MODE CHOICE ANALYSIS BETWEEN ONLINE RIDE-HAILING AND PARATRANSIT IN BANJARMAS...
Transportation Economics Term Paper
1. 1
Determinants of Time Spent Commuting to Work
Selma Dogic
Georgia Institute of Technology
April 22, 2015
Abstract
This paper takes a closer look into the factors that impact time travelled to work. It will
offer insight to the variation in travel time values across individuals and show how
occupation and family background affect ones travel time to work. As metro areas
become more populated, thanks to their lower costs of livings relative to city centers,
people are experiencing longer commutes. This leads one to question how commuters
value their drive times and what factors affect their commute time; it also sheds light on
the types of transportation policies that are relevant to citizens. This paper utilizes past
research to create a foundation for the valuation of travel time and then attempts to
quantify the findings through a regression analysis created with variables collected from
the American Community Survey. A simple ordinary least squares model was developed
and provided some deeper insight to value of time theory and prospects of new policies.
2. 2
Table of Contents
Introduction........................................................................................................................3
Literature Review .......................................................................................................... 4-5
Data ................................................................................................................................. 6-9
Model............................................................................................................................. 9-14
Policy Implications...........................................................................................................14
Conclusion/Discussion .....................................................................................................15
3. 3
Introduction
A major purpose of infrastructure investments and transportation policy is aimed
at the reduction of time travelled (Belenky, 2011). Travel time is distinguished between
paid business travel and personal travel, commuting to work is sometimes treated as a
third distinction, but it more often is included in personal travel (Belenky, 2011).
Personal travel time is estimated to be in the realm of 25% to 50% of wages; travel time
costs increase with income and are lower for children and the unemployed. In fact, the
more an individual earns, the more they are willing to pay for travel time savings.
("Travel Time - Transportation Benefit-Cost Analysis," n.d.) This makes sense, of course,
because these individuals have relatively higher opportunity costs. Thus, age and
earnings are important factors to consider when evaluating the time traveled to work
("Travel Time - Transportation Benefit-Cost Analysis," n.d.).
This type of study sheds light onto human behavior (Small, 2012), by giving a
glimpse to the decision making process. It also shows the roles that opportunity costs
play in that process.
In regards to infrastructure investment policy, travel time plays a vital role in its
evaluation; and a great bulk on recent transportation investment has been to reduce the
delay in passenger travel (Belenky, 2011). Travelers have witnessed this through the
creation of HOV lanes and toll lanes that were designed to mitigate congestion and thus
reduce passenger delays.
4. 4
Literature Review
Gary S. Becker developed a foundation of the valuation of time that has been the
basis for many other analyses. It is based on the constraint that time is spread across
work, leisure, and travel and that any travel time savings can be transferred freely
amongst either leisure or work (Small, 2012). Though this is often not the case an
important take away from the framework of time valuation is that time spent en route is
related to forgone earnings and leisure. A major portion of one’s personal time is spent on
travel; many individuals spend sixty to ninety minutes of their day in transit (Victoria
Transport Policy Institute, 2013).
Empirical results have shown that the value of commuting time averages
approximately one-half of the wage rate (Small, 2012). Thus, these values are higher
when individuals earn higher wages. There has also been evidence that driver’s value
their time more under congested conditions that under free-flowing traffic (Small, 2012).
Showing that under less pleasant conditions, traveler’s experience higher costs.
This also offers some commentary to the valuation of comfortable travel, making
note that the reduction of time traveled is also based on one’s desire to avoid
uncomfortable travel experiences (Belenky, 2011). In fact, travelers are willing to pay
higher prices to shorten the time they must travel under uncomfortable circumstances
(Belenky, 2011). Brundell-Freij showed that travel time costs are higher for these
uncomfortable and unsafe conditions (Brundell-Freij, 2006).
5. 5
Technology could be key to curtailing the higher costs of discomfort, if
technological advances further developed in-vehicle amenities and thus lowered the value
of time travelled (Small, 2012). However, there is little evidence of this and it is
particularly difficult to determining the future effects that these types of technological
advances would have on time costs (Small, 2012).
