The New Zealand Travel Survey provides data to assist in the development of policy and evaluation of programmes relating to road use and road safety.
When combined with existing crash data, crash risks for different groups of road users (including drivers, passengers, pedestrians and pedal cyclists) can be estimated.
To enable valid estimation of changes in risk and travel occurring over time, comparable survey methods were used to the previous national travel surveys of 1989/90 and 1997/98.
As travel behaviour has been shown to be strongly related to people's availability and willingness to be surveyed, personal interviews were used to gather travel data. This survey method generates the highest rate of co-operation and the most complete recording of complex travel behaviour.
An initial letter was sent from the Ministry of Transport to the households selected for interview. Included with this was a pamphlet briefly describing the aims and content of the survey.
Next, the interviewer called at the address to gather household information, explain the purpose of the survey, tell the household which were their 'travel days' (two consecutive days for which the household was to record all travel), and leave a memory-jogger for the respondents to use for recording travel.
Finally, as soon as possible after the travel days, the interviewer returned to conduct the interviews.
To enable comparison with the results of the earlier travel surveys, essentially the same questionnaire was used as in the 1997/ 98 and 1989/90 surveys. There was the further advantage of using a survey instrument that had been extremely well tested and had performed well in the field previously. Minor changes were made to update wording and response categories.
Laptop computers were used by interviewers to improve data quality and reduce the time required for the interviews.
There were two questionnaires used: one to gather information about the household and another for individual travel, demographics, alcohol usage, etc.
In addition, the interviewers had show cards for coding occupation, driving experience, ethnicity, income, drinking venue and types and quantities of alcohol consumed.
The following data were gathered (or were derived from responses):
Local government region of respondent's residence, urbanisation of respondent's residence, household structure, relationship of people in the household, number of people, number and type of household vehicles (car, motorcycle, van etc.), vehicle make and model, vehicle age, engine capacity and ownership, and response status of household.
Person: For each person in the sampled household - relationship to nominal 'head' of household, gender, age, employment, income, driving experience, number of road crashes, number of trips, ethnicity, marital status, whether they drank alcohol on travel days, and location of workplace/school.
Trip: For each trip made by sampled people on the travel days - trip purpose, mode (as driver/passenger/pedestrian/cyclist etc), date, time, origin and destination grid references, age and gender of people in the vehicle, and which household vehicle was used (linked to information on vehicle make and model, vehicle age, engine capacity, ownership).
For walking trips no distance estimates were recorded because people generally find these difficult to estimate accurately and the digitised distance calculation designed for road travel was not well suited to pedestrian travel. Duration of the walking trip and the number of roads crossed were recorded instead.
Alcohol drinking sessions: For each person - times, locations and types and amounts of alcohol consumed.
Traffic crashes: For each person - crash involvement over the last two years, location of crashes, and type of crash.
A pre-pilot or "skirmish" plus a larger scale pilot were used to test and refine the survey forms (particularly the questionnaires) and procedures. A skirmish was run very early in the project to test the wording of the questionnaires, using a small sample of households, including the extremes in socio-economic levels. The results of the skirmish led to minor changes in question wording.
The pilot test was essentially a dress-rehearsal of the main survey, used to test the interviewing procedures, the adequacy of the training, the field work, and validation interviews as well as to examine the response rates.
Interviewers' training and supervision
All interviewers underwent a three-day training session after completing an extensive home study exercise. This was necessary as the travel behaviour being recorded was relatively complex and the interviewers needed to be well-acquainted with the 60-page interviewers' manual.
They were personally supervised once during the training (which involved carrying out actual Travel Survey interviews), once during their first week of interviewing and once (at random) after that.
Checks were also made of 10 percent of each interviewer's questionnaires that the surveyed household had actually been visited by the interviewer and that each respondent had in fact been interviewed personally, and one or two items of data were also checked.
Stratification
The sample strata and substrata were geographically based using Statistics NZ definitions for the 1996 Census of Population and Dwellings: the strata were the 14 Local Government Regions, further stratified into Main Urban Areas (at least 30,000 population), Secondary Urban Areas (population between 10,000 and 30,000) and rural (including Minor Urban Areas with population less than 10,000 and all other rural areas).
The sample sizes per Local Government Region were proportional to 1996 Census populations except for the following:
Sample frame and sampling method
Survey costs were minimised (while maximising the utility of the data collected) by constructing the survey so that interviewers did not need to travel long distances between households. Meshblocks (geographical units varying in size from a city block in urban areas to extensive tracts of land in rural areas) were used as the first stage sampling units and were sampled independently within the strata.
The sampling frame for meshblocks consisted of the 1996 Census list of meshblocks. The meshblocks were sampled with probability proportional to size without replacement where size was defined as 1996 Census population.
To compile an up-to-date sampling frame of households within the sampled meshblocks, these meshblocks were visited and all dwellings were listed together with street addresses.
