since I use the estimated values of these fixed effects as
dependent variables in a second stage regression, it was important to have them in easily
interpretable units. The fixed effects in a LPM are in units of probability; however this is
not the case for the logit or probit model. Cluster-robust standard errors are used for all
models. Clusters are defined at a school zone level in order to control for
heteroskedasticity, as well as, spatial and serial correlation in the unobserved
characteristics that influence participation.
Numerous structural characteristics of the home are utilized in the 1 st stage of the
model, which give a detailed picture of the physical house and the composition of its lot.
However, the only observable household characteristics in this stage pertain to water
consumption and length of residency within the home. In order to control for
unobservable heterogeneity in these neighborhood demographics, I use elementary school
enrollment zones as a spatial fixed effect. These dummy variables serve to absorb the
average values of sociodemographic attributes and other unobserved spatially distributed
correlates of WSL participation. I then take these school enrollment zone effects into the

18

2nd stage, where they are regressed on the aforementioned Census demographic variables
to understand what factors influence these effects.
I consider several models of increasing complexity. In the first model the
probability of participation incorporates the observable structural characteristics of a
specific home, within a specific spatial zone, over a period of time. The model takes the
form:
𝑃𝑖𝑡 = 𝑋′𝑖𝑡 𝛽 + 𝐷′𝑖𝑡 𝛿 + Τ′𝑖 𝛾 + 𝜀𝑖𝑡

(1)

Where P represents the probability of participation of i, an individual household, at a
specific period of time t. 𝑋 ′ is a vector of house structure and length of residency
variables for a home, 𝐷′ is a vector of year-specific dummy variables. There are 2 main
sets of dummy variables. One set indicates the rebate value that was active at a given
time. The other set are vintage bins that are based on regulations faced by new
construction over different time periods as well as past findings about the efficiency of a
home’s water infrastructure (Brelsford & Abbott, 2017b). Homes are grouped into these
bins based on their build year.
𝑇′ is a vector of time trend variables. In Model 1 the T are variables years elapsed
and years elapsed squared. The second model adds to the first by modifying the time
trend variables. In Model 2, I utilize linear splines with three knots placed over the
twelve-year period of the study, at years 2003, 2007, and 2009. These years are chosen
because they coincide with changes in both the rebate value (see Figure 2) and water
rates. These spline variables replace the time variables in Model 1.
In Model 3, a neighborhood fixed-effect is added to Model 2, which is associated
with elementary school enrollment zones and takes this form:

19

𝑃𝑖𝑡 = 𝑋′𝑖𝑡 𝛽 + 𝐷′𝑖𝑡 𝛿 + Τ′𝑖 𝛾 + 𝛼𝑠 + 𝜀𝑖𝑡

(2)

Here 𝛼𝑠 represents the spatial fixed effects of elementary school enrollment zone s. This
was done under the assumption that there are unobservable spatial relationships amongst
households located within these school zones and as such, error terms within these zones
are likely to be correlated with observed covariates due to these shared characteristics.
Given that I do not observe household-level sociodemographic data, I utilize these fixed
effects to absorb the average effects of demographic variability across neighborhoods.
The final model in the first stage, Model 4, is similar to Model 3. It is also a
fixed-effect LPM except in this model I add household-level economic variables from the
monthly billing data.

The model’s form is:

𝑃𝑖𝑡 = 𝑋′𝑖𝑡 𝛽 + 𝐷′𝑖𝑡 𝛿 + Τ′𝑖 𝛾 + 𝐶′𝑖𝑡 𝜃 + 𝛼𝑠 + 𝜀𝑖𝑡 (3)
C represents a vector of economic variables for a household that has a one-year lag. I
utilize a lag to avoid issues of simultaneity (joint causation) between water consumption
levels and WSL participation decisions. The lag also accounts for the fact that
households, when contemplating long-run investments, like landscape remodeling, are
likely to draw upon recent historic data to project future benefits and costs of the
investment. Finally, the lag accommodates the time it takes from the initial application to
the program to fulfilling the requirements to qualify for the rebate.
In the second stage I begin the analysis of the school zone fixed effects from the
most complete first-stage model (Model 4). The second stage model’s form is:
𝛼̂𝑠 = 𝜓𝑠 + 𝑍𝑜𝑛𝑒𝑠 ′𝜆 + 𝜈𝑠

(4)

20

Here 𝜓𝑠 .is the intercept. 𝑍𝑜𝑛𝑒𝑠 is a vector of demographic variables that was calculated
from weighted U.S. census data. The second stage is estimated using heteroskedasticityrobust standard errors.
Spatial fixed effects are defined at the school attendance zone, whereas U.S.
Census data are available at the block-group. In general, each school zone contains
multiple block-groups. I weight the block-groups in each attendance zone to acquire more
representative estimates of the presence of a demographic characteristic within a school
zone. This is calculated by estimating the number of people or households with a
demographic characteristic per acre of the Census block-group. Using these estimates in
conjunction with the amount of acres of each block-group that falls within a specific
school zone I calculate the expected the number of people/households with the
characteristic in the zone. I then use the results with the total population of the zone to
determine the weighted average.

Figure 3 Shows the block level marginal price of water and the changes of price between 2000-2012.
Prices before 2003 were set to different tier levels, thus prices during this time are estimated based on
pre 2003 prices to fit post 2003 tier levels.

