Download - Household Interview Survey
Presentation on HIS
WorkshopOctober 18, 2010
Shogo Uchida, JICA Study Team
There is no royal road to learning.
Before this study
There is no officer’s road to technical transfer
Contents of PresentationBasic Concept of “Person Trip Survey”Person Trip Survey in JapanPerson Trip Survey in JICA StudiesBasic Concept of OD MatrixDemand Forecast and Interview Items Sampling of HIS in Karachi
Basic Concept of Person Trip Survey
What is “Person Trip Survey”?Survey for trip information of people:
What kind of person makes a tripWhenFrom where to whereFor what purposeBy which transport mode
All trip information for a person in a dayHousehold Interview Survey (HIS) is the main part of PT Survey
Home
Office
Customer
Shop
(1) Commuting (2) Business
(3) Business(4) Shopping
(5) To Home
Home Office
Walk Bus Rickshaw
1 Trip
What is “trip”
Minibus
Why?Base for transport analysis to formulate the transport master plan
Analysis of people’s trip behaviorAnalysis of trip flow in relation with urban structure and land useBasic data for Demand Forecast
Ave. no. of trips by age, gender, income level, ..etc
Mode Preference
Bus Rickshaw Taxi Private Car Walk
Peak time of travel
Ave. travel time
Ave. trip length Ave. no. of transfer
Trip purpose
Analysis of People’s Trip Behavior
Person Trip Survey in Japan
50 years history
Person Trip Surveys in JapanOfficial Survey by Local Government
Need approval from Ministry for Internal AffairsSubsidy from GOJ
Major Cities (more than 40)1962 -Every 10 yearUsed for transport master planMethod: household drop-off/pick-upSampling: “household” list
Person Trip Surveys in Japan
Person Trip Survey in Japan
Person Trip Surveys in Japan
Trip rate:Around 3.0 (who made trips)Around 2.5 (total)
Person Trip Survey by JICA in the World
More than 30 cities
Person Trip Surveys by JICA in the World
Characteristics of JICA PT SurveysShort periodUnreliable socio-economic data
Adjusted by screen line surveySubcontract to local firms
No experience of local firms in PT surveysWith socio-economic surveyOpinion surveyDirect interview method
Issues of JICA PT SurveyRapid growth and developmentRapid motorization
Rapid changes in transport pattern O/D becomes old in several years.
Technical transfer SurveyDemand forecast model
Basic Concept of OD Matrix
What is OD matrix and how to produce from samples
A-ZoneB-Zone
C-Zone
D-Zone
Bus Terminal
Traffic Flow on Bus RouteTraffic Flow on Bus Route
A-ZoneB-Zone
C-Zone
D-Zone
Desired RoutesDesired Routes
Potential corridors will be identified.
Analysis of People’s Movement
Total
ZoneN
…
Zone2
Zone1
TotalZoneN…Zone2Zone1
Origin-Destination Matrices Generation
AttractionAttraction
by
Mode
Purpose
Income level
Age
Gender
Etc.
OD – Basic data for transport analysis
Expansion of samplesPopulation size (total number) of trips is unknown.Sampling rate of the trip information can not be calculated.
Trips are expanded based on “person”
Total
12.1%2.4%1.8%1.5%D Town
2.4%18.2%3.6%4.2%C Town
1.8%3.6%24.2%3.0%B Town
1.5%4.2%3.0%12.1%A Town
TotalD TownC TownB TownA Town
Expansion factor (EF)= 1/ (sample rate of person)
1 trip of Mr. A is converted to EF trips
- Karachi HIS: 1 trip 80 trips(40,000*5)/16,000,000 = 80
Screen Line SurveyTraffic count along a “screen line”, which divides the survey areaTo count the total number of traffic crossing the lineTo check the O/D matrix produced from HIS
Lyari River was selected as the screen line3.5 million persons cross the screen line per day (2008, JICA)
Screen Line
Cordon Line SurveySamples of HIS are only residents:
movement of visitors is not included
O/D interview survey at “cordon lines”: enter-exit points
RoadsAirportsPortsStations
Karachi:Highway (3 points)AirportStation
JICA 2008 Study90,000 visitors from/to highway cordons15,000 visitors by railway10,000 visitors by air
Vehicle O/DHIS only collects trips by residents.Other surveys or models are needed to estimate Vehicle O/D.
