       http://jtlu.

org
. 6 . 1 [2013] pp. 63–72 doi: 10.5198/jtlu.v6i1.291

What is mixed use? Presenting an interaction method for measuring land use mix

Kevin Manaugh a Tyler Kreider

McGill University McGill University

Abstract: In recent decades, the mixing of complementary land uses has become an increasingly important goal in transportation and land
use planning. Land uses mix has been shown to be an influential factor in travel behavior (mode choice and distance traveled), improved health
outcomes, and neighborhood-level quality of life. However, quantifying the extent to which a given area is mixed-use has proven difficult.
Much of the existing research on the mixing of land uses has focused on the presence and proportion of different uses as opposed to the extent
to which they actually interact with one another. This study proposes a new measure of land use mix, a land use interaction method—which
accounts for the extent to which complementary land uses adjoin one another—using only basic land use data. After mapping and analyzing
the results, several statistical models are built to show the relationship between this new measure and reported travel behavior. The models
presented show the usefulness of the approach by significantly improving the model fit in comparison to a commonly-used land use mix index,
while controlling for socio-demographic and built form factors in three large Canadian cities (Vancouver, Toronto, and Montreal). Our results
suggest that simple, area-based, measures of land use mix do not adequately capture the subtleties of land use mix. The degree to which an
area shows fine-grained patterns of land use is shown to be more highly correlated with behavior outcomes than indices based solely on the
proportions of land use categories.

1 Introduction use motorized transportation to reach employment, commer-

cial, or leisure destinations, a fine-grained mixing of residen-
The idea of maximizing—and measuring—the extent to which tial, commercial, and recreational land uses might allow local
the built form can encourage active transportation has become residents to walk or bike to desired destinations. However, de-
a key issue for planners. Increased pedestrian and cycling ac- spite being a foremost concern for planners and policy makers,
tivity potentially offers an easy and cost-effective manner to many existing methods to measure the extent to which an area
improve human health, limit the adverse effects of transporta- is mixed rely on ungrounded assumptions. This paper proposes
tion-related pollution, and increase neighborhood interaction a new method of measuring land use mix using only basic land
and sense of place (Wilkinson and Marmot 2003). For these use data and GIS techniques. By measuring the extent to which
reasons, many regional and municipal transportation plans complementary land uses actually interact with one another, it
now place a focus on providing communities where walking is argued that this method more accurately describes the degree
is a viable alternative to the car. Many municipalities make the of land use mix in a given area.
goal of mixing land uses an explicit part of land use plans (for This research provides a new tool that can aid researchers
example: City of Montreal 2005; City of Toronto 2006; City and practitioners in studying the links between transportation
of Los Angeles 2008). In fact, land use mix plays a central role and land use. The thesis put forth in this paper is that a mea-
in major urban planning and design concepts (Smart Growth, sure that depends on proportions of various land uses in a given
New Urbanism, TOD). However, “mixed-use”, for the most area misses a key point: two areas could have identical propor-
part, remains an elusive, intangible concept. As Handy (2008) tions of land uses, yet be distributed in vastly different ways,
has shown, goals that do not have clear indicators to measure meaning that giving them a similar land use “score” could be
success, can often lose priority in the planning or implementa- extremely misleading. This work has relevance not only for
tion process to more easily measured goals. travel behavior research, but also for wider inquiry into sprawl
The link between land use mix and travel behavior has and Smart Growth, for example. As the health benefits of walk-
been widely studied. The concept is simple enough: while a ing and cycling are well documented (Rippe, Ward et al. 1988;
large, single-use residential area will often require residents to Murphy, Nevill et al. 2002), the connections to the type of de-
a
kevin.manaugh@mail.mcgill.ca
Copyright 2013 Kevin Manaugh and Tyler Kreider
Licensed under the Creative Commons Attribution – NonCommercial License 3.0.
64 JOURNAL OF TRANSPORT AND LAND USE 6.1

