Week 07: Trip Distribution

CEE 6505: Transportation Planning

Department of Civil and Environmental Engineering
Islamic University of Technology (IUT), Bangladesh
Content
• Definition and Notation
• Common methods in Trip Distribution
– Growth-Factor Model
– Synthetic Model (Gravity)
– Entropy Maximization
• Calibration of Gravity Models
• Practical Considerations
Introduction
Trip Distribution
• Where will the generated trips go?
• A distribution model tries to estimate the number of trips in
each of the matrix cells on the basis of any information
available
• Travel pattern can be represented in two ways (at least):
– Trip matrix
– Production-Attraction (P-A) basis

3
Introduction
• Trip matrix
– stores the trips made from an Origin to a Destination during a
particular time period
– Often called O-D matrix
– can be disaggregated by person type, purpose or the activity
undertaken at each end of the trip further categorizing OD matrix by HB, NHB trips,
peak hour, off peak hour trips
– Disaggregation is needed for assignment models

4
Introduction
• P-A basis
– considers the factors that generate and attract trips
– Home generally being treated as the ‘producing’ end, and Work,
Shop etc. as the ‘attracting’ end.
– cover a longer time span than O-D matrix (why? Hint: on a P-A basis
the Home end will generate two school trips and the School end will
attract two school trips during the day) OD: Home to Office kotokkhon lage
PA: Home to Office to Home kotokkhon lage (longer)
– Even the combination of intercept and home interview surveys will
fail to produce matrices where all cells have been sampled
– Modelling is required!

5
Introduction
• OD-matrix methods are normally for short-term modelling and P-A basis
methods are for long-term modelling
• In general trip distribution is treated as an aggregated modelling problem.
• However, methods exist to do the modelling in disaggregated level (up to
individual level)
• Common methods:
– Growth-Factor Methods will be covered in
our syllabus
– Gravity Distribution Model
– Entropy-Maximizing Approach
– Other locally used methods (e.g., Detroit method)
6
Definitions & Notations
• Trip Matrix - a two dimensional array of cells where rows and columns
represent each of the z zones in the study area

sum_O=sum_D=
Total

sum of every trip with destination 1 7

Definitions & Notations
• Cells of each row i contain the trips originating in that zones having destinations
in the corresponding columns (j).
• The main diagonal corresponds to intra-zone trips
• 𝑇𝑖𝑗 = the number of trips between origin i and destination j; the total array is {𝑇𝑖𝑗 }
or T. target year

• 𝑂𝑖 = the total number of trips originating in zone i

• 𝐷𝑗 = the total number of trips attracted to zone j
• 𝑃𝑖 = the number of trips produced or generated in a zone i
• 𝑄𝑗 = those attracted to zone j base year
• We shall use lower case letters, tij, oi, and dj to indicate observations from a
sample from an earlier study or base year.
• Capital letters represent the target year. 8
Definitions & Notations
• The matrices can be further disaggregated, for example, by person type (n)
and/or by mode (k)
• Tknij = trips from i to j by mode k and person type n;
• Okni = the total number of trips originating at zone i by mode k and person
type n
• Summation over sub- or superscripts will be indicated implicitly by omission

sub-category

• pkij = the proportion of trips from i to j by mode k;

• ckij = the cost of travelling between i and j by mode k. 9
OD
Definitions & Notations
• The sum of the trips in a row should equal the total number of trips
emanating from that zone; sum_trips_row =
total_production_trips
• The sum of the trips in a column should correspond to the number of trips
attracted to that zone
• The following relationships must be fulfilled:
෍ 𝑇𝑖𝑗 = 𝑂𝑖 ; ෍ 𝑇𝑖𝑗 = 𝐷𝑗
𝑗 𝑖

• Doubly constrained model: reliable information is available to estimate both

𝑂𝑖 and 𝐷𝑗 and the model satisfy both conditions
• Singly constrained model: information only about one of these constraints is
available (in most cases 𝑂𝑖 . Why?)
10
 for every other cost
that we couldn't
account for. Like +c
Definitions & Notations when we do
integration function.

• Generalized cost of travel - a measure combining all the main attributes

related to the disutility of a journey. discomfort

• The cost element may be considered in terms of distance, time or money

units.
• This is typically a linear function of the attributes of the journey weighted
by coefficients which attempt to represent their relative importance as
perceived by the traveler.
• One possible representation of this for mode k is (omitting superscript k for
Fare. For car, fare
simplicity) in-vehicle driving waiting for PT =0 but there might
Cost (in time or
time time be tolls.
money or distance).
Time is best because
money loses value.
𝑣
𝐶𝑖𝑗 = 𝑎1 𝑡𝑖𝑗 𝑤
+ 𝑎2 𝑡𝑖𝑗 𝑡
+ 𝑎3 𝑡𝑖𝑗 𝑛
+ 𝑎4 𝑡𝑖𝑗 + 𝑎5 𝐹𝑖𝑗 + 𝑎6 𝑗 + 
parking cost, for
weighting coefficients, units can walking to parking interchange time bus=0 11
vary based on unit of C_ij. lot/bus stop time (changing PT)
Definitions & Notations
𝑣 𝑤 𝑡 𝑛
𝐶𝑖𝑗 = 𝑎1 𝑡𝑖𝑗 + 𝑎2 𝑡𝑖𝑗 + 𝑎3 𝑡𝑖𝑗 + 𝑎4 𝑡𝑖𝑗 + 𝑎5𝐹𝑖𝑗 + 𝑎6𝑗 + 

