The economic choice of participation and time spent in physical activity and sport in Canada.
Humphreys, Brad R. ; Ruseski, Jane E.
Background and Motivation
Physical activity (including sport participation) is essential to
overall health. Regular physical activity can reduce the risk for many
chronic diseases including breast and colon cancer, diabetes, stroke,
heart disease, hypertension, and depression (World Health Organization,
2010). Given the importance of physical activity to overall health,
promoting regular participation in physical activity and sport is a
public health priority in many countries. In Canada, the country of
interest in this study, the prevalence of meeting physical activity
guidelines is improving but still remains low. A comparison of the
results from the 1996-1997 National Population Health Survey (NPHS) with
those from the 2005 Canadian Community Health Survey (CCHS) shows that
Canadians who report at least moderately active leisure time physical
activity rose from 43% to 52% (Gilmour, 2007). Based on data from the
CCHS, the percentage of Canadians who reported being either physically
active or moderately physically active rose a modest 2 percentage
points, from 52% in 2005 to 54% in 2011 (Statistics Canada, 2012).
Despite this improvement of participation rates based on self-reported
data, an alarmingly low percentage (15%) of adult Canadians (over age
20) meet the guidelines for sufficient physical activity based on
objective accelerometer data (Colley et al., 2011). (1) This statistic
is a motivating factor for the recently renewed Canadian Sport Policy
goal that both the number and diversity of Canadians participating in
sport will increase over the timeframe of 2012-2022 (Canadian Sport
Policy, 2012). Central to achieving such policy objectives is an
understanding of the determinants of participation in sport and physical
exercise. The objective of this study is to investigate how changes in
key economic variables (income, wage, education, and occupation),
individual characteristics (age and gender), and family structure
(marital status and presence of children) affect individual decisions
about participation and time spent in physical activity and sport.
Perhaps in response to the nearly global policy priority of
encouraging regular exercise for health benefits, a large literature in
health services research, public health, and, more recently, economics
that examines physical activity and sport participation has emerged. The
recent empirical literature in economics can be loosely placed into
three categories: 1) analyses of the determinants of physical activity
and sport (Brown & Roberts, 2011; Downward, 2007; Downward,
Lera-Lopez, & Rasciute, 2011; Downward & Riordan, 2007; Eberth
& Smith, 2010; Farrell & Shields, 2002; Garcia, Lera-Lopez,
& Srnrez, 2011; Humphreys & Ruseski, 2007; Humphreys &
Ruseski, 2011; Lera-Lopez & Rapun-Garate, 2007; Meltzer & Jena,
2010); 2) analyses of the impact of physical activity and sport on
health-related factors like self-assessed health status, health
outcomes, and health care utilization (Balia & Jones, 2008;
Contoyannis & Jones, 2004; Costa-Font & Gil, 2005; Humphreys,
McLeod, & Ruseski, 2014; Ruseski & Humphreys, 2011; Sari, 2009;
Sarma, Devlin, Gilliland, Campbell, & Zaric, 2013); and 3) the
impact of physical activity and sport participation on other factors,
like labor market outcomes and happiness (Forrest & McHale, 2011;
Huang & Humphreys, 2012; Kavetsos, 2011; Lechner, 2009; Lechner
& Downward, 2013; Pawlowski, Breuer, & Leyva, 2011; Pawlowski,
Downward, & Rasciute, 2011; Rasciute & Downward, 2010).
We restrict our attention to summarizing the findings from the
first category of empirical studies since our study contributes to this
category by developing additional evidence on the economic determinants
of sport participation using data from the CCHS Cycle 1.1. These studies
consistently find that participation in sport is affected by age,
education, household income, household structure, and ethnicity. Most
studies using cross-sectional data find that the probability of
participation in any kind of activity declines with age. Education and
income are consistently positively associated with participation in any
type of activity.
Women and married people are less likely to participate than males
and singles. The presence of children in the household has mixed effects
depending on the type of activity and the dimension (participation or
time spent) of physical activity studied. This pattern slightly changes
when one moves from overall participation to specific activities. Also,
the importance of income differs across studies. Findings with respect
to employment status are mixed. For example, Downward (2007) finds that
unemployed and part-time employed people are more likely to exercise
than employed and full-time employed people, while Humphreys and Ruseski
(2007) find that employed people are more likely to participate in
physical activity and sport. (Downward et al. [2011] and Garcia et al.
[2011] provide excellent summaries of the empirical evidence on the
determinants of sport participation.)
An important economic factor that is not as widely studied is the
opportunity cost of time. Exceptions include Anokye et al. (2011), Brown
and Roberts (2011), Christian (2009), Garcia et al. (2011), Humphreys
and Ruseski (2011), Kuvaja-Kollner et al. (2012), Maruyama and Yin
(2012), Meltzer and Jena (2010), and Pawlowski et al. (2009). Pawlowski
et al. and Anokye et al. examine the effect of travel time costs on
participation while Christian studies the impact of commuting time costs
on time spent in physical activity and other non-work activities.
Kuvaja-Kollner et al. use labor market position (working or retired) as
the opportunity cost of time in a study of the effect of the time cost
of physical exercise on the amount of physical activity undertaken.
Meltzer and Jena and Maruyama and Yin use income as a proxy for the
opportunity cost of time to evaluate the effect of earnings on exercise
intensity. Our paper adds to this limited group of studies by using
information in the CCHS to construct a wage variable. The wage variable
is used as a proxy for the opportunity cost of time in our study, which
allows for a more direct evaluation of the effect of the opportunity
cost of time on the frequency and duration of sport participation. Our
findings about the effect of the opportunity cost of time extends the
analyses in Brown and Roberts (2011), Garcia et al. (2011), and
Humphreys and Ruseski (2011). Brown and Roberts and Garcia et al. also
construct or derive wage variables as measures of the opportunity cost
of time. Humphreys and Ruseski do not have the data to construct a wage
variable and use education as a proxy for the opportunity cost of time.
We estimate a double hurdle model of the decisions to participate
and time spent in leisure time sports activities. The theoretical
framework motivating the empirical analysis is the economic model of
participation and time spent in physical activity developed by Humphreys
and Ruseski (2011). This model emphasizes that decisions about physical
activity are made on two margins: the extensive margin governing the
participation decision, and the intensive margin governing the time
spent decision, conditional on participation. This distinction has
implications for the way observed correlates of practicing sport affect
the participation and time spent decisions.
This model has been used to motivate the empirical analysis in some
recent economic studies of sport participation and physical activity
(Brown & Roberts, 2011; Eberth & Smith, 2010; Eisenberg &
Okeke, 2009; Lera-Lopez & Rapun-Garate, 2007). Our study is
primarily an extension of Humphreys and Ruseski (2011) but differs from
it and adds to the larger existing literature in four important ways.
First, it uses data from Canada rather than the United States. Canada
differs from the United States with respect to national sport policy.