21

6.1 Results and Interpretations – Cohort Analysis
In order to get a better understanding of how changes and rebate value and water
rates influenced who participated in the program, participants are split up into 4 groups.
Each group identifies the value of the rebate at the time of enrollment of the participant.
Thus cohort 1 holds all participants who enrolled at a rebate value of $.50, cohort 2 are
those who signed up at $1, cohort 3 at $2, and finally cohort at $150.
In an examination of the WSL cohorts I first look at pervious area, which for
simplicity I will refer to as “yard size.” A Kruskal-Wallis test shows that there is a
statistically significant difference in the yard size by different cohorts (𝜒 2 =111.014, p <
.01). Figure 4 shows a small decrease in the mean yard size from cohort 1 – 3, but then a
slight rebound in cohort 4. Thus as the rebate increased household yards became smaller
on average. When the rebate dropped from $2 per sq. ft. to $1.50 in cohort group 4, yard
size increased, albeit only slightly. There is a similar trend found in consumption, seen in
also in Figure 4, which was also found to be significant (𝜒 2 = 489.53, p < .01). Since
yard size is assumed to correlate with water-use, these similarities are not surprising. The
size of the yard, assuming the yard contains grass, will determine the amount of water
needed to maintain it. However, it is important to recognize that my analysis does not
take into account the effects of the depletion of potential applicants from the sample pool
over time. Further analysis is required to understand what role if any this played in
altering the composition of the different cohorts.

22

Figure 4 (Top-left) Shows pervious area cohorts. (Top-right) Shows indoor area of each cohort. (Bottomright) Shows difference in avg. seasonal water bill of cohorts. (Bottom-left) Shows the average monthly
consumption of cohorts.

Average seasonal difference is used as a proxy for the cost of outdoor water use
of a home. Similar to yard size there is a statistically significant difference in outdoor
water cost between cohorts (𝜒 2 = 412.748, p < .01). Figure 4 shows groups 1 and 4 had
similar average cost associated with outdoor water use, as did cohort 2 and 3. However,
referring back to the average consumption, cohort 4 consumes less on average compared
to cohort 1. This shows that while cohort 4 uses less water on average, their cost of
outdoor water use is virtually the same as cohort 1. This similarity in cost is most likely
due to the increase in price of water, specifically on high water use. As seen in Figure 3,
each cohort faced an increasing marginal price on water, with the largest increases
occurring on water use over 20 kgal. These increases that target high water consumption
happen first in 2003, then again in 2007 and 2008.

23

Interestingly, the amount of lawn converted does not follow the same trend as
yard size. Figure 5 shows that in cohort 2 the mean amount converted is higher than
cohort 1. So while cohort 1 had larger yards on average, cohort 2 converted more of their
yard. This may be due to the rebate increase making larger conversions more cost
efficient; especially since the water prices saw a significant increase in 2003. Cohorts 3
sees a dip in conversion area, where as cohort 4 has a slight increase.

Figure 5 Illustrates the average amount of grass converted of a household in each cohort

24

Table 3: 1st Stage Summary Statistics
Participants
Consumption (kgal)
21.64
(15.61)
[0, 330.74]
Summer Marginal
2.32
(US Dollars)
(.53)
[.85, 3.58]
Winter Marginal
2.12
(US Dollars)
(.52)
[.86, 3.65]
Seasonal Difference
32.88
(US Dollars)
(37.79)
[187.51, 875.73]
Pervious Area (m2/1000)
.52
(.47)
[.013, 17.27]
2
Impervious Area (m /1000)
.32
(.14)
[0, 2.14]
Indoor Area (m2/1000)
.197
(.077)
[.054, 1.23]
Bedrooms
3.46
(.804)
[0, 9]
Bath
2.28
(.62)
[0, 7.5]
Build Year
1988.62
(12.19)
[1925, 2011]
Log value
10.76
(.54)
[8.93, 14.18]
Park 1/4mi
.06
(.24)
[0, 1]

Non-Participants
14.02
(14.66)
[0, 3009.82]
2.17
(.76)
[.85, 3.58]
2.04
(.68)
[.85, 3.65]
16.79
(32.26)
[-513.74, 5744]
.39
(.41)
[.0008, 132.55]
.26
(.13)
[0, 9.65]
.19
(.083)
[.025, 3.47]
3.38
(.81)
[0, 12]
2.24
(.66)
[0, 14]
1992.35
(14.85)
[1912, 2012]
10.66
(.62)
[5.298, 16.43]
.056
(.23)
[0, 1]

All Households
14.48
(14.83)
[0, 3009.82]
2.18
(.75)
[.85, 3.58]
2.05
(.67)
[.85, 3.65]
17.74
(32.62)
[-513.74, 5744]
.4
(.408)
[.0008, 132.55]
.27
(.13)
[0, 9.65]
.18
(.079)
[.025, 3.47]
3.38
(.81)
[0, 12]
2.22
(.64)
[0, 14]
1989.91
(14.64)
[1912, 2012]
10.61
(.61)
[5.298, 16.43]
.06
(.24)
[0,1]

* The figures in the rows for each variable are in this order:
mean, (std. dev.), and [min, max]. All participant statistics are
based on pre-WSL observations.

25

6.2 Results and Interpretations of 1st Stage: House Structure
When looking at what differentiates homes of participants from non-participants,
it is important to note that all observations for participants after enrollment are omitted
from the data set. This brings the total number of observations down to 2,158,762 and
allows for all results to be based on pre-treatment observations. One difference that can be

gleaned from Table 3 is pervious area. This distinct difference in pervious area size
between participants and non-participants can also be visualized in Figure 6. However,
the amount of indoor area is relatively the same on average. Participants only have
slightly larger homes on average, which can be seen in Figure 6.

Figure 6 (Left) Comparison of pervious area between participants and non-participants. (Right)
Comparison of indoor space between the two groups.