velopment patterns that might support these activities are vital. who used a version of this index in a classic mode choice study,
Let us take a slightly over-simplified example to intro- which found that walking, biking and transit trips were posi-
duce the concept. Imagine three different neighborhoods with tively correlated with the measure. The entropy index can be
roughly 33 percent each of residential, commercial, and park/ expressed as:
recreational land use. In the first neighborhood, a single-use
−Σ(Aij ln Aij)
residential area borders a park which, in turn, borders a large
commercial area. The second neighborhood, however, distrib- ln Nj
utes the same proportion of land uses in a more fine-grained Where:
manner. The third exhibits an even more fine-grained mix of • Aij = Percent of land use i in census tract j
uses, with almost every parcel of each land-use lying adjacent • Nj = Number of represented land uses in census tract j
to another complementary use. The assumed distance to either
a commercial and recreational destination from any given resi- Many variations exist as to how land uses are grouped.
dential location is much smaller in B or C than in the A. In For example, Frank and Pivo (1994) divided retail, office,
fact, one could argue that B is “twice as mixed” as A as the aver- entertainment, and institutional into separate groups and pe-
age distance from any residential parcel to either commercial or nalized for the existence of single family homes while giving
recreational land use is roughly half of that in A. While these positive values for multi-family homes. However, other work
would score identically in an index based solely on proportions has used more general land use categories (i.e., residential,
of land use, the transportation implications of each could be commercial, and park). This is also dependant on the avail-
drastically different. This is illustrated in Figure 1. ability of detailed land use maps. In addition, this index has
been studied at many geographical scales, including different
Residential census aggregations, metropolitan regions, and buffers around
Park
Commercial individual households.
A related measure, the Herfindahl index (sometimes Her-
findahl-Hirschfield, or HHI) has its roots in economic analysis
Figure 1. Three different mixes of land uses that would score identically in and is defined as a concentration measure. It has been cited in
an entropy mix of land uses. studies dealing with monopoly situations in the marketplace
and industry concentration (Ordover, Sykes et al. 1982). The
Therefore, the goals of this research are 1) to develop and Herfindahl index is simply the sum of squares of the propor-
present a land use interaction measure using only basic GIS tion of different component parts. For example, an area with
techniques and land use layers and 2) to evaluate this measure five equally distributed land uses (20% each) would score a
by examining its relationship with reported walking and bik- 2,000 out of a possible 10,000. An area of only one land use
ing behavior. This study presents a relatively straight-forward would score a 10,000.
GIS-based methodology that arguably captures the travel im- While the two equations vary in appearance, Van Eck &
plications of land use patterns in a more effective and mean- Koomen (2008) found that the entropy and Herfindahl index
ingful way. By taking into account the proximity and location are virtually indistinguishable when standardized and mapped.
of complementary land uses, this new measure is expected to For the sake of simplicity, we are only comparing the entropy
better explain the extent to which an area has mixed land uses. index in this study. The Herfindahl index has been used in a
number of studies, including recent research on the influence
2 Background of built form factors on pedestrian injury severity levels (Clif-
ton, Burnier et al. 2009).
As the literature on measuring land use mix is vast, this review Past research has, in fact, addressed some of the as-
will focus on the most prevalent methods in order to evaluate sumptions made in using these indices. For example
our proposed approach. One of the most common indices of Hess, Moudon et al. (2001) identify three main issues with
measuring land use mixing comes from outside the fields of ge- any mixed use measure based solely on proportions of vari-
ography and urban planning: the entropy index. This measure ous uses: 1) land use types are not differentiated—a perfect
has been in used to study biodiversity and is used in fields as mix of industrial and park might score identical to the same
varied as ecology and communication and can be traced to the proportions of residential and commercial; 2) a measure of
work of Shannon (1948). Its use in urban planning and land proportion misses any sense of interaction (whether they
use studies goes back at least as far as Frank and Pivo (1994), are adjoining or separated by barriers of some sort); and 3)
What is mixed use? Presenting an interaction method for measuring land use mix 65