Canada's federal Department of Canadian Heritage is concerned with
sport participation at the grassroots level as is Sport Canada (Doherty
& Clutterbuck, 2013). On the other hand, the United States has no
federal counterpart and the promotion of widespread sport participation
is left to state and local organizations (Ruseski & Razavilar,
2013). Second, it analyzes the participation and time spent decisions
for specific sports rather than examining participation and time spent
in any activity. An analysis of different sports and physical activities
allows us to assess if and how the determinants of sports and physical
activity participation vary across activities. Other studies that look
at the frequency and/or duration in specific activities include Downward
(2007), Downward and Riordan (2007), Eberth and Smith (2010), Farrell
and Shields (2002), and Humphreys and Ruseski (2007). Third, the CCHS
allows us to use wages, rather than some other proxy like education or
income to measure the opportunity cost of time. Finally, the CCHS data
reflect the fact that many Canadians choose not to participate in sport
or be physically active. Our econometric approach is to account for the
large number of zeros observed in the measures of physical activity by
estimating a "full double hurdle" model (Jones, 2000) of
participation and time spent practicing sport. The "full double
hurdle" model allows for factors that affect participation and
factors that affect time spent to have different signs and for
correlation in the equation error terms. Most empirical studies of
physical activity and sport participation do not employ this empirical
approach. For example, Downward and Riordan (2007) and Humphreys and
Ruseski (2007) both adopt the Heckman self-selection approach to
estimate the participation and time spent equations. Humphreys and
Ruseski estimate a Cragg model. (2) Garcia et al. (2011) first estimate
a probit model for sport participation and then estimate a linear system
of demand for sport and other leisure time activities using seemingly
unrelated regression. Eberth and Smith (2010) use the "copula
approach" to model the participation and time spent decisions.
We focus our analysis on seven sports and physical activities:
walking, swimming, exercising at home, cycling, running, golfing, and
weight lifting. We choose these sports because they have the highest
participation rates. We find that individuals with higher income are
more likely to participate in these activities but, conditional on
participating, spend less time. This finding is important even though
numerous studies have found a positive correlation between income and
physical activity because it suggests that the income effect works
differently on the extensive and intensive margins. Our model also shows
that the effect of a change in the opportunity cost of time can be
decomposed into an income and substitution effect just as a change in
the price of a good can be decomposed in this manner. These effects work
in opposite directions and are empirically testable. Our results
generally suggest that the income effect of a change in the opportunity
cost of time dominates the substitution effect. Our findings with
respect to the effect of income and the opportunity cost of time are
consistent with the findings of Humphreys and Ruseski (2011) using US
data. These results can help inform the design of policy interventions
aimed at increasing participation in physical activity. Finding a
positive income effect on the extensive margin suggests that consumers
will respond to economic incentives to initiate sports programs.
Theoretical Framework
The theoretical model motivating our econometric analysis is the
economic model of participation and time spent in sport developed by
Humphreys and Ruseski (2011), which is summarized here. In this model,
the decision governing participation in sport and physical activity is a
two-part decision. First an individual must decide to participate; for
example, to go for a walk or go for a swim. Second, having made that
decision, the individual must determine how much time to spend walking
or swimming.
The mechanisms underlying these two separate, but related,
decisions are not explicitly considered here but could potentially be
handled in an extension to this model. Intuitively, the nature of the
activity should influence how choices are made along the participation
(or extensive) margin and the time spent (or intensive) margin. For
example, playing a round of golf entails getting a tee time at a golf
course and possibly coordinating with friends in order to have a
foursome. The decision to play a round of golf is likely affected by
different factors than the decision to go for a walk that essentially
entails putting on walking shoes and stepping out the door.
Individuals maximize utility by allocating time to participation in
sport and all other activities (such as sleeping, sedentary leisure,
working for pay, and working at home) and purchasing a bundle of goods
and services subject to time and budget constraints. The utility
function is U(a, t, z) where a represents the individual's decision
to participate in sport; t is the amount of time spent per episode of
sport activity, conditional on participation; and z represents the
individual's decision to engage in all other activities, including
work.
The budget constraint is Y = [F.sub.a] + [c.sub.a] at + [c.sub.z]z
where Y is money income; [F.sub.a] is the fixed cost of engaging in
physical activity; [c.sub.a] is the variable cost associated with
engaging in sports; and cz is the cost all other goods and services. The
budget constraint includes both fixed and variable costs associated with
participating in sports. The fixed costs are one-time costs incurred to
participate in sports but do not depend on how many times the
individuals participate, such as the yearly membership fee at a golf
club. Variable costs are costs that depend on the amount of time or the
number of times the individual engages in physical activity, such as a
golf coach's fee.
The time constraint is [[T.sup.*] = at + [theta]z where [T.sup.*]
is the time available for consumption activities such as sports and
[theta] is time spent consuming z. [T.sup.*], t, and [theta] are
measured in the same units such as hours. If T is the total time
available for work and all other activities, then [T.sup.*] = T - h
where h is time spent working. If individuals can choose the amount of
hours they work, then h is endogenous and wage earnings w can be
expressed as follows: wh = w(T - at - [theta]z) where wages are shown in
terms of total time available (T) and time spent in activities other
than work. Any time spent not working reduces earnings; thus, w can be
viewed as the opportunity cost of time spent engaged in non-work
activities. The full budget (or income) constraint is [y.sub.0] + wT =
[F.sub.a] + [p.sub.a] at + [p.sub.z]z where [y.sub.0] is exogenous
income; wT is potential income if individuals spend all of their time
working; pa = [c.sub.a] + w is the full cost of participating in sports
activities; and [p.sub.z] = [c.sub.z] + [theta]w is the full cost of
participating in other activities. Notice that the full budget
constraint includes w, the opportunity cost of time.
Consumers choose a, t, and z to maximize utility subject to the
full budget constraint. Recall that z encompasses all other activities
that an individual chooses to spend time doing including work in the
labor force and work in the home. In this sense, work is part of the
choice set for all individuals. The first order conditions describing
the utility maximizing choices of a, t, and z and the comparative static
analysis of the effect of changes in income and the opportunity cost of
time (measured by wages) on sport participation decisions are provided
in the technical appendix in Humphreys and Ruseski (2011). As is the
case in any comparative static analysis, the effect of changes income
and wages on the participation and time spent decisions are analyzed
holding all other inputs and their respective prices constant.
Consider first the effect of a change in income on the
participation and time spent decisions. The direction of the effect of a
change in income on both the participation and time spent decisions is
ambiguous. In both cases, it depends on the relationship between the
marginal utility of participating (or time spent) in physical activity
and the marginal utility from other non-leisure activities like meals or
watching television. Next consider the effect of a change in the
opportunity cost of time on the participation and time spent decisions.
In this case, the comparative static expressions have two components
that are analogous to, but not as straightforward as, the income and
substitution effects of a change in the price of a market good. It is
important to note that this income effect results from decomposing the
overall effect of a wage change into the income and substitution effects
of that change. It is different from the income effect arising from a
change in income.
In the standard consumer theory model of product demand,
individuals purchase products. An increase in the price of a good
effectively decreases the consumer's real income and, therefore,
purchasing power, so we expect the consumer to purchase less of the
good. In addition, the income effect of the price change is greater as
the importance of the good in the consumer's budget increases.
