TTE 4274:

Transportation Engineering Systems

Instructor: Sabreena Anowar, PhD

Department of Civil, Environmental and Construction Engineering
University of Central Florida

1
Today’s Outline
• Methods for analyzing travel demand
• Trip generation
• Approaches
• Solve problems
• Trip distribution

2
Trip Generation Approaches
• Two major modeling approaches
• Regression models
• Linear regression
• Regression models for count data
• Cross-classification models (category analysis)
• Separate models are estimated for
• Trip productions
• Trip attractions

3
Factors Affecting Trip Generation
• Home based trip productions
• # of trips = f (socio-demographics, land use, accessibility)
• Non-home based trip attraction
• # of trips = f (space available for industrial, commercial
and other services, accessibility)
• Socio-demographics – household size, income,
vehicle ownership
• Land use – population density, no/type of units,
distance of zone from Central Business District
(CBD), zonal employment level
• Accessibility – to transit, to services
4
Linear Regression
• Objective of trip generation modeling is to develop
an expression that predicts
• How many trips will be made
• These models typically assume a linear form
• Number of trips, 𝑇𝑖 is a function of various
socioeconomic and/or residential/commercial
characteristics

5
Linear Regression
Application of model

𝑫𝒂𝒕𝒂𝒔𝒆𝒕

Identifying predicted number of trips

𝒚 = 𝒇𝒏(𝒙)

𝑫𝒆𝒑𝒆𝒏𝒅𝒆𝒏𝒕 𝑰𝒏𝒅𝒆𝒑𝒆𝒏𝒅𝒆𝒏𝒕
𝑽𝒂𝒓𝒊𝒂𝒃𝒍𝒆 𝑽𝒂𝒓𝒊𝒂𝒃𝒍𝒆 6
Linear Regression
• Mathematically,
𝑇𝑖 = 𝑏0 + 𝑏1 𝑥1𝑖 + 𝑏2 𝑥2𝑖 + ⋯ + 𝑏𝑘 𝑥𝑘𝑖

𝑇𝑖 = number of vehicle-based trips of a given type (e.g., work

or shopping) in some specified time period made by
household 𝑘
𝑥𝑘𝑖 = characteristic (e.g., income, car ownership, number of
household members) of household 𝑖
𝑏𝑘 = coefficient estimated from traveler survey data and
corresponding to characteristic 𝑘

7
 Least Square Estimates
• Coefficients are typically estimated by method of
least squares (LS) using data from traveler surveys

Y Ti Observed
value
Shopping trips, Ti

i = Random error

Tˆi  b0  b1 X i
Observed value
X
People in the household, X
n
LS minimizes  ε i2  ε12  ε 22  ...  ε 2n Where: i = Ti - (b0+b1X8i )
i 1
Example Problem
vehicle
persons/hh, Predicted e
trips/day, xy x2
x 𝑦ෝ = y-𝑦ො
y
1 3
1 2
2 3
2 5
2 10
2 1
2 7
3 6
3 10
3 12
3 11
4 11
4 14
4 10
5 12
6 18
8 12
8 16
9 15
10 16
82 194 1001 476 9
Example Problem
vehicle
persons/hh, Predicted e
trips/day, xy x2
x 𝑦ෝ = y-𝑦ො
y
1 3 3 1
1 2 2 1
2 3 6 4
2 5 10 4
2 10 20 4
2 1 2 4
2 7 14 4
3 6 18 9
3 10 30 9
3 12 36 9
3 11 33 9
4 11 44 16
4 14 56 16
4 10 40 16
5 12 60 25
6 18 108 36
8 12 96 64
8 16 128 64
9 15 135 81
10 16 160 100
82 194 1001 476 10
 Example Problem
20

