"ws0 v0">58 0.82 0.25 0.58 0.35 1.08
lif_cyc2 2.68 3.78 4.12 1.16 1.43 0.17
lif_cyc3 3.06 2.72 4.60 0.71 1.52 2.63
lif_cyc4 3.47 2.04 0.43 2.09 4.39 2.54
lif_cyc5 7.15 8.82 4.49 1.48 4.69 7.44
lif_cyc6 1.90 2.86 2.42 1.05 0.37 0.93
loghhsize
− −
For Ts
6.30 5.68 5.85 1.36 1.70 0.26
nworkers 0.47 0.25 2.52 0.33 4.13 4.40
nvehicles 5.13 7.66 0.46 2.50 7.12 12.72
income 4.51 7.74 9.13 4.00 5.88 1.95
ndrivers 12.05 10.81 5.99 2.69 9.22 7.52
constant 3.56 4.30 4.80 0.58 1.04 0.51
otal Trip
loghhsize 23.18 31.28 6.02 13.67 24.81 15.88
nworkers 27.81 32.93 23.68
α = 0.05 Critical value is ± 1.96
67.0159.12 24.24
income 34.51 34.12 34.55 71.60 72.13 5.45
ndrivers 24.04 27.84 31.67 59.75 66.84 3.06
constant 13.00 17.67 19.29 45.16 51.39 0.99
Copyright © 2012 SciRes. JTTs
J. L. MWAKALONGE ET AL.
292
to as the transfer bias. Lfferences mag-
nitudeges in travel behavior and their respective
n Data Using Models
Usua
text; a
modeability to replicate travel in the estimation
arger disignify the
of chan
explanatory variables from the estimation to application
time period. In general, the t-statistics yielded by the
total travel models were larger compared to those yielded
by the non-motorized models.
3.3. Prediction of Non-Motorized and Total
Travel in the Estimatio
Estimated on 1990, 1995, 2001, and 2009
Data
lly, little is known to a modeler about future con-
hence, the modeler selects the best model based on
l’s cap
context. Thus, in this study, the estimated models were
applied to predict non-motorized and total travel, each in
its respective estimation dataset. The Mean Absolute
Error (MAE) and Percent Error (PE) were used to assess
the models’ capability in forecasting travel in the estima-
tion data. The MAE measure has been used by other
studies [18,19]. MAE is calculated as:
1
1
ˆ
N
iii
i
N
i
i
yyw
MAE
w
where
is the observed number of non-motorized trips pro-
by household i,
the predicted number of non-motorized trips pro-
ctor, and
N
s are presented in Table 5, which
sh increased in magni-
tuor total travel,
th
le
6,
ic
Constant
i
y
duced
ˆi
yis
duced by household i,
wi is the weighting fa
is the sample size, all in a given year.
The prediction result
ows that the error measure, MAE,
de over time for non-motorized travel. F
e MAE value was lower in 1990 and higher in 1995.
Further analysis was done to identify in which part of
the US the models were unable to predict non-motorized
travel demand. The results, which are presented in Tab
consistently showed that all of the models studied
over-predicted non-motorized travel in the South Central
tion on estimation data.
Coefficient R-Squared MAE
Table 5. Outputs for pred
Model Trip Type
Non-Motorized0.0920 0.911 2.43E-09 1
1990 Total 5.76E-09 1 0.3102 3.747
Non-Motorized 1.06E-09 1 0.0906 0.955
1995 Total 1.85E-08 1 0.3083 4.767
Non-Motorized 1.79E-09 1 0.0749 1.214
2001
Total 1.01E-08 1 0.3448 4.523
Non-Motorized 1.32E-09 1 0.0851 1.396
2009
Total 2.88E-08 1 0.3344 3.912
Table 6d predicted otorized trips byivision for each a year. . Observed antotal non-m dnalysis
Model New Middle East South East Wes
Pacific
Observed 212 1118
England Atlantic North
Central
North
Central Atlantic South
Central
South
Central
Mountain
West t
738 2241 1235 291 787 163 382
Predicted
1990
Prd 1.