Although similar, the income effect of a change in the opportunity cost
of time is more complex because it involves a labor-leisure trade-off
and because it has the opposite effect on the consumer's real
income. This occurs because an increase in the opportunity cost (or
price) of time means a higher wage and an increase in real income. If
sport participation is a normal good, then we would expect the income
effect of an increase in the opportunity cost of time to be positive
because we would expect individuals to trade-off labor for leisure in
response to an increase in real income. On the other hand, the
substitution effect is negative, which means that the likelihood of
participating in sport decreases as the opportunity cost of time
increases. The comparative static predictions of the model are ambiguous
because the substitution and income effect move in opposite directions.
The effect of a change in the opportunity cost of time on both
participation and time spent in sport is positive if the income effect
dominates the substitution effect or negative if the substitution effect
dominates the income effect.
In summary, the model motivating our empirical analysis (taken from
Humphreys and Ruseski [2011]) describes consumers' decisions about
participating in physical activity and sport and time spent for all
other activities, including work. Decisions about sport participation
are affected by changes in the relative price of time and market goods.
The signs of comparative static expressions of the effect of changes in
income and the opportunity cost of time on participation and time spent
in sports activities cannot be theoretically determined. We now turn to
empirically analyzing the effect of economic factors and individual
characteristics on decisions about participating and time spent in
sports and physical activities.
Econometric Analysis of Participation and Time Spent in Sports and
Physical Activities
Data Description
We use data from the CCHS Cycle 1.1 Public Use Microdata File
(PUMF) in the empirical analysis. The survey is a cross-sectional survey
that includes information on health status, health care utilization, and
health determinants for a nationally representative sample of Canadians.
Until recently, the CCHS operated on a two-year cycle. The CCHS Cycle
1.1 was conducted between September 2000 and November 2001 and included
persons aged 12 or older. Seasonal effects were eliminated by randomly
dividing the sample to ensure that each month of the year was properly
represented for each region of the country. (Statistics Canada, 2002).
The survey includes data on leisure time physical activity (primarily
sports activities), work-related physical activity, smoking and drinking
habits, eating habits, chronic conditions, general health status, and
health care utilization. The survey also includes data on demographic
factors like age, gender, marital status, ethnicity, and household
composition, and on economic factors like income and labor market
participation. This makes the CCHS data an ideal setting for analyzing
the effect of economic factors and individual characteristics on
participation in sports.
The CCHS Cycle 1.1 survey included 130,880 people. The questions
about participation in different physical activities specify leisure
time. The basic physical activity question in the CCHS survey is:
Have you done any of the following in the past three
months?--Walking for exercise, gardening, swimming, bicycling, popular
or social dance, home exercises, ice hockey, ice skating, inline
skating, jogging or running, golfing, exercise class or aerobics,
downhill skiing, bowling, baseball or softball, tennis, weight training,
fishing, volleyball, basketball, other or no activity.
Respondents could indicate up to three "other"
leisure-time physical activities. We initially define participation in
sports and physical activities using this survey question. The CCHS asks
further questions about the number of times individuals participated in
the various physical activities and how much time (in minutes) they
spent per episode. The question asking about frequency of participation
is:
In the past three months, how many times did you activity--e.g.
walk for exercise?
The question about duration elicits an approximation of about how
much time individuals spent on each episode of reported physical
activity. The possible response categories are: 1 = 1 to 15 minutes; 2 =
16 to 30 minutes; 3 = 31 to 60 minutes; 4 = more than one hour. These
data provide enough detail to construct an estimate of the total time
spent participating in sports activities in the past three months. We
constructed a measure of minutes spent per episode by setting each
categorical response to the mid-point of the range as follows:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII]
Table 1 contains summary statistics on frequency, participation
rates, number of participation episodes, and minutes spent per episode
for the sample of adults used in our econometric analysis. The analysis
sample contains 99,322 observations after dropping respondents under the
age of 18, accounting for missing values, and dropping observations with
unrealistically high (greater than $500 per hour) or low (less than the
minimum wage in the respondent's province) hourly wages.
Walking is by far the most frequent activity with 62.51% of the
sample reporting walking for exercise at least once in the past three
months. Participating in more than one of these physical activities is
relatively common in the CCHS. Table 2 shows the number of activities
that respondents reported participating in during the past three months.
Approximately 63% of the sample reported participating in multiple
activities. Participating in more than four activities is relatively
uncommon.
Description of Sample and Variables
Since we are interested in examining the economic factors and
individual characteristics that affect the decisions about participation
and time spent in sports activities, we use a subsample of the CCHS
Cycle 1.1 in our empirical analysis. First, Table 1 shows considerable
heterogeneity in the types of physical activities individuals
participated in, participation rates, number of episodes, and time spent
per episode. An empirical analysis using aggregated data for all
possible activities will mask potentially important variation in the
effect of economic factors and individual characteristics on decisions
about participation and time spent in sports and physical activities. It
may well be the case the effect of income on the decision to run is
different than its effect on the decision to ski since running entails
few monetary costs while skiing is relatively expensive. Rather than
make ad hoc decisions about how to group activities, we focus our
empirical analysis on adult participation in seven of the most common
activities that are clearly sports or physical activities: walking,
swimming, cycling, running, home exercise, golf, and weight lifting.
Figure 1 shows the frequency of participation and average amount of
time spent in the CCHS for these seven activities. The seven activities
displayed in Figure 1 differ in important ways that might affect
participation and time spent. About one third of the sample reported
participating in at least one of these seven activities at least one
time in the past three months. The number of times that each participant
reported taking part in these activities exhibits quite a bit of
variation. Home exercise incorporates a wide variety of exercise
activities that can be done in the home, like running on a treadmill or
doing yoga. Walkers and home exercisers participated the most
frequently, and swimmers and golfers the least frequently. On the other
hand, walkers and home exercisers spend less time per episode of
activity than swimmers and golfers. These differences in frequency of
participation and time spent likely reflect differences in the total
cost of participation in each activity. Home exercise does not require
leaving the house, and can be done in any weather. For most individuals,
swimming and golfing require travel to a pool or golf course and paying
a fee to participate, thereby raising the total cost of participation.
Golfing is also time intensive as it takes a few hours to play 18 holes.
Cycling, weight lifting, and running frequency falls in between these
two extremes. Cycling and running require going outside and also require
some equipment.
[FIGURE 1 OMITTED]
Figures 2 and 3 illustrate the frequency of participation
distribution for the activities in this sample. For purposes of
constructing these figures, walking, home exercise, and weight lifting
are grouped together as physical activities and running, cycling,
swimming, and golfing are grouped together as sports activities. Note
that the distribution of frequency of participation reported in the
sample shows considerable skew for the four sports activities. Most of
participants report participating only a few times in the previous three
months, but a small number of participants report very frequent
participation. Taking swimming as an example, Figure 3 shows that 60% of
the swimmers in the sample reported swimming 10 or fewer times in the
past three months. A small number of swimmers report participating 60 to
90 times over the past three months, which corresponds to daily, or
nearly daily participation. It is also possible that the frequency
distributions of participation reflect some respondent recall bias since
three months is a fairly long time period over which to remember how
many times they participated in any one activity.