The estimates for the four first-stage models are presented in Table 3. Almost all
the variables are significant across all 4 models at a 1 percent level, with the exceptions

26

of Pervious Area, Bedrooms, Pool, Park 1/4mi, and Vintage 1984. Out of those 5
exceptions, only Vintage 1984 is insignificant across all models.
Pervious Area is significant at a 1 percent level in the first 3 models, but loses
some significance in Model 4. Its marginal effect also drops slightly in Model 3 due to
the addition of fixed-effects, but then decreases in Model 4 to approximately one-third
the effect of the other models. This is not surprising as Model 4 includes economic
variables that are capturing water use (particularly the seasonal difference variable),
which is correlated with the size of pervious area (𝜌= .4). This allows for the conclusion
that while Pervious Area is not highly significant in the final model, it still is
economically relevant. Indoor Area had a relatively large negative effect on
participation. An increase of 1 standard deviation in indoor area, holding constant
pervious and impervious outdoor area, has a -0.004 effect on the probability of
participation. By comparison, an increase of 1 standard deviation in pervious outdoor
area has less than a 0.0004 effect on participation. While these effects may initially
appear small, it is important to remember that the overall annual probability of
participation for a household is 0.0065.
Vintage bins, aside from Vintage 1984, have a notable effect on participation,
both positively and negatively. Using homes built before 1960 as the base of the model,
it shows that those constructed between 1985 and 1992 are found to have an increase of
approximately 0.003 on the probability of participation. Those built between 1993
through 1997 had a slightly less positive effect of 0.0025. Homes built after 1997 have a
negative effect on participation, with homes built between 1998 and 2002 having a 0.0026 effect of participation and those built after 2002 having the largest negative effect

27

at -0.0079. This suggests that newer homes are less likely to participate; this is likely
driven by smaller shares of green landscaping, on smaller lots (Brelsford and Abbott
2017b) in newer developments. This large negative effect found in Vintage 2012 makes
sense since this represents homes built after 2003. In 2003 a law was passed that didn’t
allow any turf in front yards of new construction. More efficient plumbing technology
may have also reduced participation by reducing the household’s indoor water
consumption. For perspective, there is a 0.011 difference in probability between a home
that falls into the Vintage 1992 bin and one that is built after 2003 in Vintage 2012.

Table 3: 1st Stage Estimation Results
Model 1
Pervious Area
0.00345***
(0.000657)
Impervious Area
0.0134***
(0.00223)
Indoor Area
-0.0303***
(0.00326)
Pool
0.000199
(0.000358)
Bedrooms
0.00122***
(0.000384)
Bath
-0.00142***
(0.000317)
Log Value
0.00649***
(0.000629)
Residency
0.000123***
(0.0000188)
Resident < 2yr
0.00193***
(0.000283)
Park 1/4mi
-0.000718
(0.000601)
Rebate 2003 (Dummy for
0.0125***
rebate for 2003- 2006)
(0.00131)
Rebate 2007 (Dummy for
0.0164***
rebate for 2007)
(0.00138)
Rebate 2008 (Dummy for
0.0185***
rebate for 2008- 2012)
(0.00158)
Vintage 1984
0.000537
(0.000614)
Vintage 1992
0.00441***

Model 2
0.00344***
(0.000657)
0.0134***
(0.00223)
-0.0303***
(0.00325)
0.000193
(0.000358)
0.00122***
(0.000384)
-0.00142***
(0.000317)
0.00651***
(0.000629)
0.000125***
(0.0000188)
0.00205***
(0.000283)
-0.000718
(0.000601)
0.0129***
(0.00111)
0.0150***
(0.00141)
0.0254***
(0.00192)
0.000533
(0.000613)
0.00440***

Model 3
0.00302***
(0.000629)
0.0102***
(0.00137)
-0.0224***
(0.00323)
-0.00000961
(0.000339)
0.000681**
(0.000280)
-0.00109***
(0.000249)
0.00556***
(0.00104)
0.000131***
(0.0000175)
0.00206***
(0.000284)
-0.000664
(0.000561)
0.0129***
(0.00112)
0.0150***
(0.00141)
0.0254***
(0.00191)
0.000719
(0.000794)
0.00289***

Model 4
0.00107*
(0.000569)
0.00600***
(0.00118)
-0.0299***
(0.00375)
-0.00119***
(0.000350)
0.000566**
(0.000257)
-0.00120***
(0.000248)
0.00401***
(0.000970)
0.0000993***
(0.0000172)
0.00250***
(0.000288)
-0.000850*
(0.000479)
0.0108***
(0.00104)
0.0147***
(0.00140)
0.0264***
(0.00196)
0.000746
(0.000733)
0.00296***

28

Vintage 1997
Vintage 2002
Vintage 2012
Year
Year^2

(0.00109)
0.00327***
(0.000994)
-0.00210***
(0.000534)
-0.00872***
(0.000631)
0.00172***
(0.000272)
-0.000195***
(0.0000150)

(0.00109)
0.00326***
(0.000997)
-0.00210***
(0.000535)
-0.00875***
(0.000630)

(0.000977)
0.00259***
(0.000688)
-0.00326***
(0.000677)
-0.0107***
(0.000901)

(0.000919)
0.00252***
(0.000648)
-0.00261***
(0.000612)
-0.00794***
(0.000727)

0.000255***
(0.0000967)
-0.000110
(0.000258)
-0.00703***
(0.000501)
-0.00168***
(0.000126)

0.000247**
(0.0000986)
-0.0000850
(0.000248)
-0.00701***
(0.000499)
-0.00168***
(0.000125)

-0.0731***
(0.00592)

-0.0713***
(0.00585)