18 Ti  18 σ𝑥σ𝑦
σ 𝑥𝑦 −
𝑏= 𝑛
16
𝜺 2 σ𝑥 2
σ𝑥 −
14 𝑛
𝑏 = 1.4707
vehicle trips/day

12

10
𝑇෠𝑖 = 12.5
σ𝑦 σ𝑥
8 𝑎= −𝑏∗
𝑛 𝑛
6 y = 1.4707x + 3.6702
𝑎 = 3.6702
4

0
0 1 2 3 4 5 6 7 8 9 10 11
Persons/hh

11
Example Problem
vehicle
persons/hh, Predicted e
trips/day, xy x2
x 𝑦ෝ = y-𝑦ො
y
1 3 3 1 5.1 -2.1
1 2 2 1 5.1 -3.1
2 3 6 4 6.6 -3.6
2 5 10 4 6.6 -1.6
2 10 20 4 6.6 3.4
2 1 2 4 6.6 -5.6
2 7 14 4 6.6 0.4
3 6 18 9 8.1 -2.1
3 10 30 9 8.1 1.9
3 12 36 9 8.1 3.9
3 11 33 9 8.1 2.9
4 11 44 16 9.6 1.4
4 14 56 16 9.6 4.4
4 10 40 16 9.6 0.4
5 12 60 25 11.0 1.0
6 18 108 36 12.5 5.5
8 12 96 64 15.4 -3.4
8 16 128 64 15.4 0.6
9 15 135 81 16.9 -1.9
10 16 160 100 18.4 -2.4
82 194 1001 476 - 12 0.0
Example Problem
Example 1 A large retirement village has a total retail
employment of 120 jobs. All 1600 of the households in this
village consist of two (2) nonworking family members with
household income of $20,000. Assuming that shopping and
social/recreational trip rates both peak during the same hour,
predict the total number of peak-hour trips generated by this
village using the following trip generation models:
Y1 = 0.12+0.09X1 + 0.011X2 - 0.15X3
Y2 = 0.04 +0.018X1 + 0.009 X2 + 0.16X4
Y1 = Shopping trips per household,
Y2 = Social/recreational trips
X1 = household size
X2 = household income (in 1000’s dollars)
X3 = employment (jobs in 100’s)
X4 = non-working members 13
Example Problem
Example 2 Consider the retirement village described in previous
problem. Determine the amount of additional retail employment
(in the village) necessary to reduce the total predicted number of
peak-hour shopping trips to 200.
Village characteristics:
Retail employment: 120
Households: 1600 with two family members and household
income of 20,000.
Model (shopping trips):
Y1 = 0.12+0.09X1 + 0.011X2 - 0.15X3
Y1 = Shopping trips per household
X1 = household size
X2 = household income (in 1000’s dollars)
X3 = employment (jobs)
14
Cross-Classification Method
• Regression models
• Computationally convenient
• However, their use of zonal totals or zonal averages can
mask important variations in trip making between
households within a zone
• An alternative is to group households based on
important characteristics believed to be correlated
with trip generation – cross-classification method
• Income, Household size, vehicle ownership levels
• Popular method in practice

15
Cross-Classification Method
• General procedure
• Each cell of the cross-classification table contains a trip
rate, 𝑟𝑛 applicable to the households that fall into that cell
• Trip rates are derived from travel survey data
Households Income level
Household size <10,000 10,000 to 30,000 >30,000
1 10 100 50
2 10 200 50
3+ 30 100 50

HBW trip rate Income level

Household size <10,000 10,000 to 30,000 >30,000
1 1.5 2.5 2.5
2 2.5 4 5
3+ 3 5 7

16
Cross-Classification Method
• General procedure
• Multiply the appropriate trip rates in each cell with
number of households, 𝐻𝑖𝑛 in the corresponding cell
• Compute the sum over all cells to get zonal trip
generation, 𝑇𝐺𝑖
𝑁

𝑇𝐺𝑖 = ෍ 𝐻𝑖𝑛 × 𝑟𝑛
𝑛=1

Total Home-Based Work Trip Rates

Income
Household size <10000 10000-30000 >30000
1 10*1.5 = 15 100*2.5 = 250 50*2.5 = 125
2 10*2.5 =25 200 * 4 = 800 50*5 = 250
3+ 30*3 = 90 100*5 = 500 50*7 = 350
17
Cross-Classification Method
• Advantages
• Simple, easy to understand, less aggregated method
• Groupings independent of the zoning system in study area
• No prior assumptions about the shape of the relationship
(linear/otherwise) between explanatory variables and trip
rates
• Disadvantages
• Data requirements increase heavily with the number of
explanatory variables
• No statistical fit measures – difficult to assess model
performance
• Assumption that the trip rates are stable over time!!
18
Example Problem
• Example 3 Consider a zone that is located in a suburban
area of a city. Number of dwelling units of the area is 60.
Percentages of HBW, HBO and NHB trips are 17, 51 and
32, respectively. Determine the number of trips per day
generated in this zone for each trip purpose assuming
the following characteristics:
Distribution of Households Across Different Income Categories

