02-literature.Rmd

# Literature {#literature}

Understanding the equity benefit distribution of park access requires us to
consider multiple literatures. First, we discuss theories of justice as they have been applied to transportation policy to introduce our conceptualization of equity. Next, we consider the
disparity in park utility perception among different populations. We
subsequently consider quantitative techniques to evaluate the access that
individuals have to park facilities. Finally, we consider recent research 
documenting and analyzing street conversions instigated by the COVID-19
pandemic.

## Equity and theories of justice

The pursuit of equity, as it pertains to distributions of costs and benefits, is
necessary, though insufficient, to the pursuit of justice, which
@fainstein_just_2010 argues should be a central pursuit of urban policy and
planning. @fainstein_just_2010 draws on theories of justice to propose that a
just city is one that is equitable, diverse, and democratic.
@schweitzer_environmental_2004 discuss the literature on environmental justice
and transportation in terms of cost- and benefits-based claims of injustice
(both of which relate to equity under Fainstein's framework of the
just city) and process-based claims (which relate to democracy under
Fainstein's framework).

@taylor_paying_2009 apply theories of justice to categorize equity-based
arguments in support of various transportation finance mechanisms.
@pereira_distributive_2017 similarly survey the moral philosophy literature
and apply it to transportation policy evaluation. Both papers emphasize that
claims of distributional (in)justice must specify _what_ is being distributed
in addition to what that distribution should be be. Taylor and Norton (2009)
argue that the distributions of revenue collection, expenditures, and benefits
from use of transportation infrastructure must be considered together.
@pereira_distributive_2017 apply Rawlsian egalitariansim [@rawls2001justice]
and capabilities approaches [@sen2014development; @nussbaum2001women] to argue
that the transportation policy should focus on the distribution of
accessibility, broadly defined as the ability to both access the
transportation system and use it to access destinations. They join
@martens_justice-theoretic_2012 in calling for further research to arrive at a
operational definition of accessibility that best aligns with a
justice-theoretic approach.

In this study, we follow @pereira_distributive_2017 in focusing on accessibility
and adopt a utility-based accessibility measure, as we discuss in the following
sections. While our analysis does not rely on a strict definition of an ideal
distribution of accessibility, we start from a proposition rooted in both
Marxian ethics [@heller_theory_1976] and liberation theology
[@john_paul_ii_centesimus_1991] that an equitable distribution is one that
favors vulnerable and marginalized populations.

## Sociodemographic variation in park utility

The idea that different racial, ethnic, or cultural groups have different
recreational styles, and might thus have different needs and preferences for
parks and open space, has been thoroughly discussed in the leisure studies
literature. @husbands1995ethnicity offer a detailed review of that research as
of the mid-1990s. In general, explanations for racial and ethnic differences in
park use can be classified into two categories: cultural and lifestyle
differences on the one hand, and discrimination and marginalization on the
other.

@byrne2009nature summarize literature in the former category, noting that Black
park users have been described as preferring more social, sports-oriented
spaces, relative to white park users who prefer secluded natural settings
[@washburne1978black; @hutchison1987ethnicity; @floyd1999convergence;
@gobster2002managing; @payne2002examination; @ho2005gender]; Asian park users
are described as valuing aesthetics over recreational spaces
[@gobster2002managing; @payne2002examination; @ho2005gender]; and Latino park
users are said to value group-oriented amenities like picnic tables and
restrooms [@baas1993influence; @hutchison1987ethnicity; @irwin1990mexican].

In an observational and survey-based study of park users in Los Angeles,
@loukaitou1995urban found a high-level of enthusiasm for park use among Hispanic
residents. While she found, consistent with prior research
[@baas1993influence; @hutchison1987ethnicity; @irwin1990mexican], that
Hispanic park users showed a preference for passive recreation, she found that
to be the case for all other user groups as well. She also found that Hispanic
park users were the most likely to actively appropriate and modify park space,
for example, by bringing items from home. She found that Hispanic park users
tended to visit parks as family groups; African American park users tended to
visit parks as peer groups; Caucasian park users tended to visit parks alone;
and Asian residents were least likely to visit parks, even in a predominantly
Asian neighborhood. Interviews with local elderly Asian residents (Chinese
immigrants) suggested that a lack of interest in American parks was rooted in
perceptions of the ideal park as "an aesthetic element of gorgeous design,"
leaving them unimpressed with poorly landscaped American parks emphasizing
recreational functions.