reversing the proportions of land use (i.e., 30 percent residen- mixing (Brown, Yamada et al. 2009). Brownson, Hoenher et
tial and 70 percent commercial to 30 percent commercial and al. (2009) offer a thorough review of previous studies on the
70 percent commercial) would have no effect on the entropy correlations between land use mix and other neighborhood-
or Herfindahl score even though the travel implications of the level characteristics and travel behavior.
two patterns could certainly be quite different. Other authors have used simple binary measurements of
Finally, and related to the last point above, while it might land use presence. Kerr, Frank et al. (2007), in addition to a
seem to initially make intuitive sense, we question the fact that land use mix index, used dummy variables for any commer-
an even land use split is considered superior to an uneven one. cial use and any recreation/park use. Interestingly, these binary
For example, two land uses split 50/50 would score higher than variables were found to have both a higher coefficient and
a mix of 65/35. Previous work has in fact referred to this even significance level with physical activity than the land use mix
mix as a “perfect” land use mix (Rajamani, Bhat et al. 2003, measure. On the other side of the spectrum, some previous
among others). To our knowledge, the literature seems to be work has looked at completely disaggregated land uses based
lacking in a real theoretical underpinning of this key point. on parcel-level data or point-level business and retail informa-
That there is a link between land use mix and behavior has be- tion (Hess, Moudon et al. 2001). While we acknowledge that
come relatively widely accepted. For this reason, land use mix these methods can be extremely accurate, particularly in mea-
has become a central goal of municipal plans, and a key aspect suring vertical mixing of land uses, the data is often difficult
of such design paradigms as Smart Growth, New Urbanism, to access and can be data-intensive and time-consuming with
and transit-oriented development. In addition, these problem- which to work. The method proposed here uses basic land use
atic manners of measuring mix play a central role in common layers that are widely available in most academic and municipal
walkability indices (Lee and Moudon 2004; Frank, Schmid et contexts.
al. 2005; Lee and Moudon 2006) Therefore, it is absolutely It should be noted that, despite the potential issues
vital that this elusive concept is measured as accurately as pos- mentioned above, these land-use measurements have suc-
sible. cessfully been linked to behavior and health outcomes. For
A particularly good example of a partial solution to some example, researchers have found a correlation between en-
of the above issues is found in Rajamani et al. (2003). The au- tropy measures of land use mix and increased physical activity
thors utilize a series of distribution quotients that take into ac- (Kockelman 1997; Cerin, Leslie et al. 2007; Frank, Sallis et
count the ratio of each land use to the number of housing units al. 2009). A detailed review of the link between land use and
in the neighborhood. They also use a slightly modified land use transportation is provided by Badoe and Miller (2000) who
diversity index proposed by Bhat & Gossen (2004). However, found that land use mix, along with other built form variables,
this modified land use diversity index suffers from many of the is often found to significantly influence automobile use, albeit
same shortcomings by focusing on land use proportions with- sometimes with a very weak effect. They also recognize, howev-
out controlling for interaction. er, that gaps exist in this area of the literature, and make specific
Other methods exist to attempt to capture the elusive mention of weaknesses in data and methodology.
qualities of land use mix; however, as these measures, particu-
larly the entropy index, are so widely used, this brief literature
review focused on them. Kockelman (1997), in an influential 3 Methodology and data
study, used both the entropy measure and introduced a “dis-
similarity index” which values the extent to which adjoining Since our goal is to compare two measures of land use, the
land differs in use, (see also (Cervero and Kockelman 1997)). first step of this research involves generating the entropy score
However, while sharing similar concerns to our own, this mea- and the proposed interaction measure in Canada’s three larg-
sure works at an aggregate grid cell level based on the most est metropolitan regions. In order to calculate the two differ-
prevalent land use in a given cell, and therefore misses the ac- ent measures of land use mix, land use maps from Ontario,
tual extent to which the distinct uses interact. In addition, this Quebec, and British Columbia, developed by DMTI Spatial
measure does not take into account the actual type or extent (2007), were used in a geographic information system (GIS)
of mixing; it only measures whether adjoining grid cells differ which allows the measurement and manipulation of spa-
from a central cell. A study of health and obesity outcomes re- tial data. This work was undertaken using ESRI ArcMAP
lated to the walkability of respondents’ neighborhoods, found version 9.3. These maps use seven categories to describe ur-
that the presence of certain land uses at a walkable distance ban land use: residential, commercial, institutional and
explains much more behavior than the relative proportion or governmental, resource and industrial, park and recreational,
66 JOURNAL OF TRANSPORT AND LAND USE 6.1