2,158,762
0.006

2,158,762
0.006

-0.0595***
(0.0100)
X
2,158,762
0.006

0.00176***
(0.000211)
-0.000928***
(0.000292)
-0.00901***
(0.000626)
-0.00131***
(0.000110)
0.00364***
(0.000303)
0.000909***
(0.000208)
0.000113***
(0.00000985)
-0.0000973***
(0.0000145)
-0.0516***
(0.00942)
X
2,148,518
0.008

Year: (.,4)
Year: (4,8)
Year: (8,10)
Year: (10,.)
Summer Marginal
Winter Marginal
Seasonal Difference
Consumption
Constant
Fixed-Effects
Observations
R2

29

6.3 Results and Interpretations 1st Stage: Time Trends and Rebates

Marginal Effect on Participation

0.0250
0.0200
0.0150
0.0100
0.0050
0.0000

Figure 7 Shows the combined influence of the summation of the time splines and rebate dummies on the
probability of participation over time.

Figure 7 adds the predicted values from the spline regression in Model 4 on top of
the contemporaneous dummy variable for the rebate cohort. This specification allows
both smooth trends in WSL participation within each cohort and discontinuous jumps
between cohorts. From 2000 to 2002 there is a steady increase on the probability of
participation. In 2003 there is a steep jump in likelihood of participation that coincides
with the new rebate of $1. After this bump, the time trend in participation slows until
2007, when WSL rebates increase to $2. Here there is another increase in participation
similar to 2003. In 2008 the rebate decreases to $1.50, but even though the value drops
there is another increase in the probability of participation. After the 2008 rebate there is
a decrease in probability each year until 2012. This may imply that participation is likely
to occur in the presence of a new rebate value that is not necessarily tied to the rebate
increasing. This could be due to the shock of the value change on households.
Households who may have been contemplating participation are motivated by the price
30

change. However, as the price change gets further removed they are less likely to act.
During this time period there were other events, such as the 2008 financial crisis, that
may have also influenced participation. Since I am unable to separate the effects of these
events from those of the water policy changes, I cannot make any strong conclusions as
to the overall effect of the rebate change on participation.

Figure 8 (Left) Displays the difference in the avg. monthly water bill in the summer from the winter.
(Right) Shows the difference in avg. monthly consumption between the 2 groups.

6.4 Results and Interpretations 1st Stage: Water Consumption
Water consumption and economic variables play a significant role in explaining
participation rates. When looking at consumption, Table 3 shows that participants used 7
kgal more per month on average before enrollment than non-participants. This is
visualized by the right hand-side graph in Figure 8 where the 50-percentile line for
participants almost perfectly aligned with the 75-percentile line for non-participants.

31

Another characteristic that stands out is Seasonal Difference, or the cost of
outdoor water use. Table 2 shows that the mean Seasonal Difference for participants
(mean = 32.88) is approximately twice that of non-participants (mean = 16.79). This is
also reflected in the left hand side graph in Figure 8. In other words, households that
chose to participate paid double the amount per month on average for outdoor water use
than non-participants.
In Model 4 the effects of water usage and economic variables have been added to
the regression. While the highest price of water per kilo-gallon paid has a positive effect
on enrollment for both winter and summer, the summer price has a notably greater effect
at 0.0036 for every additional dollar added to the marginal price per kgal. The difference
in the average seasonal bill between summer and winter, a proxy for outdoor water costs,
has a positive effect on participation. An increase of 1 standard deviation in seasonal
difference has 0.0037 effect. These results suggest that the cost of outdoor water use is
primary factor in influencing households to participate.
Household consumption has a negative effect on participation. Since the seasonal
difference is controlling for outdoor water use and marginal pricing, consumption is
essentially representing indoor water use holding the price of water constant. An increase
of 1 standard deviation of consumption decreases the likelihood of participating by
approximately 0.001. This makes sense as the main benefits of conversion for a
household is in the savings received from the reduction in outdoor water use. Thus
households with high indoor water use would not have the same incentive to participate.
In terms of the main driving forces behind participation, it would be fair to say
that water use and its associated costs would be high on the list. Both the price of water,

32

especially in the summer, and the costs of outdoor water use have positive effects of
participation. These costs can be connected to yard size, which as previously discussed
in Sec. 6.2 are larger for participants. This suggests that the program is strongly selecting
toward homes with high outdoor water use.
6.5 Results and Interpretations 1st Stage: Residency &Parks
Length of ownership of a home, which is represented by Residency, is found to be
significant and positive. The effect on participation is relatively small, as a decade of
residency has an effect of less than 0.001. Resident <2yr, which captures the probability
of participation within the first two years of purchasing a home, is also positive and has
relatively large effect on the probability of participation. A new homeowner in their
second year of residency has an increased probability of participation of 0.0026. In a
comparison of the 2 variables, I find that while every year of residency in a home does
increase the likelihood of joining the WSL program, a new homeowner has a higher
probability of enrollment in the first two years than a resident of 25 years.
These somewhat surprising results of the effects of residency, is similar to those
found in a study done in Phoenix, AZ (Larson et al., 2017). Larson et al. found that longterm residents prefer grass lawns in comparison to residents that were newer to the area.
The results here somewhat mirror those results, as I find that new homeowners in the
Vegas area have a notably higher probability of participation and foregoing their grass
than long term residents do. I originally assumed that long-term residents would have a
greater knowledge of local water issues in comparison to new homeowners and therefore
expected a weaker effect from new ownership.