@byrne2009nature criticize such scholarship as having grossly exaggerated
ethno-racial differences in park use and preferences, and suggest a model for
explaining park use based on four elements: Sociodemographic characteristics;
park amenities and surrounding land uses; historical/cultural context of park
provision (including development politics and discriminatory land-use
policies); and individual perceptions of park space (including safety and
sense of welcome).

@byrne2012green applies a cultural politics theoretical frame to why people of
color are underrepresented among visitors to some urban parks. Focus groups of
Latino residents of Los Angeles emphasized the importance of parks to
children. Participants described visiting parks with their children and the
positive and negatives associations that parks evoked of their own childhood
memories of parks and wilderness. Participants described barriers to visiting
parks including distance, inadequate or poorly maintained facilities, and fear
of crime. They cited a lack of Spanish-language signage not only as a barrier
to understanding but also as a signal that a park was not intended to serve
Spanish speakers. Participants also expressed that they did not feel welcome
in parks located in high-income or predominantly white neighborhoods because
they expected that other park users would have racist attitudes, that a more
boisterous Latino 'recreational style' would not be tolerated, or that there
would be other behavioral norms they were not aware of.


## Defining and measuring park accessibility

"Accessibility" is an abstract concept that describes how easily an individual
can accomplish an activity at a particular space. Though not strictly
quantifiable, the idea of quantifying access is tempting and has been
frequently attempted. @Handy1997 identify three broad types of accessibility
measures: cumulative opportunity or isochrone measures, gravity-based measures,
and utility-based measures. @Dong2006 follow the same basic classification
approach as Handy and Neimeier, illustrating mathematically how the three
different types of measures can be collapsed into each other. @GEURS2004127
group cumulative opportunity and gravity-based measures into a single category
that they refer to as location-based measures. In this, @GEURS2004127 rely on
the distinction that utility-based measures incorporate revealed preferences of
individuals for particular destination characteristics (including but not limited
to distance and location) while location-based measures are entirely geo-spatial 
in their definition.

Cumulative opportunity measures are calculated by counting the number of origins
or destinations within a threshold travel cost of a location (where "cost" might
be some combination of distance, travel time, and/or monetary cost of travel). A
strength of cumulative opportunity measures lies in their simplicity and
intuitive interpretation, and this may explain why @boisjoly2017get have found that
it is universally used in published metropolitan transportation plans. However, they may
be too simple, especially with regard to trip costs near the threshold. An example 
of a cumulative opportunity measure might be the number of parks within a ten-minute
walk of a person's home, or the number of households living within ten minutes of a
park. This measure would imply that a household living immediately adjacent to a park
has the same access to it as one that lives nine minutes away, but that a household
living eleven minutes away has no access to it.

Gravity-based accessibility measures take a similar approach to cumulative
opportunity measures, but theoretically include all possible destinations and
weight them according to the travel cost that they impose, based on an impedance
function (often a negative exponential calibrated to observed trip
distributions). Cumulative opportunity measures may be considered a special case
of gravity-based measures, where the impedance function takes the form of a
binary step function that equals zero after a cutoff travel cost (which is why
@GEURS2004127 classify them both as location-based).

A major advantage of gravity-based accessibility measures lies in their
consistency with travel behavior theory: Gravity-based measures have their roots
in the trip distribution step of the traditional four-step travel demand
forecasting method, where trips originating in a particular zone are distributed
among destination zones, proportionate to each zone's gravity-based
accessibility. In spite of their theoretical advantages over cumulative-opportunity
measures, the practical advantages of gravity-based measures are less clear. Based on
their finding that cumulative-opportunity measures and gravity-based measures area
highly correlated, @boisjoly2016daily argue that the greater simplicity of
cumulative-opportunity measures makes them appropriate for transportation planning
applications.

@paez2012measuring distinguish between positive and normative accessibility indicators,
where positive indicators are based on information about the degree to which
destinations are observed to be accessible and normative indicators also incorporate a
normative judgment of the degree to which destinations _ought to be_ accessible. They
develop a normative location-based accessibility measure incorporating average trip
lengths into a cumulative opportunities measure, varying the isochrone threshold by
socioeconomic characteristics. 