Where,
𝑣
𝑡𝑖𝑗 is the in-vehicle travel time between i and j;
𝑤
𝑡𝑖𝑗 is the walking time to and from stops (stations) or from parking area/lot;
𝑡
𝑡𝑖𝑗 is the waiting time at stops (or time spent searching for a parking space);
𝑛
𝑡𝑖𝑗 is the interchange time, if any;
𝐹𝑖𝑗 is a monetary charge: the fare charged to travel between i and j or the cost of using the car for that journey,
including any tolls or congestion charges (note that car operating costs are often not well perceived and that
electronic means of payment tend to blur somehow the link between use and payment);
𝑗 is a terminal (typically parking) cost associated with the journey from i to j;
 is a model penalty, a parameter representing all other attributes not included in the generalized measure so
far, e.g. safety, comfort and convenience;
𝑎1……6 are weights attached to each element of cost; they have dimensions appropriate for conversion of all
attributes to common units, e.g. money or time
12
Definitions & Notations
• Generalized cost of travel can be in money unit or time unit
• If it is in money unit (a5=1) then 𝑎1 is interpreted as the value of time (or
more precisely the value of in-vehicle time)
• 𝑎2 and 𝑎3 would be the values of walking and waiting time
• In general, 𝑎2 and 𝑎3 are taken as 2-3 times higher than 𝑎1
• 𝑎1……6 are often provided externally to the modelling process, sometimes
specified by government. This presumes stability and transferability of
values for which there is, so far, only limited evidence.
• What are the theoretical and practical advantages in measuring generalized
cost in time units? (Hint: effect of income levels increasing)

13
Growth Factor Methods
• Growth-factor methods update the trip matrix by using a growth factor –
one for the whole matrix or a different one for each zone.
• Three methods:
– Uniform growth factor
– Singly constrained
– Doubly constrained
• Uniform growth factor: If the only information available is about a general
growth rate τ for the whole of the study area, then we can only assume that
it will apply to each cell in the matrix as:
𝑇𝑖𝑗 = τ . 𝑡𝑖𝑗 for each pair i and j

14
Example: Uniform Growth Factor
Example 5.A Trips originating from zone 1, 2, 3 of a study area are 78, 92 and
82 respectively and those terminating at zones 1, 2, 3 are given as 88, 96 and
78 respectively. If the growth factor is 1.3 and the cost matrix is as shown
below, find the expanded origin-constrained growth trip table.

Base year trip matrix

1 2 3 𝒐𝒊
1 20 30 28 78
2 36 32 24 92
3 22 22 34 82
𝒅𝒋 88 96 78 252
15
Example: Uniform Growth Factor
Solution 5.A
The equation of uniform growth rate is
𝑇𝑖𝑗 = 𝑓 × 𝑡𝑖𝑗
where f is the uniform growth factor 𝑡𝑖𝑗 is the previous total number of trips, 𝑇𝑖𝑗 is the
expanded total number of trips.
Given growth factor = 1.3, Therefore, multiplying the growth factor with each of the cells in
the matrix gives the solution as shown below
1 2 3 𝑶𝒊
1 20×1.3=26 39 36.4 101.4
2 46.8 41.6 31.2 119.6
3 28.6 44.2 33.8 106.2
𝑫𝒋 101.4 124.8 101.4 327.6 16
Example: Growth Factor Methods
Example 5.1 Consider the simple four-by-four base-year trip matrix of table
below. If the growth in traffic in the study area is expected to be of 20% then
calculate the future trip distribution matrix.

1 2 3 4
1 5 50 100 200 355
2 50 5 100 300 455
3 50 100 5 100 255
4 100 200 250 20 570
205 355 455 620 1635
17
Example: Growth Factor Methods
Example 5.1

Answer:
1 2 3 4
1 6 60 120 240 426
2 60 6 120 360 546
3 60 120 6 120 306
4 120 240 300 24 684
246 426 546 744 1962
18
Growth Factor Methods
Singly Constrained Growth-Factor Methods
• information is available on the expected growth in trips originating in or
attracted to each zone.
• Origin specific (τi) or destination specific (τj) growth factors can be
employed to obtain the trip matrix:

𝑇𝑖𝑗 = τ𝑖 . 𝑡𝑖𝑗 for origin specific factors

𝑇𝑖𝑗 = τ𝑗 . 𝑡𝑖𝑗 for destination specific factors

19
Example: Growth Factor Methods
Example 5.2 Target year trip generation along with base year trip matrix is
given below. Derive the trip matrix for design year.

Target
1 2 3 4
𝑶𝒊
1 5 50 100 200 355 400
2 50 5 100 300 455 460
3 50 100 5 100 255 400
4 100 200 250 20 570 702
205 355 455 620 1635 1962

20
 Example: Growth Factor Methods
Solution 5.2
The equation of uniform growth rate is 𝑇𝑖𝑗 = 𝜏𝑖𝑗 . 𝑡𝑖𝑗
Here, 𝜏𝑖 = 𝑂𝑖 Τ𝑜𝑖 , i.e., =400/355=1.13
Hence, 𝑇12 = 1.13 x 5 = 5.6
Target
1 2 3 4
𝑶𝒊
1 5.6 56.3 112.7 225.4 400 400
2 50.5 5.1 101.1 303.3 460 460
3 78.4 156.9 7.8 159.6 400 400
4 123.2 246.3 307.9 24.6 702 702
257.7 464.6 529.5 701.2 1962 1962
21
Growth Factor Methods
Doubly Constrained Growth Factors
• information is available on the future number of trips originating and terminating in
each zone, i.e., different growth rates for trips in and out of each zone and
consequently having two sets of growth factors for each zone, say τ i and Ƭj.
• A poor solution is using an average growth factor:

• Fratar (US) or Furness (elsewhere) method introduces two balancing factors as Ai

and Bj as follows:
𝑇𝑖𝑗 = τ𝑖 . 𝑡𝑖𝑗 . 𝑗 . 𝐴𝑖 . 𝐵𝑗
• Or, incorporating the growth rates into new variables ai and bj