Table 3 summarizes the sample means of the economic and demographic
characteristics of participants in each of the seven activities. The
final column contains averages for the entire sample for comparison. We
include age, sex, marital status, education, hourly wage, household
income categories, and the presence of children under age 12 in the
household as covariates in our statistical models. Both personal and
household income are reported in ranges in the CCHS. The ranges in the
survey are less than $15,000, between $15,000 and $30,000, between
$30,000 and $50,000, between $50,000 and $80,000, and greater than
$80,000. Following Ruhm (2005), the level of income for each individual
is coded as the midpoint of the range reported, or 150% of the unbounded
top range. We use personal income and hours worked per week to construct
a wage variable that we use as a proxy for the opportunity cost of time.
[FIGURE 2 OMITTED]
[FIGURE 3 OMITTED]
Walkers, swimmers, and home exercisers contain more females;
cyclists, golfers, weight lifters, and runners contain more males.
Participants tend to be younger than the general population. Runners and
weight lifters are the youngest groups of participants, and home
exercisers the oldest. The percentage of individuals with a college
degree in all of the sports and physical activities is higher than that
of the general population. Similarly, individuals in white collar jobs
(reported occupation "management," "professional,"
"technical," or "administrative") comprise a larger
proportion of participants in physical activities than in the general
population. The reported personal and household incomes are above the
sample average for all activities. Employment rates among participants
is higher than the general population. The hourly wage of participants
in all activities is higher than overall sample average hourly wage.
Econometric Methods
Table 2 shows that 82% of the individuals in the sample report
participating in physical activity in the previous three months. Of the
99,322 individuals in our sample, 81,552 were physically active in the
previous three months and 17,770 were not. Both the indicator variable
for participation in physical activity and sport and the variable for
the amount of time spent in physical activity in the previous month
contain a large number of zeros. Our econometric analysis of
participation in sport must account for the large number of zeros
observed in the data.
We assume that the zeros present in the data represent
"genuine zeros" as discussed in Jones (2000), meaning that the
observed non-participation in physical activity in this sample is the
result of the utility maximizing choices of sampled individuals. The
alternative explanation for the zeros is non-observable response that
occurs due to censoring. In the case of participation in physical
activity, censoring would take place if the time period considered in
the survey instrument was so short that some individuals who would
normally participate did not participate during that time period. We
assume that the three-month time period referred to in the CCHS is long
enough to avoid the non-observable response problem.
Both Jones (2000) and Amemiya (1984) discuss the appropriate
econometric techniques for dealing with zeros that are the result of
utility maximizing decisions in survey data but use different
terminology. We adopt the terminology used by Jones where "genuine
zeros" call for either a two-part model, when the participation and
intensity decisions are independent, or a hurdle model when they are
related. The key difference between a two-part model and a hurdle model
is that the participation and time spent equations are estimated
separately in a two-part model but simultaneously in a hurdle model.
Hurdle models are further differentiated based on assumptions about
dependence of the error terms. A double hurdle model with dependence,
also known as the Cragg model (Cragg, 1971), allows for factors that
affect participation and factors that affect time spent to have
different signs but assumes independence of the error terms. A double
hurdle model with dependence, sometimes referred to as the "full
double hurdle model," allows for factors that affect participation
and factors that affect time spent to have different signs and for
correlation in the equation error terms. This means that the
unobservable factors affecting participation and time spent in physical
activity can be correlated. We first estimate the parameters of the full
double hurdle model using data from the CCHS. The parameters of this
model can be estimated using a standard maximum likelihood approach
under the assumption that the equation error terms are drawn from a
normal distribution. We test the hypothesis of independent equation
errors ([H.sub.0] : p = 0). The results of the Wald test indicate that
the equation error terms of the participation and time spent equations
are not correlated for running, home exercise, swimming, and weight
lifting. The double hurdle models for these activities are then
re-estimated assuming that the equation errors are independent.
A technical issue in estimating double hurdle models is the use of
exclusion restrictions to identify the model. Exclusion restrictions are
not strictly required for identification but including the same set of
explanatory variables in both equations of the model creates
difficulties in identifying the parameters and obtaining convergence.
Imposing exclusion restrictions can help to improve identification.
There are examples of double hurdle models estimated with and without
exclusion restrictions in the literature. An exclusion restriction means
excluding one or more variables from the time spent equation that appear
in the participation equation. However, since the double hurdle model is
not an instrumental variables estimator, the excluded variables should
not be viewed as instruments for participation. No theoretical guidance
exists to aid in the determination of exclusion restrictions. We use two
variables, an indicator variable for individuals who walk to work, and
an indicator variable for individuals who reported that their health
status improved significantly or somewhat significantly over the past
year, to identify the participation decision. With respect to walking to
work, the intuition is that getting exercise through walking to work
will likely influence the first hurdle decision of participation in
additional leisure time physical activity. Somebody who walks to work
may consider that to be sufficient physical activity and may choose not
to engage in other physical activities. However, once the first hurdle
is passed, walking to work may not influence the decision about how much
time to spend per episode of physical activity. Similarly, a person who
experienced improvement in health status may be more likely to decide to
participate in physical activity but the improved health status may not
influence the time spent decision. We also estimated the double hurdle
models using only improvement in health status as an exclusion
restriction and achieved convergence with qualitatively similar results.
Results and Discussion
The estimation results for the participation equation are shown by
physical activity and sport in Table 4. The results for the time spent
equation are presented in Table 5. The tables contain parameter
estimates and asymptotic z-statistics for a two-tailed test of the null
hypothesis that the parameter is equal to zero. We included
province-specific effects in the models but do not report the results in
Tables 4 and 5. (3)
The model identifies two economic factors, income and the
opportunity cost of time as measured by hourly wage, as potentially
important determinants of participation and time spent in physical
activity. We begin by discussing our empirical findings about the effect
of changes in income and the opportunity cost of time on participation
and time spent in physical activity. We then turn our attention to our
empirical findings about the effect of individual characteristics like
age, sex, education, and family structure on these decisions.
Effect of Income on Participation and Time Spent
Our findings with respect to the effect of income on participation
and time spent are mixed. Tables 4 and 5 show that the parameter
estimates on income are mainly positive and significant in the
participation equation but are mainly negative and significant in the
time spent equation. We find that individuals with higher income are
likely to participate in swimming, golfing, weight lifting, and running
but income does not affect decisions about walking, home exercise, or
cycling. A different story emerges when evaluating the effect of income
on time spent. Time spent walking, exercising at home, golfing, weight
lifting, and running decreases with income but income does not affect
time spent cycling and swimming. Regardless of activity, the magnitude
of the effect of income on time spent is not large. (4)
Our findings with respect to the effect of income on participation
and time spent highlight the importance of recognizing that decisions
about participation in sports and physical activity are made on two
margins: the participation margin, and the time spent margin. The
empirical results indicate that the effect of a change in income on
participation is positive but negative on the optimal amount of time
spent engaged in physical activity. This finding is important for
informing policy because it suggests that consumers will respond
differently to economic incentives to be physically active depending on
whether the participation or time spent margin is targeted by the
incentive.