Meshblocks were surveyed in random order within Regions. One in eight households from sampled meshblocks was sampled. A systematic sample of households was taken by randomising a list of all households within sampled meshblocks and sampling every eighth household from this list. Introductory letters were sent to a computer-generated list of households prior to the interviews of household members. Sampled households and household members from whom responses could not be obtained were not replaced by other respondents, but were imputed for using data obtained from other similar respondents (see below).
Allocation of travel days
The households selected according to the sampling scheme were each allocated two consecutive travel days (i.e. days about which the household members should report their travel). The travel days were allocated to the sample of households in a fashion that maintained a wide geographical spread (of areas being surveyed) at any given time of the year but was compatible with a restricted number of survey interviewers, each surveying at a rate of approximately three households per day. An even spread by day-of-week was maintained by systematic allocation of travel days.
Coverage
The sampling frame consisted of all New Zealand households, excluding some sparsely populated remote areas in Westland, East Coast of the North Island, Southland and Northland.
Guests at hotels and motels were not surveyed as it was assumed that this group of people had a chance of being sampled at their home residence. There were also considered to be difficulties in gaining access to these people for interviews, particularly as the survey method required more than one visit (see above) and was not compatible with short stays at motels/hotels.
Inmates of prisons and patients of hospitals were also not surveyed. Bias due to non-response was minimised by requiring a minimum of four attempts (made at different times of the day) to contact people who were not at home.
Nevertheless, the failure to make contact with respondents who are not at home together with the exclusion of visitors and people staying at hotels/ motels (some research indicates that these people tend to travel more than the average), means that the estimates of distance travelled derived from this household survey may slightly under-estimate the total travel in New Zealand. However, this can be estimated from other sources.
There will also be some underestimation of travel by professional drivers (and hence by vehicles such as taxis and trucks). This is due to a combination of the household-based sampling that excluded accommodation used by long-distance drivers, and the under-representation of people who travel as they are not at home when surveyed.
The estimation of distance travelled is central to the calculation of exposure to crash risk, and hence the key variable of this survey.
For all their recorded trips, respondents were asked to provide addresses of the origin and destination of each leg of the trip in a format that could be used in the automated calculation of trip distances. Critchlow Associates were contracted to generate automated map co-ordinates for each address and then to calculate distance based on the shortest (in terms of travelling time) route between the origin and destination addresses.
Some approximations needed to be used when a street number was not valid (the closest valid address was used) or did not exist (the mid-point of the street was used). Where a route was used that deviated from the shortest route (e.g. a scenic drive), the interviewers recorded an intermediate address along the route taken to show that a longer route was taken.
A number of addresses that could not be automatically digitised (i.e. encoded as map co-ordinates) were digitised manually by referring to street maps. Where there was insufficient detail or errors in the recording of the address, the respondent's own estimate of the trip distance (which was recorded for all non-pedestrian trips) was used as the best distance measure.
For pedestrians, the most commonly used exposure measures are time spent walking and number of roads crossed, both of which were recorded. Distance walked was not considered to be able to be calculated using the same algorithm as for driving or cycling, hence there was no distance measure calculated for pedestrian trips.
Since the sample is not a simple random sample of the population, a simple mean or total of the sample observations is not appropriate for estimating population means and totals.
Weighted means and totals are used, where the weights are approximately equal to the reciprocals of the probability of selection of the respondents. Weights are also used to reduce the inevitable bias due to non-response.
Sampling errors are calculated using SAS, accounting for two-stage stratified sampling.
Information about motor vehicle crashes is extracted from the Ministry of Transport's database of coded information derived from Traffic Crash Reports.
When an injury crash is reported, it is usually attended by a police officer. The reporting officer's primary duties are to prevent further injury and to help those injured.
The next duty is a legal one, to ascertain whether anyone involved in the crash has committed an offence. After dealing with these other duties, the officer completes a Traffic Crash Report.
The Traffic Crash Report is examined and coded by traffic engineers and by administrative staff of Land Transport New Zealand. This coded information is loaded on to a computer, edited and checked. Further details can be found in the annual summary of crash statistics, Motor vehicle crashes in New Zealand (e.g. Land Transport Safety Authority, 2002).
Hospitalisation data are used for tables of cyclist injuries in non-motor vehicle crashes and of risks for different ethnic groups. These refer to the number of people admitted to hospital as a result of a crash and are supplied by the New Zealand Health Information Service.
1: 2001 Census data were not available at the time that the sample was selected, although these data were able to be used during the weighting of the survey data.
2: The probability proportional to size sampling method used, due to Sunter (1977), is described in Sarndal et al (1992: p94).
Speed kills because of the sudden stop. On impact, your internal organs and brain are moving forward at the same speed as before the crash. At high speeds they are smashed against your outer skeleton and rupture or haemorrhage.