33

An alternative hypothesis is this is due to a “remodeling” effect that is associated
with new home ownership. New homeowners are likely to remodel their homes to suit
their needs. In this process homeowners might have a higher willingness to opt for
conversion as part of the remodel. Conversions during remodeling may also be
preferable to new owners whose intentions are to rent or “flip” the home to turn a profit,
as xeric landscaping requires less maintenance. While I have discussed the possibilities
of a “remodeling” effect and a preference of long-term residents for grass lawns as
separate, I would argue that it is some combination of these factors that drives this “new
resident” effect on participation.
It’s worth noting that in this study I use time elapsed since the date of last sale of
a home to establish a proxy for residency in the area. It does not take into account
whether the homeowner was moving from within the area or from outside of the area.
Thus, further research should examine the dynamics between various residency attributes,
landscape choices, and implications for water demand.
The first stage of the model yields another interesting result, for the outcome of
Parks ¼ mi. My initial hypothesis was that parks would act as a substitute for lawns, and
therefore, households that are within walking distance to parks would be more likely to
participate in the program. However, the presence of a park is found to be insignificant
in earlier iterations of the current models. What is found to be significant, though only at
a 10 percent level, is the presence of a park that has a water feature, such as a pool or
lake, within a ¼ mi. of a home. However, these parks had a negative effect on
participation implying that people who chose to live near these types of parks do so
because they value their green spaces, lawns included.

34

6.6 Results and Interpretations 2nd Stage: Demographics
Table 4: 2nd Stage Estimation Results
Model 1
Low Income
-0.00548
(0.0124)
Low Mid Income
0.0126
(0.0103)
Middle Income
0.0207
(0.0137)
High Mid Income
0.0300**
(0.0128)
High Income
-0.00944
(0.0103)
HH Size 3&4
0.0157
(0.0145)
HH Size 5&up
-0.0390**
(0.0183)
HS Diploma
0.00417
(0.0136)
Some College
0.0110
(0.0113)
Bachelor Degree
0.0154
(0.0231)
Graduate Degree
-0.0420*
(0.0253)
White
-0.0195
(0.0165)
Black
-0.0108
(0.0145)
Asian
-0.0531**
(0.0249)
Hispanic
0.0144
(0.0132)
Own Home
-0.00756
(0.00464)
Age 18to29
-0.0232
(0.0303)
Age 30to44
0.00650
(0.0206)
Age 45to64
0.0530**
(0.0216)
Over 65
0.000460
(0.0193)
Constant
-0.000131
(0.0206)
Observations
134
R2
0.544
Robust standard errors in parentheses
* p<.1, ** p<.05, *** p<0.01

35

Figure 9 presents the mean annual household probabilities of WSL participation
associated with living in a specific school zone. In looking at Figure 1 the zones that
have a high cumulative percentage of participants match up extremely well with the areas
of high probability of participation in Figure 9.
Figure 10 illustrates the fixed-effects for each school zone. The fixed effects are
in units of the annual probability of participation and have a mean of zero. As such, they
represent the shared “neighborhood” anomaly in the probability of the WSL participation,
relative to the average neighborhood. A positive value of 0.01 reflects that homes in a
particular zone participate at a rate of 0.01 more on average than predicted by observable
household characteristics alone. Figure 10 shows that an empirically significant portion
of the variability in participation rates in Figure 9 is explained by the covariates in the
first stage regression. However, there are still some “hot” and “cold” zones. Specifically,
zones to the Northwest and Southeast participate at a higher rate than predicted by the
first stage variables alone. These “hot” broadly coincide with zones of higher than
normal participation in Figure 9. Interestingly, the more central zones are comparative
“cold” spots.
Figure 10 shows these central “cold” spots cluster around the Vegas Strip. This
area is the older part of Las Vegas, with the mean build year of the homes in these zones
falling between the 1950s and 1970s. The city was built out from this central area,
therefore the father out from this central area, the newer the neighborhoods tend to be.
The majority of the “hot” zones, including the ones in the Northwest and Southeast were
built after the mid-1990s. However, since house vintage was controlled for in the 1 st

36

stage this leads me to believe there is some other unobserved characteristic driving this
variability.

Figure 9 Illustrates the average annual probability of household participation in each school enrollment
zone

37

Figure 10 Displays the first-stage fixed-effects of each school zone.

In an attempt to explain this spatial variability I regressed the spatial fixed effects
as a function of Census demographics. The complete results for the 2 nd stage model can
be found in Table 4. The coefficients in the second stage model are the effects of
neighborhood-level demographics on the neighborhood anomaly in annual participation.
Unlike the 1st stage models, here the majority of variables are not significant – perhaps
due to the small number of observations. There are a few significant variables, however.
Individuals in school zones with larger shares of households earning a salary between
$100,000 and $150,000 are more likely to participate by 0.003 for every 10% increase.
This effect may be attributed to the type of homes that attract people in this income
bracket. Homes that could be afforded in this income range would most likely have a

38

yard that could be converted. However, the income is perhaps not so high that incentives
and water price wouldn’t influence their decision to participate.
Areas that have high amount of large households (5 people or more) had a
negative effect of 0.0039 on participation for a 1% increase. This makes sense since
larger households most likely have children. Having a grass-covered area for children to
play without leaving the property could be seen as more important that gains made from
lawn conversion.
I also found that areas that had a high population of people between the ages of 45
and 64 had a positive effect on participation rates. This had the largest positive effect
with a 1% increase resulting in a 0.0053 increase. This could be due to people within this
age range being better suited to invest in their homes. Individuals younger than this are
likely to have young children and other financial commitments (i.e. student loans) that
take precedence over lawn conversion. Once an individual is older than 65 they be
retired or nearing retirement and therefore less likely to make substantial home
investments. Moreover, individual within this “middle-age” age range are less likely to
have young children and are often at the peak of their lifetime earnings. Therefore, an
investment like lawn conversion that would lower future water costs, would be appealing.
The other 2 variables that are found to be significant and have a negative effect
are area with a larger Asian population and a larger share of residents with graduate
degrees. The interpretation of these effects is not clear and may be driven by unobserved
neighborhood variables correlated with these sociodemographic characteristics more than
the characteristics themselves.