We do not interpret our results as causal evidence of the effect of
income on physical activity because, due to data limitations, we have
not addressed the potential endogeneity of income in the model. Most
previous studies of physical activity do not treat income as a
potentially endogenous variable in the empirical analysis. This is
likely the case because it is difficult to find suitable instruments for
income in survey data. Three recent exceptions include Lechner (2009),
Humphreys and Ruseski (2011), and Kosteas (2012). Lechner and Kosteas
use propensity score matching to account for the endogeneity between
participation in physical activity, including sport and earnings, while
Humphreys and Ruseski take an instrumental variables approach. Our
results, uncorrected for endogeneity, are consistent with the
endogeneity corrected results in Humphreys and Ruseski (2011) where a
change in income has a positive effect on participation but a negative
effect on time spent. Given this consistency in results across the two
studies, we do not believe that the findings about the effect of income
on physical activity in this study are spurious.
Effect of Opportunity Cost of Time on Participation and Time Spent
Referring again to Tables 4 and 5, we turn next to the results for
the hourly wage, which is a measure of the opportunity cost of time.
Recall from the model that the effect of a change in the opportunity
cost (or price) of time can be decomposed into an income and
substitution effect. The income effect arising from this price change
decomposition is different from the income effect arising from a change
in earnings. A higher opportunity cost of time makes non-work related
activities more costly and reduces the amount of time spent
participating in those activities; therefore, reducing time spent
participating in physical activities as the hourly wage increases
indicates that the substitution effect dominates the income effect.
Conversely, a positive relationship between hourly wage and time spent
in physical activity is suggestive of a dominating income effect.
Participation in sports entails at least some monetary costs and people
with higher incomes have greater financial means to participate.
The hourly wage is generally positive and significant in both the
participation and time spent equations, suggesting a dominating income
effect. The effect of a change hourly wage on time spent differs
depending on the activity. It is positive and significant for cycling,
swimming, golfing, and running; negative and significant for walking;
and insignificant for home exercise and weight lifting. Regardless of
sign and significance, the effect of a change in the hourly wage on time
spent is small.
It is possible that some of the effect of the opportunity cost of
time on the participation and time spent decisions is reflected in the
education and white collar job variables. A positive relationship
between income and education has been widely documented in the economics
literature. Evidence shows that more educated people tend to have higher
paying (and probably white collar) jobs and higher hourly wages and,
therefore, higher opportunity costs of time. We allow education to have
a nonlinear effect on participation and time spent in sport by including
two indicator variables. Education College is an indicator variable that
is equal to 1 if the individual completed college and 0 otherwise and
Education--High School is an indicator variable that is equal to 1 if
the individual graduated from high school and 0 otherwise. Graduating
from college has a strong positive effect on the participation decision
across all activities except running. People in white collar jobs are
less likely to participate in all of the activities. Completing high
school is not an important factor in explaining the participation
decision. Conditional on participation, occupation and education have
mainly strong positive effects on time spent. People with white collar
jobs spend between 4.7 and 33.5 minutes more per week engaged in sport
and physical activity than people in other types of jobs. People with a
high school or college education spend between 9 and 43 minutes more per
week playing sports or being physically active than people with less
than a high school education. If occupation and education are picking up
an opportunity cost of time effect, then these results, together with
the hourly wage results, provide further evidence of a dominating income
effect.
Effect of Individual Characteristics and Family Structure on
Participation and Time Spent We turn next to the influence of age, sex,
marital status, and the presence of young children in the household on
sport and physical activity. The effect of age on participation differs
across sports and physical activities, but conditional on participation,
time spent tends to decline with age. The decrease in time spent varies
across the sports and physical activities, ranging from a decrease of 9
minutes over three months for home exercisers to 69 minutes for weight
lifters. In the participation equation, older people are more likely to
walk, exercise at home, swim, cycle, and lift weights but are less
likely to run. Golfers are different in that age does not affect either
the participation or time spent decisions. These results further
highlight the importance of distinguishing between the participation
(extensive) and time spent (intensive) margins when evaluating the
effect of age on being physically active. Most cross-sectional studies
that examine the effect of age on only the time spent in physical
activity find a negative relationship between age and participation.
Treating the decision as a two-part decision suggests that the
mechanisms underlying the relationship between age and physical activity
are complex.
Distinguishing between decisions on the extensive and intensive
margins furthers our understanding of the mechanism underlying the
effect of sex on physical activity A positive association between being
male and physical activity has been found in many studies. Our results
indicate that the effect of sex on physical activity behavior is more
complex than men simply being more physically active than women. On the
extensive margin, we find that women are more likely to participate in
all of the sports and physical activities except for golf. The positive
association between being male and physical activity occurs on the
intensive margin, but only for some sports and activities. Women spend
more time walking, exercising at home, and swimming than men. Men spend
more time cycling, golfing, lifting weights, and running. These results
are largely consistent with the results in Humphreys and Ruseski (2007).
They find that males are more likely to participate in activities that
take more time like group sports and outdoor recreation activities
whereas females are more likely to engage in less time consuming
activities.
The effect of family structure on sport participation and physical
activity are measured by marital status and the presence of young
children in the household. The effect of being married and having young
children in the household suggests that family structure plays an
important role in decisions about sport participation and physical
activity, particularly on the intensive margin. Again, golfers appear to
be different. Married people are more likely to play golf and spend more
time playing than single people. Otherwise, marriage does not affect the
participation decision. Marriage does play a role in the time spent
decision. Married people spend less time in home exercise, cycling,
lifting weights, and running but more time swimming and golfing than
single people.
The effect of having young children in the household varies across
sports and activities. We do not find a relationship between having
young children and participation for walking, exercising at home,
cycling, or swimming but we find that people with young children are
less likely to golf, lift weights, and run. Conditional on
participation, people with young children spend more time cycling and
swimming but less time in the other activities. These results indicate
that married couples and households with young children have different
demands on their time and different opportunity costs of time than
unmarried and childless people. The increase in time spent in cycling
and swimming when there are young children in the house is not
surprising since these are common activities for families to do
together.
Summary and Policy Implications
This research examines participation and time spent in seven common
sports and physical activities: walking, home exercise, cycling,
swimming, golfing, weight lifting, and running by empirically examining
testable implications from our consumer choice model. A number of
interesting conclusions emerge from the analysis. Our findings about the
effect of income on participation and time spent are mixed. Income does
not have an effect on participation or time spent across all activities.
However, patterns do emerge among the statistically significant
variables. When significant, people with higher income are more likely
to participate but, conditional on participation, spend less time. Using
the wage rate as a proxy of the opportunity cost of time, we find some
evidence that the income effect dominates the substitution effect as the
opportunity cost of time increases.
With respect to the income, age, and sex variables, we find it is
important to recognize that decisions about participation in sports and
physical activity are made on two different margins: the participation
margin and the time spent margin. Cross-sectional studies that are based
on single equation models consistently find a positive relationship
between income and participation, that older people spend less time
engaged in physical activity, and that males are more likely to
participate than females. Our results are consistent with these results
but the new insight here is establishing where the link is occurring. In
the case of income, we find that people with higher income are more
likely to choose to participate in sport and physical activity but,
conditional on participation, devote less time participating. The
positive relationship between income and physical activity is occurring
in the participation decision rather than the time spent equation.
Similarly, we find that the effect of age on participation differs
across sports and physical activities, but that time spent declines with
age, suggesting that the well-documented negative relationship between
age and participation is occurring the time spent, rather than the
participation equation. These results suggest that programs aimed at
increasing participation in older populations and encouraging continued
participation over the life cycle might be particularly effective.