While traditional four-step travel demand models distribute zonal trips based on
a gravity-based accessibility model, the travel demand modeling profession has
shifted more recently towards a destination choice framework that distributes
trips based on discrete-choice regression models. @mcfadden1974measurement applied
discrete choice models to urban travel demand to predict mode choice, and modern
disaggregate activity-based models apply them to all travel behavior choices,
including to select among alternative routes or alternative destinations 
[@de2011modelling]. Though the application of random utility models to destination
choice is not new [see @anas1983discrete], the increasing availability of computing
resources makes estimating and applying discrete choice models on large alternative sets
in a practical context more feasible.

Destination choice models estimate the probability of selecting a particular
destination among a set of alternatives based on the relative attractiveness, or
_utility_, of each alternative. Utility may be a function of distance or travel
time alone (in which case, a utility-based accessibility measure might be quite
similar to a location-based measure), but the function can also incorporate
other destination characteristics that lead one destination to be more
highly-valued and used than another. 

Although @paez2012measuring focus on location-based measures in classifying
accessibility metrics as positive or normative, they also describe utility-based
measures as "essentially positivistic" (p. 143) and critique them in terms of
practicality, noting that, in order to limit the size of the choice set to be
computationally manageable, the analyst must either aggregate destinations to a higher
spatial resolution or calculate utility based on a randomly-selected subset of all
available alternatives.

@siddiq_tools_2021 review 54 different accessibility tools utilizing a variety of
different accessibility metrics and classify accessibility measures as either
people-based or place-based. They define people-based measures as those that account for
individual traveler characteristics, perceptions, and constraints. This includes all
utility-based measures, as well as variations on cumulative opportunity measures (such
as those proposed by @paez2012measuring) that incorporate socioeconomic data on
travelers. They argue that people-based measures are most appropriate for equity
analysis.

### Location-based measures of park accessibility

ParkScore [@parkscore2019], developed by the Trust for Public Land, is a popular
measure of park accessibility that starts from a cumulative opportunity measure
(the share of the population that resides within a 10-minute walk of a green
space) and adjusts this value based on the total city green space, investment,
and amenities weighted against the socioeconomic characteristics of the
population outside of the 10-minute walk threshold. The resulting score is a
convenient quantitative tool in estimating the relative quality of green space
access across cities [@Rigolon2018]. ParkScore may be less useful at identifying
the comparative quality of access within a city, particularly since the vast
majority of residents in dense areas like San Francisco (100\%) and New York City
(99%) may live within the binary 10-minute walk threshold. The Centers for
Disease Control and Prevention (CDC) has developed an "Accessibility to Parks
Indicator" along similar lines [@Ussery2016], calculating the share of the
population living within a half-mile of a park for each county in the U.S.

Urban scholars have used gravity-based measures to explore the
spatial distribution of park access across Tainan City, Taiwan
[@chang2011exploring] and to estimate the relationship between park access and
housing prices in Shenzhen, China [@wu2017spatial].

Some scholars have used location-based measures of park accessibility to
evaluate equity in park access. @chang2011exploring use a gravity-based measure
to determine that low-income neighborhoods have less access to parks than
higher-income neighborhoods in Tainan City, Taiwan. @bruton2014disparities
conduct a neighborhood-level analysis of park amenities in Greensboro, North
Carolina, and find that low-income neighborhoods tend to have parks with more
picnic areas, more trash cans, and fewer wooded areas, but they do not address
the question of the extent to which different populations might value these
different amenities. @kabisch2014green find that neighborhoods in Berlin with
high immigrant populations and older populations likewise had less access to
parks, and they pair these findings with survey results suggesting that these
disparities are not consistent with the preferences expressed by those
populations.

### Utility-based measures of park accessibility



A utility-based measure of park accessibility can incorportate characteristics of parks in addition to travel time or distance, including park size, cleanliness, or the availability of particular
amenities. The degree to which these park and trip attributes influence the
destination utility can be estimated statistically using survey data.

Though destination choice utility models have not commonly been used to measure
park accessibility, scholars have acknowledged that park accessibility metrics
should be linked with park use, since a park that has many visitors must by
definition be accessible to those visitors. @McCormack2010 provide a
comprehensive review of this literature; it is sufficient here to note that most
studies find park use to depend on a complicated interplay between park size,
maintenance, facilities, and travel distance. Many of these attributes are
incorporated into ParkIndex [@Kaczynski2016], which estimates the resident park
use potential within small grid cells by applying utility preference
coefficients estimated from a survey in Kansas City.

