CHAPTER 6.0 HOUSEHOLD SURVEY METHODS DESIGN AND SAMPLING DECISIONS

Note: significant portions of this chapter come from the 2008 NCRP Report 571, with review and contributions by Kouros Mohammadian and Harry Timmermans.

6.1 Survey Design

The survey design task requires planners of household travel/activity surveys to address a series of successively more detailed survey design issues, beginning with the determination of the survey’s role in the sponsoring agency’s long-term planning processes, and including the selection of the best survey methods and data collection techniques. Each survey design decision needs to be guided by the agency’s time and budget constraints and by the practical realities facing transportation agencies today.

The section summary shown above reviews the key issues that the survey team needs to address during the survey design phase of the project. These issues frame the lengthy discussion of design issues that follows.

6.1.1 The Data Needed from the Household Travel/Activity Survey and the Survey Design Implications of the Required Data Analyses

As is the case for any survey effort, the design of a household travel/activity survey needs to be informed by the foreseeable uses of the collected data. At the beginning of the design task, the sponsoring agency should define the goals of the data collection effort. Most household surveys are used as inputs into broadly-defined planning applications, such as the development or refinement of regional travel demand models. Other survey efforts are designed for more narrowly focused analyses, such as infrastructure project analysis. For instance, the California Department of Transportation (Caltrans) sponsored a recent analysis of the feasibility of high-speed rail service in California which utilized a household survey technique. Some household travel/activity surveys are designed as part of a larger data collection effort. Others are essentially stand-alone analyses.

It is recommended that the sponsoring agency develop a Statement of Goals for the survey effort that can be used as a guide in survey design. This statement should describe:

•  The data needs that have led the agency to conclude that survey work is necessary;
•  The expected analyses and uses of the survey data; and
•  The guiding principles of the data collection.

The Statement of Goals should be as detailed as possible in defining potential analyses that will rely on the household survey data, because many types of analyses will not only determine survey question content, but will also have implications on the choice of overall survey strategies and survey methods.

The Statement of Goals is a valuable document for ensuring that the data collection effort provides the necessary information to the sponsoring agency. It can be used throughout the survey planning process to help make survey design decisions, and to provide staff not directly involved with the survey and other agencies and firms with information on the survey project.

A wide range of analysis issues affect the overall design of household travel/activity surveys. Prior to the household travel/activity survey, the survey team should define specific analyses to the maximum extent possible. Some analysis issues that affect the overall design of household/travel activity surveys which survey teams have recently considered include the following:

•  Are the needed data related to entire households or to individuals within households?
•  Are the needed data cross-sectional in nature (data representing a single point in time), or are the likely analyses going to rely on longitudinal analyses?
•  Do the likely uses of the data involve traditional trip-based analyses or activity-based modeling approaches?<!--[endif]-->
•  How complete do the travel/activity data need to be?
•  Do the likely analyses require only revealed travel behavior data, or is there a need to obtain hypothetical choice and attitudinal information, as well?
•  Are the needed data specific to certain seasons of the year?
•  Are the needed data specific to certain time periods?

Survey teams should consider analysis issues such as these very early in the survey design process because the selection of survey methods and techniques will be influenced by them. The implications of each of the analysis issues listed above are described below.

6.1.2 The Household versus the Individual as the Basic Unit of Analysis

Household travel/activity surveys are usually complex surveys. Provided that special care is applied in the organization and expansion of the data for analysis, the survey data can typically be analyzed using several different units of analysis, including households, individuals, vehicles, and trips or activities.

However, before any data are collected or any analyses are performed, the survey team needs to define what the basic unit of analysis will be, the household or the individual.

For analyses that treat study area households as the tripmaking unit and the travel decision-making unit, it is necessary to collect survey data about entire households. For instance, trip generation models are generally developed at the household level, and thus need survey information on all trips made by household members over some period of time. On the other hand, analyses that are based on individuals’ travel behavior require the survey team to collect data on only a representative sample of study area residents, so data on only one household member are needed. Stated-response household surveys often will seek out a single individual within a household.

The distinction between these two types of analyses is of critical importance in survey design. Household travel/activity surveys used to obtain information on entire households are usually longer, much more complicated, and more burdensome for respondents than surveys that obtain similar information for only a single household member. Among the issues that need to be addressed for the household-based data collection are:

•  Procedures for identifying individuals within the household, and for distinguishing between them throughout the survey data collection;
•  Procedures for communicating with each household member or having household members communicate through a designated spokesperson; and
•  The potential need for proxies, in which one member of a household answers the survey questions for a member who cannot, either because he or she is too young or because he or she is unavailable.

Person-based survey efforts are far less complicated. The key issue for person-based surveys is how to select the proper household member for the survey. It is widely acknowledged that asking the household members who answer the door, open the mail, or answer the telephone to participate in a survey leads to a non-representative sample of a study area, and so travel surveyors have used a number of techniques for randomly selecting a household member. For instance, a common approach is to select the household member who is the next to have a birthday.

The survey team should determine whether the survey analyses will be based on household-based analyses or person-based analyses. The survey effort is significantly reduced if the latter is true, but more importantly, the analysis needs for household travel/activity surveys typically require that all household members be included in the survey effort.

6.1.3 Cross-Sectional versus Longitudinal Analyses

Traditional travel models rely almost exclusively on cross-sectional data, so household travel/activity surveys which are designed to capture people’s behaviors and attitudes at a single point in time have always been the most appropriate data collection tool. However, in recent years, researchers have recognized that many of the behaviors that travel models attempt to forecast are actually related to people’s decisions over time (Hensher, 1985)( Lawton and Pas, 1995). The renewed interest in how people’s behaviors change over time has led to the use of longitudinal survey designs, such as panel studies, cohort studies, trend studies, and before-after studies.

From analyses conducted thus far, it appears that longitudinal data collection efforts, in general, and panel studies, in particular, hold a great deal of promise for travel demand modeling. Chapter 13.0 discusses survey-related issues related to the emerging and promising use of longitudinal analyses and surveys, but as that section discusses, if longitudinal analyses of household travel/activity survey data are anticipated, the survey team needs to be prepared to make a continuing commitment to high quality data collection, and to expending significantly more resources to address:

•  Additional complexity of the survey recruitment;
•  Sample maintenance and replacement;
• Wider scope of survey questions;
•  Use of responses in past waves to frame questions;
•  Attrition (for panel surveys);
•  Weighting of longitudinal data; and
•  Additional reporting requirements.

Decisions about how to incorporate these issues into the survey design will certainly affect the cost and time estimates for the survey, and may also help determine which survey methods and techniques to use.

6.1.4 Trip versus Activity Analyses This section needs a discussion of how ABM utilize travel/activity survey data and how this informs the survey design KM

For many years, transportation planners have recognized the fact that travel is a derived demand – the demand for travel is related to the activities from which and to which people travel. In the early 1970s, a number of researchers proposed the development of a new set of models that would predict the activities in which households would take part, and then determine the household’s future travel patterns (Chapin, 1974).

In the past few years, there has been growing interest in looking at activity-based modeling again. One region is currently developing a prototype activity-based model system, and a great deal of research is underway to improve the state-of-the-art in this field (Kitamura et al., 1995). Activity-based model systems differ from conventional modeling approaches in that they predict the numbers and types of activities households will perform, then rely on either a set of behavioral rules or a series of econometric equations to forecast how household members will schedule and travel to and from activities. The models relay on stochastic microsimulation techniques to forecast activity and travel patterns. ADD KM

The choice between trip-based analyses and activity-based analyses has a basic effect on the design of household travel/activity surveys. The survey team needs to determine whether to collect detailed information on people’s trips directly, or to collect information on people’s activities and their travel to and from the activities (assuming the respondents need to travel to the activity).

