Consumer Research in Urban Transportation: Some Methodological Issues

ABSTRACT - The nature of the transportation market has important methodological implications for consumer researchers. This paper points up opportunities for improving sample design, considers alternative methods of data collection, and discusses selection of measurement techniques. In the context of developing importance weights for multi-attribute models of modal choice, a review is made of priority evaluation and other methods.

Citation:

Christopher H. Lovelock (1976) ,"Consumer Research in Urban Transportation: Some Methodological Issues", in NA - Advances in Consumer Research Volume 03, eds. Beverlee B. Anderson, Cincinnati, OH : Association for Consumer Research, Pages: 407-415.

Advances in Consumer Research Volume 3, 1976      Pages 407-415

CONSUMER RESEARCH IN URBAN TRANSPORTATION: SOME METHODOLOGICAL ISSUES

Christopher H. Lovelock, Harvard University

ABSTRACT -

The nature of the transportation market has important methodological implications for consumer researchers. This paper points up opportunities for improving sample design, considers alternative methods of data collection, and discusses selection of measurement techniques. In the context of developing importance weights for multi-attribute models of modal choice, a review is made of priority evaluation and other methods.

PREVIOUS RESEARCH TRADITIONS

Historically, transportation research has been dominated by engineers and economists. Engineering was the discipline which designed and built the physical facilities; it used analogies from the physical sciences to describe and evaluate traffic flows; and it employed often simplistic mathematical models to project the future demand for travel by alternative modes.

Economists were intrigued by the nature of the transportation industry, with its "perishable" product that could not be stored, its widely fluctuating demand levels over time, and its high fixed to variable cost ratios, as well as by the fact that the marginal costs of carrying an extra passenger under conditions of excess capacity tended to approach zero. These characteristics posed--and still pose--fascinating pricing problems for both passenger transit and public highway facilities.

Economic utility theories, meantime, provided the basis for development of econometric models which attempted to explain and predict consumers' choice of travel modes in terms of a very limited number of variables, notably price and travel time.

A NEW MARKETING ORIENTATION

In recent years, however, transportation studies have increasingly employed concepts and techniques drawn from the behavioral sciences.

In part, this represents recognition by transportation specialists of the insights to be gained from studying consumer behavior. It also reflects a growing interest by consumer researchers in applying their skills to new areas, outside such traditional fields as private sector marketing and political polling.

Finally, it represents a significant change in the whole transportation environment. Planners are no longer thinking in passive terms of projecting future travel demand patterns and then developing new facilities to meet forecast demand. Instead they are considering how demand patterns may he influenced and different modal choice behavior encouraged. This calls for an approach which considers the problem of moving people, with all their individual idiosyncrasies, rather than one of simply facilitating the movement of mindless vehicles. At the same time, a new breed of transit manager is emerging, who no longer speaks of riders as did one old timer, who was wont to refer to them as "seated freight."

With government encouragement, the transit industry is adopting a new marketing orientation. Transit is now seen as a consumer product which must be sold against fierce competition. Efforts are therefore being made to tailor transit operations more closely to consumer needs and preferences, while monitoring and evaluation programs are being established to measure the effectiveness of particular managerial strategies in the transportation marketplace. [See, for example, U.S.! Department of Transportation (1975).]

Consumer Research Issues and Applications

As a result, research techniques developed in consumer marketing and political polling are finding increasing application in the urban transportation field. They are being used to determine not merely the size of the travel market but also ways in which it is segmented: to evaluate public attitudes towards both proposed and existing facilities; and to develop a better understanding of the factors underlying modal choice behavior and how consumers' decisions may be influenced.

It is encouraging to see marketing research tools and concepts being applied in the transportation field. During the past few years, a growing number of research studies have been conducted, often yielding significant findings. However, I believe that there is still plenty of room for further refinement of research procedures. Indeed, without such refinement we are unlikely to advance our understanding of modal choice behavior much beyond its present level.

In the balance of this paper, I would like to look briefly at some of the methodological issues facing urban transportation researchers. My objectives are, first, to point up some opportunities for improving sample design; second, to review the problem of selecting appropriate measurement techniques; and, third, to consider alternative methods of communicating with subjects.

These three issues are, of course, interrelated. The nature of the sample design may facilitate or impede certain forms of communication with members of the sample population; the characteristics of people in the sample frame may constrain the type of research instrument used and the measures employed; while the measures used may serve to determine--or be constrained by--the method chosen for communicating with respondents.