Finally, we find that, with the exception of golf, women are more likely
to participate in all of the sports and physical activities but that the
effect of sex on time spent differs across sports.
Distinguishing between the participation and time spent margins is
also important in examining the effect of family structure on physical
activity decisions. We find that, with the exception of golf, marriage
does not affect the participation decision. Being married does have an
effect on the amount of time spent but the effect differs across
activities. Married people spend less time in home exercise, cycling,
lifting weights, and running but spend more time swimming and playing
golf than single people. The effect of having young children on
participation and time spent is also complex. We find that people with
young children are less likely to play golf, lift weights, and run;
however, conditional on participation, people with young children spend
more time participating in family- oriented activities like riding bikes
and swimming. These results provide further evidence that policy
interventions designed to target these sub-populations are likely to be
more effective than a "one size fits all" policy.
References
Amemiya, T. (1984). Tobit model: A survey. Journal of Econometrics,
24, 3-61.
Anokye, N., Pokhrel, S., Buxton, M., & Fox-Rushby, J. (2012).
The demand for sports and exercise: Results from an illustrative survey.
The European Journal of Health Economics, 13, 277-287. Balia, S., &
Jones, A. (2008). Mortality, lifestyle and socio-economic status.
Journal of Health Economics, 27, 1-26.
Brown, H., & Roberts, J. (2011). Exercising choice: The
economic determinants of physical activity behaviour of an employed
population. Social Science and Medicine, 73, 383-390. Canadian Sport
Policy (2012). Canadian sport policy 2012. Technical report.
Christian, T. J. (2009). Opportunity costs surrounding exercise and
dietary behaviors: Quantifying trade-offs between commuting time and
health-related activities. Social Science Research Network.
Colley, R. C., Garriguet, D., Janssen, I., Craig, C. L., Clarke,
J., & Tremblay, M. S. (2011). Physical activity of Canadian adults:
Accelerometer results from the 2007 to 2009 Canadian health measures
survey. Health Rep, 22, 7-14.
Contoyannis, P., & Jones, A. (2004). Socio-economic status,
health and lifestyle. Journal of Health Economics, 23, 965-995.
Costa-Font, J., & Gil, J. (2005). Obesity and the incidence of
chronic disease in Spain: A seemingly unrelated probit approach.
Economics and Human Biology, 3, 188-214.
Cragg, J. (1971). Some statistical models for limited dependent
variables with application to the demand for durable goods.
Econometrica, 39, 829-844.
Doherty, A., & Clutterbuck, R. (2013). Chapter 24: Canada. In
K. Hallmann & K. Petry (Eds.), Comparative sport development:
Systems, particiaption andpublicpolicy (pp. 323-342). New York, NY:
Springer.
Downward, P. (2007). Exploring the economic choice to participate
in sport: Results from the 2002 General Household Survey. International
Review of Applied Economics, 21, 633-653.
Downward, P., Lera-Lopez, F., & Rasciute, S. (2011). The
zero-inflated ordered probit approach to modelling sports participation.
Economic Modelling, 28, 2469-2477.
Downward, R, & Riordan, J. (2007). Social interactions and the
demand for sport: An economic analysis. Contemporary Economic Policy,
25, 518-537.
Eberth, B., & Smith, M. (2010). Modelling the participation
decision and duration of sporting activity in Scotland. Economic
modelling, 27, 822-834.
Eisenberg, D., & Okeke, E. (2009). Too cold for a jog? Weather,
exercise, and socioeconomic status. The B.E. Journal of Economic
Analysis & Policy, 9, 1-30.
Farrell, L., & Shields, M. A. (2002). Investigating the
economic and demographic determinants of sporting participation in
England. Journal of the Royal Statistical Society (Series B), 165,
335-348.
Forrest, D., & McHale, I. (2011). Subjective well-being and
engagement in sport: Evidence from England. In P. Rodriguez, S. Kesenne,
& B. R. Humphreys (Eds.), The economics of sport, health and
happiness: The promotion of well-being through sporting activities (pp.
184-199). Cheltenham, UK: Edward Elgar Publishing.
Garcia, J., Lera-Lopez, F., & Suarez, M. (2011). Estimation of
a structural model of the determinants of the time spent on physical
activity and sport. Journal of Sports Economics, 12, 515-537.
Gilmour, H. (2007). Physically active Canadians. Technical Report
3. Statistics Canada.
Huang, H., & Humphreys, B. (2012). Sports participation and
happiness: Evidence from US microdata. Journal of Economic Psychology,
33, 776-793.
Humphreys, B., & Ruseski, J. (2007). Participation in physical
activity and government spending on parks and recreation. Contemporary
Economic Policy, 25, 538-552.
Humphreys, B. R., McLeod, L., & Ruseski, J. E. (2014). Physical
activity and health outcomes: Evidence from Canada. Health Economics,
23, 33-54.
Humphreys, B. R., & Ruseski, J. E. (2011). An economic analysis
of participation and time spent in physical activity. The BE Journal of
Economic Analysis & Policy, 11, 1-47.
Jones, A. M. (2000). Health econometrics. Handbook of health
economics, 1, 265-344.
Kavetsos, G. (2011). Physical activity and subjective well-being:
An empirical analysis. In P. Rodriguez, S. Kesenne, & B. R.
Humphreys (Eds.), The economics of sport, health and happiness: The
promotion of well-being through sporting activities (pp. 213-222).
Cheltenham, U K : Edward Elgar Publishing.
Kosteas, V. (2012). The effect of exercise on earnings: Evidence
from the NLSY. Journal of Labor Research, 33, 225-250.
Kuvaja-Kollner, V., Valtonen, H., Komulainen, P., Hassinen, M.,
& Rauramaa, R. (2013). The impact of time cost of physical exercise
on health outcomes by older adults: The DR's EXTRA Study. The
European Journal of Health Economics, 14, 471-479.
Lechner, M. (2009). Long-run labour market and health effects of
individual sports activities. Journal of Health Economics, 28, 839-854.
Lechner, M., & Downward, P. (2013). Heterogeneous sports
participation and labour market outcomes in England. IZA Discussion
Paper 7690, Bonn.
Lera-Lopez, F., & Rapun-Garate, M. (2007). The demand for
sport: Sport consumption and participation models. Journal of Sport
Management, 21, 103-122.
Maruyama, S., & Yin, Q. (2012). The opportunity cost of
exercise: Do higher-earning Australians exercise longer, harder, or
both? Health Policy, 106, 187-194.
Meltzer, D., & Jena, A. (2010). The economics of intense
exercise. Journal of Health Economics, 29, 347-352.
Pawlowski, T., Breuer, C., & Leyva, J. (2011). Sport
opportunities and local well-being: Is sport a local amenity? In P.
Rodrguez, S. Kesenne, & B. R. Humphreys (Eds.), The economics of
sport, health and happiness: The promotion of well-being through
sporting activities (pp. 223-241). Cheltenham, UK: Edward Elgar
Publishing.
Pawlowski, T., Breuer, C., Wicker, P., & Poupaux, S. (2009).