open area, and water. This dataset was chosen for several rea- that are single-detached. Census tracts were used as they are
sons. As this data is available and consistent throughout the designed to approximate neighborhoods and therefore an ap-
country, it was seen to be an ideal source to compare both propriate size to measure land use mix, yet small enough that a
land use patterns and travel outcomes on a wide scale. Oth- sufficient number exist in each metropolitan area for statistical
er, finer scale datasets were considered for parts of the Mon- modeling with significant local variation.
treal metropolitan region, but as these were not available to In order to preserve continuity with the theory behind
us for the three regions under study, thus the more general land use mix and only compare measures that differ in method,
seven-category land use data was chosen. In addition, this three categories were defined from the land use dataset. The
broad data was used for the express purpose of demonstrating underlying assumption of this research is that shorter travel
that a more accurate measure of land use mix can be generated distances—allowing active transportation choices—can be
using readily-available data. It is recognized that different data generated by the mixing of complementary land uses. In order
sources may yield varying results. to operationalize this for this study, we have divided the seven
Additional data were collected at the census tract level in land uses into three categories. Residential was left as its own
the metropolitan regions of Toronto, Montreal, and Vancou- category; commercial, institutional and governmental, and
ver from Statistics Canada (StatsCan 2006). These variables resource and industrial were collapsed into a second category
include population density, percentage of population under 35 to represent commercial and employment destinations within
years old, percentage of dwellings built before 1946, median each census tract; and park and recreational was collapsed with
household income, and percentage of residential dwellings water into a third category to represent opportunities for leisure

Figure 2 Interaction method procedure

What is mixed use? Presenting an interaction method for measuring land use mix 67

and recreation. Open areas were removed from the dataset, as Background section. The result was a number assigned to each
unused space is not expected to contribute to land use mix. census tract ranging from 0 to 1, where 1 is a perfect mix and 0
Although this method of categorization may differ from that comprises only 1 use (or entirely open area). The standardized
used in some previous studies that measure land use mix using Z-score of this value was then calculated so that results using
the entropy index (Frank and Pivo 1994), one reason for doing the entropy method would be comparable with that using the
so was so that better land use scores would be given to census interaction method described below.
tracts containing all three categories, rather than penalizing for The interaction method proposed here uses the same land
the lack of more specific categories (such as institutional and use data as were used to generate the entropy index. The land
governmental). In addition, the three-group approach pre- use dataset is represented by polygons. These polygons are then
sented here does have a precedent (Bhat and Gossen 2004). converted into lines using basic GIS tools, representing the in-
Importantly, this approach does not count interaction between teractions between two distinct uses. The only interaction lines
industrial and government, for example, as a complementary that are preserved, however, are those between two comple-
mix. The intention is to capture those areas where the mixing mentary uses, meaning that boundaries between any use and
of residential, commercial and economic, and recreational uses open area are eliminated. The total lengths of these interaction
would be thought to lead to the potential of active transporta- lines are determined for each census tract, and this value is di-
tion. vided by the area of the census tract, minus open space. As
The entropy index was calculated using the three land was done for the entropy index, the Z-score is calculated based
uses described above. The formula used can be found in the on the mean and standard deviation of interaction length per

Figure 3 Z-scores of entropy and interaction methods for the three metropolitan areas
68 JOURNAL OF TRANSPORT AND LAND USE 6.1