22
Growth Factor Methods
Doubly Constrained Growth Factors
• Steps in solution:
1. Set bj = 1
2. With bj solve for ai to satisfy trip generation constraint
3. With ai solve for bj to satisfy trip attraction constraint
4. Update matrix and check for errors
5. Repeat steps 2 and 3 till convergence
• Condition for convergence:
෍ τ𝑖 ෍ 𝑡𝑖𝑗 = ෍ 𝑗 ෍ 𝑡𝑖𝑗 = 𝑇
𝑖 𝑗 𝑖 𝑖
23
Growth Factor Methods
Doubly Constrained Growth Factors
• Perfect convergence is not common. But error must be lower than the
accepted threshold obtained by:

𝐸 = Σ 𝑂𝑖 − 𝑂𝑖1 + Σ|𝐷𝑗 − 𝐷𝑗1 |

• This method is often called a ‘bi-proportional algorithm’

24
Example: Doubly Constrained Growth Factor
Example 5.B The base year trip matrix for a study area consisting of three zones is
given below. The productions from the zone 1,2 and 3 for the horizon year is
expected to grow to 98, 106, and 122 respectively. The attractions from these zones
are expected to increase to 102, 118, 106 respectively. Compute the trip matrix for
the horizon year using doubly constrained growth factor model using Furness
method.
Base year trip matrix
1 2 3 𝑜𝑖
1 20 30 28 78
2 36 32 24 92
3 22 34 26 82
𝑑𝑗 88 96 78 252 25
Example: Doubly Constrained Growth Factor destination_growth_balance_factor
origin_growth_balance_factor
From, target_Oi/base_Oi
𝑇𝑖𝑗 = 𝑡𝑖𝑗 × 𝑎𝑖 × 𝑏𝑗 , assuming 𝑏𝑗 =1.0 =98/78=1.26

1 2 3 𝒐𝒊
1 20× 1.26 = 30 × 1.26 28 × 98
25.2 = 37.8 1.26 =35.28
2 36 × 1.15 32 × 1.15 24 × 1.15 106
= 41.4 = 36.8 = 27.6
3 22 × 1.49 34 × 1.49 26 × 1.49 122
= 32.78 = 50.66 = 38.74
𝒅𝟏𝒋 99.38 125.26 101.62
𝑫𝒋 102 118 106
sum_column=99.38 26
Example: Doubly Constrained Growth Factor
target_Dj
𝐷𝑗
Now, 𝑏𝑗 = and 𝑇𝑖𝑗 = 𝑡𝑖𝑗 × 𝑏𝑗 , here
𝑑𝑗1

dj_assuming_bj_1 102
𝑏1 = = 1.03
99.38
118
𝑏2 = = 0.94
125.26
106
𝑏3 = = 1.04
101.62

27
Example: Doubly Constrained Growth Factor
So,
1 2 3 𝑜𝑖 𝑂𝑖
1 25.2×1.03= 37.8 ×0.94= 35.28 × 1.04 98.18 98
25.96 35.53 = 36.69
2 41.4×1.03= 36.8 ×0.94= 27.6 × 105.93 106
42.64 34.59 1.04=28.70

3 32.78 ×1.0 50.66 ×0.94 38.74 × 1.04 121.67 122

3=33.76 =47.62 = 40.29
2 iteration er
beshi korte bolbe
na exam e
𝑏𝑗 1.03 0.94 1.04
ei value abar next iteration e use
𝐷𝑗 102 118 106 korbo naki b_j=1 e assume korbo
NEED TO VERIFY 28
Example: Doubly Constrained Growth Factor

1 2 3 𝑜𝑖 𝑂𝑖
1 25.96 35.53 36.69 98.18 98
2 42.64 34.59 28.70 105.93 106
3 33.76 47.62 40.29 121.67 122
𝑑𝑗 102.36 117.74 105.68 325.78
𝐷𝑗 102 118 106 326

29
Example: Doubly Constrained Growth Factor
Now, we can calculate the error:

Error = ෎|𝑂𝑖 − 𝑜𝑖 | + ෍|𝐷𝑗 − 𝑑𝑗 |

So,
Error = |98.18-98|+|105.93-106|+|121.67-122|+|102.36-102|+|117.74-
118|+|105.68-106| = 1.32
if less than given
threshold --> we
stop

30
Example: Growth Factor Methods
Example 5.2 Target year trip generation along with base year trip matrix is
given below. Derive the trip matrix for design year.
Target
1 2 3 4
𝑶𝒊
1 5 50 100 200 355 400
2 50 5 100 300 455 460
3 50 100 5 100 255 400
4 100 200 250 20 570 702
205 355 455 620 1635

Target 𝑫𝒋 260 400 500 802 1962

31
Example: Growth Factor Methods
Answer

Target
1 2 3 4
𝑶𝒊
1 5.25 44.12 98.24 254.25 401.85 400
2 45.30 3.81 84.78 329.11 462.99 460
3 77.04 129.50 7.21 186.58 400.34 400
4 132.41 222.57 309.77 32.07 696.82 702
260 400 500 802 1962

Target 𝑫𝒋 260 400 500 802 1962

32
Growth Factor Methods
• Advantages
– Simple to understand
– make direct use of observed trip matrices and forecasts of trip-end growth
• Limitations a lot of changes can
– Only suitable for short-term horizon happen within 5
years
– the methods do not take into account changes in transport costs due to
improvements (or new congestion) in the network
– Heavily depends on the accuracy of the base-year trip matrix
– if parts of the base-year matrix are unobserved, they will remain so in the
forecasts trip matrix e kono cell = 0 hoile future trip matrix eo 0 e thakbe
– Not common to use in the analysis of policy options involving new modes, new
links, pricing policies and new zones because growth factor doesn't account for changes 33
(e.g., land-use change)
trucks entering Dhaka then leaving
through highways (e.g., Mymensingh-
Dhaka highway)

When we use Growth Factor then?