SAMPLE DESIGN

My first concern centers around, but is not confined to, issues relating to sample design. I sometimes wonder if many researchers who approach transportation from a consumer goods orientation truly understand the nature of the transportation "product."

Blumer (1969) has argued that methodology embraces the entire scientific quest and not merely some selected portion or aspect of that quest, such as data analysis. He emphasizes that "the entire act of scientific inquiry is oriented and shaped by the underlying picture of the empirical world that is used." My concern is that many researchers do not have as clear and realistic a picture of the transportation world as is needed to conduct incisive research.

In evaluating modal choice decisions, there has been a tendency to conduct research and build models which see urban travel choices as lying simply between automobiles and public transportation. In practice, of course, there may also be other options such as walking or bicycling. Perhaps more serious is the practice of lumping all modes of public transportation together under one heading when comparing them with car travel (e.g. Paine et al., 1967; McMillan and Assael, 1968, 1969).

Many cities offer multiple-mode transit systems and it is most unrealistic to assume that consumers perceive different public modes such as buses, trains, rapid transit, streetcars and even ferries as broadly similar. My own research indicates that people see bus and train travel as differing sharply from one another on several important attributes (Lovelock, 1973). There is a risk that respondents may assign to the block term "public transportation" the characteristics of that mode which performs worst (or best) on any given attribute, particularly if questions are phrased in terms of asking them "how satisfied" they are with transit's performance on specific attributes.

Recently, I have come to the conclusion that it may also be dangerous to generalize within a given mode. To borrow a refrain from recent aspirin commercials, "all bus journeys are not alike." Many transportation studies distinguish between the nature of travel for different purposes, but they do not consider possible differences in the nature of the travel experience between, say, different routes.

Two Scenarios

Consider the following two scenarios. Mr. X and Ms. Y, who are quite closely matched demographically, both live in the same neighborhood, approximately the same distance from their respective bus stops. Ms. Y has a seat and enclosed shelter at her stop, whereas Mr. X has neither and is exposed to the elements.

Mr. X takes the #19 bus to his office. It is a notoriously unpunctual service. In its wisdom, the transit company habitually puts on an ageing vehicle with cracked windows, broken suspension and torn seats, driven by a bad-tempered operator who takes it out on passengers and vehicle alike. The bus is always packed and Mr. X has to stand the whole way. The route follows city streets up and down several steep hills, with many stops for congestion, traffic lights and embarking or disembarking passengers. The five mile journey takes Mr. X anywhere from 15-30 noisy, uncomfortable minutes, depending on the state of the traffic, the weather and the mood of the driver. Because of crowding and inadequate ventilation, the bus often smells like a zoo by the time Mr. X reaches his destination, usually feeling exhausted and either too hot or too cold.

Ms. Y also travels five miles to her job which, luckily for her, is in a suburban complex in the opposite direction. Buses on the #23 route run on time, are almost brand new and their drivers generally friendly and courteous. Running counter to the main commute direction, the bus is half empty and the route follows a smooth, limited-access highway with little traffic. The seats are comfortable, the heating and air-conditioning work, and it's a pleasant, 10-minute ride. By a quirk of the fare zone boundaries, Ms. Y's fare is 25 cents versus 40 cents for Mr. X. In a recent survey, Mr. X gave bus travel a rotten rating on almost every attribute and indicated that he and his wife were thinking of buying a second car so he could drive to work. Ms. Y, by contrast, gave bus travel high marks all around and indicated that she planned to continue using it and was promoting it to all her friends.

Sample Design and Service Characteristics

Obviously, I have painted an extreme case in this fictional tale. But it does illustrate the possibility for wide variations in respondent attitudes towards what are effectively two very different products, although usually treated by researchers as one and the same thing. The nature of the "augmented product" varies widely in urban transportation and, like many services, is difficult to standardize. I believe that few other fields can offer the incautious researcher such opportunities for comparing apples with oranges, bananas with cucumbers, and coconuts with pumpkins.

What I am arguing for, therefore, is a much more incisive approach to sample design in urban transportation surveys. Researchers might usefully seek to stratify sample populations according to the observed conditions under which they generally travel. This calls for careful identification of subjects' most frequently made journeys, with special attention to the characteristics of routes which they use (or potentially might use) on a regular basis. Note that this applies to auto users as well as transit users. A car driver who regularly sees Mr. X's elderly #19 bus grinding up the hill is likely to form a poorer impression of bus travel than does the driver who sees the sleek new #23 speeding along the expressway.