Figure 4 Close-ups of two census tracts showing methodological discrepancies

area. In this manner, a census tract with residential and com- order to understand its usefulness.
mercial uses adjoining one another will score higher than one Before presenting a statistical analysis of this measure, a
with these land uses being distant. comparison map is shown to highlight the potential useful-
The series of maps shown in Figure 2 present a general ness of this approach. Figure 4 shows two census tracts in the
image of the steps taken in generating the interaction measure. Montreal region: the first scores high on the entropy measure
The uppermost image shows a close up of a single census tract and quite low on the interaction method, while in the second
in Montreal with the original land use data. The middle image map, the opposite is true (Note: Z-scores are shown; these es-
shows these land uses dissolved into three general categories: sentially measure the number of standard deviations away from
residential; commercial, government and institutional, and the mean, in other words, the uppermost area scores well below
resource and industrial; and park and water. The final image average using the interaction approach and well above average
shows both the interaction between complementary land uses using the entropy method). While the top does have a mixing
and the land use boundaries that were discarded. In this way, of land uses, both their type and geographic location would
the interaction between residential land use and commercial arguably not have the same travel implications as in the lower
activity is counted, while residential land use adjoining open figure where, although the proportion of land uses might not
space is not. be even, their proximity to one another could have a major
effect on daily travel behavior. For example, nearly every resi-
4 Initial findings dential location in this census tract is within an 800 meter walk
of a commercial land use, while in the first figure, residential
Using the Z-score of the entropy index and the land use mix land is separated from the park by a large institutional land use.
value generated from the interaction method, six chlorop- Importantly, the lack of actual interaction between these large
leth maps were generated to visually compare the outcomes areas of institutional, residential, and park is reflected in the
from each method in Toronto, Montreal, and Vancouver. low score in the interaction method, but the entropy measure
The method used for classification in the maps was quin- overestimates the mixing of uses by only using the relative pro-
tile (five equal classes). Thus, the maps are easily comparable portions of land use.
across measures and regions. As can be seen in Figure 3, the
two maps vary dramatically. The entropy index is displayed 5 Statistical application
in a seemingly random manner, with many census tracts in
the far peripheries of the cities having very high entropy index 5.1 Model Specification
scores. Initial visual inspection seems to confirm the usefulness
of the interaction approach, as the left hand side agrees with In order to test the validity of the proposed method of measur-
subjective experience of the three cities. This initial mapping ing land use mix (the interaction method), a series of simple
seems to confirm some of the issues present in the entropy OLS models are designed. The unit of analysis is the census
measure, however, this will need to be statistically validated in tract as defined in 2006, and the dependent variable in these
What is mixed use? Presenting an interaction method for measuring land use mix 69

models is the percentage of people who use walking or bicy- ploys the entropy index, and the third model introduce the
cling as their primary mode of transportation to work. While interaction method. With these nine models, it is possible 1) to
many studies have studied the expected negative effect of land identify the differences in explanatory power between entropy
use mix on automobile use (see Badoe and Miller (2000) for and interaction, and 2) to observe the differences among the
a review), walking and biking were chosen in this model so three metropolitan areas under study in determinants of travel
that transit accessibility would not be a necessary variable in behavior. It should be noted here that all variables are included
the models. This allows for models that utilize only data readily in all models even if a better model could be achieved by omit-
available from the Canadian census. The independent variables ting variables. This is in order to show the differences between
chosen for the model are a group of variables that have been cities in terms of what affects travel behavior.
shown in the literature to be associated with travel behavior at
the neighborhood scale. Table 1 Descriptive Statistics of Model Variables
Population density is measured as the number of people per Variable Toronto Montreal Vancouver
square meter. Age is represented by percentage of the popula- x s x s x s
tion under 35 years of age. This is chosen to account for ar- Walk/bike (%) 6.03 7.88 9.03 8.95 6.40 7.33
eas that have a high amount of young people, which could Pop. Density 0.0050 0.0050 0.0057 0.0052 0.0041 0.0041
be expected to increase levels of walking and cycling. Socio- (/m2)
<35 years (%) 44.48 6.45 39.84 6.04 42.72 5.91
economic status is measured using median household income,
Pre-WWII (%) 11.15 20.04 15.29 20.05 7.67 11.76
measured in units of $10,000 CAD. Finally, two additional
Income 6.95 2.59 5.12 2.22 6.02 1.83
urban form variables (aside from land use mix) are examined. ($10,000)
The first is percentage of housing stock that was built before Single-detached 44.79 29.60 14.34 14.64 41.06 27.31
1946, which serves as a proxy for neighborhood characteristics (%)
such as street grid pattern. The other is the percentage of hous-
ing stock that is single-detached, a variable often used in travel In order to show differences amongst the three regions under
behavior studies (Boarnet and Greenwald 2000) and here hy- study, some descriptive statistics for the population of each
pothesized to lead to decreased active transportation. This data metropolitan region as a whole are reported in Table 1. The de-
is obtained from the 2006 census of Canada (StatsCan 2006). pendent variable, the percentage of people who regularly walk
It is recognized that the dependent variable used in this paper is or bike to work, is highest in Montreal at 9% of the popu-
not ideal as a measurement for the amount that people actually lation, and just over 6% in Toronto and Vancouver. Single-
walk or bike, but it is used here due in part to lack of data other detached homes are over 40% in the latter two cities, while
than those from the census but also to demonstrate the appli- only 14% in Montreal. Income is also significantly higher in
cability of land use mix in a model using only readily available Toronto and Vancouver than in Montreal. Vancouver has the
data. The literature has shown, in fact, that neighborhood-level lowest population density and amount of pre-WWII housing,
land-use mix could have more of an effect on non-work trips. while Montreal has the highest of both. Finally, the percent
However, as the intention is to show a large-scale application of people under 35 years is similar in all three cities, ranging
of the new land-use mix measure, using widely available census between 40% and 44%.
data was seen as the best solution and future work is planned
to test this measure on a smaller scale using, for instance, actual 5.2 Results
pedestrian counts on individual street segments and origin-
destination surveys. In addition, other travel outcomes, such as The results from the nine models described in the previous sec-
the percentage of people who take transit, are highly correlated tion are reported in Table 2, grouped first by city and then by
with transit accessibility and reliability and were seen to be out land use mix variable. The R2 values, displayed at the bottom of
of the scope of this study. the models, show that the degree to which variation in the pro-
To examine the effect of the two land use mix variables on portion walking and biking is explained changes only slightly
modeling the degree to which people walk or bike, the en- when entropy is added to the base model (0.67%, 0.87%, and
tropy index and interaction method are used for the three 3.47% increases for the three cities, respectively), but dras-
most populous metropolitan areas in Canada: Toronto, tically when interaction is used as the land use mix variable
Montreal, and Vancouver. Three models are built for each of the (26.05%, 11.36%, and 12.11% increases). The changes in t-
cities. The first model uses the dependent variables described statistics for the three models are also very telling. While the
above without a land use mix variable, the second model em- entropy index is always statistically significant, its prominence
70 JOURNAL OF TRANSPORT AND LAND USE 6.1

