2.0 The Generic Travel Survey Process

2.1 The Survey Implementation Process

Most travel surveys (and other types of surveys, as well) follow a common implementation procedure. Backstrom and Hursh-Cesar (YEAR??) divide the generic survey implementation process into the 20 steps listed in Table 2.1.¹ These 20 steps can be classified into the five general stages shown in the figure (WHICH FIGURE?):

• Survey planning,

• Survey design,

• Field implementation,

• Data preparation, and

• Data analysis.

This manual concentrates on the three middle stages: survey design, field implementation, and data preparation. The survey planning and data analysis stages are related to decisions about the scopes and the forms of the travel models and other analyses that are to be developed using survey data. For this manual, it is assumed that the agency has either developed a detailed plan for the formation or revision of an existing modeling system or has identified a particular set of survey data needs.

Figure 2.1 shows the many functional relationships between the steps in the implementation of a generic travel survey. Decisions made at each point of the process affect many other elements of the survey effort. Once the need for a new travel survey has been identified, an agency needs to proceed with three tasks: compiling useful background information for the survey effort, designing the overall survey effort based on the recognized data needs and available data sources, and organizing the survey

Table 2.1

Figure 2.1

**WORDS MISSING HERE ... team and survey resources. Decisions made during the survey design process allow the survey team to develop a sampling plan and to develop the preliminary survey instruments.

The field implementation stage of the travel survey involves the training of fieldworkers, conducting a survey pretest, and the actual survey data collection. The survey pretest allows the survey team to rethink survey design issues, including the overall design of the survey, the survey sampling procedures, and the survey instruments. In addition, the pretest can identify areas where additional fieldworker training would be helpful.

The results of the data collection effort are then fed into the data preparation stage. The survey results are coded, entered, and cleaned. Then the survey results are manipulated into useful formats for data analysis.

Chapters 6.0 through 12.0 apply the generic organization of survey steps to discuss specific types of travel surveys. In the remainder of this chapter, we briefly elaborate on each of the survey implementation steps.

2.2 The Survey Design Stage

Assembling Background Information

Prior to embarking on any survey data collection effort, an agency should carefully review existing data sources. Available travel data can be used in a number of ways, including:

• In Lieu of New Survey Work – Travel surveys are often expensive and time-consuming. If appropriate, existing data sources can be used instead of the survey data. Of course, the quality and timeliness of existing data sources should be examined to determine whether new survey data collection may be avoided.

• For Developing Survey Samples – Existing information about the survey population can be used to develop more efficient samples so survey costs can be reduced and/or accuracy can be enhanced.

• For Validating Survey Results – One of the best ways to assess the validity of new survey results and detect potential biases in the data is to compare the survey results with other available information. This process can be accomplished with formal statistical analyses or with non-formal comparisons. Many statistical validation routines are commonly performed during the development of new travel models.

Background data sources for each type of travel survey are described in Chapters 6.0 through 12.0 of this manual, but generally, the most common sources of background information for travel surveys include:

• U.S. Census Population and Housing Summary Tape Files (STF1, STF2, STF3, STF4);

• U.S. Census Transportation Planning Package (CTPP);

• U.S. Census Public Use Microdata Sample (PUMS);

• National Household Travel Survey (NHTS) and its predecessor the Nationwide Personal Transportation Survey (NPTS);

• Previous local survey efforts; and

• Traffic counts and transit passenger counts.

Survey Design

In the survey design portion of the travel survey implementation process, the survey team needs to determine the best overall survey approach to obtain the particular data items required for the expected analyses. The survey designers need to accomplish two general tasks during the survey design effort:

• Selection of the general survey methods; and

• Establishment of budget and time constraints for the survey effort.

Ideally, the appropriate survey methods would be determined first, and then the budget and time constraints would be defined based on the selected approach. Usually, however, the survey method is selected with advance knowledge of the likely budget and schedule limitations. Thus, the challenge is to design the survey method or methods that will yield the most cost-effective data collection effort.