Household surveys can be divided into three types in this regard:

•  Trip-based surveys that directly gather information on people’s trips over some period, using either diary methods or recall methods;
•  Activity-based surveys that gather information on activities to which respondents need to travel during a set time period; and
•  Time Use-based surveys that gather information on all activities in which respondents participate during a set time period.

The primary advantages of trip-based surveys are that they use the most efficient data collection approach, in terms of survey time and respondent burden. Respondents are asked directly about the subject of interest, their travel. The primary disadvantage of the approach is that typically the only information gathered on why the respondent is traveling is a non-detailed trip purpose. These surveys do not typically provide the information to examine the activities that people perform which produce their travel.

Activity-based surveys were developed as a means to improve upon traditional trip-based surveys. Surveyors have found that people do a better job remembering and recording trip information when they are asked about what they did rather than simply about where they went (Stopher, 1992). Of course, the survey also needs to query respondents about their travel to and from activities, so these questionnaires require more information than the trip-based approach. This translates into more work for respondents and longer data retrieval questionnaires, which in turn is likely to translate into higher non-response rates and more complaints about the survey effort. Although surveys of this type are commonly referred to as activity surveys, they are generally not suitable for activity-based modeling, because they do not provide the full set of activities for respondents.

The final type of household survey, the time use-based survey, asks respondents to record all of their activities over some period of time. These include activities that take place within people’s homes as well as those to which respondents need to travel. The surveys also collect the travel data for any trips between activities. These surveys provide a basis from which either traditional trip-based modeling approaches or activity-based modeling approaches may be developed, and because respondents are asked to record all of their activities over the time period, the number of trips that are accidentally left out is likely to be smaller than either of the other types of surveys.

On the other hand, the time use-based surveys are necessarily much longer than the other types and the respondents are asked to supply a great deal of personal information. These surveys have been found to be too invasive by a number of potential respondents in the regions where they have been fielded. A recent household activity survey in New Hampshire attempted to record both in-home and out-of-home activities, but respondent complaints to the Department of Transportation led the survey team to revise the data collection and analysis approach so only out-of-home activities were collected.

New generation travel demand models, like TRANSIMS, will likely require very detailed time use-based survey data.

6.1.5 The Comprehensiveness of the Travel/Activity Data

Many analyses require that all travel within a time period for a household (or for a person within a household) be reported in the survey data. For instance, to measure daily household trip generation rates, the survey team would want a full accounting of the trips made by each sample household in a 24-hour period. On the other hand, some analyses focus on a few specific trips made by household members, such as work trips or trips by certain modes.

The distinction between these two types of analyses has a very important impact on the household travel/activity survey because it is the primary determinant of whether formal travel or activity diary procedures are needed, or whether the use of respondent recall will be sufficient.

If the survey team is seeking a complete listing of travel and activities for a household, as is often the case, recent household travel/activity survey experience would suggest that the team use either travel or activity diaries, and not rely on respondent recall questions. As Richardson, Ampt, and Meyburg suggest, one needs only to try to remember in detail what they did, and where they went yesterday to realize that it is extremely difficult to obtain reliable and complete information using recall survey questions.⁶ By the early 1980s, the use of travel diaries supplanted recall surveys for collecting travel model input data. for the simple reason that they are more effective at capturing people’s trips. Diary methods consistently outperform trip recall questions in capturing: Short trips, Off-peak trips and Non-work trips. Travel and activity diaries are thought to be better than recall methods in these instances because:

•  Respondents are asked to complete diaries for a pre-specified future time period, so they are probably more cognizant of their travel during the particular time period than they would otherwise be; and
•  Respondents are asked to record travel and activities as they occur, so the likelihood of forgetting an activity or trip is reduced.

The many diary types are discussed in Section 6.6. The use of the less-expensive and more simple recall method may be a better approach when:

•  The survey team is interested only in certain types of travel and activities;
•  The survey data are not being used to develop travel volume estimates for the person or household; or
•  The survey team is willing to weight trip rates according to known (or estimated) volumes.

The recall method may also serve as a backup approach to try to get households who have refused to participate in a more detailed diary survey.

If limited travel or activity information is needed from a household, then a recall technique might be successful. With careful questioning, and perhaps interviewer probing, respondents can generally be induced to remember specific trips in the recent past, particularly if the trips can be defined specifically (e.g., a trip between home and work) or are somehow noteworthy or unique for the respondent.

The need (or lack of need) for diaries has an important effect on the selection of the survey method. If diaries are required, then the chosen survey method will be required to have a mail or in-person component to physically get the diaries to the household. If only recall methods are required, the household travel/activity survey can be accomplished with a single-contact interview (in-person or telephone) or a simple mail survey.

6.1.6 Stated Response Analyses

The emphasis of household travel/activity surveys has traditionally been to collect people’s actual travel behavior (their revealed travel preferences), but as the analysis demands on the survey data are being increased, travel surveyors have begun to experiment with collecting hypothetical choice information from household travel/activity surveys.

Stated response survey questions can provide the survey team with information such as:

• How people are likely to react to changes in transportation services and infrastructure;
•  How people are likely to react to potential new government policies; and
•  Confirmation (or rejection) of revealed preference modeling results.

This emerging use of stated response techniques and the survey design issues related to them are described in Chapter 13.0 The key design issues include:

• The added costs and complexity of designing survey questions;
•  The potential need for delivering stated-response survey materials to respondents;
•  The added burden on respondents of figuring out and answering the stated-response questions; and
•  The different analysis requirements of such survey data.

The inclusion of stated response questions on a household travel/activity survey will affect the decision of the best survey method since these questions work best when an interviewer is available to answer respondent questions and to explain the sometimes complex instructions. In addition, including these exercises will affect the sample selection and the need for advanced fieldworker training.

6.1.7 Seasonal Analyses

In general, household travel/activity surveys capture travel conditions over a small period of time during a year. Usually, the data and analyses are extended to look at other times of the year by factoring trips using travel volume data, but because it is generally acknowledged that people’s travel patterns vary between seasons along many dimensions, including trip purpose, duration, frequency, and destination choice, it is highly likely that the household travel or activity data represent the time period for which they were collected to a much higher degree than other periods.

Traditionally, household travel/activity surveys are conducted in either the Spring or Fall. These seasons coincide with the most common traffic data collection periods. In addition, they represent time periods when schools are in session, and when potential respondents are least likely to be away from their homes on vacation.

However, the survey team should consider the analyses that will need to be performed before scheduling the household/activity survey. For the past five years, a primary driving force behind travel demand modeling has been the need to better measure and track air quality. Most regions concerned with air quality issues are most interested in Summer conditions (due to increased ozone levels) and Winter conditions (due to cold start emissions). Nevertheless, most surveys and models continue to be for the Spring or Fall, because they seek to capture specific “average” or “peak” conditions. Agencies whose primary concerns are air quality-related should determine which season is the most important to have accurate travel data, and schedule the survey accordingly.

Some recent household travel/activity survey efforts have collected data from respondents in more than one season in a year. The survey data are allowing the sponsoring agencies to compare travel between seasons, and the resulting analyses of the data are likely to describe “average” travel conditions better than a single season survey would. Unfortunately, this approach is not cost-free. First, many agencies do not have the luxury to add six or nine months on to the survey development schedule to spread out the data collection. Second, the cost per completed survey is likely to be higher since there are economies-of-scale related to many survey cost components. For most common survey methods, it is less expensive to conduct one large household survey than several smaller ones. Third, to perform seasonal comparisons, the total sample size is likely to have to be higher, further increasing costs.