1. For external trips (E-I, E-E, I-E)
because for example, considering
both the dynamics of Dhaka and
Rajshahi is very difficult.
2. When we need to predict within a
very short horizon (say 2 years)

To properly consider the dynamics of

Dhaka city, we need to use
something much more complex,
which is GRAVITY MODEL (next
slide)
Synthetic or Gravity Model
that's why we use
The Gravity Distribution Model it for internal-
internal trips

• Synthetic models estimate trips for each cell in the matrix without directly
base year OD matrix
using the observed trip pattern is not needed like in land-use changes
GF method

• Proves its efficiency when important changes in the network take place.

• They start from assumptions about group trip making behaviour and the
way this is influenced by external factors such as total trip ends and
cost, distance
distance travelled.
based on future
year trips

34
 Synthetic or Gravity Model
The Gravity Distribution Model
• The best known of these models is the gravity model, originally generated
from an analogy with Newton’s gravitational law as:
F=(m_1*m_2)*G/d^2

• Pi and Pj are the populations of the towns of origin and destination, dij is
the distance between i and j, and α is a proportionality factor (with units
trips·distance2/population2)
35
Synthetic or Gravity Model
The Gravity Distribution Model
• Gravity model can be generalized by assuming that the effect of distance
or ‘separation’ could be modelled better by a decreasing function, to be
specified, of the distance or travel cost between the zones as:
𝑇𝑖𝑗 =  𝑂𝑖 𝐷𝑗 𝑓(𝑐𝑖𝑗 ) used instead of d^2 to
better relate trips with
travel cost. Travel cost
beshi hoile trips kombe.
• where f(cij) is a generalized function of the travel costs with one or more
parameters for calibration
• f(cij) is often called ‘deterrence function’ because it represents the
disincentive to travel as distance (time) or cost increases
• α is the single proportionality factor
36
Synthetic or Gravity Model
The Gravity Distribution Model parameter (does
not change in
• Popular versions of the function f(cij) are: future)
which
𝑓(𝑐𝑖𝑗 ) = 𝑒𝑥𝑝(𝑐𝑖𝑗 ) exponential function one to
−𝑛 use?
𝑓(𝑐𝑖𝑗 ) = 𝑐𝑖𝑗 power function Depends
𝑛 on TLD.
𝑓 𝑐𝑖𝑗 = 𝑐𝑖𝑗 𝑒𝑥𝑝(𝑐𝑖𝑗 ) combined function
• First two each has one parameter for calibration.
• Third one has two parameters for calibration (which ones?) n and beta
• These parameters are estimated so that the results from the model reproduce, as
closely as possible, the trip length (cost) distribution (TLD) of the observations
• the greater the number of parameters, the easier it is to obtain a closer fit with the
sampled trip length distribution parameter joto beshi model toto beshi accurate, much
like LLMs (ChatGPT)
• Assumption: β does not change in future. 37
Synthetic or Gravity Model
The Gravity Distribution Model
• Graphical forms of popular versions of the function f(cij):

38
 korbo na

Synthetic or Gravity Model

Singly and Doubly Constrained Model
• The single proportionality factor α is replaced by two sets of balancing
factors Ai and Bj as:
𝑇𝑖𝑗 = 𝐴𝑖 𝑂𝑖 𝐵𝑗 𝐷𝑗 𝑓(𝑐𝑖𝑗 )
• Optionally, further, subsume Oi and Dj as:
𝑇𝑖𝑗 = 𝑎𝑖 𝑏𝑗 𝑓(𝑐𝑖𝑗 )
• Now,
Σ𝑖 𝑇𝑖𝑗 = Σ𝑖 𝐴𝑖 𝑂𝑖 𝐵𝑗 𝐷𝑗 𝑓(𝑐𝑖𝑗 )
• But,
Σ𝑖 𝑇𝑖𝑗 = 𝐷𝑗
39
Synthetic or Gravity Model
Singly and Doubly Constrained Model
• Hence,
𝐷𝑗 = 𝐵𝑗 𝐷𝑗 Σ𝑖 𝐴𝑖 𝑂𝑖 𝑓(𝑐𝑖𝑗 )
• And, similarly:
𝐴𝑖 = 1൘෍ 𝐵𝑗 𝐷𝑗 𝑓(𝑐𝑖𝑗 )
𝑗

𝐵𝑗 = 1൘෍ 𝐴𝑖 𝑂𝑖 𝑓(𝑐𝑖𝑗 )
𝑖
• For singly constrained models, one of the balancing factors (Bj for origin-
constrained and or Ai for destination constrained) is assumed to be 1.0
40
Synthetic or Gravity Model
Singly and Doubly Constrained Model
• Typical trip length distribution (TLD) in urban area:
• In order to calibrate deterrence Deterrence
function tries to
represent this
functions, the parameters are graph as closely
as possible

estimated so that the results from the

model reproduce, as closely as possible,
the trip length (cost) distribution (TLD)
of the observations
41
Gravity Model
Example 5.3 The productions from zone 1, 2 and 3 are 98, 106, 122 and attractions
1
to zone 1,2 and 3 are 102, 118, 106. The function is 𝑓(𝑐𝑖𝑗 ) = 2
𝑐𝑖𝑗

The cost matrix is as shown below 1.0, 1.2, 1.8, ...