Regardless of whether the survey team is constrained by available resources, the selected survey methods should be based on anticipated data analysis needs. The decisions to be made in this regard include:

• Which survey methods are appropriate, and which are likely to be the most effective at obtaining the needed modeling data? As this manual will demonstrate, there are several different methods for completing each type of travel survey. The key design issue facing the survey team is which survey method is the most effective for obtaining the needed survey data. The strengths and weaknesses of different survey methods are discussed in Chapter 3.0. Later chapters discuss the choice of survey methods for specific types of surveys.

• Which individuals or establishments should be included in the sample population? The respondent population for travel survey efforts will depend on the particular survey data needs of an agency. For instance, if a survey is to be conducted as part of an effort to predict the effects of a potential transit fare increase, the best survey population is likely to be current transit riders. If the survey is being used to predict the effects of transit improvements, the best survey population would include both current transit riders and potential users.

• What sampling frame or frames are available for sampling the relevant survey population? Do these sampling frames imply the use of any particular survey method, or exclude the use of any? To conduct a survey, one needs to somehow enumerate (or list) the sample population so that a sample can be drawn. It is rare to find a perfect sampling frame, so the survey team needs to find or develop the best possible frame given available resources. Many lists are related to a particular survey method. For instance, some transit agencies maintain lists of riders and their telephone numbers that have been obtained from previous surveys. If such a list is to be used as a sampling frame, then the survey would have to employ a telephone method.

• Should special measures be employed to enhance the likely respondent cooperation rate? Travel surveyors have employed many mechanisms to increase survey response rates, including monetary incentives, letters of encouragement, and the personalization of survey materials for individual respondents. Since these mechanisms generally increase the cost of surveys, survey teams need to determine which of these to employ in a particular survey.

• What procedures are needed to ensure that groups that are difficult to survey are included in the survey effort? What needs to be done to assure that respondents’ language and literacy limitations do not significantly bias survey results? Travel survey populations often include a number of subpopulations that are particularly difficult to survey. Travel survey designers need to take steps to encourage members of these groups to participate to the maximum extent possible.

• What techniques are available (and which should be used) for obtaining the survey response data given a particular survey method? Travel surveyors have conducted travel surveys using a number of data retrieval techniques. Mail surveys (and other self-administered surveys) have been conducted using standard questionnaires, bubble forms (similar to the forms used to record answers on standardized tests, like the SAT), and scratch-off forms, in which respondents can complete without a pen or pencil. Personal interview and telephone survey results can be recorded using pencil-and-paper techniques or by using a computer-assisted approach (CAPI or CATI).

Organization

In the organizational stage of the survey implementation process, the survey team defines the logistical requirements of the survey effort, and determines how the available survey resources are to be allocated. The organization task includes determining the following:

• Staff needs (including numbers and required special skills);

• The need for consultant contract help;

• The level of expertise required of interviewers and other fieldworkers;

• Field supervision needs;

• The need for special facilities;

• Equipment needs;

• Coordination needs (local agencies, police, etc.); and

• Citizens’ participation and publicity needs.

Sampling

Almost all travel surveys rely upon sampling techniques in which a part of a total population is queried to make inferences about the population as a whole. Sampling a population, rather than conducting a full population census, has the following advantages:²

• Economy;

• Speed and timeliness;

• Feasibility; and

• Quality and accuracy (data collection for a census may be so difficult that the quality of data would be poorer than sample data).

The challenge facing the survey team is to select a sampling approach and sample sizes that will enable the development of reliable, accurate transportation demand models without overspending on an expensive data collection effort.

In theory, developing a statistically reliable survey sample involves the following steps:

• Identification of the survey population (or universe);

• Identification of sampling frame and selection of sampling procedures;

• Determination of necessary precision (sampling error) for one or more specific data items being collected;

• Calculation of sample size; and

• Estimation of necessary resources.

In practice, the survey design process will likely be constrained by the available resources from the beginning. While an agency might be able to reduce costs by reducing the amount of data collected for each unit or adopting procedures from elsewhere, the steps would still likely have to be modified to more closely resemble the following:

• Determination of available resources;

• Identification of survey population;

• Identification of sampling frame and selection of sampling procedures;

• Determination of maximum sample size based on procedures and resources;

• Determination of survey precision for one or more data items being collected; and

• Assessment of the adequacy of precision levels.