19
Example Problem

20
 Example Problem
• Solution
Total number
of surveyed
households

21
Example Problem
• Solution

22
Example Problem
• For low income category, number of trips by trip
purposes are
• HBW = 19×0.17 = 3
• HBO = 19×0.51 = 10
• NHB = 19×0.32 = 6
• For medium income category, number of trips by trip
purposes are
• HBW = 232×0.17 = 40
• HBO = 232×0.51 = 118
• NHB = 232×0.32 = 74

23
Example Problem
• For high income category, , number of trips by trip
purposes are
• HBW = 415×0.17 = 71
• HBO = 415×0.51 = 212
• NHB = 415×0.32 = 133
• Total trips, by trip purposes
• HBW = 666×0.17 = 113
• HBO = 666×0.51= 340
• NHB = 666×0.32= 213

24
Count-Data Models
• Linear regression is a popular method
• Issue
• Models produce fractions of trips for a given time period
• Fractions of trips are not realistic
• A modeling approach that gives the probability of
making a non-negative integer number of trips (0, 1,
2, 3, …) may be more appropriate
• Poisson model

25
Problem with Linear Regression
persons/ vehicle
e
hh, trips/day, xy x2 Predicted 𝑦ෝ
= y-𝑦ො
x y
1 3 3 1 5.1 -2.1
1 2 2 1 5.1 -3.1
2 3 6 4 6.6 -3.6
2 5 10 4 6.6 -1.6
2 10 20 4 6.6 3.4
2 1 2 4 6.6 -5.6
2 7 14 4 6.6 0.4
3 6 18 9 8.1 -2.1
3 10 30 9 8.1 1.9
3 12 36 9 8.1 3.9
3 11 33 9 8.1 2.9
4 11 44 16 9.6 1.4
4 14 56 16 9.6 4.4
4 10 40 16 9.6 0.4
5 12 60 25 11.0 1.0
6 18 108 36 12.5 5.5
8 12 96 64 15.4 -3.4
8 16 128 64 15.4 0.6
9 15 135 81 16.9 -1.9
10 16 160 100 18.4 -2.4
82 194 1001 476 - 0.0

26
Count-Data Models
• Poisson model
−𝜆 𝑇𝑖
𝑒 𝜆𝑖𝑖
𝑃 𝑇𝑖 =
𝑇𝑖 !
𝜆𝑖 = exp(𝛽0 + 𝛽1 𝑥1 +𝛽2 𝑥2 +…+𝛽𝑛 𝑥𝑛 = exp(𝛽𝑥𝑖 )

𝑇𝑖 = # of trips made in some specified time period by household 𝑖

𝑃(𝑇𝑖 ) = probability of household 𝑖 making exactly 𝑇𝑖 trips
𝑒 = base of the natural logarithm (𝑒 = 2.718)
𝜆𝑖 = Poisson parameter for household 𝑖 , which is equal to
household 𝑖’s expected number of trips in some specified time
period, 𝐸[𝑇𝑖 ]
27
Count-Data Models

Source: https://www.quora.com/What-is-a-Poisson-Distribution-example 28
Example Problem
Example 4 Consider a Poisson regression model for the number
of social/recreational trips generated during a peak-hour period
that is estimated by

BZi = -0.75 + 0.025 (household size) + 0.008 (annual household

income in thousands of dollars) + 0.10 (no. of non-working
household members).

Suppose that a household has five members (3 of whom work)

and annual income of $100,000. What is the expected number of
peak-hour social/recreational trips, and what is the probability
that the household will not make a peak-hour social/recreational
trip?
29
Trip Distribution
I Trip Generation J

Oi Dj
Trip Distribution
I J
Tij
Tij,auto

I Mode Split J

Tij,transit
30
Trip Distribution

31
Trip Distribution

32
Trip Distribution
• What is being modeled?
• Destination choice of individuals to pursue activities
• What factors influence trip distribution?
• Trip cost
• Travel time: in-vehicle, out-of-vehicle, waiting time, including
congestion
• Actual cost: transit fare, tolls, parking, gasoline
• Desirability of attraction zone
• Quantity of opportunities, zonal attributes