Offhand, I do not know of any survey research which has attempted to be this specific. However, my research in the San Francisco Bay Area showed that respondents in different counties had significantly different perceptions of bus travel in their local areas, whereas there were no significant differences in their perceptions of car travel. These findings were consistent with my own more objective evaluations of the relative quality of service provided by the three different transit operations under review (Lovelock 1973).

Sample Design and Geographic Location

Surveys which seek to measure, say, public attitudes towards new tax proposals for financing transit should realistically sample the entire population of prospective taxpayers. However, surveys which seek to determine consumer satisfaction with specific attributes of public transportation, with a view to better understanding modal choice behavior, need to focus on geographically specific sample populations.

In a 1968 study, McMillan and Assael asked respondents, a high proportion of whom lived many miles from the nearest transit route, how satisfied they were with specific attributes of public transportation. This is hardly guaranteed to produce useful findings. It is an established fact that accessibility to a transit route of a traveler's origin and destination points is a significant factor in consumers' modal choice behavior (Lovelock, 1973, 1975). If we want realistic and meaningful evaluations of public transit modes against the private car, then a strong case can be made for confining our research to sample populations whose location provides a reasonable transit alternative.

Ideally, this calls for taking a map of the area where the survey is to be conducted, marking in transit routes, and then drawing contour lines around each stop or station to indicate accessibility in terms of travel time to the stopping point. Studies suggest that most commuter rail patrons are located within a maximum of ten minutes' drive from the station, while the effective catchment area (or "transit envelope") for local bus service lies within 1000 feet or a two block radius of the bus stop (Bonsall, 1971; Lovelock, 1973; Metro Toronto Area Regional Transportation Study, 1969).

Excluding Unwanted Subjects

Depending on the objectives of the survey, it may also be necessary to narrow the sample frame on other than locational grounds.

Supposing we are studying the relationship between attitudes, beliefs and modal choice behavior for commute travel. Obviously, we want to exclude people who do not commute at all. This means removing retired people and those without regular jobs from the sample. We also need to exclude those whose jobs or physical condition clearly restrict them to a single mode, as well as those who do not have access to a car.

If we are using personal or telephone interviews, some preliminary questions may result in early termination of the interview on the grounds that a particular respondent falls outside the sample frame. Mail questionnaires can either be discarded after completion on the basis of responses to key questions, or subjects may be advised in the cover letter that they should only complete and return the survey if they meet certain criteria. Since the latter method may sensitize respondents or be misinterpreted by desired subjects, the former approach is recommended.

SELECTING MEASUREMENT TECHNIQUES

The next issue I want to examine is that of selecting measurement techniques, particularly as these apply to measuring consumer attitudes.

During the past decade, attitudinal surveys have proliferated in transportation research. Stearns (1975) notes that "Attitude surveys have been widely used, frequently on an 'ad hoc' basis, for a variety of transportation planning and evaluation purposes. Their casual use has been due to a widespread need to measure motivational and evaluative factors and to estimate potential responses to alternative transportation services. These applications have been made despite a lack of a unified methodology integrating available and suitable techniques from several disciplines."

Unlike some areas of research in fields such as psychology or sociology, there are no well-tested "off-the-shelf" scales which researchers can use for measuring consumer attitudes in urban transportation. Development of common scales which could be used in various studies, conducted at different times and in different locations, might improve the generalizability of transportation research findings. While transportation studies have yet to reach this point, some researchers are beginning to replicate aspects of previous studies, which is a step in the right direction.

At a recent workshop, comprising participants drawn from transportation planning, transit system operation, market research and the social sciences (Stearns, 1975), three roles were outlined for attitudinal survey techniques in transportation planning and evaluation:

(1) A marketing role which would measure transportation system and service preferences, market knowledge, and experience with the service;

(2) An evaluative role to measure qualitative responses to transportation system alterations; and

(3) A planning role to measure local interests as an input to planning transportation systems.

Attitude Components

[This discussion is drawn primarily from Day (1973).]

When discussing attitudes, it is important to distinguish between their different components. There is wide acceptance of the concept that attitude structures have three component parts.