1.0 1.2 1.8

1.2 1.0 1.5
1.8 1.5 1.0

42
Gravity Model
Solution 5.3
Steps of solving Gravity model
1. Set 𝐵𝑗 = 1, find 𝐴𝑖 using future_dest_trip
1
𝐴𝑖 =
෍ 𝐵𝑗 𝐷𝑗 𝑓(𝑐𝑖𝑗 ൯
𝑗
0

2. Find 𝐵𝑗 using equation future_origin_trip

1
𝐵𝑗 =
෍ 𝐴𝑖 𝑂𝑖 𝑓(𝑐𝑖𝑗 ൯
𝑖
The A_i we just
calculated 43
Gravity Model
3. Compute the error as

Error = ෌ |𝑂𝑖 − 𝑂𝑖1 | + ෍|𝐷𝑗 − 𝐷𝑗1|

A clear steps of Gravity model calculation is given from next slide

44
Gravity Model
The cost matrix is
1.0 1.2 1.8
1.2 1.0 1.5
1.8 1.5 1.0
1
The function 𝑓(𝑐𝑖𝑗 ) = 2 can be written as
𝑐𝑖𝑗

1 1 1
12 1.22 1.82
1 1 1 1 0.69 0.31
= 0.69 1 0.44
1.22 12 1.52 0.31 0.44 1
1 1 1
1.82 1.52 12
45
Gravity Model
D_j 102 118 106
1 1
𝐴𝑖 =
i j 𝐵𝑗 𝐷𝑗 𝑓(𝑐𝑖𝑗 ) = 2 𝐵𝑗 𝐷𝑗 𝑓(𝑐𝑖𝑗 ൯ ∑ 𝐵𝑗 𝐷𝑗 𝑓(𝑐𝑖𝑗 ൯ ෍ 𝐵𝑗 𝐷𝑗 𝑓(𝑐𝑖𝑗 ൯
𝑐𝑖𝑗 𝑗

1 1 102 1 102
1 2 1 118 0.69 81.42 216.28 0.00462
3 1 106 0.31 32.86
1 1 102 0.69 70.38
2 2 1 118 1 118 235.02 0.00425
3 1 106 0.44 46.64
1 1 102 0.31 31.62
3 2 1 118 0.44 51.92 189.54 0.00527
3 1 106 1 106
46
 O_i
98

Gravity Model 106

122

1 1
𝑩𝒋 =
j i 𝐴𝑖 𝑂𝑖 𝑓(𝑐𝑖𝑗 ) = 2 𝐴𝑖 𝑂𝑖 𝑓(𝑐𝑖𝑗 ൯ ∑𝐴𝑖 𝑂𝑖 𝑓(𝑐𝑖𝑗 ൯
෍𝐴𝑖 𝑂𝑖 𝑓(𝑐𝑖𝑗 ൯
𝑐𝑖𝑗
1 0.00462 98 1 0.4523
1 2 0.00425 106 0.69 0.3117 0.9618 1.0397
3 0.00527 122 0.31 0.1978
1 0.00462 98 0.69 0.3124
2 2 0.00425 106 1 0.4505 1.0458 0.9562
3 0.00527 122 0.44 0.2829
1 0.00462 98 0.31 0.1404
3 2 0.00425 106 0.44 0.1982 0.9815 1.0188
3 0.00527 122 1 0.6429
47
Gravity Model A_1 O_1
B_3 D_3 f(c_13)

𝑇11 = 0.00462 × 98 × 1.0397 × 102 × 1 = 48.01

𝑇12 = 0.00462 × 98 × 0.9562 × 118 × 0.69 = 35.24
𝑇13 = 0.00462 × 98 × 1.018 × 106 × 0.31 = 15.157
𝑇21 = 0.00425 × 106 × 1.0397 × 102 × 0.69 = 32.96
𝑇22 = 0.00425 × 106 × 0.9562 × 118 × 1 = 50.83
𝑇23 = 0.00425 × 106 × 1.018 × 106 × 0.44 = 21.40
𝑇31 = 0.00527 × 122 × 1.0397 × 102 × 0.31 = 21.14
𝑇32 = 0.00527 × 122 × 0.9562 × 118 × 0.44 = 31.92
𝑇33 = 0.00527 × 122 × 1.018 × 106 × 1 = 69.43

48
Gravity Model
known
predicted
Final Table
1 2 3 𝐴𝑖 𝑂𝑖 𝑂𝑖1
1 48.01 35.24 15.157 0.00462 98 98.407
2 32.96 50.83 21.40 0.00425 106 105.19
3 21.14 31.919 69.43 0.00527 122 122.489
𝐵𝑗 1.0397 0.9562 1.0188
𝐷𝑗 102 118 106
𝐷𝑗1 102.11 117.98 105.99

49
Gravity Model
known

Error = ෎|𝑂𝑖 − 𝑜𝑖 | + ෍|𝐷𝑗 − 𝑑𝑗 |

predicted

Error = |98-98.407|+|106-105.19|+|122-122.489|+|102-
102.11|+|118-117.98|+|106-105.99|= 2.03
very low error

50
Gravity Model
• Analogous to the Newton’s law of gravity:
𝑃𝑖 𝐴𝑗
𝑇𝑖𝑗  replacement of
𝑖𝑗 cost function for
PA based model
𝑇𝑖𝑗 = number of trips produced in zone i and attracted to zone j
• Directly proportional to:
- Total number of trips produced in zone i (𝑃𝑖 )
- Total number of trips attracted to zone j (𝐴𝑗 )
• Inversely proportional to:
- Impedance to travel between zones i and j (𝑖𝑗 )
- Impedance = spatial separation, travel time, travel cost, etc.
51
Gravity Model: Production-Attraction Based
𝑃 𝑖 𝐴𝑗
𝑇𝑖𝑗 = 𝑐𝑖  ; 𝑐𝑖 = constant of proportionality
𝑖𝑗