33
Trip Distribution
• Approaches commonly used
• Gravity model
• Growth factor model
• Intervening opportunities
• The gravity model is most commonly used
• Simple and accurate
• Uses the attributes of the transportation system and land
use characteristics

34
Gravity Model
• Analogue of Newton’s law of gravity

35
Gravity Model
• The analogy used in human system is
• Number of trips between two zones (an origin-destination
pair) is directly proportional to the product of the number
of trip productions at the origin and number of trip
attractions at the destination and inversely proportional to
the spatial separation between the areas
• Casey (1955)
𝛼𝑃𝑜𝑝𝑖 𝑃𝑜𝑝𝑗
𝑇𝑖𝑗 = 2
𝑑𝑖𝑗

36
Gravity Model
• Improvement
𝛼𝑃𝑖 𝐴𝑗
𝑇𝑖𝑗 =
𝑑𝑖𝑗 𝑛
• Further generalization
T𝑖𝑗 = 𝛼𝑃𝑖 𝐴𝑗 𝑓(𝐶𝑖𝑗 )𝐾𝑖𝑗
𝑇𝑖𝑗 = trips produced at 𝑖, attracted to 𝑗
𝑃𝑖 = trip ends produced at 𝑖
𝐴𝑗 = trip ends attracted to 𝑗
𝑐𝑖𝑗 = generalized cost of trip from 𝑖 to 𝑗
𝐾𝑖𝑗 = desirability of a trip from 𝑖 to 𝑗 relative to other combinations of 𝑖 and 𝑗

37
Generalized Cost
• There are different costs associated with trips
• Time
• In-vehicle travel time (IVTT)
• Out-vehicle travel time (OVTT)
• Waiting time (transit)
• Transfer time
• Costs
• Fare
• Terminal cost (Parking)
• Distance
• A generalized cost is estimated combining all the
costs associated with a trip
38
Generalized Cost
• 𝑓(𝐶𝑖𝑗 ) is a generalized function of the travel costs
• Also, referred to as the deterrence function or
friction factor
• Commonly used functional forms
Exponential function = exp −𝛽𝑐𝑖𝑗
Power function = 𝑐𝑖𝑗 −𝑛
Combined function = 𝑐𝑖𝑗 𝑛 exp −𝛽𝑐𝑖𝑗

39
Example Problem
• Plot the power and exponential cost functions for
distance (as the only cost variable) in the deterrence
function for values of 𝑛 =1.2 and 𝛽 = 0.1

40
Example Problem
• Plot the power and exponential cost functions for
distance (as the only cost variable) in the deterrence
function for values of 𝑛 =1.2 and 𝛽 = 0.1

41
Example Problem
• Plot the power and exponential cost functions for
distance (as the only cost variable) in the deterrence
function for values of 𝑛 =1.2 and 𝛽 = 0.1

42
Different Cost Functions

43
Types of Gravity Models
• There are two broad types of gravity models
• Singly constrained model
• Production (origin) - constrained model
• Attraction (destination) - constrained model
• Doubly constrained (production-attraction) model
• Conservation of trips
𝑁

෍ 𝑇𝑖𝑗 = 𝑃𝑖
𝑗=1
𝑁

෍ 𝑇𝑖𝑗 = 𝐴𝑗
𝑖=1
44
Constraints

45
Types of Gravity Models
• In order to meet the constraints, proportionality
factor is replaced by balancing factors
T𝑖𝑗 = 𝛼𝑃𝑖 𝐴𝑗 𝑓 𝐶𝑖𝑗 𝐾𝑖𝑗
=> T𝑖𝑗 = 𝐵𝑖 𝐶𝑗 𝑃𝑖 𝐴𝑗 𝑓 𝐶𝑖𝑗 𝐾𝑖𝑗
• Conservation of productions and attractions can be
used to estimate the balancing factors
𝑁 𝑁