The Sample Design Workshop of the Second International Conference on New Survey Methods in Transport (AUTHOR & YEAR?) found that:

“Little formal effort is usually given to the analyses of sample size requirements, primarily because budgets were almost invariably set prior to the technical involvement of the transportation analyst. These budgets reflect the expectations of administrators as to the cost of particular items of information, rather than any explicit considerations of required accuracy.”³

The Workshop also identified the difficulty of determining necessary precision levels as a reason for first identifying the resource constraints, and then determining the resulting survey parameters:

“The idea that there might be a single optimal allocation of resources was seen as an oversimplification, in that the end use of the data is generally an input to a large number of very different, but interdependent analyses. No single, most important output could be defined, so no natural criterion for optimality could exist.”⁴

The basic principles of sampling for travel surveys are outlined in Chapter 5.0 of the manual. Later sections identify the most commonly used survey populations, sampling frames and sampling procedures for each type of travel survey. These sampling approaches are all variations and combinations of the most common sampling procedures:

• Simple Random Sampling – With this approach, sampling units are drawn randomly from the sample frame.

• Systematic Sampling – With this approach, rather than randomly selecting from the sampling frame, the analyst selects sampling units in sequences separated by a preset interval. Provided that the sampling frame order is relatively unbiased, this approach is essentially equivalent to the simple random sample. Often, in personal intercept surveys, fieldworkers are instructed to approach every `nth’ person passing a certain point. Similarly, random-digit-dialing (RDD) telephone surveys are often conducted by calling every `nth’ telephone number within a prespecified set of telephone exchanges.

• Stratified Sampling – If data are available to segment the survey population into subpopulations (or strata) prior to the sample selection, then this sampling approach may help to reduce sampling error or to reduce the amount of data collection needed. Household travel surveys commonly use stratified samples, based on measures such as household size and automobile availability.

• Cluster Sampling – With this approach, the sampling units are actually groups (clusters) of the survey elements rather than individual elements. All of the units within a selected cluster may be included in the sample, or a second stage subsample may be drawn from the chosen cluster. Workplace/establishment surveys are examples of cluster surveys. Some small number of establishments are first selected from the population of all establishments within a study area. Employees and visitors are then sampled within the selected establishments.

• Choice-based Sampling – When the data analyses require significant representation of a group or groups which are difficult to locate in the population at large, a sample may be drawn on the basis of the outcome of one of the choice processes under study. A common example of choice-based sampling is the collection of data from transit users on board transit vehicles for use in the development of mode choice models.

A number of textbooks deal with the details of survey sampling.^5,6,7  (THESE CITATIONS ARE FROM 1965 THROUGH 1979 & NEED UPDATING)

Drafting and Constructing

The earlier steps of the survey implementation process dictate the types of survey forms that will be needed and the respondent information that will be required from the survey. Of particular importance will be the survey administration methodology; specifically, whether responses to survey questions will be completed by respondents or by trained interviewers. In the questionnaire drafting and construction step, the survey team develops the data collection instruments. The following tasks are needed for developing the actual survey instruments:

• Identification of the required survey instruments and related forms;

• Selection of the types and forms of the questions that will best address the data needs;

• Formulation of wording for the survey questions;

• Determination of the best sequencing for the questions;

• Refinement of the questionnaire to ensure that all questions are effective and necessary; and

• Design of the layout of the survey instruments.

2.3 The Field Implementation Stage

Pretesting

All travel surveys should be tested extensively before they are actually undertaken. Nearly all survey researchers stress the necessity and importance of pretesting questionnaires, but this is the stage of the survey implementation process which is most likely to be squeezed out due to time and cost pressures.⁸ If the pretest is conducted appropriately, the surveyor will be able to improve the survey effort on a number of different fronts, including:

• Refining fieldworker and interviewer procedures and logistics;

• Testing and revising question wording, sequencing, and formatting;

• Comparing alternative approaches to gathering certain data items;

• Identifying unexpected responses and respondent behavior;

• Estimating the survey completion time; and

• Developing preliminary estimates of the variance in key variables to help establish final sample sizes.

If possible, travel survey pretests should be conducted in three steps: the office pretest, the questionnaire pretest, and the survey dry-run.