The cognitive or perceptual component represents a person's information about an object. Pieces of information can be broadly classified as either beliefs in the existence of an object or evaluative beliefs about an object. Techniques for measuring the former center on awareness measures, such as aided and unaided recall. Evaluative beliefs, meantime, provide information on the comparative judgments consumers make between alternative brands or products, notably as these relate to an individual's perceptions of specific attributes.

The affective or feeling component deals with a person's overall feelings of like or dislike for a situation, object, person or concept. By measuring consumer preferences, researchers may hope to obtain an understanding of the respondent's ideal product.

The conative component is concerned with an individual's intentions. Most research on attitudes has emphasized either their explanatory or predictive value, although the utility of the construct depends on achieving both. What has been termed the explanatory or structural approach deals with the cognitive or affective components, whereas predictive studies focus on the relationship between affect, intentions and overt behavior.

Multi-Attribute Attitude Models

Only recently have researchers come to appreciate the complexity of the transportation "product." There is now a recognition that consumers may choose between alternative modes on the basis of a variety of different attributes.

At last year's ACR meeting, I proposed a microanalytic model of the modal choice decision process (Lovelock, 1974). This saw consumers as trying to satisfy a particular travel need by first specifying the characteristics of the trip itself, then specifying the "ideal" modal attributes required for this trip, next evaluating the perceived characteristics of alternative modes against this "ideal" solution, and finally selecting that mode which provided the best perceived match.

Wilkie and Pessemier (1973) define a multi-attribute product as "a bundle of attributes leading to costs and benefits of differential desirability to individuals or segments of the market." Some of these attributes may be more important to consumers than others and in evaluating alternative products, consumers are often forced to make trade-offs between different attributes. It is essential that transportation planners and managers should not only be able to identify the attributes desired by consumers in modal choice decisions, but should also understand the degree of importance attached to each of these various attributes and whether or not different consumer segments set different priorities.

Identifying Salient Attributes

What attributes do travelers consider important in selecting a mode of transportation? A number of studies have researched this issue and have come up with broad]y similar conclusions.

Solomon, Solomon and Silien (1968) reviewed five different research studies conducted in various locations in the United States between 1962 and 1967.- Depending on the study, subjects were asked either to rank specified attributes in order of importance, to list attributes they considered important, or to rate specified attributes on Likert scales. Solomon et al. noted that, despite significant differences in methodology and often widely varying characteristics in the sample populations, the ranking of modal choice criteria was unusually consistent across all five surveys. They summarized these criteria, in order of importance, as safety, reliability, time savings, cost, convenience and comfort.

Among the more sophisticated of these five studies was that of Paine et al. (1967) who surveyed 1,021 respondents in Baltimore and Philadelphia. Based upon a thorough review of previous research findings they developed an attitude instrument which would measure consumer attitudes along 33 variables believed to relate to modal choice decisions. Respondents were first asked to rate the importance of each item on a 7-point scale. Items consisted of descriptive phrases and were scaled from "Not at All Important" to "Of Greatest Importance.'' Factor analysis was subsequently employed to reduce these 33 items to a more parsimonious set of variables. From this analysis the researchers were able to suggest the basic attributes of a generalized ideal system.

McMillan and Assael (1968, 1969), conducted a national survey of 2,500 respondents replicating several aspects of the earlier Paine study. They used a very similar format, with 7-point scales and self-administered questionnaires, but only 15 descriptive phrases instead of 33.

A somewhat different approach to measuring importance of transportation attributes was followed by Golob et al. (1972). As part of their research, they conducted a home interview survey in the suburban city of Warren, Michigan. Some 800 respondents completed a questionnaire using the method of paired comparisons to evaluate 32 transportation system characteristics.

The results of this study, which was undertaken after Solomon et al.'s review, yielded very similar findings to previous studies, with the results emphasizing the attributes of punctuality/reliability, comfort, convenience, travel time and cost (in approximately that order). It may be noted that Golob et al. excluded attributes relating to safety from their study, on the grounds that safety of the system was not realistically subject to trade-off in transit system design. This raises the issue for researchers of deciding which characteristics considered "important" by respondents are actually salient attributes that should be included in a multi-attribute model of modal choice.