To summarize, the selection of the survey season (or seasons) should be based on the following considerations:

•  Are the expected analyses of the survey data seasonal in nature, like air quality analyses?
•  Are travel patterns in a particular season predictable based on another season’s travel patterns and available interseasonal travel volume information?
•  How do respondent contact and cooperation rates vary by season for the different survey methods? What effects do these variations have on survey cost?
•  Do time and budget constraints preclude the possibility of collecting the household travel/activity survey data over two or more seasons?

6.1.8 Analysis Time Periods

Just as some analyses are related to particular seasons, many transportation modeling analyses are related to particular days of the week and hours of the day. Based on the anticipated analyses, the survey team has three important survey design decisions to make with regard to analysis time periods:

•  How much travel or activity data are needed from each respondent or respondent household?
•  For which weekday time periods are data needed?
•  Are data for weekends also needed?

Based on the analysis needs for the survey data, the survey team must determine the days and hours for which travel or activity information will be sought. In the U.S., household surveys have traditionally asked that respondents record travel or activities over a 24-hour period. Some smaller survey efforts, including surveys in Keene, NH (1991) and Southeastern New Hampshire/Southern Maine (1992) and a survey on Staten Island (1990), have asked for this information only for peak travel periods, but most survey efforts collect the full day information, even when only peak-hour analyses are conducted. In Europe, some travel diary periods are as long as two weeks, but European respondents are generally much more tolerant of survey efforts than North American respondents. Diary periods of this length are not likely to be successful in the U.S.

A few recent major household travel/activity survey efforts in the U.S. have asked for the data for 48-hour periods. These surveys have sought to describe day-to-day variation in activities and travel behavior. Although some of the second day trip information is duplicative of the first day information, the surveyors have found that the multi-day survey data better explains day-to-day variation in household and personal trip generation rates, and provides more mode choice data. In addition, the multi-day diaries can provide the survey team with insights about travel behavior that one-day diaries cannot.

The primary reservations that surveyors express about multi-day diaries is that the increased respondent burden of the multi-day diaries will lead to higher fatigue levels and higher non-response rates. In addition, many surveyors worry that because of the fatigue factor, respondents would be more likely to under-report trips and activities on the later days of the multi-day diary. Research on the subject confirms that this is a problem for longer travel periods (seven days or more), but the evidence on two and three day diary periods is less conclusive. Lawton and Pas have found that two-day diary periods are not subject to declining trip reporting.⁷ On the other hand, the recent Dallas-Fort Worth pretest data showed that the second day of the 48-hour diary had significantly fewer reported weekday trips than the first day of the 48-hour diary. The pretest also recorded slightly fewer reported trips in the first 24 hours of the two-day diary than in the 24-hour diary.

The collection of multi-day diary data complicates the development of 24 hour travel models because the daily trip patterns within individual diaries are not independent. Analysts need to account for the dependencies in developing 24-hour travel models, or estimate models for the multi-day period that corresponds to the diary length. Because the recent multi-day diaries have only just been completed, it has not yet been shown whether and how the additional data improve travel models.

In addition to deciding the duration of the diary period, survey teams must also consider for which days of the week to seek the travel and activity data. Most transportation planning analyses have traditionally sought to describe an average weekday’s conditions. This has led most surveys teams to seek travel information for Tuesdays, Wednesdays, or Thursdays of non-Holiday weeks. In recent years, a number of agencies have identified the need for analyses based on special conditions, such as weekends or Friday afternoon peak periods. Household travel/activity surveys need to reflect these analysis requirements, and travel survey teams need to consider the effect that these special investigations have on required sample sizes, respondent requirements, and survey cost.

6.1.9 Selection of the Survey Method

Once the effects of the likely analyses of the survey results are well-understood, the most basic survey design issue for the household travel/activity survey is the selection of the survey method to be used. This decision needs to be based on the strengths and weaknesses of the different survey methods and the overall goals of the survey team. Most survey implementation issues that will be encountered (and are discussed in this chapter) will relate back to the basic selection of the survey method, and conversely the selection of the survey method should be guided by the survey team’s preliminary evaluation of later key issues.

6.1.10 The Components of A Household Travel/Activity Survey

To define the universe of available survey method options for household travel/activity surveys, it is useful to consider the fieldwork components of a household survey separately. Household surveys consist of the following key components:

•  Screening and Recruitment – Enlisting the cooperation of potential respondents and ensuring that a contact meets the geographic and demographic requirements of the study (as needed by the travel demand models);
•  Distribution of Materials – Delivery of survey forms and related documents to respondents; and
•  Collection or Retrieval of Survey Responses – Obtaining the survey responses from the respondents.

The different survey methods can be defined by how they accomplish each of these component tasks. Some methods combine the basic components. Others do not include one of the components. However, decisions about the three main components, screening and recruitment, distribution of materials, and collection of survey responses, define the available survey methods.

6.1.11 Commonly Used Household Survey Methods

Based on the strengths and weaknesses of the data collection procedures for each survey component, surveyors have applied many combinations of recruitment, materials distribution, and data retrieval in their survey designs. As Figure 6.1 shows, combining all the different methods for each component of the household travel/activity survey yields more than a dozen feasible survey methods.

The household travel/activity survey team may want to consider the strengths and weaknesses of each feasible method for their particular survey effort. However, because most of these methods have not been proven to be efficient for household travel/activity surveys, this Manual focuses on only a few of the methods listed above.

Tables 6.2 through 6.7 summarize the most relevant household travel/
activity survey methods. Table 6.2 discusses the simple single contact telephone survey, and Table 6.3 describes the basic mail survey. Table 6.4 and Table 6.5 describe the two most common combinations of mail and telephone survey methods for household travel/activity surveys, the telephone-mailout-mailback survey and the telephone-mail-telephone survey. These four survey methods are the primary focus of the remaining discussion of household travel/activity surveys.

Table 6.6 describes the simple in-home survey that was commonly used in the 1960s household travel surveys. Table 6.7 summarizes the two stage in-home survey method that was developed as an extension to the traditional in-home survey when the need for travel diaries was recognized. As Tables 6.6 and 6.7 indicate, in-home methods for household travel/ activity surveys are probably relevant only in very specialized situations. The use of these two methods is not recommended for most new travel surveys.

6.1.12 Selection of Data Collection Techniques for Household Travel/Activity Surveys

Once the survey team has selected one or more methods for further survey design, the next survey design task is to determine the best data collection techniques for each method.

As we discuss below, the quality of data collection using mailback surveys can be enhanced by a number of design factors, but the data collection techniques for these types of surveys are essentially the same. The respondent is expected to complete the survey materials as instructed and to send the completed forms back to the survey team.

Figure 6.1 Some Feasible Household Survey Methods

Table 6.2 Household Travel/Activity Survey Methods: The Single-Contact Telephone Survey

Table 6.3 Household Travel/Activity Survey Methods: The Mail Method

Table 6.4 Household Travel/Activity Survey Methods: The Telephone-Mail-Telephone Survey

Table 6.5 Household Travel/Activity Survey Methods: The Telephone-Mailout-Mailback Survey

Table 6.6 Household Travel/Activity Survey Methods: The Single Contact In-Home Survey

Table 6.7 Household Travel/Activity Survey Methods: The Two-Stage In-Home Survey

Collecting data by interviewing respondents, either by phone or in person, can be accomplished in more than one way. If the survey team is considering one of the interview techniques for the household travel/activity survey, they will need to make the following decisions about the data collection techniques:

•  Is the use of qualitative survey techniques viable or desirable for the survey effort?
•  For telephone surveys, should centralized interviewing facilities be used?
•  Should computer-assisted interviewing techniques be employed?

6.1.13 Qualitative Survey Techniques

Typical travel surveys are designed to be highly structured. Whenever possible, respondents are asked to answer closed-ended questions that have predetermined response categories. Sometimes a few open-ended questions are included in the surveys, but usually only when absolutely necessary. For almost all travel modeling applications, a highly structured survey instrument is necessary or at least highly desirable.