Office Pretest

The office pretest, whether conducted formally or informally, is likely to be the best mechanism for identifying problems with the questionnaire and with specific questions. Many researchers feel that the best way to discover potential survey problems before they occur is to have colleagues or other experts not involved directly in the survey design review the questionnaire and proposed procedures.⁹ Two recent studies of pretest error detection rates support this thesis. These empirical studies found that pretests that rely solely on samples drawn from the ultimate target population have fairly low error detection rates.^10,11

Questionnaire Pretest

Although it is an important element of testing the questionnaire, the office pretest will be insufficient in most cases. Travel survey questionnaires should be tested on non-experts because they are often confusing to people without knowledge of transportation planning, and because the surveys often rely upon respondents’ understanding of technical (and sometimes ambiguous) terms and expressions. For instance, a common challenge in travel surveys of all types is to get respondents to use a consistent definition for the term, “trip.” The questionnaire pretest is the stage of the survey implementation process that ensures that respondents are answering in a consistent manner.

During the questionnaire pretest, respondents are administered the survey, and are asked to describe any problems or areas of confusion that they encounter. Often, these pretests are personally administered even when the ultimate survey will not be. It is becoming increasingly popular to conduct this portion of the pretest as part of a formal or informal focus group. This allows the analyst to observe first-hand how respondents react to the survey.

There are two procedures used to determine the respondents’ reactions to the questionnaire. In the “protocol” method, the respondent is asked to think out loud as the questionnaire is being completed. In the “debriefing” method, the respondent completes the questionnaire and then talks about the questionnaire afterwards.¹² The protocol method is generally the better approach for identifying problems with specific questions; the debriefing approach is the better approach for identifying question sequencing and respondent tiring. Some surveyors have split the pretest so that some respondents use the protocol approach while others are debriefed.

The Survey Dry-Run

The final step of the pretesting task is to complete the survey on a small number of respondents in an identical manner to the full survey effort. Ideally, the pretest would encompass the whole range of survey tasks from sample selection to data analysis. This will ensure that all aspects of the survey effort are in place, and are operating as expected prior to the beginning of any data collection.

Sometimes, when the survey dry-run goes well, and the pretest timing and sample design are consistent with the larger effort, the data can be used in the general survey effort. However, surveyors need to allocate their time and money resources so that they are prepared to make changes in the questionnaire and procedures based on the outcome of the survey dry-run. If problems are found, the questionnaire(s) and/or survey procedures should be modified and re-tested if necessary. Ideally, this means that pretests would be completed well in advance (often a month or more) of the actual survey.

Training and Briefing

The quality of the fieldworkers, and interviewers in particular, will have a major effect on the success of the survey effort. Assembling a group of well-trained and consistent fieldworkers is not an easy task because of the nature of the work. Survey fieldwork is low-paying, generally part-time, intermittent work requiring the individual to have schedule flexibility, often during the evenings. The job requires excellent communication skills and reasonably good reading and writing skills. In addition, some travel surveys require fieldworkers to have a high level of mobility.

In many cases, direct supervision of fieldworkers is minimal, so they must be carefully trained and briefed on the travel survey effort. Training involves teaching or re-teaching the basic fieldwork skills necessary for a survey of the type being conducted. Training should include the following topics:^13,14

1. Use of survey quota sheets, maps, building layouts, etc. (for field surveys).

2. Procedures for observing or counting individuals and for recording the information (field surveys).

3. Procedures for contacting potential respondents and presenting the study to them (or reminding them of the study).

4. Procedures for screening potential respondents based on survey quotas or other criteria to determine whether to ask a particular person to participate.

5. The conventions used in the design of the questionnaire with respect to wording and skip instructions so that interviewers can ask the questions in a consistent and standardized way.

6. Procedures for using computer-assisted questionnaires (CAPI and CATI surveys).

7. Procedures for probing inadequate answers in a non-directive way.

8. Procedures for recording answers to open-ended and closed questions.

9. Rules and guidelines for handling the interpersonal aspects of the interview in a non-biasing way.

The fieldworker briefing provides the fieldworkers with specific information regarding the particular study. Generally, it is extremely helpful for the sponsoring agency to be directly involved in the briefing of fieldworkers, because agency personnel will best be able to describe the importance of the survey effort to their future efforts. Briefing issues include the following:^15,16

1. Specific purposes of the project, including the sponsorship, the general analysis goals, and anticipated uses of the research. Fieldworkers need this information because they will need to provide respondents and others with appropriate answers to questions and because this information will help fieldworkers enlist cooperation.