39

Conclusion:
The analysis of participation in the WSL program identifies key factors that
influence a household’s likelihood of enrollment. Here I find that larger yards, high
water, the cost of water and specifically, the cost of outdoor water use are all influential
in who chooses to participate. I also show how changes in rebate value, both increasing
and decreasing, as well as water price are correlated with the characteristics of
households that enroll. These are the primary drivers of participation found in the first
stage of this study. While residency and parks yielded some interesting results, I would
consider these as suggestive, but worthy of further investigation.
Though the results do not bring about strong connections between parks and
residents’ willingness or unwillingness to give up their lawns, there are more avenues that
can and should be investigated. This study is limited to only analyzing homes in relation
to distance to parks and park types, but not all green space are considered. Also, while
this study looks at homes that converted from grass to xeric landscaping, there is no data
that identified a homes landscape choice outside of the program. Adding these factors
may shed some light on the effects of greenspace on participation.
In the 2nd stage I examine school zone “neighborhood” effects. Here I find there
is significant evidence to show that neighborhood characteristics affect a household’s
probability of participation. Zones that have a high share of people between the age of 45
and 64, as well as those with a high middle income have a positive effect on participation.
On the other side of the spectrum, school zones with a high amount of large households
(over 5 people) have a negative effect. While these results do explain some of these
neighborhood effects there is still more that needs to be known in order to complete the

40

picture. Overall a more in-depth look at these effects and what the drivers are behind
them needs to be taken in order to better understand how they influence participation.
One factor of note in this study is that the 2008 financial crisis was not controlled
for since households were, to a varying extent, likely affected by the event. This made
creating a control group to test the effects of the crisis on participation infeasible. It is
also unclear what effect it had on participation. The crisis could have had a positive
effect as it put pressure on households to save money, thereby increasing participation.
This hypothesis is consistent with the significant increase in participation in 2008 shown
in Figure 7. However, the spike in participation may also have been caused by the
contemporaneous increase in the rebate. Nonetheless, the number of foreclosures that
occurred would suggest that there were also homes with high probabilities of
participation that did not enroll during this period due to bank ownership. Thus the
overall effect of the crisis on participation is unknown.
While I have focused on the WSL program as a non-price program, the first-stage
results do suggest how rebate programs can be used jointly with a price based approach
to achieve higher levels of conservation through increased participation. The strong
effects of summer marginal water prices and the difference between summer and winter
water bills on participation suggest that changes in structure of water rates can
significantly impact participation in the non-price program. Therefore price and non-price
policies can be complementary; utilizing the changes in price in conjunction with a rebate
can increase participation more than would be achievable with the rebate alone. Potential
negative distributional effects of price increases on low-income households can be
mitigated by careful design of block rates while also providing high marginal prices that

41

are likely to primarily affect homes with large outdoor water use. Relatively low-income
households with large amounts of water intensive landscaping can take advantage of
generous rebates to replace their landscapes – providing transition mechanism for these
households to avoid excessive water bills.
While here I look at the WSL program itself and how value changes of the rebate
and the price of water affected participation, there is much room for further research.
The SNWA created numerous programs during this time period and continues to do so.
It would be worthwhile to extend the analysis to take into account the effect of
interactions between these programs. Marketing, educational programs, and other
information shocks could play a vital role in understanding why people chose to
participate in the program. The availability of other rebate programs, such as the pool
cover rebate, and the order in which they are initiated could also be important in
understanding participation rates and the efficiency of the program.
In addition to expanding the effect of other policies on participation, a more indepth look at the participants themselves and the role of social interactions in driving
their water use is needed. Social characteristics have been found in numerous studies to
be an important factor in whether someone chooses to voluntarily participate in a
conservation program (Allcott, 2011; Attari 2014; Guerin et al., 2000; Lubell et al.,
2017). People’s social network and how they perceive themselves in comparison to their
neighbors plays a role in decision-making. Social norms of an individual or a specific
group of people can be a strong motivator in whether or not they participate in a program.
The level of trust one has in the administrator of a program can also be a factor. This

42

attribute would be particularly interesting, since the SNWA has a history of prioritizing
transparency and building trust with the communities it works with.
As water scarcity becomes a more pressing issue in arid cities, devising more
effective water management practices is imperative for achieving sustainability goals.
Understanding the drivers of participation can help policymakers to craft more effective
and targeted policies that satisfy water conservation targets in a cost-effective manner. It
can also show how multiple policies – including combinations of water pricing and nonprice policies – can be combined to augment participation, improve the selectivity, and
mitigate undesirable distributional outcomes of individual programs. This knowledge
can help water managers sustain their water supplies in an era of increasing scarcity.