There are limited examples of researchers using a destination choice model to
predict recreation attractions. @Kinnell2006 apply a choice model to a survey of
park visitors in New Jersey, and estimate the relative attractiveness of park
attributes including playgrounds, picnic areas, and park acreage weighed against
the travel disutility and the relative crime rate at the destination. In a
similar study, @Meyerhoff2010 model the urban swimming location choice for a
surveyed sample. In both studies, the researchers were attempting to ascertain
which attributes of a recreation generated the most positive utility, and
therefore which attributes should be prioritized for improvement. Though neither
was attempting to understand relative park accessibility, @macfarlaneNYC applied
the @Kinnell2006 estimates in an exploration of utility-based park accessibility
and its relationship to aggregate health outcomes.

One primary obstacle to estimating discrete-choice models on the park
destination problem has been the lack of sufficiently detailed, trip-level data
on park users. Most destination choice models in practice are estimated from
household travel surveys that must focus on all trip purposes, and necessarily
group multiple recreation and social trips together [@nchrp716]. However, the
advent of large-scale mobile device networks and the perpetual association of
unique devices with unique users has given researchers a new opportunity to
observe the movements and activity location patterns for large subsets of the
population [@Naboulsi2016]. Such passively collected location data --- sometimes
referred to as part of a larger category of “Big Data” --- is a by-product of other
systems including cellular call data records  [e.g., @Bolla2000;
@Calabrese2011], probe GPS data [@Huang2015], and more recently Location Based
Services (LBS) [@Roll2019; @Komanduri2017]. LBS use a network of mobile
applications that obtain the users’ physical location at different points in the
day. Commercial vendors repackage, clean, and scale these data to population or
traffic targets and provide origin-destination flows to researchers and
practitioners. @Monz2019, for example, demonstrate that passive device data can
accurately estimate trip flows to natural recreation areas.

A number of methods have been proposed to develop destination choice
information from these passive data. @Bernardin2018 employs a passive
origin-destination matrix as a shadow price reference in an activity-based
location choice model, iteratively adjusting the calibration parameters of the
choice utilities to minimize the observed error between the passive data and the
modeled predictions. @tf_idea uses the passive flow data as a probabilistic
sampling frame to recreate individual trips through simulation. A similar method
developed by @Zhu2018 uses the passive dataset directly, sampling 10,000 random
trips from GPS traces of taxi trips in Shanghai and estimating a destination
choice model. Employing the passive data set in this way provides the authors an
opportunity to examine the choices of a large sample of a small population (taxi
passengers). The @Zhu2018 methodology could be extended to other situations
where collecting a statistically relevant survey sample would be prohibitively
difficult, but where passive device location data reveals which destinations
people choose among many observable options.

## Street Conversion Equity Analysis

In their analysis of over one thousand reallocations of street space that
occurred in response to the global COVID-19 pandemic, @combs2021shifting find
that a plurality created additional space for walking, cycling, and recreation,
although some reallocated space to commerce (e.g. outdoor dining and shopping)
or converted short-term parking to urban freight or food delivery.

If we define an urban park as a public space that is designated for the purpose
of recreation, exercise, and social gathering, then the rapid reallocation of
street space to accommodate recreation and active travel could be characterized
as a proliferation of small urban parks. Researchers at the Trust for Public
Land have explicitly described the reallocation of street space from cars to
pedestrians as a strategy to relieve pressure on parks [@hussain_parks_2020] and
have suggested that these actions should (and, in New York, had failed to)
prioritize areas that would otherwise have low access to parks
[@compton_parks_2020]. @fischer_covid-19_2021 have likewise done an equity
analysis of street reallocation from vehicles to pedestrians in three mid-sized
Canadian cities and found that interventions were generally more common in
places with higher proportions of white residents and fewer children. The
analyses by both @compton_parks_2020 and @fischer_covid-19_2021 were both based
on proximity alone rather than on utility-based accessibility measures.

Of course, streets that reallocate space for active travel and recreation do not
have the amenities or general character of most parks, and classifying them as
equivalent to their greener peers in an accessibility analysis would be
erroneous for many reasons. But a utility-based accessibility framework would
allow us to discount these street parks for the amenities they lack while also
considering the benefits proffered by their availability and proximity. Further,
we can model these tradeoffs with statistical weights determinable through
observing park trip distribution patterns revealed through passive mobile device
data.