2. Description of everything in the fieldworker kit (for field surveys).

3. Description of how the forms are to be completed (pencil-and-paper approach) or how responses are to be recorded on the computer (CAPI and CATI).

4. The specific approach that was used for sampling, again to provide a basis for answering respondent questions. In addition, there may be some training required in how to implement the basic sample design.

5. Details regarding the purposes of specific questions.

6. The specific steps that will be taken with respect to confidentiality, and the kinds of assurances that are appropriate to give to respondents.

7. Detailed description of procedures to follow if problems are encountered.

8. Procedures for contacting field supervisors (field surveys) or central location supervisors (telephone surveys).

In addition to providing fieldworkers with an understanding of the procedures to be employed in the study, the training and briefing sessions should motivate fieldworkers to believe that the survey work and their efforts are important, and that the highest quality data are needed. The participation of personnel from the sponsoring agency in the training/
briefing sessions may help fieldworkers understand the importance of their work.

Under ideal circumstances, the training and briefing sessions should impart the following attitudes to the fieldworkers:¹⁷

• This Job is Important – Stress the importance of this particular study: how it is intended to contribute to the public good, solve problems, and improve the community.

• I Must Follow Instructions – Teach the importance of following instructions, the necessity of proper field procedures, and the importance of consistency.

• Biases can Cripple Data – Teach fieldworkers about the destructive role of the biases that they can bring into the research effort.

• Research is Important – Communicate the value of research: how re-search information improves our ability to make decisions, to solve problems, to contribute to the common goals of society, and to save money and resources.

• Surveys Work – Stress that surveys can be valid, reliable measures of people’s information, attitudes, preferences, and behavior.

• People Like to Participate in Surveys – Help fieldworkers understand that they are not snoops or irritants: many people like to express their opinions, they know about polls, they are usually flattered to be chosen, and they are curious about how it all works.

• I Am a Professional – Each fieldworker should believe: “I have a job to do; I am a professional being paid for services rendered.”

• Randomness Works – Each fieldworker should believe: “no matter what I think about who should be part of the sample, we are likely to get a better (more fairly representative) sample by relying on chance and the survey selection procedures, rather than personal decisions about whom to interview.”

• The Respondent is Entitled to Courtesy – Each fieldworker should understand that “I must respect the people whose time I am using, and I must treat all respondents with equal courtesy.”

• The Respondent is Entitled to Privacy – Warn that respondents usually are more comfortable expressing themselves privately on some issues, so fieldworkers must help to ensure that privacy.

Interviewing and Questionnaire Distribution

The data collection fieldwork tasks of the survey implementation process are where the considerable planning efforts of the survey designer are actually put to the test. Unfortunately, these data collection tasks are usually those over which the designer has the least control. They are also the tasks where the greatest amount of uncorrectable bias can enter the process. Major interviewer and fieldworker errors are very costly because they may require redoing part or all of the fieldwork, but small errors by interviewers and fieldworkers may be as bad. These smaller errors are often undetectable, and may greatly increase the level of bias in the survey results, unbeknownst to those analyzing the data.

To ensure that fieldworkers are performing the necessary survey functions in the non-biasing and consistent manner for which they have been trained, adequate supervision is essential. For some types of surveys, such as telephone surveys, supervision techniques have been developed that help to identify problems and help interviewers to correct them immediately. For instance, central site telephone survey supervisors can maintain up-to-the-moment statistics on interviewer completion rates, average completion times, and item non-response levels. Telephone survey supervisors can usually monitor individual interviews if problems with particular interviewers are detected. However, the cost and logistics of supervision for some types of surveys, such as in-home personal interviews, can be prohibitive.

Some key issues for fieldwork supervisors in evaluating the fieldwork as it occurs are the following:

• Are the survey response rates and cooperation rates different than expected prior to the survey fieldwork?

• Are the costs per completed interview different than expected prior to the survey?

• Is the quality of the completed questionnaires and interviews – response rates, validity of responses, legibility – as expected?

• Are survey fieldwork procedures working adequately? Are staff being utilized efficiently?

• Are fieldworkers completing their tasks consistently?