However, some planners outside of the U.S. have found that removing the tight structure of the interview is an effective way to obtain information about how respondents actually think and believe about certain issues . These planners have developed and applied household travel surveys that rely on unstructured (or semi-structured) interactions between respondents and interviewers.

Qualitative surveys (or interactive surveys) are in-depth interviews where respondents’ answers are used to guide the format and topics of the interview. They are similar to focus group discussions, except they are conducted on a one-on-one basis, either in-person or by telephone. Interviewers probe and ask supplementary questions about the most interesting topics raised in the interview, and in some cases the interviewer will ask purposely biased questions to challenge the strength of a response or to clarify the respondent’s opinions. Typically, interviewers work from discussion guides, rather than questionnaires, and the interviews are tape recorded so that responses can be analyzed in detail at a later date (Jones, 1985).

Unfortunately, the costs related to qualitative surveys and the special skills needed to perform them often make these surveys infeasible. For interactive surveys to be successful, the interviewers have to be highly skilled and must understand the survey topic and the issues facing the sponsoring agency. Therefore, the number of interviewers that are able to perform these types of surveys is small. In addition, the analyses of the tape recorded interviews requires special talents and a significant amount of time.

In general, travel demand models are designed to use structured data, so transportation planners typically do not see any reason to perform qualitative surveys. However, special household travel/activity surveys that are seeking to obtain large amounts of opinion and attitude data could benefit from an interactive approach. Some of the next generation travel models, like TRANSIMS, will likely benefit from the data available from qualitative surveys.

6.1.14 The Use of Centralized Telephone Interviewing Facilities

A telephone interviewer can complete his or her task from virtually any telephone. Many early telephone surveys were conducted from interviewer homes or offices. In these cases, each telephone interviewer was given a subset of the telephone numbers in the sample. The interviewer would contact as many of the households as possible, and then after a pre-specified time they would deliver the completed survey instruments to survey managers.

A very serious problem with this approach is that the ability to supervise interviewers as they conduct the surveys is lost. The survey team needs to rely on the skill and professionalism of the interviewers to conduct the surveys correctly and without biasing results.

For this reason, it is recommended that all travel telephone surveys be conducted from centralized locations with supervisors. Supervisors are able to observe interviewers while they work to ensure that they are following procedures correctly, and if problems are identified, they can be rectified immediately. Interviewers are able to ask questions if needed, and if a respondent wishes to speak to a supervisor to verify the authenticity of the survey or to complain, they can be easily transferred. In addition, interviewers at centralized locations are able to learn from each other as they conduct the interviews.

Using a centralized telephone interviewing facility also allows the survey team to establish regulations on telephone interviewing hours. A common complaint that telephone survey respondents (and non-respondents) have is that they were contacted too late at night or at an inconvenient time. Professional marketing research firms usually have guidelines with regard to calling times. For instance, many firms avoid making calls after 9:00 p.m. Different limits on calling times are likely to be appropriate for different survey populations, so survey teams need to establish the regulations individually for each study.

The central telephone survey location can be either a professional marketing research interviewing facility or a temporary facility fashioned out of an agency’s or firm’s office. Since most telephone interviewing is conducted in the evenings and on weekends, it is possible to transform an office into a primitive telephone interviewing facility during off-hours and then switch back to an office in time for regular business hours. Open plan offices with individual phone lines, which are currently quite common, are especially easy to turn into telephone interviewing facilities.

The facilities are not ideal, however, because monitoring calls and providing general supervision are somewhat difficult. In addition, interviews that require toll calls are best handled from professional facilities with WATS lines and other more sophisticated telephone equipment.

6.1.15 The Use of Computer-Assisted Interviewing Techniques

In the past, the most common technique used to record the results of personal interviews and telephone interviews was the pencil-and-paper interview (PAPI), in which:

1. Interviewers record answers on survey instruments or on interview schedules;
2.  Trained coders translate the answers into codes, and record them on coding sheets; and
3.  Data entry specialists enter the codes into a computer data file.

However, the widespread availability of desktop and notebook computers has led to the development and wide acceptance of computer-assisted telephone interviewing (CATI) and computer-assisted personal interviewing (CAPI) software. Most household travel/activity surveys in the last few years that have involved telephone interviewing have been performed using CATI techniques.

6.1.16 CATI Advantages

CATI reduces the three step data collection-coding-data entry process into one automated, on-line procedure. A computer screen prompts an interviewer to ask a question, then the interviewer records the response, and the computer codes it and saves it to a data file.

CATI techniques have the following interviewing advantages (Fowler, 1988)( Jones and Polak, 1993):

1.  They can be designed to permit the entry of only legal codes in any particular field (preventing many data entry errors);
2.  They can be used to check entries to make sure that they are consistent with other previously entered data (preventing data inconsistencies);
3.  They automatically route interviewers through the interview (ensuring that respondents are asked all the relevant questions and are not asked ones that should be skipped);
4.  They can use information from previous questions or previous interviews to make interview questions or the sequencing of questions specific to a particular respondent; and
5.  They can be used to help combine the survey’s data collection and management functions; for example, once a telephone interviewer has finished with one respondent, the CATI system can check whether she or he has arranged to return a call to another number, or search the non-contacted numbers for instances where the current time has not yet been tried.

Appendix C shows an example of a recent household travel survey that illustrates the advantages of using a CATI approach.

In addition to improving interviewing capabilities and reducing editing and coding requirements, CATI systems have a number of other advantages, including (Nicholls):

Sample Management – The CATI system maintains the sample status of each case and links input data to the interview and output record.

•  On-Line Call Scheduling and Case Management – The CATI system sets the priority, sequence, and timing of calls.
•  On-Line Monitoring – The CATI system is able to reproduce any interviewer’s screen at a supervisor’s terminal where audio monitoring may occur as well.
•  Automatic Recordkeeping – The CATI system stores information on on-line calls, their outcomes, response rates, and interviewer productivity, and makes the information accessible to managers in on-line and printed reports.

6.1.17 CATI Disadvantages

As noted above, CATI systems are now commonly used for household travel/activity surveys. However, despite the advantages of the computer-assisted techniques discussed above, there are also negative aspects of the computer-assisted technologies.

First, CATI surveys require a great deal of lead time so that they can be programmed to produce the desired range-checking, question sequencing, and calculations. The CATI programs need to be perfect before the survey is fielded, because interviewers will not generally be able to fix them as they go along. Testing and debugging complex CATI programs could take several weeks and require well over a person-month to complete.

Second, even though the systems can be taught to accept only answers that fall within an acceptable range, they cannot control the quality of data entry. When a CATI interview is completed, the only record of the interview is the data file. There are no source records like in pencil and paper interviews to verify that the survey data was entered into the computer accurately. In addition, it is often difficult for interviewers to include special notes or extra information.

Third, if the CATI program is not carefully designed so that interviewers can avoid collecting duplicative information and can insert missing information from previous responses, the CATI interview can take longer than a pencil-and-paper interview. The paper-and-pencil phone retrieval of Portland’s two-day diary survey took about the same respondent time as the CATI retrieval of a one-day diary pretest in Dallas/Forth Worth. The diary format for the Dallas survey was subsequently modified to allow the CATI data collection to run much more smoothly.

Finally, CATI systems are highly specialized software routines. Most agencies do not have the resources to develop software of this nature in house, so by selecting to use computer-assisted methods, a survey team is probably also ensuring that they will need to enlist marketing research contractors for the survey effort. Most CATI household travel/activity survey efforts have used the commercial packages that the marketing research contractors purchased or licensed and have customized for collecting data.