Reliability also plays a key role in the valuation of time. Costs increase with less
reliability and uncertainty in driving conditions (delays en route) (Cohan and Southworth,
1999). The reduction of reliability causes an increase in time travelled, as individuals
account for extra time in their commute to mitigate the uncertainty of their trip.
Unfortunately, there are many unobservable variables that affect time traveled to
work, such as preferences and attitudes. Preferences are increasingly being incorporated
into in-depth analysis, particularly through the use of state-preference questionnaires in
which respondents offer solutions to hypothetical situations (Small, 2012). Yet, attitudes
are much harder to quantify and include in an analysis. Though incorporating stated and
revealed preferences could develop individual attitudes into the model, it is not clear
whether a model would benefit from any statistical efficiency (Small, 2012).
6. 6
Data
The following data has been collected from the US Census Bureau’s American
Community Survey 2004 Public Use Microdata. Below is a table summarizing the
collected cross-sectional variables.
Variable
Observations
Mean
Standard
Deviation
Minimum
Maximum
Minutes
Travelled
to
Work
1,194,354
10.43109
18.44492
0
200
Mode
of
Transportation
to
Work
1,194,354
0.8256689
1.971638
0
12
Occupation
1,194,354
2,664.38
2,937.56
9
9,920
Total
Personal
Earnings
1,194,354
17,672.86
36,593.4
-‐10,000
729,000
Number
of
Vehicles
1,194,354
2.091158
1.096379
0
6
Number
of
Children
1,194,354
0.9690862
1.27915
0
13
Marital
Status
1,194,354
2.847479
1.858984
1
5
Age
1,194,354
38.34378
22.94143
0
93
Sex
1,194,354
1.518616
0.4996535
1
2
Mode
of
travel,
occupation,
earnings,
number
of
vehicles,
number
of
children,
and
marital
status
are
used
as
explanatory
variables
in
the
model;
they
are
used
to
explain
the
variation
in
minutes
traveled
to
work.
Mode
of
travel,
occupation,
and
marital
status
are
categorical
variables.
Number
of
children,
number
of
vehicles,
and
earnings,
on
the
other
hand,
are
continuous
variables.
Below
is
a
table
describing
the
data
collected.
Minutes
travelled
to
work.
From
the
table
above
we
see
that
the
average
time
travelled
to
work
is
ten
and
a
half
minutes.
The
standard
deviation
shows
how
spread
out
the
data
points
are;
for
minutes
travelled
to
work
this
is
almost
twenty
minutes.
Interestingly,
the
time
travelled
varies
from
no
minutes
(for
those
who
are
unemployed
or
work
from
home)
to
two
hundred
minutes
–
almost
three
hours.
Mode
of
transportation
to
work.
This
variable
has
thirteen
categories,
taking
on
values
from
zero
to
twelve.
If
the
observation
is
missing
for
this
variable
it
would
take
on
7. 7
the
variable
zero.
The
value
one
is
devoted
to
personal
vehicle
travel,
ten
for
individuals
who
walked
to
work,
eleven
to
those
who
worked
at
home,
and
twelve
to
other
modes.
The
pattern
among
the
categories
appears
to
be
that
more
comfortable
modes
of
transportation
take
on
lower
values,
thus
the
higher
the
value
for
this
category
the
more
uncomfortable
the
mode
is
and
it
is
expected
that
the
less
time
one
would
spend
in
transit.
Occupation.
Like
mode
of
transportation,
occupation
is
divided
amongst
several
categories.
The
values
are
indicative
of
the
US
Standard
Occupation
Codes,
where
each
numerical
value
represents
a
specific
occupation
within
a
grouping.
The
code
111011
represents
chief
executives,
while
490000
codes
represent
production
workers.
Within
this
category,
it
seems
that
codes
with
a
greater
numerical
value
represent
occupations
with
relatively
lower
wage
rates.
In
a
regression
analysis,
one
would
expect
that
the
lower
this
variable
is,
the
less
time
spent
commuting
to
work
since
these
individuals
have
greater
value
their
travel
time.