Alpert (1971) emphasizes that not all attributes rated as important by consumers are necessarily determinant, in terms of differentiating one product (or mode) from another. For instance, safety may well be taken for granted by many travelers when selecting among alternative modes of urban transportation. Alpert suggests three broad approaches for identifying determinant attributes, namely direct questioning, indirect questioning (including motivation research and covariate analysis), and observation and experimentation. His own research (on attitudes towards moderately priced pens) indicated the most efficient approach to be direct dual questioning, where respondents are asked to rate product attributes in terms of (1) how important each is thought to be in determining choice, and (2) how much difference is perceived among competing products in terms of each attribute.

Perhaps we need to review this issue of salience more carefully as it relates to safety, since consumers may be concerned not merely with issues relating to safety of the vehicle on the track or highway but also with personal safety from the standpoint of being mugged at the bus stop.

We also need to consider the extent to which the relative importance of different attributes may vary according to the nature of the trip being made and the personal characteristics of the traveler. Paine et al. and McMillan and Assael found some differences between requirements for work and non-work trips. Golob et al. (1972) looked at differences between demographic groups and found some variations in the rank order of attributes for single people under 20, for the elderly and for low income groups, but not for any of the other segments they broke out. A later study by Golob, Dobson and Sheth (1974) showed limited differences according to respondents' sex and age; low income respondents (under $6,000) evidenced different concerns from other income groups, and there were also variations between central city residents and suburbanites. However, to my knowledge, no in-depth study has yet attempted to examine variations in importance ratings with reference jointly to both trip purpose and traveler characteristics. I believe this constitutes an important research opportunity.

One issue which should be raised here concerns the most appropriate technique to employ when measuring importance. While experience shows a variety of different methodologies to have generated similar rankings, a purely ordinal scale does not yield the same insights as interval scale data.

The advantage of the method of pair comparisons employed by Golob et al. (1972) is that it forces respondents to compare each item separately against all other items, whereas Likert or semantic differential scales do not force such choices and allow for ties in ratings. However, the problem with this first method is that the number of pair comparisons increases at an exponential rate as the number of stimuli (n) increases, the formula being n(n-1)/2. Although relatively straightforward, the process can quickly become time consuming and boring, leading to respondent fatigue.

When measuring consumer preferences for 32 different modal characteristics, Golob et al. were able to reduce the number of paired comparisons required from a single matrix of 496 to nine smaller matrices, each related to a specific group of characteristics and totaling only 168 choices. To provide a common basis for measuring the relative importance of all characteristics, several of them were included in more than one group. Nevertheless, correspondence with the researchers revealed that a 30-page questionnaire was required for these 168 paired choices and that respondents needed between 30 and 75 minutes to complete the questionnaire, with an interviewer in attendance to monitor the process.

It is evident that the paired comparisons method quickly becomes unwieldy for rating more than a limited number of attributes, and may need closer supervision than Likert or semantic differentials scales. As a result, it appears to be of limited usefulness in large-scale surveys where the researcher has a variety of different questions to ask respondents and/or is operating under budget constraints which make supervised completion of questionnaires at home an unrealistic alternative. On the grounds of economy, speed and simplicity, I would therefore argue for use of Likert and semantic differential scales in transportation attitude studies.

Use of Constant Sum Scales

An alternative approach to measuring salience which also forces trade-off variations among attributes is the constant sum method, in which subjects are required to allocate a fixed number of points between each of the specified attributes. While I am not aware of any published transportation research employing this technique in the United States, it has been used for transportation and related studies in Britain (Hoinville, 1973; Hatch and Flack, 1974). I consider it a sufficiently interesting methodology to merit more detailed discussion.

Hoinville argues that the main limitation of most attitude measures is that respondents are not forced, as they are in a behavioral situation, to trade off some preferences against others. He asks, "Can people behave in a serious and responsible way when removed from the pressure of a real behavioral decision?"

The problem, as Hoinville sees it, is to develop a measuring device which maintains simplicity for respondents and yet approaches the complex reality of multi-choice situations. Re developed some interesting priority evaluation techniques for his own research into community preferences concerning such environmental variables as traffic noise, parking availability, pollution levels, etc. The range of choices offered respondents was illustrated pictorially on a board containing 30 small pictures. Different drawings represented 3 separate standards for each of the 10 environmental variables, and were ranked from "poor" on the left to "good" on the right. These drawings had been pretested for comprehension and rank ordering by traditional interviewing methods. Although some of the concepts were admittedly difficult to depict, it could be argued that pictures conveyed more than could just a brief verbal description.