It is possible to combine CATI and PAPI techniques within the same survey effort. Some recent household travel/activity surveys have used CATI techniques for recruitment, but PAPI techniques for data retrieval. It is also possible to combine the techniques within the same survey, such as by using CATI to retrieve household and person record information and PAPI to collect trip and activity diary information.

6.1.18 Procedures to Enhance the Accuracy of Household Travel/Activity Surveys

Along with determining the survey methods and data collection techniques to be used for the household travel/activity survey, the survey team needs to consider the different available procedures for managing survey bias and inaccuracy. Chapters 4.0 and 5.0 identified the following sources of survey bias:

•  Misidentification of the survey population;
•  Imperfect sampling frames and sampling loss;
•  Non-response;
•  Poor questions and survey instruments;
•  Fieldworker and interviewer errors; and
•  Coding, data entry, and data processing errors.

Procedures to minimize the last three items in household travel/activity surveys are discussed in detail later in the chapter, but if the survey team is to effectively reduce the biases associated with the first three sources of bias, it will be necessary to address them while the survey is first being designed. Procedures to improve household travel/activity survey accuracy should be viewed as integral to the survey design, rather than “extras,” and the costs associated with these procedures should be considered before the final selection of survey methods and techniques are made.

6.1.19 Procedures for Improving the Identification of the Survey Population

The survey population for a household travel/activity survey is either the collection of all households within a study area or some collection of the people who live within those households. In designing the household travel/activity survey, the survey team must:

•  Ensure that the anticipated analyses can be accomplished with household-based data, as opposed to data based on other sampling units like trips within a particular analysis corridor; and
•  Define the boundaries of the study area for which analyses will be required.

In most cases, these concerns will have been addressed prior to the detailed household travel/activity survey design. Presumably, the anticipated analyses have led to the need for household-based data because otherwise, different (and usually less expensive) types of surveys would be considered. In addition, the study area for the survey is usually set independently of the survey design effort based on particular analyses needs and political boundaries. Definition of the study area boundary is discussed in a greater detail in Chapter 7.0.

Most regional agencies define the geographical extent of their survey by county (political) boundaries. Often this is expedient since Metropolitan areas are defined by county boundaries, and it is easy for executive boards to understand. In cases where a county may extend very far beyond the urbanized area boundary, a cordon line may be used to determine areas for inclusion or exclusion in the survey.

6.1.20 Procedures for Improving the Sampling Frame and For Reducing Sampling Loss

It is likely that the household travel/activity survey team will be faced with an imperfect sampling frame. Because all the most common address-based and telephone-based sampling frames are designed for other purposes, it is not surprising to find that they often need to be cleaned, edited, and augmented for the survey effort. The most common procedures for improving the sampling frame for a household travel/activity survey include:

• Field validation of address-based data sources;
•  Combination and cross-checking of two or more sampling frame databases; and
•  Special efforts to include identifiable underrepresented groups.

6.1.21 Procedures for Reducing Non-Response

Survey non-response is commonly categorized into unit non-response, referring to the failure of potential respondents to reply to the survey as a whole, and item non-response, referring to respondents’ failure to respond to particular items on the survey. Methods to reduce item non-response are discussed later in the description of questionnaire design. Methods to reduce unit non-response are described in this section.

Four general approaches are commonly used for reducing unit non-response in household travel surveys:

•  Pre-notification of the survey effort;
•  Survey follow-up techniques;
•  Offering potential respondents tangible incentives to complete the survey; and
•  Response facilitators (elements of the mail or telephone surveys that decrease the likelihood that potential respondents will refuse to participate).

These approaches can all have a large effect on the overall design and cost of the household travel/activity survey effort.

6.1.22 Pre-Notification as a Method of Improving Survey Response

Pre-notification of the household travel/activity survey consists of contacting potential respondents by telephone or mail prior to soliciting participation in the survey. The pre-notification contact is used by surveyors to build respondent interest in the survey effort, and to help allay respondent doubts about the validity of the survey. There is evidence that pre-notification improves survey response rates, response speeds, and response quality (Hornik, 1982). Another potential benefit of pre-notification is that it can provide an early measure of likely response rates and non-response trends.

Pre-notification can be used for household surveys with mail, telephone, or in-person methods, or any combination. In theory, the pre-notification contact can be accomplished in any of the three common ways:

1.  Telephone pre-contact;
2.  Pre-contact with a letter, brochure, or postcard; and
   
3.  Face-to-face personal pre-contact.

However, in general, the cost of in-person surveying precludes this approach as a pre-notification procedure. In addition, it is not common for telephone pre-notification to be used prior to household surveys with telephone recruitment. In this situation, mail pre-notification or no pre-notification at all are more commonly used. The short recruitment call probably achieves many of the same goals of the telephone pre-notification.

Pre-notification of some type is usually always warranted in the case of mail surveys and surveys with in-home recruitment. Since the sampling frames for these surveys are usually address-based, respondent phone numbers are generally not known. Therefore, the most common approach is to send a postcard or letter of introduction.

In a sense, pre-notification is a sales technique to convince potential respondents to participate in the survey effort. Consequently, the most successful pre-notification efforts tend to employ sales techniques.

A few recent household travel/activity surveys have used formal pre-notification techniques, and those that have seem to have benefited from it. For instance, prior to conducting recruitment calls for their travel survey, the Metropolitan Washington Council of Governments (MWCOG) sent out an introductory letter signed by the directors of the Departments of Transportation in the region. The letter simply provided an overview of the survey and the study, and asked for the recipients’ participation in the upcoming survey. MWCOG estimates that the pre-notification letter increased survey participation by between five and ten percent .

6.1.23 Survey Follow-up Techniques for Improving Survey Response

One of the most effective ways to reduce survey non-response is to follow-up with respondents who do not complete the survey. Survey follow-up procedures are generally used with mailback surveys, but the concept can be applied to telephone and in-home surveys, as well.

6.1.24 Survey Follow-Up for Mail and Telephone-Mailout-Mailback Surveys

Mail survey follow-up techniques are used for two reasons:

•  To clarify responses on returned questionnaires (corrects item non-response); and
•  To convert refusals and other non-responses into completed usable responses (corrects overall non-response).

Follow-Up for Item Non-response and for Clarification of Responses in Mail and Telephone-Mailout-Mailback Surveys

Clarification of responses is generally done by phone to expedite the process and to ensure that the corrected/edited responses are adequate. In telephone-mailout-mailback designs, the respondent has been recruited by phone, so it is relatively easy to recontact him or her to ask about specific responses (provided that the responses with the problems do not require the respondent to have any survey materials on hand).

Many recent travel survey efforts have used this technique to clarify and correct spurious, suspicious, or out-of-range answers. In general, the surveyors found that the number of clarification calls needed was small, and that because most problems were quickly corrected or clarified, most follow-up calls were short.

To clarify or correct the responses on simple mailout-mailback surveys by telephone, it may be necessary to request telephone contact information from respondents. Ironically, asking for this information to correct item non-response may actually increase the overall non-response rate because of people’s confidentiality concerns. Surveyors may be able to determine some respondents’ telephone numbers from telephone directories and/or reverse directories, but if this approach is adopted, the surveyor must understand that she or he could end up with different quality data for those with listed numbers and those without listed numbers. Since the problems that need to be clarified will probably be minor, the potential bias is generally ignored.