Total
Personal
Earnings.
The
average
annual
earnings
for
this
sample
were
$17,672.
However,
with
a
relatively
large
standard
deviation
this
value
may
not
depict
a
full
picture
of
the
data
set.
The
median
earnings
for
this
sample
were
relatively
low,
only
$2,300;
the
US
Bureau
of
Labor
Statistics
has
recently
published
median
earnings
for
the
US
being
$41,392
as
of
the
fourth
quarter
of
2014.
This
may
be
an
indicator
that
the
sample
population
is
not
completely
representative
of
the
population
in
general.
Interestingly,
there
are
individuals
in
the
sample
that
make
close
to
a
million
dollars
($729,000)
while
other
individuals
seemingly
have
negative
earnings.
The
negative
earnings
may
not
necessarily
represent
individuals
who
have
lost
money,
although
this
is
sometimes
the
case.
A
value
of
-‐10,000
is
representative
of
individuals
that
are
younger
than
fifteen
years
old,
and
thus
are
ineligible
(under
labor
laws)
to
work
and
earn
taxable
wages.
Value
greater
than
-‐10,000
and
less
than
zero
represent
losses
incurred
by
individuals
in
the
data
set.
This
variable
is
incredibly
8. 8
important
in
the
regression
analysis,
because
as
has
been
discussed
above,
time
is
valued
as
a
percentage
of
wages;
making
note
that
the
more
one
earns,
the
more
they
value
their
time
and
the
less
time
they
prefer
to
spend
en
route.
Number of vehicles. The average number of vehicles owned for this sample is
two. This shows that an average household has two vehicles to its disposal.
Number of children. An average household within the dataset has approximately
one child. One would imagine that the more children one has, the less time they would
spend in transit, as their preferences of spending time shift towards home and family. The
relationship between children and time travelled to work is likely to be negative.
Marital Status. This variable is categorized among from one to five. One indicates
a married individual, one with a current spouse, while the other categories represent
individuals with no spouse (this could be the result of a death, divorce, or having never
been married). For the same reasons as above, the lower a number the more of an
incentive to spend time with family. This leads one to think that married individuals have
higher time values; so, a lower value for this variable should have a negative impact on
time travelled to work.
Age. The average age in the sample population is close to thirty-eight. Assuming
that on average, income rises with age, one could assume that minutes travelled to work
would fall as age increases. In other words, age and minutes travelled to work have a
negative relationship.
Sex. The variable sex was included in the data set for experimental purposed only.
Evidence shows that men, on average, earn more than women. Using this as a basis, one
would be led to believe that men on average travel shorter time periods to work. As the
9. 9
data is categorizes (1 for men and 2 for women), this variable should have a positive
effect on minutes traveled to work.
Model
A simple ordinary least squares regression was utilized in this paper. The
regression modeled travel time to work as the dependent variable, while all other
variables previously discussed served as explanatory variables.
Minutes
Traveled
to
Work
Explanatory
Variable
Coefficient
Standard
Error
t
P>|t|
95%
Confidence
Interval
Age
-‐0.0768264
0.0010463
-‐73.42
0
-‐0.0788772
-‐0.0747756
Occupation
0.0013342
0.00000541
246.44
0
0.0013235
0.0013448
Total
Personal
Earnings
0.0001715
0.000000754
377.6
0
0.0001706
0.0001724
Marital
Status
-‐0.7632183
0.0117898
-‐64.74
0
-‐0.7863259
-‐0.7401106
Sex
-‐0.4687807
0.0304302
-‐15.41
0
-‐0.5284229
-‐0.4091385
Number
of
Children
-‐0.6890591
0.0141573
-‐48.67
0
-‐0.716807
-‐0.6613113
Mode
of
Transportation
to
Work
0.3605778
0.0078724
45.8
0
0.3451482
0.3760075
Number
of
Vehicles
0.2243699
0.013992
16.04
0
0.1969461
0.2517937
Constant
9.57783
.1029292
93.05
0
9.376092
9.779567
N
1,194,354
R2
0.2272
10. 10
Under OLS, the model is assumed to be homoscedastic, which means that the
variance of the error terms is constant. To visually verify whether this is true, the model’s
residuals were plotted against its fitted values. Below is the visual representation.