in the model is markedly low. The interaction variable, on the lation density shows a marked decrease in explanatory power in
other hand has the highest t-statistic in Toronto and Vancou- all three cities, though it is still fairly powerful in the Vancouver
ver of all the independent variables, and is second-highest in model. Another interesting finding is that the percentage of
Montreal. The coefficients are also much higher for interaction people under 35 years of age has the expected effect in Mon-
than for entropy, which are made comparable by the use of treal (albeit statistically insignificant), with more young people
their Z-scores rather than their actual values. contributing to more walking, while in Toronto and Vancou-
Furthermore, what can also be drawn from the models is the ver, the effect is the opposite. This peculiarity disappears in To-
difference in what independent variables are important be- ronto, however, when the interaction variable is used. While
tween the three cities. In terms of land use mix, the interaction there may be a way of explaining this, the variable was insig-
measure improves the model fit most drastically in Toronto nificant in Montreal and Toronto, and does not seem to be a
while in Montreal the difference is the lowest in all aspects very powerful indicator of walking and biking. The percentage
(though it should be mentioned that interaction is still an im- of single-detached houses has the strongest effect in Toronto,
provement over the entropy index in Montreal). Aside from but is relatively weak in Montreal. Similarly, income is only a
land use mix, the most important variable in Toronto, Mon- powerful variable in Montreal, where lower income areas are
treal, and Vancouver in explaining walking and biking is the indicative of more walking and biking.
percentage of houses built before 1946 (World War II). Popu-
Table 2 OLS models on percentage active transport as primary mode for work trips at
the census tract level
Toronto
Variable No mix variable Entropy Interaction
β t β t β t
Pop. Density (km2) 2.48 4.79 2.74 5.15 0.93** 1.86
<35 years -0.039 -1.13 -0.025 -0.73 0.049 1.50
Pre-WWII 0.14 12.65 0.14 12.55 0.10 9.70
Income ($10,000) 0.0017 1.5 0.0017 1.56 0.0023* 2.20
Single-detached -0.081 -7.34 -0.074 -6.45 -0.043 -4.04
Entropy (Z) - 0.0046* 2.04 -
Interaction (Z) - - 0.031 12.01
Constant 0.074 4.33 0.063 3.55 -0.068 -3.46
Adjusted R2 0.3271 0.3293 0.4123
Montreal
Variable No mix variable Entropy Interaction
β t β t β t
Pop. Density (km2) 2.93 4.71 3.04 4.89 1.04** 1.69
<35 years 0.0079 0.2 0.015 0.38 0.052 1.39
Pre-WWII 0.21 15.55 0.21 15.05 0.16 10.85
Income ($10,000) -0.0066 -4.8 -0.0065 -4.77 -0.0047 -3.58
Single-detached -0.034 -1.31 -0.018 -0.67 0.010 0.42
Entropy (Z) - 0.0068 2.83 -
Interaction (Z) - - 0.033 10.08
Constant 0.076 4.66 0.071 4.33 0.062 3.99
Adjusted R2 0.4815 0.4857 0.5362
Vancouver
Variable No mix variable Entropy Interaction
β t β t β t
Pop. Density (km2) 10.09 13.36 10.52 14.13 3.46 3.37
<35 years -0.1* -2.44 -0.078** -1.94 -0.060 -1.59
Pre-WWII 0.15 7.3 0.15 7.46 0.14 7.13
Income ($10,000) -0.0016 -0.79 -0.0014 -0.72 -0.00068 -0.37
Single-detached -0.036* -2.38 -0.019 -1.27 -0.040 -2.88
Entropy (Z) - 0.011 4.48 -
Interaction (Z) - - 0.033 8.74
Constant 0.078 3.91 0.06 2.97 0.086 4.64
Adjusted R2 0.5655 0.5851 0.6340
Statistically significant at 99%, *95%, **90%
What is mixed use? Presenting an interaction method for measuring land use mix 71