Travel time spending behavior in recreational sports--An econometric
approach with management implications. European Sport Management
Quarterly, 9, 215-242.
Pawlowski, T., Downward, P., & Rasciute, S. (2011). Subjective
well-being in European countries on the age-specific impact of physical
activity. European Review of Aging and Physical Activity, 8, 93-102.
Rasciute,S., & Downward, P. (2010). Health or happiness? What
is the impact of physical activity on the individual? Kyklos, 63,
256-270.
Ruhm, C. (2005). Healthy living in hard times. Journal of Health
Economics, 24, 341-363.
Ruseski, J., & Humphreys, B. (2011). Participation in physical
activity and health outcomes: Evidence from the Canadian Community
Health Survey. In P. Rodrguez, S. Kesenne, & B. R. Humphreys (Eds.),
The economics of sport, health and happiness: The promotion of wellbeing
through sporting activities (pp. 7-32). Cheltenham, UK: Edward Elgar
Publishing.
Ruseski, J. E., & Razavilar, N. (2013). United States. In K.
Hallmann & K. Petry (Eds.), Comparative sport development: Systems,
participation and public policy (pp. 311-322). New Yo r k, NY: Springer.
Sari, N. (2009). Physical inactivity and its impact on healthcare
utilization. Health Economics, 18, 885-901.
Sarma, S., Devlin, R. A., Gilliland, J., Campbell, M. K., &
Zaric, G. S. (2013). The effect of leisure-time physical activity on
obesity, diabetes, high bp and heart disease among Canadians: Evidence
from 2000/01 to 2005/06. Canadian Centre for Health Economics Working
Paper #2013-01.
Statistics Canada (2012). CANSIM Table 105-0501, Health indicator
profile. Technical report. Tremblay, M. S., Warburton, D. E., Janssen,
I., Paterson, D. H., Latimer, A. E., Rhodes, R. E.,... Dugan, M. (2011).
New Canadian physical activity guidelines. Applied Physiology,
Nutrition, and Metabolism, 36, 36-46.
World Health Organization (2010). Physical inactivity: A global
public health problem. Retrieved from
http://www.who.int/dietphysicalactivity/factsheet_inactivity/en/
Endnotes
(1) Canada's physical activity guidelines state that adults
aged 18-64 years should accumulate at least 150 minutes of moderate-
to-vigorous-intensity aerobic physical activity per week, in bouts of 10
minutes or more to achieve health benefits (Tremblay et al., 2011).
These guidelines are different from the definitions of active and
moderately active in the CCHS.
(2) The Cragg model, proposed by Cragg (1971), is a double hurdle
model that allows for the same factors (for example, education) that
affect both participation and time spent to have different signs but
assumes independence of the equation error terms.
(3) The full set of results is available from the authors on
request.
(4) Although participation in hockey is low (only 2.97% of the
sample), we did estimate a model for hockey because it is a high profile
sport in Canada. We find that the effect of changes in income and wages
are positive and statistically significant in both the participation and
time spent decisions.
Brad R. Humphreys [1] and Jane E. Ruseski [1]
[1] West Virginia University
Brad. R. Humphreys, PhD, is an associate professor of economics.
His research interests include the economic impact of professional
sports teams and facilities, the effect of regulations on
intercollegiate athletics, and the economic determinants of sport
participation.
Jane E. Ruseski, PhD, is an associate professor of economics. Much
of her current research studies the socioeconomic determinants of health
and (un)healthy behaviors, the effect of health behaviors on outcomes,
and the mechanisms underlying health behaviors.
Table 1. Distribution of Physical Activities
Activity Frequency Participation
Rate
Walking 62082 62.51%
Gardening 42165 42.45%
Home Exercise 21718 21.87%
Cycling 16211 16.32%
Swimming 16112 16.22%
Dancing 14105 14.20%
Golf 10558 10.63%
Fishing 9324 9.39%
Weight Lifting 8923 8.98%
Running 7748 7.80%
Other 1 7652 7.70%
Other 2 7138 7.18%
Bowling 6711 6.75%
Aerobics 5902 5.94%
Softball 3854 3.88%
Skating 3729 3.75%
Inline Skating 3307 3.33%
Hockey 2947 2.97%
Skiing 2572 2.59%
Volleyball 2469 2.49%
Basketball 2438 2.45%
Tennis 2180 2.19%
Other 3 1274 1.28%
Activity Times Minutes Spent
Participated per Episode
Walking 46.08 37.57
Gardening 24.51 53.09
Home Exercise 39.78 25.88
Cycling 20.11 43.86
Swimming 14.03 45.63
Dancing 6.93 64.93
Golf 10.69 73.76
Fishing 7.65 72.59
Weight Lifting 30.20 41.41
Running 23.58 33.03
Other 1 19.33 62.72
Other 2 20.22 63.80
Bowling 6.06 70.51
Aerobics 23.20 50.03
Softball 10.03 69.17
Skating 6.52 55.30
Inline Skating 10.61 49.52
Hockey 13.24 67.38
Skiing 6.14 72.83
Volleyball 7.45 63.78
Basketball 10.16 53.25
Tennis 9.47 62.37
Other 3 19.32 62.62
# of Observations 99,322
Table 2. Distribution of Number of Activities
Number of Activities Frequency Percent
0 17,770 17.89
1 19,206 19.34
2 18,801 18.93
3 14,618 14.72
4 9,961 10.03
5 6,647 6.51
6 4,252 4.28
7 2,878 2.90
8 1,916 1.93
9 1,287 1.30
10 827 0.83
11 584 0.59
12 288 0.29
13 204 0.21
14 120 0.12
15 71 0.07
16 34 0.03
17 23 0.02
18 11 0.01
19 4 0.00
# of Observations 99,322
Table 3: Summary Statistics on Participants
Home
Variable Walkers Exercisers Cyclists
% Male 40.61% 38.68% 54.74%
Age 49.4 47.6 41.9
% Married 59.94% 58.87% 62.06%
% HS Graduate 18.02% 17.78% 17.12%
% College Graduate 49.87% 54.83% 59.35%
% Employed 61.93% 65.36% 80.12%
% in "White Collar" Jobs 34.26% 38.22% 44.28%
Hours Worked 40.0 40.0 41.1
Personal Income (000s) 32.582 34.145 40.007