෍ 𝑇𝑖𝑗 = ෍ 𝐵𝑖 𝐶𝑗 𝑃𝑖 𝐴𝑗 𝑓 𝐶𝑖𝑗 𝐾𝑖𝑗 = 𝐴𝑗

𝑖=1 𝑖=1

46
Types of Gravity Models
𝑁 𝑁

෍ 𝑇𝑖𝑗 = ෍ 𝐵𝑖 𝐶𝑗 𝑃𝑖 𝐴𝑗 𝑓 𝐶𝑖𝑗 𝐾𝑖𝑗 = 𝐴𝑗

𝑖=1 𝑖=1
 𝐶𝑗 𝐴𝑗 σ𝑁
𝑖=1 𝐵𝑖 𝑃𝑖 𝑓 𝐶𝑖𝑗 𝐾𝑖𝑗 = 𝐴𝑗
1
 𝐶𝑗 = σ𝑁
𝑖=1 𝐵𝑖 𝑃𝑖 𝑓 𝐶𝑖𝑗 𝐾𝑖𝑗

• In similar vein,
1
𝐵𝑖 =
σ𝑁
𝑗=1 𝐶𝑗 𝐴𝑗 𝑓 𝐶𝑖𝑗 𝐾𝑖𝑗

47
Production Constrained Model
• Conservation of productions is used and the other
constraint is relaxed
• For instance, for production constrained model, the
known amounts of production in each zone is
distributed to attraction opportunities in competing
destination zones considering 𝐶𝑗 = 1.
1
𝐵𝑖 = 𝑁
σ𝑗=1 𝐴𝑗 𝑓 𝐶𝑖𝑗 𝐾𝑖𝑗
𝑃𝑖 𝐴𝑗 𝑓 𝐶𝑖𝑗 𝐾𝑖𝑗
T𝑖𝑗 =
σ𝑁
𝑗=1 𝐴𝑗 𝑓 𝐶𝑖𝑗 𝐾𝑖𝑗

48
Attraction Constrained Model
• Conservation of attractions is used and the other
constraint is relaxed
• Considering 𝐵𝑖 = 1
1
𝐶𝑗 = 𝑁
σ𝑖=1 𝑃𝑖 𝑓 𝐶𝑖𝑗 𝐾𝑖𝑗

𝑃𝑖 𝐴𝑗 𝑓 𝐶𝑖𝑗 𝐾𝑖𝑗
T𝑖𝑗 =
σ𝑁
𝑖=1 𝑃𝑖 𝑓 𝐶𝑖𝑗 𝐾𝑖𝑗

49
Example Problem
• Let us consider a 4 zone system with 1972 shopping trips.
Zone 1, zone 2, zone 3 and zone 4 produces 400, 460, 400,
and 702 trips, respectively. Each zone contains a shopping
mall with 260, 400, 500, and 808 attractions, respectively.
Attracter Zones
Producer Productions in
Zones Target
1 2 3 4
1 ? ? ? ? 400
2 ? ? ? ? 460
3 ? ? ? ? 400
4 ? ? ? ? 702
Attractions in
Target 260 400 500 802 1972
50
Example Problem
• An exponential form is assumed for the deterrence function
and the 𝛽 = 0.10. 𝑓 𝐶𝑖𝑗 = exp(−0.1𝐶𝑖𝑗 )

Attracter Zones
Producer Zones
1 2 3 4
1 3 11 18 22

2 12 3 12 19

3 15.5 13 5 7

4 24 18 8 5

51
Example Problem
• Step 1:
• Calculate the deterrence function values 𝑓 𝐶𝑖𝑗 =
exp(−0.1𝐶𝑖𝑗 )

Attracter Zones
Producer Zones
1 2 3 4
1 0.74 0.33 0.17 0.11

2 0.30 0.74 0.30 0.15

3 0.21 0.27 0.61 0.50

4 0.09 0.17 0.45 0.61

52
Example Problem
• Step 2:
• Start with the production constraint model and match
as needed
𝑃𝑖 𝐴𝑗 𝑓 𝐶𝑖𝑗 𝐾𝑖𝑗
T𝑖𝑗 =
σ𝑁
𝑗=1 𝐴𝑗 𝑓 𝐶𝑖𝑗 𝐾𝑖𝑗
• 𝐾𝑖𝑗 = 1, unless otherwise specified
𝑃𝑖 𝐴𝑗 𝑓 𝐶𝑖𝑗
T𝑖𝑗 =
σ𝑁
𝑗=1 𝐴𝑗 𝑓 𝐶𝑖𝑗