6 Discussion and conclusions References

This work presents a first step towards conceptualizing a new Badoe, D. A. and E. J. Miller (2000). “Transportation-land-
manner of measuring land-use mix that is arguably more ap- use interaction: Empirical findings in North America, and
plicable to land use, transportation, and public health stud- their implications for modeling.” Transportation Research
ies than those currently in wide use. The results clearly show Part D: Transport and Environment 5(4): 235-263.
that, while controlling for census-tract level demographic and Bhat, C. and R. Gossen (2004). “A mixed multinomial log-
built form factors, land use interactions explain more variation it model analysis of weekend recreational episode type
in walking and cycling behavior than existing methods. This choice.” Transportation Research Part B 38(9): 767-787.
suggests that the focus that the entropy index places on bal- Boarnet, M. and M. Greenwald (2000). “Reproducing other
ance of land uses is misplaced, and that equal proportions of urban areas’ empirical test results in Portland, Oregon.”
uses are somewhat arbitrary in predicting travel outcomes. The Transportation Research Record(1722): 27-37.
interaction method, on the other hand, is useful in capturing Brown, B., I. Yamada, et al. (2009). “Mixed land use and
the degree to which complementary land uses actually meet, walkability: Variations in land use measures and relation-
which was shown in this paper to have a greater effect on the ships with BMI, overweight, and obesity.” Health and Place
explanatory power of land use mix when modeling walking 15(4): 1130-1141.
and cycling. Brownson, R. C., C. M. Hoehner, et al. (2009). “Measuring
The subtlety of measuring land use mix is a complex and im- the built environment for physical activity: State of the sci-
portant topic that, though widely studied, has much room for ence.” American Journal of Preventive Medicine 36(4): S99-
improvement. One important goal for land use and transpor- S123.
tation planning is finding measures that more accurately and Cerin, E., E. Leslie, et al. (2007). Applying GIS in physical
efficiently measure land use mix for particular travel outcomes activity research: Community ‘walkability’ and walking be-
or trip purposes. Perhaps specific land use mixes would impact haviors. GIS for Health and the Environment: Development
travel for different trip purposes. For example, while the pres- in the Asia-Pacific Region With 110 Figures. P. C. Lai and A.
ence of parks might not impact work trips, it might play an S. H. Mak. Berlin, Springer.
important role in generating leisure and social trips. Cervero, R. and K. Kockelman (1997). “Travel demand and
In addition to using more relevant travel outcomes, such as the 3D’s: Density, diversity and design.” Transportation Re-
pedestrian counts and non-work walking trips from origin/ search Part D: Transport and Environment 2(3): 199-219.
destination surveys, further refinement of the actual technique City of Los Angeles (2008). Long range transportation plan.
could be explored. For example, certain interactions (e.g., resi- City of Montreal (2005). Montreal Master Plan.
dential/commercial) might yield a greater impact on travel out- City of Toronto (2006). Toronto Official Plan.
comes than others (e.g., industrial/water). Weighting different Clifton, K., C. Burnier, et al. (2009). “Severity of injury re-
types of interactions could increase the explanatory power of sulting from pedestrian-vehicle crashes: What can we learn
the interaction method in these types of models. It is hoped from examing the built environment? .” Transportation Re-
that this analysis will add to the ongoing discussion and re- search Part D: Transport and Environment 14: 425-436.
finement of this essential topic across the many fields that it DMTI Spatial (2007). CanMap® Streetfiles. Markham,
touches. DMTI Spatial.
Frank, L. and G. Pivo (1994). “Impacts of mixed use and den-
7 Acknowledgements sity utilization of three modes of travel: single-occupant ve-
hicle, transit, and walking.” Transportation Research Record
The authors wish to acknowledge Prof. Ahmed El-Genei- 1466: 44 - 52.
dy and Prof. Nancy Ross for assisting with the paper’s frame- Frank, L. and G. Pivo (1994). “Impacts of mixed use and den-
work and for providing useful feedback on the direction of the sity utilization of three modes of travel: single-occupant
research. This research was partly funded by the Fonds québé- vehicle, transit, and walking.” Transportation Research Re-
cois de recherche sur la société et la culture (FQRSC). Thanks cord(1466): 44 - 52.
also to the anonymous reviewers whose input greatly improved Frank, L. D., J. F. Sallis, et al. (2009). “The development of a
the paper. walkability index: Application to the neighborhood qual-
ity of life study.” British Journal of Sports Medicine 44(13):
924-933.
72 JOURNAL OF TRANSPORT AND LAND USE 6.1