43

Works Cited
Abbott, Joshua K., and H. Allen Klaiber. “Is All Space Created Equal? Uncovering the
Relationship between Competing Land Uses in Subdivisions.” Ecological
Economics, vol. 70, no. 2, 2010, pp. 296–307.,
doi:10.1016/j.ecolecon.2010.09.001.
Allcott, Hunt. “Social norms and energy conservation.” Journal of Public Economics,
vol. 95, no. 9-10, 2011, pp. 1082–1095., doi:10.1016/j.jpubeco.2011.03.003.
Attari, S. Z. “Perceptions of Water Use.” Proceedings of the National Academy of
Sciences, vol. 111, no. 14, Mar. 2014, pp. 5129–5134.,
doi:10.1073/pnas.1316402111.
Barrett, Greg. “Water Conservation: The Role Of Price And Regulation In Residential
Water Consumption.” Economic Papers: A Journal of Applied Economics and
Policy, vol. 23, no. 3, 2004, pp. 271–285., doi:10.1111/j.17593441.2004.tb00371.x.
Bennear, Lori S., et al. “Municipal Rebate Programs for Environmental Retrofits: An
Evaluation of Additionality and Cost-Effectiveness.” Journal of Policy Analysis
and Management, vol. 32, no. 2, 2013, pp. 350–372., doi:10.1002/pam.21692.
Black, J. Stanley, et al. “Personal and contextual influences on househould energy
adaptations.” Journal of Applied Psychology, vol. 70, no. 1, 1985, pp. 3–21.,
doi:10.1037/0021-9010.70.1.3.
Boomhower, Judson, and Lucas W. Davis. “A Credible Approach for Measuring
Inframarginal Participation in Energy Efficiency Programs.” Journal of Public
Economics, vol. 113, 2014, pp. 67–79., doi:10.1016/j.jpubeco.2014.03.009.
Brelsford, Christa, and Joshua K Abbott. How Smart is Water Smart Landscapes? . 22
May 2017a.
Brelsford, Christa, and Joshua K. Abbott. “Growing into Water Conservation?
Decomposing the Drivers of Reduced Water Consumption in Las Vegas, NV.”
Ecological Economics, vol. 133, 2017b, pp. 99–110.,
doi:10.1016/j.ecolecon.2016.10.012.

44

Brelsford, Christina M. “Whiskey is for drinking - water is for fighting over: population
growth, infrastructure change, and conservation policy as drivers of residential
water demand.” Arizona State University, 2014.
Cameron, Adrian Colin., and P. K. Trivedi. Microeconometrics Using Stata. Stata Press,
2010.
Clark, W. A. V. “Demographic Change, Attendance Area Adjustment and School System
Impacts.” Population Research and Policy Review, vol. 6, no. 3, 1987, pp. 199–
222., doi:10.1007/bf00122782.
Datta, Souvik, and Sumeet Gulati. “Utility Rebates for ENERGY STAR Appliances: Are
They Effective?” Journal of Environmental Economics and Management, vol.
68, no. 3, 2014, pp. 480–506., doi:10.1016/j.jeem.2014.09.003.
Datta, Souvik, and Massimo Filippini. “Analysing the Impact of ENERGY STAR Rebate
Policies in the US.” Energy Efficiency, vol. 9, no. 3, Aug. 2015, pp. 677–698.,
doi:10.1007/s12053-015-9386-7.
Drescher, Michael, et al. “Private landowners and environmental conservation: a case
study of social-Psychological determinants of conservation program participation
in Ontario.” Ecology and Society, vol. 22, no. 1, 2017, doi:10.5751/es-09118220144.
Ferraro, Paul J., and Juan José Miranda. “Heterogeneous treatment effects and
mechanisms in information-Based environmental policies: Evidence from a
large-Scale field experiment.” Resource and Energy Economics, vol. 35, no. 3,
2013, pp. 356–379., doi:10.1016/j.reseneeco.2013.04.001.
Frondel, Manuel, et al. “The impact of information on landowners’ participation in
voluntary conservation programs – Theoretical considerations and empirical
evidence from an agri-Environment program in Saxony, Germany.” Land Use
Policy, vol. 29, no. 2, 2012, pp. 388–394., doi:10.1016/j.landusepol.2011.08.003.
Gillingham, Kenneth, et al. “The Rebound Effect and Energy Efficiency Policy.” Review
of Environmental Economics and Policy, vol. 10, no. 1, 2015, pp. 68–88.,
doi:10.1093/reep/rev017.

45

Glenn, Diana T., et al. “Tools for evaluating and monitoring effectiveness of urban
landscape water conservation interventions and programs.” Landscape and
Urban Planning, vol. 139, 2015, pp. 82–93.,
doi:10.1016/j.landurbplan.2015.03.002.
Guerin, D. A., et al. “Occupant Predictors of Household Energy Behavior and
Consumption Change as Found in Energy Studies Since 1975.” Family and
Consumer Sciences Research Journal, vol. 29, no. 1, Jan. 2000, pp. 48–80.,
doi:10.1177/1077727x00291003.
Harrison, Christian. “Water Use And Natural Limits In The Las Vegas Valley: A History
Of The Southern Nevada Water Authority.” University of Nevada, Las Vegas,
2009.
Hartman, Raymon S. “Self-Selection Bias in the Evolution of Voluntary Energy
Conservation Programs.” The Review of Economics and Statistics, vol. 70, no. 3,
1988, p. 448., doi:10.2307/1926783.
Hoicka, Christina E., et al. “Residential energy efficiency retrofits: How program design
affects participation and outcomes.” Energy Policy, vol. 65, 2014, pp. 594–607.,
doi:10.1016/j.enpol.2013.10.053.
Howarth, Richard B, et al. “The economics of energy efficiency: insights from voluntary
participation programs.” Fuel and Energy Abstracts, vol. 41, no. 6, 2000, p. 416.,
doi:10.1016/0140-6701(00)94775-1.
Koontz, Tomas M. “Money Talks?But to Whom? Financial Versus Nonmonetary
Motivations in Land Use Decisions.” Society &amp; Natural Resources, vol. 14,
no. 1, 2001, pp. 51–65., doi:10.1080/08941920117246.
Langpap, Christian. “Conservation Incentives Programs for Endangered Species: An
Analysis of Landowner Participation.” Land Economics, vol. 80, no. 3, 2004, p.
375., doi:10.2307/3654727.
Larson, Kelli L., et al. “Residents’ Yard Choices and Rationales in a Desert City: Social
Priorities, Ecological Impacts, and Decision Tradeoffs.” Environmental
Management, vol. 44, no. 5, 2009, pp. 921–937., doi:10.1007/s00267-009-93531.