The chart above indicates that the model, in fact, has heteroscedastic characteristics. To
formally test this, a Breusch-Pagan test was conducted to detect whether any linear form
of heteroscedasticity exists. The results indicated that the, indeed, heteorscedasticity is
present. To compensate for this issue, the regression was conducted again using robust
standard errors. By using robust standard errors, the assumption that error terms are
independent and identically distributed is relaxed, thus making the robust standard errors
more reliable. The results are displayed below.
11. 11
Minutes
Traveled
to
Work
Explanatory
Variable
Coefficient
Robust
Standard
Errors
t
Age
-‐0.0768264
0.0009478
-‐81.06
Occupation
0.0013342
0.00000632
211.14
Total
Personal
Earnings
0.0001715
0.000000827
207.36
Marital
Status
-‐0.7632183
0.0137621
-‐55.46
Sex
-‐0.4687807
0.029799
-‐15.73
Number
of
Children
-‐0.6890591
0.0136054
-‐50.65
Mode
of
Transportation
to
Work
0.3605778
0.0119117
30.27
Constant
9.57783
0.1081094
88.59
N
1,194,354
R2
0.2272
Note that number of vehicles has been deleted as an explanatory variable as it
offers little intuitive explanation for its effects on time travelled to work and its omission
from the model doesn’t affect R-squared at all.
As noted by the R-squared value above, the independent variables explain about
twenty-three percent of the variability on the minutes travelled to work. The p-values for
all independent variables were zero, meaning that each variable is statistically different
from zero at the five percent level. Though each variable is of statistical significance, it
may not necessarily be economically significant in that a variable’s impact on the
dependent variable may be relatively small.
From the chart above, it is obvious that age and minutes travelled to work have a
negative relationship. A one-year increase in age will, on average, decrease minutes
traveled to work by .077; showing that, as people grow older they get closer to work. The
relationship between the two variables was expected. Simply put, the older one grows,
the more they value their time. This could be a representation of their growing
opportunity costs or it could be a reflection of their earnings, making it hard to
12. 12
distinguish whether older people value their time more because they have less of it or
because their earnings are relatively higher.
Occupation has a positive relationship with minutes driven to work; this result
was also expected. As previously noted, occupations were classified under standard
occupational codes; where codes of higher values often represent occupations with
relatively lower wage rates. Thus, individuals classified under higher occupational codes
earn less, under assumption they also value their time less, and subsequently they take
more time to get to work. It is important to note that the economic effect that occupation
has on the dependent variable is much lower than expected.
The same can be said for the variable “total personal earnings” with respect to
economic impact. Surprisingly, though, earnings have a positive impact on time travelled
to work. A large portion of the text has been dedicated to relaying the message that
income positively affects time valued so that individuals with higher incomes value their
time more; yet, the data do not seem to relay the same message and seems
counterintuitive.
Marital status has a negative impact on minutes driven to work. A value higher
than one for this variable indicates an individual without a spouse, making the
relationship between this independent variable and the dependent variable interesting. It
would seem that an individual with a spouse would value their time more and in turn
would spend less time commuting to work; while individuals without a spouse would
exhibit opposite behavior. These results may be indicative of human behavior and
preferences for which this simple model could not account. This variable had the greatest
economic impact.
13. 13
The coefficient on sex shows that men on average spend more time commuting to
work than women. This variable, too, had a relatively larger economic effect on commute
time.
The coefficient on the number of children has a significant economic impact
relative to many of the other variables. The coefficient is also negative, showing that the
more children one has, the less time they spend commuting to work.
As far as the variable “mode of transport to work” is concerned, intuitively it was
estimated that this variable would have a negative impact on time travelled to work since
higher values of the variable indicated less comfortable modes of transportation.
However, the model displays an opposite effect. This is likely due to the fact that the less
comfortable modes of transportation require longer commutes on average than personal
automobiles.