Household Income (000s) 53.114 53.302 56.455
Hourly Wage 14.2 15.3 19.0
% Young Children 23.4% 24.7% 33.3%
Participants 62,080 21,717 16,211
Weight
Variable Swimmers Golfers Lifters
% Male 43.88% 68.21% 56.76%
Age 42.0 44.7 38.0
% Married 64.82% 67.28% 54.86%
% HS Graduate 18.32% 19.98% 17.01%
% College Graduate 61.08% 60.87% 65.00%
% Employed 78.16% 81.01% 87.92%
% in "White Collar" Jobs 44.66% 46.71% 52.36%
Hours Worked 40.8 44.2 42.2
Personal Income (000s) 38.895 48.606 44.639
Household Income (000s) 56.069 54.147 54.285
Hourly Wage 18.4 20.9 21.2
% Young Children 38.8% 25.9% 28.1%
Participants 16,108 10,558 8,923
Variable Runners Overall
% Male 57.55% 46.16%
Age 36.4 50.1
% Married 57.50% 59.94%
% HS Graduate 16.30% 18.07%
% College Graduate 65.78% 46.32%
% Employed 90.17% 56.86%
% in "White Collar" Jobs 53.70% 30.76%
Hours Worked 42.2 41.0
Personal Income (000s) 45.140 32.002
Household Income (000s) 52.929 52.488
Hourly Wage 21.7 12.7
% Young Children 33.0% 23.3%
Participants 7,748 99,322
Notes: % Young Children: children under age 12 in household; Hourly
wage for employed persons only
Table 4. Parameter Estimates and z-statistics-Participation Equation
Home
Variable Walkers Exercisers Cyclists
Age 0.0131 *** 0.0181 *** 0.0203 ***
(8.37) (7.06) (4.23)
Male -0.532 *** -0.731 *** -0.874 ***
(-12.89) (-10.22) (-6.57)
Married -0.0121 0.0302 -0.064
(-0.31) (0.41) (-0.51)
Wage 0.807 0.302 *** 0.347 ***
(0.88) (8.68) (8.41)
Household Income (000) 0.00067 0.0017 -0.000139
(1.34) (1.96) (-0.14)
White Collar Job -2.417 *** -1.941 *** -1.947 ***
(-23.11) (-12.43) (-9.42)
Education--College 0.210 *** 0.477 *** 0.302 **
(4.9) (6.1) (3.19)
Education--High School -0.0141 0.216 * 0.147
(-0.28) (2.31) (1.27)
Young Children -0.0802 -0.0661 -0.109
(-1.36) (-0.65) (-0.97)
Improvement in Health 0.560 *** 0.820 *** 0.370 **
Status (8.74) (7.1) (3.1)
Walk to Work 4.040 *** 4.454 2.775 ***
(5.2) (1.06) (6.06)
Participants 62,080 21,717 16,211
Log likelihood -596497.6 -222800.9 -168473.6
-0.08 0.029 0.216
Wald Test 15.84 *** 0.38 16.91 ***
Variable Swimmers Golfers
Age 0.0147 * -0.000584
(2.42) (-0.29)
Male -0.768 *** -0.0352
(-7.66) (-0.60)
Married -0.0386 0.425 ***
(-0.31) (8.22)
Wage 0.487 *** 0.544 ***
(3.35) (4.19)
Household Income (000) 0.00362 ** 0.00453 ***
(3.27) (6.76)
White Collar Job -1.849 -1.987 ***
(-11.09) (-9.00)
Education--College 0.444 *** 0.707 ***
(4.15) (12.1)
Education--High School 0.301 * 0.501 ***
(2.5) (7.15)
Young Children 0.0216 -0.464 ***
(0.19) (-4.75)
Improvement in Health 0.418 *** 0.0182
Status (3.49) (0.26)
Walk to Work 1.892 *** 0.609 ***
(4.92) (11.77)
Participants 16,108 10,558
Log likelihood -162617.9 -112042.7
0.045 0.0491
Wald Test 0.94 174.02 ***
Weight
Variable Lifters Runners
Age 0.00810 * -0.0149 ***
(2.21) (-3.43)
Male -0.386 *** -0.401 ***
(-4.82) (-3.40)
Married -0.167 0.177
(-1.80) (1.56)
Wage 0.0807 *** 0.149 ***
(8.72) (7.31)
Household Income (000) 0.00223 ** 0.00245 *
(3.28) (2.5)
White Collar Job -0.344 ** -1.027 ***
(-2.92) (-5.94)
Education--College 0.190 * 0.188
(2.21) (1.72)
Education--High School -0.025 0.0172
(-0.25) (0.13)
Young Children -0.188 * -0.228 *
(-2.42) (-2.13)
Improvement in Health 0.935 *** 0.460 ***
Status (10.59) (4.03)
Walk to Work 0.678 *** 1.335 ***
(9.43) (9.84)
Participants 8,923 7,748
Log likelihood -98010.5 -82862.6
0.028 -0.023
Wald Test 0.23 0.1
Participation: Indicator variable for participation in sport or
physical activity
* p <0.05; ** p <0.01; *** p <0.001
Table 5. Parameter Estimates and z-statistics-Time Spent Equation
Home
Variable Walkers Exercisers Cyclists
Age 1.780 ** -9.357 *** -30.20 ***
(2.67) (-14.58) (-36.22)
Male -441.7 *** -261.9 *** 553.7 ***
(-25.08) (-14.44) (27.26)
Married -4.269 -93.37 *** -65.22 **
(-0.23) (-4.98) (-2.99)
Wage -1.655 *** -0.0443 3.418 ***
(-3.57) (-0.10) (6.99)
Household Income (000) -0.663 *** -0.725 *** -0.25
(-3.42) (-3.72) (-1.18)
White Collar Job 5.638 110.2 *** 181.6 ***
(0.28) (5.29) (8.12)
Education--College 216.4 *** 294.4 *** 270.8 ***
(10.29) (13.41) (10.83)
Education--High School 132.7 *** 107.4 *** 55.25
(5.2) (4.04) (1.82)
Young Children -294.2 *** -119.1 *** 82.36 ***
(-13.05) (-5.24) (3.43)
Participants 62,080 21,717 16,211
Log likelihood -596497.6 -222800.9 -168473.6
P -0.08 0.029 0.216
Wald Test 15.84 *** 0.38 16.91 ***
Variable Swimmers Golfers
Age -17.18 *** 0.52
(-20.27) (0.54)
Male -99.05 *** 929.0 ***
(-6.54) (39.31)
Married 56.38 ** 165.6 ***
(3.18) (6.69)
Wage 2.229 *** 2.827 ***
(5.68) (5.85)
Household Income (000) -0.278 -1.855 ***
(-1.58) (-8.40)
White Collar Job 56.70 *** 241.2 ***
(3.38) (10.29)
Education--College 355.0 *** 361.7 ***
(15.17) (12.44)
Education--High School 177.1 *** 336.2 ***
(6.78) (9.86)
Young Children 323.5 *** -231.9 ***
(17.88) (-8.67)
Participants 16,108 10,558
Log likelihood -162617.9 -112042.7
P 0.045 0.0491
Wald Test 0.94 174.02 ***
Weight
Variable Lifters Runners
Age -68.93 *** -50.03 ***
(-34.56) (-29.56)
Male 831.2 *** 505.6 ***
(17.43) (17.43)
Married -261.9 *** -118.7 ***
(-5.10) (-3.62)
Wage 1.185 3.465 ***
(1.27) (4.97)
Household Income (000) -2.561 *** -2.787 ***
(-6.36) (-10.15)
White Collar Job 391.2 *** 401.8 ***
(8.07) (13.66)
Education--College 513.9 *** 408.4 ***
(8.52) (10.78)
Education--High School 250.2 *** 80
(3.52) (1.77)
Young Children -375.8 *** -114.6 ***
(-7.49) (-3.65)
Participants 8,923 7,748
Log likelihood -98010.5 -82862.6
P 0.028 -0.023
Wald Test 0.23 0.1
Time spent: exercise minutes last three months
* p <0.05; ** p<0.01; *** P<0.001