• Are survey response rates and cooperation rates significantly lower or higher than the average for a specific fieldworker (a low response rate will affect survey cost while a high response rate might suggest improper survey techniques)?

If problems are detected during the fieldwork process, the survey team should be ready to modify the procedures or retrain (or replace) fieldworkers, as necessary. Such modifications are generally very challenging since the available information on which to make decisions is limited, and the underlying reasons for the detected problem(s) are usually not obvious.

2.4 The Data Preparation Stage

Coding and Data Entry

During the coding step of the survey process, the raw survey data are translated into codes usable for model development and presentation of results. The objective of the survey team in this step is to “unambiguously assign each survey answer to one and only one analytically meaningful code.”¹⁸

Most travel surveys rely on a three-step process for converting responses to usable data:

1. The respondent or fieldworker records a response;

2. The coder translates the response into a pre-specified code; and

3. The data entry specialist keys the response into a database.

Many survey instruments can be designed to be “self-coding” without damaging the clarity of the questions. These surveys greatly reduce or eliminate the work of the coder. In addition, the interactive processing available with computer assisted telephone interview (CATI) and computer assisted personal interview (CAPI) surveys reduce these steps to a single automated step. Computer assisted survey techniques also allow interviewers to check for the reasonableness of responses vis-à-vis other responses in the interview, and to correct problems prior to losing contact with the respondent.

Of particular interest are recent improvements in technologies to perform geocoding, the translation of locational survey data into a usable format. Chapter 14.0 discusses geocoding issues in some detail.

Editing and Cleaning

Once the survey data have been entered, the survey team should systematically analyze the results to identify data problems. Three editing and cleaning tasks can be conducted:

1. Simple data cleaning to correct coding and data entry problems;

2. Validation of survey responses; and

3. Application of analytical techniques to reduce non-response.

The first step is to verify the completeness of each record. Next, each data field should be checked to make sure only legal codes are entered. Then, the analyst should evaluate the internal consistency of the responses to related questions.

Manual checking of the survey data by surveyors or editors is essential, but writing specialized programs to perform automated checking of data files is usually also worthwhile. Such programs can consistently perform intra- and inter-record data checks that would be impossible to perform manually. For example, a survey record might include the beginning and ending time of an activity or trip with associated “am” or “pm” codes. If one of the codes is mis-keyed, the resulting data could imply that the activity or trip ended before it began, even though all data passed specified range checks.

Likewise, a household included in a home interview survey might have five members. If data for one of the household members was skipped in data entry, the error would never be caught via simple range checks of specific data items. Survey data that are “clean” in terms of range checks for specific data items often contain illogical data. Such errors can affect resulting data summaries and travel models.

If problems are identified with a particular response, the analyst should refer to the original sources, the completed interview sheets or the returned questionnaires. The error(s) in the database should be corrected if possible, but if the error involves more than coding and entry errors, the record will need to be marked unusable and dropped from the analysis database. As one would expect, tracking down and correcting errors in large survey databases is a long, slow, and inefficient process. It is generally much more cost-effective to spend extra time on the coding and data entry tasks to avoid large editing tasks.

CATI and CAPI surveys offer the opportunity to perform data cleaning and consistency checks on-line, while respondents are still accessible, but one of the drawbacks of computer-assisted interview surveys is that there is no source which to refer to when incorrect or inconsistent data are found in the database after the survey. Therefore, it is essential that editing, cleaning, and consistency checks be built into the programs. If operators enter invalid information during an interview, the computer should prompt them to try again.

The second editing/cleaning task that is sometimes performed is to validate a small sample of the survey responses by recontacting respondents and reviewing their responses. This process can be conducted on a random sample basis to ensure that each fieldworker completed the work he or she was supposed to complete, and to ensure that responses were completed consistently. The validation process can also be used selectively if there is some question about the work of certain fieldworkers. Finally, validation can be used for responses with identified problems to limit the number of non-usable responses.

The final editing task available to analysts is to apply statistical procedures to impute the values of missing or incorrect data elements. In almost every survey, some respondents will be unwilling or unable to answer all the questions posed to them. In addition, many respondents will knowingly or unknowingly respond to questions inaccurately. Some question types are more susceptible to item non-response and inaccuracy than others, with questions about income generally being the most problematic. Some analysts have statistically related the variable in question to other survey variables or other data sources, such as Census data, to be able to use the response more effectively in subsequent analyses.