NoNoTruck
NoNoTaxi (empty)
No (OK by CL survey)NoNon-Residents
OKOKResidents
Inside-OutsideWithin the areaTrip by
Demand Forecast and Interview Items
Why we ask the question?
Demand Forecast ModelTraditional Demand Forecast Model – 4 step demand forecast
Trip productionTrip generation/ attraction modelTrip distribution modelTrip assignment model
OD of “Trips” is the base of the modelLinkage of trips of a person is ignored.Modeling becomes simple and stable.
Trip Production Model
PaTP ⋅= a = trip rate per personP = population of Karachi
ccc PaTP ⋅= c = 1: car own householdc = 2: household without car
Simple
Complex
∑ ⋅=i
cicic PaTP
),...,,( 21 nxxxfTP = x = all available variables
i: personal attributes - Age & Gender- Age & Gender & Income- Age & Gender & Income & Job status
∑ ⋅=i
cicpicp PaTP p: trip purpose
Trip Generation/ Attraction Model
Simple
Complex
rrr PaTP ⋅= ar = trip rate per person of zone rPr = population of zone r
rcrcrc PaTP ⋅= c = 1: car own householdc = 2: household without car
∑ ⋅=i
rcircirc PaTP
),...,,( 21 nxxxfTP = x = all available variables
i: personal attributes - Age & Gender- Age & Gender & Income- Age & Gender & Income & Job status
∑ ⋅=i
rcircpircp PaTP p: trip purpose
h: peak factor
by job statusby job status5
by income groupby income group4
by age group, genderby genderby car-license status
3
by purpose
by time, by mode
AllOD
by age groupby purpose
2
by car-ownership1: Need
per person0: MustProduction, G/ALevel
Necessary Model Parameters
Necessary InformationFor modeling
Household Income LevelCar ownershipJob Status (worker, jobless, students,…)Age, genderCar licensePersonal income levelZone of householdZone of working place/ school
For surveyAccurate addressTelephone numberNameFamily structure
Trip InformationOrigin and destination (address, type)Departure and arrival timesModePurposeTransfer pointsCost (fare) of transportSelf-Drive or not
Linked trip and unlinked tripLinked trip
Trip on purposeUnlinked trip
Trip on mode
Linked trip is the base of demand forecastInfo. on unlinked trip is used to make the present OD by mode
Used for calibration
Home Office
Walk Bus Minibus Walk
1 Trip (Linked Trip)
Unlinked trips
Priority mode of a linked tripOD of Linked Trip:
By purpose clearBy mode not clear
OD by mode is made using “Priority mode”.Priority mode is the major mode in a trip.Identification of priority mode is not easy
Simply decide by hierarchy
TrainBusCarTaxiRickshawWalk
Usage of OD by mode depends on modeling
High
Low
Sampling and Expansion
Statistical approach
Statistical background
answer of % :psize Sample :n
size Population :Ninterval confidence 95%for 96.1
)1(1
=
−⋅
−−
±=
Zn
ppN
nNZerror
Formulation of Error
2
2
2
2
25.0
)1()/11(1
1
eZ
eZ
ppN
N
n
≅
⋅−
−+
=
Necessary no. of samples(N is large, p=0.5)
0
5
10
15
20
0 500 1000 1500 2000
Err
or (%
)
No. of samples
Error and no. of samples under 95% confidence interval
400 samples can achieve 5% error level
Sampling rates in Japan Sampling rate is calculated using
Sampling example:Tokyo: 2.68%Nagoya: 2.8%Osaka/Kyoto: 2.3%
modes of No.purposes of No.zones of No.:K
rate Sampling :r trips)(Total size Population :N
96.120% error relativeRSD
)1()1(
××
===
−⋅−±=
Z
rNrKZRSD
Sampling check after the surveyExample problem – Male : Female
53.9% : 46.1% in Karachi (1998 Census)560 : 440 from 1000 samples Is this OK? (Should be 539: 461 ….?)If not OK, sampling was wrong.
Chi-square test
( )∑=
−=
N
i i
ii
EEO
1
22χ
…
…
11.0709.4887.8155.9913.8415%
54321Degree of freedom
Oi : Observed valueEi : Expected value
Sampling check after the surveyExample Problem: It’s OK.