Follow-Up for Overall Non-response in Mail and Telephone-Mailout-Mailback Surveys

The second type of follow-up survey seeks to increase the overall response of the survey by reminding non-respondents that they have not yet completed the survey. Because the overall response rate for mailback surveys is generally fairly low, follow-up techniques are often used to increase the response. Fowler claims that:

“While attractive presentation of the study and good questionnaire design will help, there is no question that the most important difference between good mail surveys and poor mail surveys is the extent to which researchers make repeated contact with non-respondents” (Fowler, 1988)

There are several different follow-up approaches for mailout-mailback surveys, including:

•  Follow-Up Postcards – respondents are sent a reminder postcard stressing the importance of their responses to the survey;
•  Follow-Up Letters – respondents are sent a brief letter (usually from an elected official, such as the one who signs the cover letter for the initial mailing) restating the goals of the survey and its importance;
•  New Survey Materials – respondents are sent a new set of survey materials under the assumption that they have misplaced the original set;
•  Telephone Reminders – respondents are called, reminded about the survey and are usually asked if they need a new set of survey materials; 
• Telephone Retrieval – respondents are called and asked to provide the survey information by telephone; and
•  Combinations of any or all of the above.

The best follow-up method will depend on the available budget, available time, the initial response rate, and the surveyor’s level of concern about non-response. Experts differ on the best approach.

The following mail survey sequence is recommended (Dillman, 1978), ( Richardson, Ampt and Meyburg, 1995), ( Fowler, 1988): 

1.  Send pre-notification letter one week prior to the initial survey mailing or recruitment call;
2.  Recruitment call (if chosen method requires it);
3.  Initial survey mailing;
4.  Send postcard reminder or make telephone reminder call one week after initial mailing;
5.  Send letter and new materials three weeks after initial mailing;
6.  Send letter reminder four weeks after initial mailing; and
7.  If response rate is still unsatisfactory, after six weeks send letter and new materials, or make telephone reminder calls for respondents with listed numbers.

Peterson, Albaum, and Kerin recently compared 27 alternative pre-notification and follow-up strategies for mailout-mailback surveys (Peterson, Albaum and Kerin). Their results are particularly interesting because they used a survey instrument that was designed to generate relatively low response rates, similar to mailed household travel and activity surveys. They compared the contact strategies based on response rates and cost per completed response. Figure 6.2 summarizes some of their findings.

The simple mail survey without pre-notification or follow-up yielded a 10 percent net response rate at a cost of $6 per completed response. Introducing pre-notification increased the response rate to 11 percent (for postcard notification) and 14 percent (for letter notification), and increased survey costs to $8 per completed response. Introducing a single follow-up contact without pre-notification produced a 13 percent return at a cost of $9 per response for postcard follow-up, and an 18 percent return at a cost of $8 per response for a follow-up letter with a copy of the questionnaire. Combining pre-notification and a single follow-up contact produced response rates between 15 and 20 percent at costs between $8 and $10 per response.

As the figure shows, the most successful strategies (in terms of response rate) involved pre-notification and two follow-up contacts. The cost per completed response for these strategies are slightly higher than the simple survey effort, but the response rates were more than double the simple effort. At these low response levels, the higher response rates almost certainly would outweigh the slightly higher costs.

Richardson, Ampt, and Meyburg also conclude that pre-notification and follow-up are cost effective investments for household travel/activity mail surveys (Richardson, Ampt and Meyburg, 1995). Table 6.8 shows a cost comparison based on a 1993 Australian household mail survey. The top of the table shows the costs of a non-follow-up survey design which yields a total of 6,000 returns. The bottom half of the table shows the costs for the survey design the authors recommend that also yields 6,000 returns. The survey with the extensive follow-up is estimated to actually cost less. The survey with follow-up also has a higher response rate, perhaps reducing the amount of bias.

Figure 6.2 Alternative Contact Strategies for Mail Surveys

Table 6.8 Cost Comparison of a Household Travel Survey With and Without Survey Follow-Up

6.1.25 Survey Follow-Up for Telephone and Telephone-Mail-Telephone Surveys

For telephone and in-home surveys, overall non-response occurs because of the surveyors inability to contact potential respondents, or because potential respondents refuse to participate in the survey. Therefore, non-response-reducing strategies have been designed primarily for the recruitment stage of the survey, rather than for the retrieval follow-up stage.

While item non-response is as much or more of a problem with interview surveys as it is for mailback surveys, for the most part it is dealt with during the actual interview. If a respondent is unable or unwilling to answer a specific question, an interviewer probes for an answer or further explains the question. Follow-up contacts are not likely to improve the quality of the responses that the initial interviewer is able to get. This is particularly true if the interviewer is well-trained and the CAPI or CATI software is designed well to trap inconsistencies, illogical answers, and errors.

Still, it is relatively easy and common to recontact respondents by telephone to correct problems discovered after the interview. Since the respondent has already invested a great deal of his or her time into the interview, clarifying a few questions is generally not a problem. On the other hand, respondents are likely to get tired of re-answering questions so follow-up contacts need to be short. If a response has so many questions or problems that it would require more than a few follow-up questions, the surveyor should probably classify the response as unusable.

6.1.26 Survey Follow-up Considerations for Diary Surveys

Travel and activity diaries usually ask respondents to record their travel or activities over a pre-specified period. If a respondent fails to complete the diary during that period or immediately following the period, she or he is more likely to forget about certain travel or travel details. Consequently, in follow-up contracts, most survey teams ask respondents to consider a different upcoming day (or days) when completing the diary. While this method is probably preferable to asking respondents to remember travel and activities a day or a week or even more in the past, it is still not optimal. In expanding the data, the survey team may need to consider the differences in travel conditions between the desired and actual diary periods.

In addition, when reassigning diary periods for respondents, the survey team should consider potential inconsistencies between the original and new periods. For instance, if schools are in session during the original diary period, they should also be in session on the new date. Usually, the follow-up contact asks respondents to use the same day or days of the week as the original period as soon as possible after the original period.

6.1.27 The Use of Survey Follow-Up to Measure and Correct for Non-response Bias

The primary goal of using survey follow-up techniques in household travel/activity surveys is to reduce the level of non-response in the survey effort. Another possible advantage of conducting the follow-up is that it provides the survey team with a means to infer the characteristics of non-respondents and perhaps to even make corrections. Methods for performing these procedures are discussed in Section 6.12.

6.1.28 Incentives for Survey Methods

Surveyors often provide respondents with incentives of one type or another to motivate them to participate in their survey efforts. The most common incentives that are employed are:

•  Prepaid Cash – some denomination sent to the potential respondents with the survey materials;
•  Promised Cash – an offer in which a specified amount of money would be provided upon completion of the survey;
•  Provided Gifts – a gift, such as a pen, key ring, or refrigerator magnet, enclosed with the survey materials;
•  Promised Gifts – an offer to provide the potential respondent with a specified gift upon completion of the survey;
•  Lottery – the inclusion of the potential respondent in a lottery drawing;
•  Study Results – respondent is promised survey results upon completion of the study;
•  Charitable Contribution – prepaid or promised donation of a specified dollar amount to a charity in the name of the potential respondent.

Travel survey specialists, like their general marketing research colleagues, have mixed views on the cost-effectiveness of incentives. Their usefulness is probably related to the population of the region under study, so broad generalizations about their effectiveness are difficult to make. However, it is apparent that incentives do improve response rates and speeds in many cases. The remaining question for the survey designer is whether the benefits of incentives outweigh the investment in providing them and the potential biases that they may cause.

Based on the recent literature, the prepaid cash incentive is the most consistent incentive method for improving response rates. It is also considered the least biasing of available incentives as well as easiest to use. This conclusion is supported by evidence from household travel/activity surveys, such as the Puget Sound Transportation Panel Survey. In this effort, three incentive approaches were used; 1) no incentive, 2) $1.00 per household member prepaid, and 3) $10.00 per household promised incentive. The two groups that received incentives each had diary rates of slightly more than 60 percent, compared to a return rate of 49 percent for the group not receiving the incentive (Furse and Stewart, 1982), (Church, 1993), (Gajraj, Faria and Dickinson), (Tooley, 1995).