This approach is still much debated, however, since bias could be increased in some cases, and because many analysts believe that imputing values implies that the answer to the question is already known. The survey team has other options, including:

• Ignoring the non-response (if the sample size is sufficient without them); and

• Using a modeling variable to describe the non-response.

These and other mechanisms for reducing non-response are discussed in later chapters.

Programming and Compiling

The final steps of the survey process covered in this manual relate to preparing the survey data for modeling and other analyses. Fortunately, the task of compiling survey datasets has been greatly simplified since the days of the previous Travel Survey Manual, when piles of perforated computer cards awaited the analyst. Today, survey data are generally entered in ASCII data files which are easily read into any of the available statistical analysis software packages. Thus, the compilation task is quite straightforward.

However, there are three key programming and compiling tasks that are of special interest to the modeling analyst:

• The determination of data storage needs;

• The development of one or more survey response weighting schemes; and

• The tabulation of survey results.

Large survey efforts require a great deal of data storage. Often, the survey database from a travel survey will use in excess of 60 megabytes, excluding any constructed variables or analysis results (which could easily double the size of the file). The survey team should make rough calculations of the data storage requirements once the survey coding requirements are known, and should plan to invest in expanded storage if necessary.

The determination of survey weights is a key element of the survey analysis. Depending on the modeling analysis needs, results obtained from surveys that are designed to (or accidentally) oversample or undersample some groups of the population usually need to be weighted so that members of the subgroups are proportionally represented in the population as a whole. Some analyses will require the use of the weighted data while others will not. Often, different weighting schemes are used depending on the analysis that is being applied.

Tables 2.2 and 2.3 illustrate two weighting exercises. Table 2.2 shows the calculation of work trip mode weights for a recent household travel survey. The first columns of the table shows the modal distribution of work trips obtained from a survey of 8,346 work trips. The following columns show the actual modal distribution for the study area based on best available information, including the Census Journey-to-Work data. To make conclusions about the work trip mode split from the survey data, it is

Table 2.2

Table 2.3

necessary to weight each survey response by the ratio of the actual share divided by the survey share. For instance, the survey work trips that were made by walking need to be weighted by (191,614/6,931,237)/(33/8,346) = 6.99.

Table 2.3 illustrates another common weighting exercise. As discussed below, transit onboard surveys generally sample transit trips. To make conclusions about transit riders, one needs to weight the survey results by individuals’ frequency of transit usage. The example shows survey results for a question about the respondents’ choices of fare types. Monthly pass trips accounted for 50 percent of the surveyed trips, but because monthly pass users use transit more frequently (based on another survey question) they represent only 39 percent of transit users. Users of the data would want to use the weighted percentages to make conclusions about riders, and the unweighted percentages to make conclusions about transit trips.

The final programming task is the tabulation and cross-tabulation of the raw and weighted survey results. Because the survey team will usually soon be immersed in detailed and time-consuming model estimation efforts, a useful final step for the survey work is to produce a complete set of cross-tabulations of the survey results. The tabulations will be a useful reference source for the analyst while he or she develops the travel models, and they will provide interested, less technical parties with a great deal of information on the survey population under study. For most agencies, it is advantageous to produce travel survey results reports as quickly as possible. These reports will provide funding agencies and other interested groups and individuals with evidence that the travel survey effort was more than a purely academic effort used to develop obscure modeling parameters.

The preparation of clear, concise data documentation is an important task. Too often, final datasets are documented by a list of variables and a copy of the survey instruments. A data dictionary that clearly delineates the file format(s) and the data contained in each field should be prepared, including allowable codes and meanings. This is especially true if derived variables (such as zone numbers) or independently estimated information (such as weighting factors) have been added to the dataset.

Finally, care must be taken to ensure confidentiality of the data. Many surveys collect personal information – number of household members, ages of household members, household income, typical daily travel patterns, and household addresses. Such a database could be a valuable tool to a “high-tech” thief. Care should be taken to ensure that data files containing personal information are never distributed with detailed address or locational data (e.g., latitude and longitude).