1995
Pr1 1
2001
Pr 3 12009
PE 6.7 0.9 23 40.6 26.8 81.8 51.5 8.5 11.5
831 1,602 1,426 339 976 305 658 277 978
PE 12.6 28.5 15.5 16.5 24 87.3 72.2 30.7 12.5
Observed 3314 8008 1388 389 1196 169 1208 475 1800
edicte4.064 6.775 1,480 426 644314 2.564 375 1.667
PE 22.6 15.4 6.6 9.4 37.4 85.7 112.3 21 7.4
Observed 914 13,862 12,681 1404 5.685 523 1736 1110 4339
edicted 901 12,302 3,256,8405.903 1.060 3.298 1040 4.257
PE 1.4 11.3 4.5 31.1 3.8 102.8 90 6.3 1.9
Observed 1548 13110 3588 3283 26123 961 11110 6321 20955
edicted 1444 13229 44144615 31301747 6829 6858 18551
Copyright © 2012 SciRes. JTTs
J. L. MWAKALONGE ET AL. 293
Divisionsnclutes lxas
se redicn-md traell fo
the Pacific Division, which includes CaliThe
ckluster prmance of all the presenteddels i
data
appl
scribex (TI) was used to
, which ide staike Teand Tennes-
ee. Thmodels pted nootorizevel w
fornia.
r
la erfo mon
predicting travel in the East South Central Division may
be attributed to low sample representation compared to
other divisions. The differences in the models’ ability to
predict travel in the different divisions implies differ-
ences in travel behavior exhibited by households in the
respective divisions. Divisions with lower PE values may
have the advantage of using the national travel survey
datasets for understanding non-motorized travel patterns
and behavior in their regions and updating their travel
demand models. This may be further investigated by
formulating models for each division.
3.4. Prediction of Non-Motorized and Total
Travel in 2009 Using Models Estimated on
1990, 1995, and 2001 Data
The models estimated on the 1990, 1995, and 2001
were used to predict non-motorized and total travel in the
ication contexts. Along with the MAE value de-
ed previously, the Transfer Ind
measure how well the estimation context model predicted
the observed trips in the application context relative to
the application context model. Mathematically, TI is ex-
pressed as [20]:
2
transferred
2
local
R
TI R
The results are presented in Table 7, which illustrates
that the magnitude of the resulting error measure from
the application of the 1990 model to predict travel de-
mand in the three periodased wime. T0
model better explainedmotorized travel at the
houseeve1995or 20s indicatby
the lvalue of the trr indexe 199el
pectively. This was done to investi-
ga
-
tant
s increith the 199
non-
than fhold ll for 01, aed
arge ansfe. Th5 mod
better explained trip variability for 2009 than 2001, even
though the reverse would be expected since one would
expect fewer changes in land use, household structure,
and travel patterns in 2001 than in 2009. The 2001 model
explained travel relatively well for the 2009 local model.
With respect to total travel, all the transferred models
explained trip variation less than the local models, as re-
flected by their respective lower TI values. From Table 7,
it is evident that the total travel models transferred better
in time than the non-motorized models. It is worth noting
that the MAE value yielded by the 2009 model’s applica-
tion in estimation data was large compared to the trans-
ferred models. This may be due in part to changes in
variables explaining non-motorized travel that are not spe-
cified in the model.
Further, a comparison of the models’ ability to predict
observed trips in 2009 by household size and household
income was undertaken, and the results are depicted in
Figures 3 and 4, res
Model Application Trip Type Cons
te whether modelers should model by market segmen-
tation rather than the unified approach employed in this
study. As observed in Figure 3, the relationship between
non-motorized and total trip rate and household size for
both 2009 observed trips and predicted trips by each-
model was non-linear, with trip rate increasing monotoni-
cally with increased household size. This was expected
since the proportion of children in a household, who make
fewer trips on average than adults, increased as house-
hold size grew. In addition, the trip rate increase de-
creased with the increase in household size, which was
motorized travel in future contexts.