Frank, L. D., T. L. Schmid, et al. (2005). “Linking objectively Wilkinson, R. and M. Marmot (2003). The Solid Facts:
measured physical activity with objectively measured urban Social Determinants of Health, World Health Or-
form: findings from SMARTRAQ.” American Journal of ganization.
Preventative Medicine 28(2): 117 - 25.
Handy, S. (2008). “Regional transportation planning in the
US: An examination of changes in technical aspects of the
planning process in response to changing goals.” Transport
Policy 15(2): 113-126.
Hess, P. M., A. V. Moudon, et al. (2001). “Measuring Land
Use Patterns for Transportation Research.” Transportation
Research Record: Journal of the Transportation Research Board
1780: 17-24.
Kerr, J., L. Frank, et al. (2007). “Urban form correlates of pe-
destrian travel in youth: Differences by gender, race-ethnic-
ity and household attributes.” Transportation Research Part
D: Transport and Environment 12(3): 177-182.
Kockelman, K. M. (1997). “Travel behavior as function of ac-
cessibility, land use mixing, and land use balance: Evidence
from San Francisco Bay Area.” Transportation Research
Record: Journal of the Transportation Research Board 1607:
116-125.
Lee, C. and A. V. Moudon (2004). “Physical activity and envi-
ronment research in the health field: Implications for urban
and transportation planning practice and research.” Journal
of Planning Literature 19(2): 147-181.
Lee, C. and A. V. Moudon (2006). “Correlates of Walking for
Transportation or Recreation Purposes.” Journal of Physical
Activty and Health 3: S77 - S98.
Murphy, M., A. Nevill, et al. (2002). “Accumulating brisk
walking for fitness, cardiovascular risk, and psychologi-
cal health.” Medicine & Science in Sports & Exercise 34(9):
1468-1474.
Ordover, J., A. Sykes, et al. (1982). “Herfindahl concentration,
rivalry, and mergers.” Harvard Law Review 95(8).
Rajamani, J., C. R. Bhat, et al. (2003). “Assessing impact of
urban form measures on nonwork trip mode choice after
controlling for demographic and level-of-service effects.”
Transportation Research Record: Journal of the Transportation
Research Board 1831: 158-165.
Rippe, J., A. Ward, et al. (1988). “Walking for health and
fitness.” The Journal of the American Medical Association
259(18): 2720-2724.
Ritsema Van Eck, J. and E. Koomen (2008). “Characterising
urban concentration and land-use diversity in simulations
of future land use.” Annals of Regional Science 42: 123-140.
Shannon, C. (1948). “A mathmatical theory of communica-
tion.” Bell System Technical Journal 27: 379-423.
StatsCan. (2006). “Canadian Census.” from www.chass.
utoronto.ca.