53
Example Problem
• Step 2:
𝑃𝑖 𝐴𝑗 𝑓 𝐶𝑖𝑗
T𝑖𝑗 =
σ𝑁
𝑗=1 𝐴𝑗 𝑓 𝐶𝑖𝑗
𝑁

෍ 𝐴𝑗 𝑓 𝐶𝑖𝑗 = 60∗0.74+400∗0.33+500∗0.17+802∗0.11=497.62
𝑗=1

𝑃1 𝐴1 𝑓 𝐶11 =260*400*0.74 = 76960

𝑃1 𝐴1 𝑓 𝐶11 76960
T11 = 𝑁 = = 154.66
σ𝑗=1 𝐴𝑗 𝑓 𝐶𝑖𝑗 497.62
54
Example Problem
• Step 2: Iteration 0
Attracter Zones
Producer Productions in
Zones Target
1 2 3 4
1 154.66 106.11 68.33 70.91 400

2 55.69 211.33 107.09 85.89 460

3 25.14 49.74 140.46 184.66 400

4 20.39 59.25 196.06 426.30 702

Attractions in
Target 260 400 500 802 1972

55
Example Problem
• Step 2: Iteration 0
Attracter Zones
Producer Productions in
Zones Target
1 2 3 4
1 154.66 106.11 68.33 70.91 400

2 55.69 211.33 107.09 85.89 460

3 25.14 49.74 140.46 184.66 400

4 20.39 59.25 196.06 426.30 702

Attractions in
Target 260 400 500 802 1972

56
Example Problem
• Match production and attraction targets by iterative
row and column factoring
• Adjust with respect to production totals
• Adjust with respect to attraction totals
• Continue the process until the production and attraction
factors are almost equal to 1

• In our case, for the first step production totals will

match

57
Example Problem
• Step 2: Iteration 0
Producer Attracter Zones Row Production Productions
Zones 1 2 3 4 Total Factor in Target

1 154.66 106.11 68.33 70.91 400 1.00 400

2 55.69 211.33 107.09 85.89 460 1.00 460

3 25.14 49.74 140.46 184.66 400 1.00 400

4 20.39 59.25 196.06 426.30 702 1.00 702

Column
255.88 426.42 511.93 767.76
Total
Attraction
1.02 0.94 0.98 1.04
Factor
Attractions
260 400 500 802 1972
in Target

58
Example Problem
 Iteration 1: Correct for column factors (from
previous slide)
Producer Attracter Zones Row Production Productions
Zones 1 2 3 4 Total Factor in Target

1 157.15 99.53 66.73 74.08 397.49 1.01 400

2 56.59 198.23 104.60 89.72 449.13 1.02 460
3 25.55 46.65 137.18 192.90 402.28 0.99 400
4 20.72 55.58 191.49 445.31 713.10 0.98 702
Column
260 400 500.00 802.00
Total
Attraction
1.00 1.00 1.00 1.00
Factor
Attractions
260 400 500 808 1972
in Target
59
Example Problem
 Correct for Row Factors (from previous slide)
Producer Attracter Zones Row Production Productions
Zones 1 2 3 4 Total Factor in Target

1 158.14 100.16 67.15 74.54 400 1.00 400

2 57.95 203.03 107.13 91.89 460 1.00 460
3 25.40 46.39 136.40 191.80 400 1.00 400
4 20.40 54.72 188.51 438.38 702 1.00 702
Column
261.90 404.30 499.19 796.61
Total
Attraction
0.99 0.99 1.00 1.01
Factor
Attractions
260 400 500 802 1972
in Target

60
Example Problem
• Step 1: Calculate the deterrence function
• Step 2: Apply the production-constrained gravity
model equation
• Step 3: Match production and attraction targets by
iterative row and column factoring
• Adjust with respect to attraction totals
• Adjust with respect to production totals
• Continue the process until the production and attraction
factors are almost equal to 1 (between 0.99 and 1.01)

61
References
 Meyer, M. D. and Miller, E. J. (2001). 2nd edition. Urban
Transportation Planning, McGraw Hill, New York.
 Ortuzar, J. de D. and Willumsen, L. G. (2001). 3rd edition.
Modelling Transport, John Wiley & Sons.
 Principles of Highway and Traffic Engineering, Fifth Edition,
by Fred L Mannering, and Scott Washburn, Wiley, 2013.
 Traffic and Highway Engineering, Second Edition, by Nicholas
J. Garber and Lester A. Hoel, PWS Publishing, 1999.

62