Coefficient Transfer R2 MAE TI
Table 7. Outputs for prediction of non
Non-Motorized 0.1107 0.9247 0.0741 0.946 0.87
1995
4.881 0.92
Nd 0
2001
Non-zed
1990
2009
Non-zed
2001
Non-zed
1995
2009
Non-zed
2001 2009
Total 1.0893 1.1582 0.2833
on-Motorize 0.5223 0.7815 0.0555 1.106 .65
Total 0.7938 1.1551 0.3042 4.655 0.88
Motori 0.6447 0.9746 0.0684 1.237 0.80
Total 0.7369 1.0671 0.3213 3.910 0.96
Motori 0.3794 0.8882 0.0703 1.102 0.83
Total 0.4849 1.0208 0.3424 4.551 0.99
Motori 0.4776 1.0487 0.0733 1.238 0.86
Total 0.1068 0.8759 0.3253 4.203 0.97
Motori0.0036 1.1965 0.0800 1.331 0.94
Total 0.3501 0.8385 0.3182 4.209 0.95
Copyright © 2012 SciRes. JTTs
J. L. MWAKALONGE ET AL.
294
0
0. 5
1
1. 5
2
2. 5
3
3. 5
4
0246810
Household siz e
trip r ate
obs erved
2009
2001
1995
1990
0
5
10
15
te
20
25
02468
10
Househol d size
trip ra
obvser9ved 200
2009
2001
1995
1990
(a) (b)
Figure 3. Comparative picture of observed household trip rate in 2009 versus predicted trip rate, by household size, using
2009, 2001, 1995, and 1990 models.
0
0. 2
0. 4
0. 6
0. 8
1
1. 2
051015
Income category
trip rate
obs erved
2009
2001
1995
1990
on-motorized travel
0
2
4
6
8
10
12
14
051015
Income Category
tri p rate
obs erved 2009
2009
2001
1995
1990
Total trips
(a) (b)
Figure 4. Comparative picture of observed household trip rate in 2009 versus predicted trip rate, by household income, using
2009, 2001, 1995, and 1990 models.
also expected since some tr
With respect to non-motorized travel, the rate of in-
ewer persons. The second group, with a
m
higher for low-income and high-income households but
seholds. The low performance
by the low likelihood of this group to make non-moto-
l
travel and urban characteristics included the following:
ip purposes, such as grocery low for medium-income hou
shopping, can be done by one household member for the
whole family.
of the 1990 and 1995 models in predicting travel in 2009
for households that were more affluent can be explained
crease showed three household groups. The first group,
which had the larger rate, was composed of households
with four or f
edium trip rate increase, included households with five
or six persons, and the third group, which has the lowest
trip rate increase, was composed of households with seven
or more persons. Of all the models, the 2001 model bet-
ter predicted trips in 2009 for households with five or
fewer people than any other model. For total travel, all
else being equal, the 1995 model better predicted travel
in 2009 for a single-person household. The 1990 model
better predicted travel for two- and three-person house-
holds, the 2009 model better predicted travel for four-
person households, and the 1995 and 2001 models bet-
ter predicted travel for households with five or more per-
sons.
With respect to income, non-motorized trip rates were
rized trips in those years. This trend was not the same for
this group in 2009, when members made more non-mo-
torized trips. As observed in Figure 4, with regard to
income, the 2001 model predicted non-motorized trips
better than any of the other studied models. In contrast,
for total travel, all models exhibited a similar trend and
showed three distinct groups. As expected, the estimated
2009 model best predicted the observed travel in 2009.