Experience with monetary incentives has revealed that incentives need not be substantial. The incentive should be a small token of appreciation for the respondents’ efforts. Ideally, it should build rapport between surveyors and respondents, and it should motivate respondents to try to please the survey sponsors. Larger incentives, especially in the promised form, take on the feeling of payment for one’s time, and for complex household travel/activity surveys, even relatively high payments are not likely to be adequate compensation for many respondents.

Despite the advantages of the prepaid monetary incentive, there are conditions when another incentive type is more reasonable. Agencies may be able to provide other types of incentives more cost-effectively, or may have reasons for not wanting to provide the pre-paid incentive. Sometimes agencies can obtain suitable gifts, such as pens, maps, or refrigerator magnets, at no cost or reduced cost. Gift incentives would probably be more cost-effective in these cases. A recent household activity survey in the Boston region (an area with a high rate of state lottery participation) offered vouchers for a $1.00 state lottery ticket, in part because it did not require the agency sponsoring the survey to send cash incentives to people at a time of state government cutbacks.

Although incentives of all types are used to increase survey response, evidence suggests that incentives do not have the same appeal for all respondents. Biases can be created when incentives are used. No known studies relate incentive conditions to survey measures of respondent travel, but incentives are known to have different appeals based on the respondent’s sex, marital status, employment status, property ownership, and religion (Furse and Stewart, 1982). Therefore it is reasonable to assume that trip generation estimates could be affected by the use of incentives, as well. Some travel survey experts do not recommend the use of incentives because they feel the risk of bias outweighs the potential improvement in response.

6.1.29 Response Facilitators

Although the use of incentives is the most well-known mechanism for increasing survey response, it is likely that other survey considerations will have as large or larger effects on survey response and quality. Based on their experiences and intuitions, survey researchers have developed a number of survey response facilitators that they believe increase the likelihood of survey participation. It is not clear how much these facilitators affect response rates, because researchers have difficulty isolating them from other aspects of the survey. However, most survey designers stand by one or more of them.

During the survey design, the household survey team should decide which facilitators are most likely to be important for their survey population, and they should estimate the costs of providing them.

As Dillman points out:

Non-response is a serious problem under any circumstances. Thus each element that might help prevent it – no matter how trivial – is worthy of design considerations (Dillman, 1978)

Response facilitators include the following:

Mail Survey Response Facilitators

•  Include a cover letter signed by a high-ranking and popular elected official.
•  Personalize the survey materials for each respondent, where possible.
•  Use postage stamps on any packages sent to respondents, rather than prepaid or machine stamped mailings, so the mailing stands out from direct mail.
•  Send materials in distinctive envelopes.
•  Provide a toll-free telephone number for respondents to call in case they have questions or complaints.
• Have the return address(es) be within the region under study.
•  Have the return address(es) be for the agency or another public organization, rather than for a private firm.
• Provide the respondent with a deadline for replying to the survey.
• Provide brief reassurances of anonymity on the survey materials.
•  Provide descriptions on the survey materials of the importance of the survey and of the specific respondent’s role in the survey.

Telephone Survey Response Facilitators

•  Make sure interviewers have local accents or are relatively accent-free.
•  Provide reassurances of anonymity at the beginning of the call.
•  Provide descriptions of the importance of the survey and of the specific respondent’s role in the survey.
•  Provide a toll-free telephone number for respondents to call in case they have questions or complaints.

In-Person Survey Response Facilitators

•  Select interviewers that are of the same age groups, races, ethnic backgrounds, and social classes of potential respondents.
•  Provide reassurances of anonymity at the beginning of the interview.
•  Provide descriptions of the importance of the survey and of the specific respondent’s role in the survey.
• Provide a toll-free telephone number for respondents to call in case they have questions or complaints.

These mechanisms are all likely to help improve response rates marginally, but the survey team needs to consider the facilitators as a package. Simply selecting a few facilitators to improve response will not be as effective as developing an integrated strategy, using pre-notification, follow-up, incentives and facilitators that work well together and complement one another.

6.1.30 Output of the Survey Design Task

By the time the survey team completes the survey design task, they will have analyzed the output data needs from the household travel/activity survey, and made decisions about the survey method, data collection techniques, and the inclusion of different design elements to improve the quality of the survey results. The survey team will have a clear idea of the approach (or approaches) that will need to be pretested.

The survey design task outputs will feed directly into the sampling, survey organization, and survey materials development tasks, but, in reality, the survey design task will guide all the work conducted on the rest of the tasks.

It can be helpful at this state of the survey implementation process to prepare a detailed plan for the household travel/activity survey. The survey team will be in a position to define detailed survey procedures and to layout more accurate schedules and budgets. The detailed survey plan is a useful document for involving outside agencies and/or technical advisory committees in the development of the household travel/activity survey. In addition, the plan organizes the survey team’s tasks, and can be an effective tool for allocating responsibilities.

By the time the survey design task is winding down, it is likely that the survey team will already have gotten underway on the organization and sampling tasks, which are discussed next.

6.2 Sampling for Household Travel KM Surveys

The statistical computations needed to determine sample sizes for travel surveys are described in Chapter 5.0. For household travel surveys used to develop travel demand models such as those maintained by metropolitan planning organizations, the study population is usually known. The sampling unit is the household, and the sampling frame a list of households by telephone number or address.

Urban travel demand model systems include a number of components to be estimated for which the survey data are needed. These include:

•  Trip generation;
• Trip distribution; and
•  Mode choice.

The variables of interest are different for each of these models. The sampling requirements for each are discussed below.

6.2.1 Trip Generation

The main variable of interest for trip generation models is the number of trips generated by households for each trip purpose. The household variables generally used in trip production models include number of persons, income, auto ownership, number of workers (for work trips), and number of students (for school trips). Most variables of this type have distributions that can be obtained from census data, providing a good basis for computing sample size requirements. The census, however, provides no information on the number of trips generated (and no information at all on non-work trips).

If information on the mean and variance (or coefficient of variation) of the number of trips generated per household were available from another source – say a previous survey – the required sample size could be computed from Equation 5.7:

where:

σ² = represents the standard deviation of the population; and

SE(m) = the standard error for the mean for a given confidence level and precision level.

Smith used values from some older (1960s) household surveys to determine the coefficient of variation and computed a typical sample size requirement of about 900-1,200 households at the 90 percent confidence level and a precision level of +/-5 percent (Smith, 1979). The higher number resulted from an assumed cross-classification in the trip production model by income and auto ownership.

This analysis, however, did not take into consideration different trip purposes; ideally one would compute the required sample size for each purpose in the model and use the largest. With smaller means for the number of trips by purpose, the standard error may be smaller, resulting in larger required sample sizes.

In planning for the 1990 Bay Area household survey effort, the MTC estimated necessary sample sizes by trip purpose using data from their 1981 survey effort. Table 6.9 shows the conversion of trip rate information into sample size estimates for this effort.

Some recent household survey efforts intended for use in developing trip production models have used smaller sample sizes of around 500 households, including surveys in the Portland, Maine and Pittsburgh areas. While information on statistical levels of accuracy and precision have not been reported, it can be assumed that lower levels of one or both were found to be acceptable in these areas.

It is common practice to use a stratified sampling plan for collection of trip generation data. Since trip production models are often cross-classification models, the survey sample can be stratified according to variables in the model such as those described above. Information on existing distributions of the variables is usually available from census data, so required sample sizes can be computed. This is a good strategy for ensuring sufficient sampling of relatively small but important markets such as households without autos.

The main difficulty with such a procedure is that it is impossible to tell which stratum a household is part of until it is recruited. This can be addressed by collecting a larger sample than necessary to account for the expected number of responses in the critical cell or by screening households prior to having them complete the entire survey – basically creating cell quotas.