Next, a price tag was attached to each standard, representing the crude supply costs of arriving at each standard. Respondents were then given fifteen pegs, like radio jacks, each worth L100 and told that they must select one standard for each variable. The baseline (poor) standard was free; better standards could be "purchased" by inserting the appropriate number of pegs.

Respondents could easily assess their present position at any time, since existing choices on the board were illuminated electrically. At the outset, all baseline choices were illuminated. As the respondent "purchased'' a more desirable position by inserting pegs, so the light in the old position would be extinguished and the new one lit up. Respondents were allowed to modify their choices after studying the outcome or "pay off" resulting from these choices. The advantages of this approach included ease of use for respondents, flexibility, avoidance of excessive mental arithmetic and a very wide range of choice combinations (the 10 x 3 matrix used on the board provided for more than 3,700 possible combinations from which to choose). By changing the "costs" attached to the different alternatives, it was possible to see how respondents altered their positions on the board to generate a different set of trade offs.

Hatch and Flack used priority evaluation techniques (but without Hoinville's sophisticated electric board) to assess consumer preferences for alternative improvements in the London Underground. While they feel that the approach provides useful data on consumer priorities, they also note some general problems. In particular, they emphasize that successful use of priority research techniques rests on an assumption that consumers actually think in terms of priorities and alternatives, whether or not financial and other constraints are placed on preferences. They argue that this may not be the case and believe there is a danger that "introducing alternatives and priorities to the sample being studied educates and conditions it so that it is no longer typical of the universe it was selected to represent."

Obviously, the priority evaluation techniques developed from constant sum scales are not without their disadvantages. Except for very simple scales, they tend to require the presence of an interviewer or supervisor, with accompanying increases in costs. Nevertheless, I believe that this methodology merits further consideration for transportation research purposes on this side of the Atlantic.

Measuring Consumer Awareness

As noted earlier, the cognitive or perceptual component of attitudes represents a person's information about an object. It can be divided into beliefs in the existence of an object (i.e., awareness) and evaluative judgments about the object.

It is self evident that if a person does not know that a product (such as transit service) exists, he or she can hardly be expected to use it except under impulse purchase conditions. However, it can be argued that mere knowledge of existence is scarcely sufficient in most situations, and that consumers need to have information about certain key attributes of a product before they can make comparative judgments against competing products.

major shortcoming of economic theories of consumer behavior, and the models based upon these, is that they cannot explicitly handle the problem of imperfect information (Ratchford, 1975). It may very well be that a partial explanation for modal choice behavior patterns may be found in individuals' lack of knowledge of the availability and characteristics of alternative modes.

In the case of public transportation, prospective users may need to know about the routing of services (and, in particular, the location of stops in relation to their journey origins and destinations); what time the transit vehicle leaves and when it arrives; how much the journey costs and how the fare is to be paid.

Unlike attributes such as comfort, which are highly subjective and complex to measure (Nicolaidis, 1975),'routing, scheduling and costs characteristics can be quantified in unambiguous terms. Consequently, respondents can be surveyed to determine how knowledgeable they are about the existence and specifics of transit services in relation to these attributes and their answers graded by degree of accuracy.

If the location of respondents is known (and this speaks to issues of both sample design and delivery procedures), then each can be asked such specific questions as (1) How far is your home from the nearest bus stop? (2) Is there a train service from your neighborhood to downtown? (3) How long does the train journey take on this route during commute hours? (4) How frequently does the bus operate on this route on weekday mornings outside commute hours? (5) How much is the bus fare from your home to downtown? (6) What is the name of the transit company providing the service? and so forth.

Methodological problems for the researcher center around formulating questions which can easily be validated for each respondent (which requires identifiable rather than anonymous responses); are equally appropriate for all respondents in the sample frame (which may cover a wide geographic area); can be answered unambiguously by respondents without undue prompting; and are simple to code for subsequent analysis.

The problem of developing such measures is compounded in situations where transit services are not scheduled consistently on regular headways (e.g., every 30 minutes), respondents can choose between express and local services on the same routes, there are alternative transit routes between the same two points, and there are a variety of different fair options. However, it should not be beyond the wit of the ingenious researcher to circumvent such problems by careful attention to both sample design and questionnaire wording.