The model poses several problems besides heteroskedasticity that cause the model
itself many limitations. The first being omitted variable bias. It is likely that many
relevant variables were not included in the model and that this omission heavily swayed
the results. Such variables may be whether the individual lived in an urban area, whether
they worked in an urban area, whether they were the head of the household, and their
employment status.
There could also be spatial correlation in the model, meaning that some of the
variables may be correlated with each other based on geography. For instance, in densely
populated areas (such as the northeast) public transit is relatively easier to access and
more often available; people living in these areas also may be less likely to own several
cars per household.
14. 14
Lastly, it is likely that multicollinearity exists in the model. Multicollinearity is a
case when independent variables are correlated. This issue would be most prevalent
among the occupation and earnings variables. However many of the other variables are
also likely to be correlated, such as age and earnings or age and marital status.
Policy Implications
The model developed in this paper may not necessarily provide a clear outline of
potential transportation policies that are fitting to the information analyzed. However, it
does give a new perspective on how to approach policy development.
In theory the improvement of transit service lies in its reduction of time and
increase in comfort. This could be achieved through policies aimed at reducing traffic
congestion or frequency of service for public transit. As well as improving comfort levels
of public transit services.
Though this paper’s model may not offer much insight into transportation
policies, it does give a new perspective on the people who are most affected by policies
aimed at reducing travel time and improving travel comfort. The model shows readers
that marital status and number of children have the greatest impact on time traveled to
work. Thus policies geared towards reducing commute time are likely to impact married
adults with relatively few children. As noted previously higher earners value their time
more, yet policies that improve time spent in commute may not necessarily benefit high
earners the most. Benefits will also be greatly realized by lower income households that
do not have the capability of moving closer to work.
15. 15
Conclusion/Discussion
This paper utilizes both theory and data to prove specific theories of
transportation economics. From theory, readers have been shown that time is divided
amongst labor, leisure, and travel. Theory also suggests that individuals value time based
on fifty percent of their real wage rate. Through closer examination of the valuation of
time, it became apparent that time valuations differed based on the flow, comfort, and
reliability of travel.
The model utilized in the paper offered deeper insight into personal factors that
impact the time travelled to work and offered a small glimpse of the decision making
process. The data showed that marital status and number of children had relatively large
impacts on the minutes travelled to work. Total personal earnings and occupation appear
to be important factors in determining commute times, but the model showed that these
variables on average had small economic impacts on time travelled.
The model however posed several complications that limited it use. It omitted
several relevant and important variables and also did not include any variables that
indicated where the individuals lived or worked. It was also unable to account for
attitudes or preferences, which may greatly impact how one travels to work and
subsequently how much time they travel to work.
16. 16
References
Travel Time. (2013). In Transportation Cost and Benefit Analysis II. Victoria Transport
Policy Institute.
Litman, T. (n.d.). Valuing Transit Service Quality Improvements. Journal of Public
Transportation, 11(2). Retrieved from http://www.nctr.usf.edu/jpt/pdf/JPT11-
2Litman.pdf
Brundell-Freij, Karin. 2006. User benefits and time in road investment and maintenance:
The role of speed choice and driving comfort. Transportation Research Board Annual
Meeting. Available at www.mdt.mt.gov/research/docs/trb_cd/ Files/06-0158.pdf.
Belenky, P. (2011). Memorandum to Secretarial Officers Modal Administrators. U.S.
Department of Transportation. Retrieved from
http://www.dot.gov/sites/dot.dev/files/docs/vot_guidance_092811c.pdf
Travel Time - Transportation Benefit-Cost Analysis. (n.d.). Retrieved from
http://bca.transportationeconomics.org/benefits/travel-time2
Small, K. (n.d.). Valuation of Travel Time. Economics of Transportation, 1(1). Retrieved
from http://www.socsci.uci.edu/~ksmall/VOT review.pdf
17. 17
Becker, G. (n.d.). A Theory of the Allocation of Time. The Economic Journal. Retrieved
from http://www.unc.edu/~shanda/courses/plcy289/Becker_EJ_Time.pdf