4. Summary and Conclusions
This study investigated changes in non-motorized and
total travel and the characteristics of the traveling public
that are relevant to non-motorized modeling in 1990,
1995, 2001, and 2009. The study also investigated the
temporal transferability of linear-regression trip genera-
tion models for non-motorized and total travel under such
changes. The relevant changes in non-motorized and tota
Copyright © 2012 SciRes. JTTs
J. L. MWAKALONGE ET AL. 295
In general, high-income households made few non-
to
ing land use vari
[3] Urban Transportation Caucus, “Urban Transportation Re-
port Card, Sann, Cascade Bicycle
Club, and Chieration Transporta-
ute, The Texas A & M Uni-
rtation Planning Models: Summary
el Model Improvement Program,
HTS
5.
motorized trips in 1990 and 1995. However, the trend
changed in 2001 and 2009 for this group, with an in-
crease in non-motorized trips. Persons aged 50 and
over showed an increased demand for non-motorized
travel, whereas children aged 0 - 15 showed a de-
creasing preference for non-motorized travel over time.
With the exception of Lifecycle 2, 3, and 5, all the
household structures showed an increasing demand in
non-motorized travel over time.
Change in the composition of households was reflec-
ted in an increase in the number of workers per
household, an increase in the number of vehicles per
worker, and a decrease in the household size.
With the exception of 2009, there was an increase in
the average total number of trips made per household,
notwithstanding the decline in average household size.
This suggests that changes in the socioeconomic
structure of households and lifestyle changes were in-
fluential factors in the increase in trips.
The empirical investigation on the temporal transfer-
ability of non-motorized and total trip generation models
yielded the following findings:
The increase in vehicle ownership impacted the de-
mand for non-motorized travel negatively. However,
further analysis is required to identify the level of in-
ter-relationship between the number of vehicles owned
and the number of non-motorized trips made for a
given household.
For non-motorized travel, only the coefficient for sin-
gle-adult households with no children was stable
across all of the analysis years. For both non-moto-
rized and total travel, most model parameter estimates
were stable short term but not long term.
With respect to the models’ transferability to 2009,
the 2001 model predicted travel better than the 1990
and 1995 models. Further, the models’ ability to pre-
dict travel in future contexts decreased with increas-
ing time between estimation and application contexts.
This indicates the inability of a model to capture large
changes in urban structure and travel behavior.
In general, in all analysis years, the total travel mo-
dels transferred better than non-motorized models.
In general, despite not finding universal stability in
model parameter estimates, the models were marginally
able to replicate travel in 2009 relative to the locally es-
timated 2009 model. This study gives a general picture of
the temporal transferability of non-motorized travel com-
pared to total travel using the available national datasets.
More research is required, particularly at the regional
level, to understand a region’s specific changes in land
use and travel behavior and their influence on non-mo-
rized travel. Further, well-specified models incorporat-
ables may be appropriate where the data
are available to improve models’ ability to explain varia-
tions of non-motorized travel.
REFERENCES
[1] M. R. Baltes, “Factors Influencing Nondiscretionary Work
Trips by Bicycle Determined from 1990 US Census Me-
tropolitan Statistical Area Data, Transportation Research
Record 1538,” Transportation Research Board, Washing-
ton DC, 1996, pp. 96-101.
[2] C. R. Bhat, J. Y. Guo and R. Sardesai, “Non-Motorized
Travel in the San Francisco Bay Area,” University of
Texas, Austin, 2005.
Francisco Bike Coalitio
cago Land Bicycle Fed
tion Alternatives, 2007.
[4] Urban Mobility Report, Texas Transportation Institute,
The Texas A & M University System, Austin, 2004.
[5] D. Shrank and T. Lomax, 2009 Urban Mobility Report,
Texas Transportation Instit
versity System, Austin, 2009.