To solve for “𝑐𝑖 ” apply the production conservation equation

σ𝑘 𝑇𝑖𝑘 = 𝑃𝑖 ∀𝑖
𝑃 𝑖 𝐴𝑘 1
 σ𝑘 𝑐𝑘 𝑖𝑘 = 𝑃𝑖 ∀𝑖
𝐹𝑖𝑗 = 
𝑖𝑗

1 is called the friction factor

 𝑐𝑖 = 𝐴𝑘 ∀𝑖
σ𝑘
𝑖𝑘

𝐴𝑗
1 𝑃 𝑖 𝐴𝑗
(
𝑖𝑗) 𝐴𝑗 𝐹𝑖𝑗
𝑇𝑖𝑗 = 𝐴 (  ) = 𝑃𝑖 𝐴  𝑇𝑖𝑗 = 𝑃𝑖 [ σ ]
σ𝑘 𝑘 σ𝑘 𝑘 𝑘 𝐴𝑘 𝐹𝑖𝑘
𝑖𝑘 𝑖𝑗
𝑖𝑘
52
Gravity Model
𝐴𝑗 𝐹𝑖𝑗
𝑇𝑖𝑗 = 𝑃𝑖 [ σ ]  production-constrained gravity model
𝑘 𝐴𝑘 𝐹𝑖𝑘

𝐴1 𝐹11 𝐹11 = 1.5

[ σ4 ] = fraction of trips produced in zone 1 that are Zone 1
𝑖=1 𝐴𝑘 𝐹1𝑘 𝐴1 = 0
attracted to zone 1= 0
𝑃1 = 1000
𝐴 𝐹
[ σ4 2 12 ] = fraction of trips produced in zone 1 that are
𝑖=1 𝐴𝑘 𝐹1𝑘
attracted to zone 2= (0.3) Zone 2 Zone 3 Zone 4
𝐴3 𝐹13
[ σ4 ] = fraction of trips produced in zone 1 that are 𝐹14 = 2
𝑖=1 𝐴𝑘 𝐹1𝑘 𝐹12 = 3 𝐹13 = 1
attracted to zone 3= (0.3) 𝐴2 = 500 𝐴3 = 1500 𝐴4 = 1000
𝐴4 𝐹14
[ σ4 ] = fraction of trips produced in zone 1 that are
𝑖=1 𝐴𝑘 𝐹1𝑘
attracted to zone 4= (0.4)
53
 Gravity Model
1
𝐹𝑖𝑗 =  is called the friction factor
𝑖𝑗

• Captures the impedance to travel between zones i and j

obstruction/resistance

For example, 𝐹𝑖𝑗 = 𝑓(𝑡𝑖𝑗 )

Where, 𝑡𝑖𝑗 is the travel time between zones i and j

What is the functional form for “f”?

54
 Gravity Model
(1) Parametric forms for the friction factor function
𝐹𝑖𝑗 = (𝑡𝑖𝑗 )𝑎 𝑎 < 0

55
 Gravity Model
(1) Parametric forms for the friction factor function
𝐹𝑖𝑗 = 𝑒 𝑏𝑡𝑖𝑗 𝑏<0

56
 Gravity Model
(1) Parametric forms for the friction factor function
𝐹𝑖𝑗 = (𝑡𝑖𝑗 )𝑎 𝑒𝑏𝑡𝑖𝑗 𝑎 >0&𝑏 <0

57
 Gravity Model
(2) Non-Parametric form for the friction factor function
𝐹𝑖𝑗 = (𝑡𝑖𝑗 )𝑎 𝑒𝑏𝑡𝑖𝑗 𝑎 >0&𝑏 <0

𝐹1 𝑖𝑓 0 𝑡𝑖𝑗  5
𝐹2 𝑖𝑓 5 𝑡𝑖𝑗  10
𝐹𝑖𝑗 =
…
𝐹7 𝑖𝑓 30 𝑡𝑖𝑗  35

58
Doubly Constrained Gravity Model
𝑇𝑖𝑗 = 𝑐𝑖 𝑑𝑗 𝑃𝑖 𝐴𝑗 𝐹𝑖𝑗 ; 𝑐𝑖 𝑎𝑛𝑑 𝑑𝑗 = constants of proportionality

Applying the production conservation Applying the attraction conservation

equation, equation,
෍ 𝑇𝑖𝑘 = 𝑃𝑖 ∀𝑖 ෍ 𝑇𝑘𝑗 = 𝐴𝑗 ∀𝑗
𝑘 𝑘

෍ 𝑐𝑖 𝑑𝑘 𝑃𝑖 𝐴𝑘 𝐹𝑖𝑘 = 𝑃𝑖 ∀𝑖 ෍ 𝑐𝑘 𝑑𝑗 𝑃𝑘 𝐴𝑗 𝐹𝑘𝑗 = 𝐴𝑗 ∀𝑗
𝑘 𝑘
1
 𝑐𝑖 = σ ∀𝑖  𝑑𝑗 = σ
1
∀𝑗
𝑘 𝑑𝑘 𝐴𝑘 𝐹𝑖𝑘 𝑘 𝑐𝑘 𝑃𝑘 𝐹𝑘𝑗

59
Doubly Constrained Gravity Model
𝑇𝑖𝑗 = 𝑐𝑖 𝑑𝑗 𝑃𝑖 𝐴𝑗 𝐹𝑖𝑗
1
𝑐𝑖 = ∀𝑖
σ𝑘 𝑑𝑘 𝐴𝑘 𝐹𝑖𝑘
1
𝑑𝑗 = σ ∀𝑗
𝑘 𝑐𝑘 𝑃𝑘 𝐹𝑘𝑗

• This collapses to the production-constrained gravity model if 𝑑𝑗 = 1 ∀𝑗

• The 𝑐𝑖 and 𝑑𝑗 depend on each other. So, an iterative procedure is required

to determine 𝑇𝑖𝑗 given 𝑃𝑖 , 𝐴𝑗 and 𝐹𝑖𝑗
60
Doubly Constrained Gravity Model
Iterative procedure is required to determine 𝑇𝑖𝑗 given 𝑃𝑖 , 𝐴𝑗 and 𝐹𝑖𝑗