As an input to modal choice models, it may be useful to develop an awareness index for each consumer. For such a purpose, the researcher may wish to weight certain awareness items more heavily than others and also, perhaps, develop a scoring procedure which gives partial credit for responses within defined limits of the correct answer.

In my own research (Lovelock, 1973), I have used a simple additive awareness index, whereby respondents were given one point for each answer which was correct within certain limits (e.g., a fare within 10 cents of the actual fare), but nothing for wrong answers, and the totals then summed for each individual. While this measure discriminated quite well between users and non-users of transit (as one would expect), I believe that more sophisticated scoring and weighting procedures might improve the explanatory and predictive power of the awareness variable.

Problems With Halo Effects

Attempts to measure evaluative beliefs about specific attributes of an object may be confounded by halo effects. Several researchers have drawn attention to the problems associated with such effects in multi-attribute attitude models (Wilkie and Pessemier, 1973; Beckwith and Lehmann, 1975). The essence of the problem is that individuals who favor a particular product tend to rate it highly on all desirable attributes, while individuals who dislike it have a tendency to give it a low rating on each of the attributes listed. Wilkie and Pessemier note that "halo effects have long been recognized in personality and psychological testing as potential suppressors of importance variation. The presence of halo effects in the marketing model.. . . will confound investigations as to the dimensionality of attitude structure and impair diagnostic analyses of brand strengths and weaknesses."

In a transportation context, the result of halo effects is most likely to be reflected in superior ratings for car travel (generally the preferred mode) across all attributes and inferior ratings across the board for travel by transit modes. What can be done to counter this problem?

Wilkie and Pessemier's suggestions include adding warm-up instructions to respondents which discourage yea-saying; and rating all products (modes) within attributes (rather than rating each mode separately on all attributes). Beckwith and Lehmann recommend that care be taken to include only relevant attributes, since respondents' statements about less important attributes may be determined almost totally by the halo effect. They also urge that researchers select attitude measures which are as objective as possible. Amongst other things, this argues against use of terminology such as "how satisfied are you?" in favor of a more neutral "please rate performance." It is perhaps noteworthy that both Paine et al. and McMillan and Assael employed satisfaction measures in their respective studies, which showed respondents to rate car travel more favorably than transit on every single attribute. By contrast, Lovelock (1973) used semantic differential scales and found train and bus travel to be rated significantly better than car travel on the characteristic of safety.

COMMUNICATING WITH SUBJECTS

In this section, I want to look at the problem of selecting the most appropriate means of communicating with subjects for transportation studies.

It is important to emphasize that every research program poses somewhat different needs. While the objective function in each instance should (hopefully!) be to maximize the quality and usefulness of the research findings, the constraints involved for any given project are likely to vary widely. Non-response bias may have more serious implications in some studies than in others; cost may or may not be a key consideration; and the importance of precise sample design varies with the characteristics of the universe under study. It can never be claimed that one methodology is always "superior" to another, only that one may be more suitable under certain, specific conditions.

As mentioned earlier, the choice of procedures for obtaining information from respondents is closely related to issues of sample design and selection of measurement techniques.

In evaluating alternative methodologies, the researcher must recognize the interrelationships between such decision variables as size, characteristics and location of size, degree of detail desired in the responses, questionnaire length, response rate required, size of total budget, anticipated cost per usable response, and survey topic. Changes in any one of these variables may impact upon others. Thus, a longer questionnaire should provide more detailed results but may be more costly to produce and mail; under some circumstances it may also result in a lower response rate, thereby generating not only higher total costs but also significantly higher costs per usable response. Attempts to increase response rates by monetary incentives, careful follow-up procedures or other techniques tend to raise total costs but the average cost per response received may not raise proportionately.

Choice of Locations for Surveys

The nature of the transportation market provides some interesting opportunities for reaching potential and actual users, in that the sample population is typically defined by such readily identifiable characteristics as usage behavior or origin and destination location. The researcher's problem is to determine the most appropriate means of reaching subjects in these sample populations. Opportunities exist for surveying respondents at home, at destination locations, or even en route as they travel between the two.

Let's examine each of these alternative locations in turn, and consider their implications for sample design, selection of measurement techniques, and general nature of the research instrument.