Test: distribution of gender and age groupNeed population data by gender by age group
If expansion factor is determined by gender by age, this test is not required.
Also need reliable data by gender by age group
( ) ( ) 841.3775.1461
461440539
539560 222 <=
−+
−=χ
Limitation of HIS data Breakdown will decrease reliabilityExample:
OD (UC Level) by purpose by mode by gender by age group
maybe unreliableOD (UC Level) by mode by income level by car ownership by occupation almost impossible
Need to check relative error before using breakdownRelative Error
Within 20%
answer of % :psize Sample :n
size Population :N
)1(1
96.1pn
pN
nNpp −
⋅−−
±=Δ
Breakdown Check: Example
Trips by mode by purpose by townExample: Purpose 20% (to work, for example), Mode 12% (Car), Town (1/24) p=0.2*0.12*(1/24)=0.01 (1%)n = 10,000 (per town)0.8% < p < 1.2%
Relative error = 20% OK
In case less n (ex. 2,000) (breakdown by age, gender, etc) 0.5% < p < 1.5%
Relative error = 44% No use
Trips by mode by UCExample: Mode 20% (Minibus), UC (1/200) p=0.2* (1/200)=0.001 (0.1%)n = 400 (per UC)0% < p < 0.4%
Relative error = 300% No use
Check is available after HIS
Why “household” survey for “person” trip?
Traditional demand forecast uses “trips”.Personal interview
Only trips for his/hersHH interview
Only trips for only the members’ of the HH
Reliability of HH sampling is lower than that of personal sampling
Random sampling of “trips”is practically impossible.
HH survey can collect many samples by one visit.
Reliability can be improved by larger no. of samples.
HH survey is the best in terms of cost and time
performance.
Sampling and Expansion
Household
Member
Trip
Population samples
Household
Member
Trip
sampling
sampling(all)
All trips
Calculate expansion factor
Expansion
Expansion factorin case statistics of age and gender are not available
Data Layout of the Result1 Trip = 1 Record (is popular)
Household InformationTrip-ID Personal
Attribution Trip information Expansion Factor
Same data repeated
Origin Destination Purpose Mode-1 Mode-nDate Start
timeend time
Sampling in Karachi
As random as possible
SamplingSamples should represent the population
Sampling should be “random”Random sampling is very difficult when a list of households is not available.
In Japan, sample households are selected randomly from household lists.
In Karachi, randomness cannot be ensured; however, sampling should be as much as random.
Condition of randomness in PT SurveySample households should reflect proportion of socio-economic of UC.Area distribution of sample households should also be random.
Even if this sample reflect the socio-economic proportions of the population, trip information is not random.
triptrip
trip
trip
Sub division of UCUC is divided into 3-5 sub zones to ensure the randomness of place.This sub-divided UC is called “Survey Zone” – more than 750 survey zonesAccurate UC boundary maps are necessary.
Sampling in a sub-zoneThere are three methods:
Put identity number to every houses in maps and select samples at random (1)Visit a house for every 100 buildings (2)Pick up a representative area in the sub-zone and visit all the houses (3)
Comparison of three methods
If the sub-zone consists of similar type of houses, this is proper method.
- Randomness is ensured only if the area represents the population.- Lower area randomness
- Can avoid surveyors’preference for house selection.- Surveyors mobilization is easy.
3
Apply this method under proper supervising
- Surveyors tend to visit houses which is easy to access (in terms of hospitality and friendly)
- Randomness of place is ensured.- Field survey is simple and practical.
2
Impassible due to limited resources
- Time-consuming work- Difficult to reach the target due to low literacy of surveyors for map reading
- Randomness is ensured.1
EvaluationDisadvantageAdvantageNo
Sampling method in this StudyMixture of 2 and 3Select representative blocks in the center of a Survey ZoneDeploy enumerators along streets in the blocksEasy to supervise and organize
Date check (Editing) and CodingEditor: check the survey forms after interview
Logical checkFill in blank, if possibleDecipher hand writing
Coder: translate text to number
Find zone code from address is the major work.
Example of editing“Bus stop” near origin and destination can be translated to address. Gender can be assumed by the name.1 and 7 can be distinguished by other information.06:00 after 9:00 may be 18:00