(1) Assume 𝑑𝑗 = 1 ∀𝑗
(2) Calculate 𝑐𝑖0 = (σ𝑘 𝑑𝑘0 𝐴𝑘 𝐹𝑖𝑘 )−1 ∀𝑖
(3) Set n = 1
(4) Calculate 𝑑𝑖𝑛 = (σ𝑘 𝑐𝑘𝑛−1 𝑃𝑘 𝐹𝑖𝑘 )−1 ∀𝑗
(5) Calculate 𝑐𝑖𝑛 = (σ𝑘 𝑑𝑘𝑛 𝐴𝑘 𝐹𝑖𝑘 )−1 ∀𝑖
(6) If 𝑐𝑖𝑛  𝑐𝑖𝑛−1 ∀𝑖 & 𝑑𝑗𝑛  𝑑𝑗𝑛−1 ∀𝑗 then, 𝑇𝑖𝑗 = 𝑐𝑖𝑛 𝑑𝑗𝑛 𝑃𝑖 𝐴𝑗 𝐹𝑖𝑗
else, set n = n + 1 and Go to STEP 4

61
Doubly Constrained Gravity Model
𝐴𝑗 𝐹𝑖𝑗
𝑇𝑖𝑗 = 𝑃𝑖 [σ ]  Production-constrained gravity model
𝑘 𝐴𝑘 𝐹𝑖𝑘

• “K” factors are sometimes added the gravity-model expression to take of

special socio-economic linkages that may exist between zonal pairs that
are not captured by the friction factor
introduced to deal
with cases where
there might be
𝐴𝑗 𝐹𝑖𝑗 less or more trips
𝑇𝑖𝑗 = 𝑃𝑖 [ ] based on context
σ𝑘 𝐴𝑘 𝐹𝑖𝑘

NOTE: The “K” factors will be 1 for most zonal pairs and some other value only
for select zonal-pairs
62
Gravity Model with ‘K’ Factors
Example: Gravity model for home based work trips

Let,
𝐹13 = 𝐹14 𝑇13 =?
𝐹23 = 𝐹24 𝑇14 =?
𝐴3 = 𝐴4 𝑇23 =?
𝐴1 = 𝐴2 = 0 𝑇24 =?
63
Gravity Model with ‘K’ Factors grav_model says
trips will be evenly
distributed.
Example: Gravity model for home based work trips But based on
context, we can
see that
high_income_zon
e will not make
trips to
low_income_jobs

Let, Gravity model says Expect:

𝐹13 = 𝐹14 𝑇13 = 0.5𝑃1 𝑇13 = 𝑃1
𝐹23 = 𝐹24 𝑇14 = 0.5𝑃1 𝑇14 = 0
𝐴3 = 𝐴4 𝑇23 = 0.5𝑃2 𝑇23 = 0
𝐴1 = 𝐴2 = 0 𝑇24 = 0.5𝑃2 𝑇24 = 𝑃2 64
Gravity Model with ‘K’ Factors
Example: Gravity model for home based work trips

Let, Add “K” factors to make the model

𝐹13 = 𝐹14 𝐾13 =1 sensitive to the special linkages
𝐹23 = 𝐹24 𝐾14 =0 𝐴𝑗 𝐹𝑖𝑗 𝐾𝑖𝑗
𝑇𝑖𝑗 = 𝑃𝑖 [ ]
𝐴3 = 𝐴4 𝐾23 =0 σ𝑘 𝐴𝑘 𝐹𝑖𝑘 𝐾𝑖𝑘
forcing T_14 to be
𝐴1 = 𝐴2 = 0 𝐾24 =1 zero 65
Gravity Model with ‘K’ Factors
Issues with the “K” factor based approach:
• These are rather arbitrary constant values added to the model to replicate certain
base-year effects used for base-year
• Starts tying the gravity model to the base-year zoning system – Note that the “K”
factors are specific to zonal pairs
• Stability of the implied linkage over time is unknown:
– What if the high-income employer relocates from zone 3?
– What is the zone 3 develops as a high-income neighborhood in the future
year?
• The “K” factors are also used to capture the effects of physical barriers
to simulate a
bridge collapse
66
Gravity Model vs Growth Factor Models
Gravity Models Growth Factor Models
• Not “tied” to the base-year zoning • “Tied” to the base-year zoning system
system (when there are no “K” factors) • Requires the entire base-year production-
• Do not require the entire base-year attraction (PA) matrix as an input
production-attraction (PA) matrix as an • The forecast-year flow for a zonal pair will
input full matrix not required, we only
need to future total trips be zero if the corresponding base-year
• The forecast-year flow for a zonal pair flow is zero suffers from base year flow 0
remaining permanently 0.
can be non-zero even if the • Not sensitive to transportation system
corresponding base-year flow is zero characteristics
does not change
• Sensitive to transportation system base year trip 0 with network
characteristics (via the friction factor hoileo pera nai changes

terms)
e.g., land-use
changes 67
Aggregate vs Disaggregate Model ekhan theke poray
nai

Growth factor and gravity models are “aggregate” models

We determine what fraction of the total trips produced in a

zone are attracted to each of the candidate destinations 𝑇11 = 0 Zone 1
Fraction to
Zone 1 = 0%
𝐴2 𝐹12 𝑃1 = 1000
[ σ4 ] = fraction of trips produced in zone 1 that are
𝑖=1 𝐴𝑘 𝐹1𝑘
attracted to zone 2= (0.2)
Zone 2 Zone 3 Zone 4

The friction-factor of impedance to travel between any two 𝑇12 = 200 𝑇13 = 500 𝑇14 = 300
zonal pairs is the same for all trips/travelers Fraction to Fraction to Fraction to
Zone 2 = Zone 3 = Zone 4 =
20% 50% 30%
The models do not account for the differences in the
characteristics of the travelers
68
Disaggregate Location-Choice Model
The alternate is the disaggregate approach