Table 6.9 Calculation of Necessary Sample Sizes from Previous Trip Rate Information

6.2.2 Trip Distribution

It is now generally recognized that household travel surveys are not appropriate means of generating acceptable estimates of zone-to-zone trips (Smith, 1979). While household surveys taken in the 1960s were generally used for this purpose, they usually had much larger sample sizes, and models had fewer zones. Presently, travel demand modelers use household survey data to estimate parameters of trip distribution models rather than attempt to develop zone-to-zone trip tables directly from the survey data.

Most trip distribution models in U.S. urban areas are gravity models based on travel times. Some areas use generalized cost instead of travel time, but generally gravity models are based on one variable. With that in mind, the variable of interest is the trip length frequency distribution. Again, Equation 5.7 can be used to estimate the required sample size if the coefficient of variation and mean are known. Pearson reported in 1974 coefficients of variation of 0.53 for home-based work trips, 0.58 for home-based non-work trips, and 0.63 for non-home-based trips (Pearson et al., 1974). Using these numbers, samples sizes of about 600-700 trips per purpose would be required at the 90 percent confidence level for the +/-5 percent error level. Since households make several trips per day on average, only a few hundred households would be required to obtain a statistically significant estimate of the mean trip length. Even if travel time estimates are desired for different times of day, as long as there are not a large number of different time periods (most models use one or two, some three or four), there should be enough trips to estimate travel time distributions.

The above discussion leads to the conclusion that any survey which is sufficient for the development of trip generation models is likely also sufficient for the estimation of gravity model parameters.

Some agencies have been developing destination choice (or more accurately attraction choice) models for the purposes of estimating trip tables. These are generally logit models which are similar in function, and often variables, to gravity models. If travel time is the only parameter of these models, then the same analysis as described above for gravity models holds. However, if other variables are used in the model, the problem becomes similar to that of mode choice models, as discussed below.

6.2.3 Mode Choice

In most urban areas, the use of household travel surveys for the estimation of mode choice models is problematic for the following reasons:

•  In many areas, there are simply too few transit trips to get an accurate estimate of the distribution of important variables among transit users.
•  Unless households are recruited at transit stops or on transit vehicles, it is difficult to determine in advance whether or not there are transit trips made by the household. Therefore, a stratified sampling approach with respect to mode would be difficult to implement.
•  It is unlikely to have information about the means, standard deviations, or coefficients of variation of most of the variables in mode choice models unless another survey had collected them. These variables include fares, parking and other auto-related costs, and wait and access times. In addition, data on other variables such as demographic or area type measures are unlikely to be available weighted by trip (as opposed to by household).
•  The logit model formulation does not lend itself to simple derivation of statistical computation of sample size.

Given these problems, it is rare that a household travel survey sample size would be based on mode choice model requirements. However, in some large cities where mode choice models can be developed from survey data, it is likely that the sample size would have to be much larger than what would be required for trip generation.

It is possible to develop simple estimates of the required sample size to get a statistically significant sample of users for each mode using Equation 5.7. This is done using the sample variance s2 in the equation based on the estimated mode share. For work trips, this is generally available from census data; for non-work trips, a conservative (high) estimate of transit (or other rarely used mode) share can be used to develop a conservative estimate of the sample size requirement. Of course, if there are a large number of modes to be examined, the computation must be repeated for each one.

Stratification of the sample can be an efficient means of increasing the accuracy of the survey data for mode choice purposes. Obviously, if one could identify transit users before recruitment, better information about critical variables for transit users could be obtained. Even though this pre-selection would be very difficult, it is possible to target specific markets that are easier to define. For example, in Portland, Oregon (NuStats, Inc., 1994) the household survey was stratified to include areas near transit lines and with favorable land use characteristics for non-auto modes. Such geographic stratification is not difficult to determine using readily available data such as census data for planners familiar with transportation in the local area. In addition, the Portland survey employed choice-based sampling for one stratum, recruiting park-and-ride users at parking lots.

REFERENCES

Chapin, S.S., Human Activity Pattern in the City, John Wiley & Sons, 1974.

Church, A.H.. Estimating the Effect of Incentives on Mail Survey Response Rates: A Meta Analysis. Public Opinion Quarterly, Volume 57 (Spring 1993): pp. 62 79.

Dillman, Dan. Mail and Telephone Surveys: The Total Design Method, John Wiley & Sons, New York, 1978, p. 161.

Fowler, Floyd J., Survey Research Methods, SAGE Publications, 1988, p. 54, 130-134.

Furse, David H. and David W. Stewart. Monetary Incentives Versus Promised Contribution to Charity: New Evidence on Mail Survey Response. Journal of Marketing Research, Volume XIX (August 1982): p. 363, 375.

Gajraj, Ananda M., A.J. Faria, John R. Dickinson. A comparison of the effect of promised and provided lotteries, monetary and gift incentives on mail survey response rate, speed and cost. Journal of the Market Research Society. Volume 32, No. 1: pp. 150 151.

Hensher, David A., Longitudinal Surveys in Transport: An Assessment in Ampt. E.S., Richardson, A.J., and Brög, W. (1985). New Survey Methods in Transport, VNU Science Press: Utrecht, The Netherlands, pp. 77 78.

Hornik, Jacob. Impact of Pre-Call Request Form and Gender Interaction on Response to a Mail Survey. Journal of Marketing Research, Vol. XIX (Feb. 1982): p. 144.

Jones, Peter and John Polak, Computer-based Personal Interviewing: State-of-the-Art and Future Prospects, Journal of the Market Research Society 1993, Volume 35, No. 3, p. 222.

Jones, Peter. Interactive Travel Survey Methods: The State-of-the-Art in Ampt, E.S., Richardson, A.J. and Brög, W. (1985). New Survey Methods in Transport, VNU Science Press: Utrecht, The Netherlands, pp. 99-127.

Kitamura, R., D. Reinke, C. Lula, E.J. Pas, and R. Pendayala, Data Needs for Development of Activity based Travel Demand Models: The Implementation of AMOS for the Metropolitan Washington Council of Governments, Presentation at the 5th National Conference on Transportation Planning Methods Applications, Seattle: June 1995.

Lawton, T. Keith and Eric I. Pas, Survey Methodologies, Resource Paper for Household Travel Surveys: New Concepts and Research Needs Conference, Irvine, CA (March 1995).

Nicholls, William L. Computer-Assisted Telephone Interviewing: A General Introduction in Groves, R.M., Biemer, P.P., Lyberg, L.I., Massey, J.T., Nicholls, W.L., and Waksberg, J. Telephone Survey Methodology, John Wiley & Sons: New York, p. 378.

NuStats, Inc. “Sample Productivity Plan,” Technical Memorandum, 1994.

Pearson, D.F., et al. A Procedure for Estimation of Trip Length Frequency Distributions. Texas Transportation Institute Report No. TTI 2 10 74 17 1, prepared for the Federal Highway Administration, April 1974

Peterson, Robert A., Gerald Albaum, and Roger A. Kerin, A note on alternative contact strategies in mail surveys. Journal of the Marketing Research Society Volume 31, No. 3.

Phone conversation with Robert Griffiths of MWCOG, September 22, 1994.

Richardson, A.J., Elizabeth Ampt, and Arnim Meyburg. Survey Methods for Transport Planning, Eucalyptus Press, Melbourne 1995, p 155.

Smith , M. E., “Design of Small-Sample Household-Interview Travel Surveys.” Transportation Research Board. Transportation Research Record 701, 1979, pp. 29 35.

Stopher, Peter R., Use of Activity-based Diary to Collect Household Travel Data, Transportation, 1992, Volume 19, pp. 159 176.

Tooley, Melissa. Incentives and Rate of Return for Travel Surveys, presented at 5th Conference on Transportation Planning Applications, Seattle, April 1995.