46

Larson, Kelli L., et al. “Legacy effects and landscape choices in a desert city.” Landscape
and Urban Planning, vol. 165, 2017, pp. 22–29.,
doi:10.1016/j.landurbplan.2017.04.014.
Layton, David F., and Juha Siikamäki. “Payments for Ecosystem Services Programs:
Predicting Landowner Enrollment and Opportunity Cost Using a Beta-Binomial
Model.” Environmental and Resource Economics, vol. 44, no. 3, 2009, pp. 415–
439., doi:10.1007/s10640-009-9293-5.
Lubell, M.n., et al. “Conservation Program Participation and Adaptive Rangeland
Decision-Making.” Rangeland Ecology &amp; Management, vol. 66, no. 6,
2013, pp. 609–620., doi:10.2111/rem-d-13-00025.1.
Maggioni, Elena. “Water Demand Management in Times of Drought: What Matters for
Water Conservation.” Water Resources Research, vol. 51, no. 1, 2015, pp. 125–
139., doi:10.1002/2014wr016301.
Mansur, Erin T., and Sheila M. Olmstead. “The Value of Scarce Water: Measuring the
Inefficiency of Municipal Regulations.” Journal of Urban Economics, vol. 71,
no. 3, 2012, pp. 332–346., doi:10.1016/j.jue.2011.11.003.
Moon, Katie, et al. “Personal circumstances and social characteristics as determinants of
landholder participation in biodiversity conservation programs.” Journal of
Environmental Management, vol. 113, 2012, pp. 292–300.,
doi:10.1016/j.jenvman.2012.09.003.
Moon, Katie. “Conditional and resistant non-Participation in market-Based land
management programs in Queensland, Australia.” Land Use Policy, vol. 31,
2013, pp. 17–25., doi:10.1016/j.landusepol.2011.08.011.
Olmstead, S. M. “The Economics of Managing Scarce Water Resources.” Review of
Environmental Economics and Policy, vol. 4, no. 2, Jan. 2010, pp. 179–198.,
doi:10.1093/reep/req004.
Olmstead, Sheila M., and Robert N. Stavins. “Comparing Price and Nonprice Approaches
to Urban Water Conservation.” Water Resources Research, vol. 45, no. 4, 2009,
doi:10.1029/2008wr007227.

47

Richards, Meredith P. “The Gerrymandering of School Attendance Zones and the
Segregation of Public Schools.” American Educational Research Journal, vol.
51, no. 6, 2014, pp. 1119–1157., doi:10.3102/0002831214553652.
Rogers, Jerry R., and Erika Moonin. “Recent Southern Nevada Water History and Water
Education.” World Environmental and Water Resources Congress 2014, 2014,
doi:10.1061/9780784413548.110.
Shen, Junyi, and Tatsuyoshi Saijo. “Reexamining the relations between socioDemographic characteristics and individual environmental concern: Evidence
from Shanghai data.” Journal of Environmental Psychology, vol. 28, no. 1, 2008,
pp. 42–50., doi:10.1016/j.jenvp.2007.10.003.
Silva, Julie A., and Alfons Mosimane. ““How Could I Live Here and Not Be a
Member?”: Economic Versus Social Drivers of Participation in Namibian
Conservation Programs.” Human Ecology, vol. 42, no. 2, 2014, pp. 183–197.,
doi:10.1007/s10745-014-9645-9.
Sorice, Michael G., et al. “Increasing participation in incentive programs for biodiversity
conservation.” Ecological Applications, vol. 23, no. 5, 2013, pp. 1146–1155.,
doi:10.1890/12-1878.1.
“Southern Nevada Water Authority: Five-Year Conservation Plan 2004-2009.” Aug.
2004.
Sovocool, Kent A, et al. “An in-Depth investigation of Xeriscape as a water conservation
measure.” American Waterworks Association Journal, vol. 98, no. 2, Feb. 2006,
pp. 82–93.
Timmins, Christopher. “Demand-Side Technology Standards under Inefficient Pricing
Regimes: Are They Effective Water Conservation Tools in the Long-Run?”
Economy &amp; Environment Current Issues in the Economics of Water
Resource Management, 2002, pp. 123–146., doi:10.1007/978-94-015-9984-9_7.
Viñoles, M.v., et al. “Length of residency and water use in an arid urban environment.”
Water Resources and Economics, vol. 12, 2015, pp. 52–66.,
doi:10.1016/j.wre.2015.09.004.

48

Warner, Laura A., et al. “Classifying Residents who use Landscape Irrigation:
Implications for Encouraging Water Conservation Behavior.” Environmental
Management, vol. 58, no. 2, 2016, pp. 238–253., doi:10.1007/s00267-016-07062.
Wilson, Charlie, and Hadi Dowlatabadi. “Models of Decision Making and Residential
Energy Use.” Annual Review of Environment and Resources, vol. 32, no. 1,
2007, pp. 169–203., doi:10.1146/annurev.energy.32.053006.141137.
Wirl, Franz. “Lessons from Utility Conservation Programs.” The Energy Journal, vol. 21,
no. 1, Jan. 2000, doi:10.5547/issn0195-6574-ej-vol21-no1-4.

49