We determine what is the probability with Probability of

which each zone will be chosen as the choosing Zone Zone 1
1=0
destination by a traveler of given Individual i
characteristics

The models account for the differences in the Zone 2 Zone 4

characteristics of the travelers making those Zone 3
trips
Probability of Probability of Probability of
choosing Zone choosing Zone choosing Zone
2 = 20% 3 = 50% 4 = 70%
69
Disaggregate Location-Choice Model
𝑉𝑞𝑖𝑗 = 𝑓(𝑋𝑗 , 𝑇𝑆𝑖𝑗 , 𝑍𝑞 )

𝑉𝑞𝑖𝑗 = the “utility” (or satisfaction) obtained by traveler q currently located

at zone i by choosing zone j as the destination for the trip under
consideration
𝑋𝑗 = Characteristics of the candidate destination zone j
𝑇𝑆𝑖𝑗 = Transportation system characteristics (travel time, travel cost, etc.)
between zones i and j
𝑍𝑞 = Characteristics of the traveler q

70
Disaggregate Location-Choice Model
exp(𝑉𝑞𝑖𝑗 )
𝑃𝑟𝑜𝑏𝑞𝑖𝑗 =
σk𝐶 exp(𝑉𝑞𝑖𝑘 )
𝑞𝑖

𝐶𝑞𝑖 is the set of zones considered as candidate destination locations by the

individual q currently at zone i (choice set)

0  𝑃𝑟𝑜𝑏𝑞𝑖𝑗  1 𝑞, 𝑖, 𝑗
෍ 𝑃𝑟𝑜𝑏𝑞𝑖𝑘 = 1 𝑞, 𝑖
k𝐶𝑞𝑖

A zone with a higher utility value (relative to the other zones in the choice
set) is more likely to be chosen as the destination location 71
Disaggregate Location-Choice Model
References

• Bhat, C.R., Govindarajan, A., and Pulugurta, V. (1998) "Disaggregate Attraction-End

Choice Modeling", Transportation Research Record, Vol. 1645, pp. 60-68.

• Pozsgay, M.A., and Bhat, C.R. (2002) "Destination Choice Modeling for Home-Based
Recreational Trips: Analysis and Implications for Land-Use, Transportation, and Air
Quality Planning", Transportation Research Record, Vol. 1777, pp. 47-54.

72
Disaggregate Location-Choice Model
Example:
Model for recreational trips for the Dallas-Fort Worth Region (Pozsgay, M.A. and Bhat, C.R., 2002)

𝑉𝑞𝑖𝑗 = −1.965 𝑇𝑇𝑖𝑗 − 1.753 𝑇𝑇𝑖𝑗 ∗ 𝐴𝑔𝑒𝑞 − 0.689 𝑇𝑇𝑖𝑗 ∗ 𝐶ℎ𝑖𝑙𝑑𝑞

−0.391 𝑇𝑇𝑖𝑗 ∗ 𝑆𝑜𝑙𝑜𝑞 + 2.988 𝑇𝑇𝑖𝑗 ∗ 𝐶𝑎𝑟𝑠𝑞 + 0.399 𝑇𝑇𝑖𝑗 ∗ 𝐿𝐼𝑞

+ 0.496 𝐿𝑁[𝑅𝑒𝑡𝑗 + 0.045𝑁𝑜𝑛𝑅𝑒𝑡𝑗 ] + … …

𝑇𝑇𝑖𝑗 Travel time between zones i and j
𝐴𝑔𝑒𝑞 Age of the traveler
𝐶ℎ𝑖𝑙𝑑𝑞 Presence of children in the travelers household ( 1 = YES, 0 = NO)
𝑆𝑜𝑙𝑜𝑞 Is the trip being made solo or jointly with others ( 1 = SOLO, 0 = NO)
𝐶𝑎𝑟𝑠𝑞 Number of cars in the traveler’s household
𝐿𝐼𝑞 Is the traveler from a low income household ( 1 = YES, 0 = NO)
𝑅𝑒𝑡𝑗 Retail acerage in the zone j

𝑁𝑜𝑛𝑅𝑒𝑡𝑗 Non Retail acerage in the zone j 73

 Disaggregate Location-Choice Model
Let: 𝑉𝑖𝑗 =  𝑇𝑇𝑖𝑗 + 𝐿𝑁 𝐴𝑗
NOTE: as specified above, 𝑉𝑞𝑖𝑗 = 𝑉𝑖𝑗 𝑞

exp 𝑉𝑖𝑗 = exp 𝑇𝑇𝑖𝑗 + 𝐿𝑁 𝐴𝑗 = exp 𝑇𝑇𝑖𝑗 exp 𝐿𝑁 𝐴𝑗 = 𝐴𝑗 exp 𝑇𝑇𝑖𝑗

exp(𝑉𝑖𝑗 ) 𝐴𝑗 exp(𝑇𝑇𝑖𝑗 )
𝑃𝑟𝑜𝑏𝑖𝑗 = = Leads to a production-
σk𝐶 exp(𝑉𝑖𝑘 ) σk𝐶 𝐴𝑘 exp(𝑇𝑇𝑖𝑘 ) constrained gravity model type
𝑖 𝑖
𝐴𝑗 𝐹𝑖𝑗 expression.
=σ 𝐴𝑘 𝐹𝑖𝑘
, where, 𝐹𝑖𝑗 = exp(𝑇𝑇𝑖𝑗 )
k𝐶𝑖 Broadly, the gravity model is a
= Fraction of trips produced in i special case of the disaggregate
and attracted to j location choice model
74
At the end of the first two steps of the four-step process…..
• We have determine the person-trips Production-Attraction (PA) matrix
• Next, we will figure out the travel mode for each of these trips and
generate the vehicle-trips Production-Attraction (PA) matrix

75
 Thank You!

• Contact me
– Email: moinul@[Link]