[6] M. Replogle, “Integrating Pedestrian and Bicycle Factors
into Regional Transpo
of the State-Of-The-Art and Suggested Steps Forward,”
Urban Design, Telecommunication and Travel Forecast-
ing Conference: Summary, Recommendations and Com-
pendium of Papers, Trav
Williamsburg, 27-30 October, 1996, pp. 75-94.
[7] K. J. Crifton and K. J. Krizek, “The Utility of the N
in Understanding Bicycle and Pedestrian Travel,” Na-
tional Household Travel Survey Conference: Under-
standing Our Nations Travel, Washington DC, 1-2 No-
vember 2004.
[8] US Department of Transportation, Bureau of Transporta-
tion Statistics, Bicycle and Pedestrian Data Sources, Needs
and Gaps, Washington DC, 2000.
[9] C. R. Emond, W. Tang and S. L. Handy, “Explaining Gen-
der Difference in Bicycling Behavior,” Transportation
Research Board, Washington DC, 2009, pp. 16-2
[10] R. Cervero and M. Duncan, “Walking, Bicycling, and
Urban Landscapes: Evidence from the San Francisco Bay
Area,” American Journal of Public Health, Vol. 93, No. 9,
2003, pp. 1478-1483. doi:10.2105/AJPH.93.9.1478
[11] E. J. Kannel and K. W. Heathington, “Temporal Stability
Shiftan, “Spatial and
o. 1,
Collected in Single and
of Trip Generation Relations,” Highway Research Record,
Vol. 472, 1973, pp. 17-27.
[12] A. Cotrus, J. N. Prashker and Y.
Temporal Transferability of Trip Generation Demand
Models in Israel, Bureau of Transportation Statistics,”
Journal of Transportation and Statistics, Vol. 8 N
2003, pp. 37-56.
[13] Oak Ridge National Laboratory, “The 2001 National
Household Travel Survey,” Oak Ridge National Labora-
tory for the US Department of Transportation, 2004.
http:/nhts.ornl.gov/publications.html
[14] J. L. Mwakalonge, “Econometric Modeling of Total Ur-
ban Travel Demand Using Data
Copyright © 2012 SciRes. JTTs
J. L. MWAKALONGE ET AL.
Copyright © 2012 SciRes. JTTs
296
he Estimation of Trip Generation and
hold
, 2008, pp. 195-206. Repeated Cross-Sectional Surveys,” Ph.D. Dissertation,
Department of Civil and Environmental Engineering, Ten-
nessee Technological University, Cookeville, 2010.
[15] K. G. Goulias, R. M. Pendyala and R. Kitamura, “Practi-
cal Method for t
Trip Chaining,” Transportation Research Board, Washing-
ton DC, 1991, pp. 47-56.
[16] Oak Ridge National Laboratory, “The National House
Travel Survey,” Oak Ridge National Laboratory for the US
Department of Transportation, 2010.
http:/nhts.ornl.gov/publications.html
[17] G. F. Ulfarsson and V. N. Shankar, “Children’s Travel to
School: Discrete Choice Modeling of Correlated Moto-
rized and Non-Motorized Transportation Modes Using
Covariance Heterogeneity,” Environment and Planning B:
Planning and Design, Vol. 35, No. 2
doi:10.1068/b3360
[18] A. Hadayeghi, A. S. Shalaby, B. N. Persaud and C. Cheung,
“Temporal Transferability and Updating of Zonal Level
Accident Prediction Models,” Accident Analysis & Pre-
vention, Vol. 38, No. 3, 2006, pp. 579-589.
doi:10.1016/j.aap.2005.12.003
[19] D. A. Badoe, “An Investigation into the Long Range
erability Ana-
dels,” Transportation Re-
Transferability of Work-Trip Discrete Mode Choice Mo-
dels,” Ph.D. Dissertation, University of Toronto, Toronto,
1994.
[20] F. S. Koppelman and C. G. Wilmot, “Transf
lysis of Disaggregate Choice Mo
search Board, Washington DC, 1982, pp. 18-24.