If we can reach people while they are actually traveling. we are immediately in a position to define our sample as people who behave in a specific way of interest to us (i.e., traveling that day at a specific hour, by a particular mode on a given route). Depending on the nature of the mode and the travel conditions at the time, it may be possible to conduct on-the-spot personal interviews, ask travelers to complete a questionnaire for immediate or later return, or take individual names, addresses and/or phone numbers to allow for subsequent mail, phone or personal interviews. An interesting variation on the latter approach, occasionally used in highway traffic surveys, is to photograph cars' rear number plates, obtain names and addresses of their owners from state motor registry files, and then mail questionnaires to these people.

Most of us are probably familiar with cordon traffic surveys, where a proportion of cars passing a particular point are stopped and brief questions asked of their occupants. Similar surveys are sometimes conducted at transit stops and stations or on board moving vehicles. However, there are obvious limitations here on the type of instrument employed. Congestion on highways restricts the feasibility of traffic surveys and passenger congestion often makes it impracticable on board transit vehicles. Even where contact can be made, the amount of time which is available may be too short for all but a short, simple survey which cannot get into detailed attitudinal issues. Usually such surveys are designed to obtain descriptive information about travelers, such as their journey origins and destinations, trip purposes, key demographic characteristics and, perhaps, frequency of travel on that route by that mode.

On longer journeys by public transportation (such as trips on express buses, ferries and trains, or airline flights), the situation may be different. In these situations, the passenger is effectively a "captive," more likely to have a seat, less subject to other distractions and perhaps even responsive to the opportunity to pass the time by completing a written questionnaire.

The appropriateness of the second alternative, reaching people at their destination (which for our purposes may be considered the location reached on the outbound journey from the traveler's home residence), depends largely on the nature of the destination. There are obvious differences between interviewing respondents (a) as they arrive at a bus or rail terminus; (b) arrive at a bus stop; (c) leave their car at a parking lot; (d) start a shopping expedition at a particular location; (d) arrive at a theatre or sports stadium; or (f) arrive at work. Realistically, the first five situations probably allow for no more than a brief survey (if that), or for collecting names and addresses for subsequent follow-up.

Surveys at work locations have more potential, provided cooperation can be obtained from the employer in question. Possibilities include completion of either personal interviews or self-completion questionnaires on company time; or distribution of questionnaires by the employer for completion by employees during their own time, with return either through the mails or via the employer. However, employers may be unwilling to provide the researcher directly with a list of employee names and addresses for independent follow-up.

Because of the time constraints and other distractions inherent in en route and destination surveys, it is reasonable to anticipate that when conducting detailed attitudinal surveys, subjects' homes will be the preferred choice of the transportation researcher.

Evaluating Alternative Interviewing Media

Three basic media of communication are generally available to the researcher wishing to conduct interviews with subjects at their home residences, namely personal interview, telephone interview and unsupervised completion of a questionnaire by the respondent. Variations on these three basic approaches include: the use of advance postcards or phone calls to inform people that a mail or personal survey will soon be conducted; a choice between using the mails or personal drop-off and pick-up for delivery and collection of questionnaires; and including a self-completion questionnaire as part of a personal interview session.

The major criteria usually employed in selecting between these alternatives include cost per completed response, the length and nature of the questions to be posed, the format in which responses are desired (e.g. open-ended versus forced choice), the desire to avoid various types of bias, the characteristics of the sample, and the time frame available for completion of the research. In this section of the paper, I propose to briefly review the appropriateness of each of the three major alternatives from the standpoint of attitudinal research within a transportation context.

Telephone Surveys

Telephone surveys are attractive to many survey re--searchers. They are normally less costly and more expeditious than face-to-face interviews, while still providing for interaction between interviewers and respondents. They generally achieve higher rates of response than either mail or personal contact and can be validated by monitoring or callback (Glasser and Metzger, 1972). However, I believe that telephone interviews may present problems which make them a less-than-satisfactory means of gathering detailed attitudinal data for transportation surveys.

A key problem relates to the risk of sample bias resulting from households without telephones and households with unlisted telephone numbers. Glasser and Metzger (1975) indicate that nationally, 19.2% of household subscribers had unlisted numbers in March, 1974. Potentially of serious concern to transportation researchers is the incidence of non-listings among specific demographic and geographic groups. It is particularly high in the West census region (26.1%); in counties located in the five largest metropolitan areas (29.0%) and the next 21 largest metro areas (23.5%); in households with incomes between $5,000 - $9,000 (25.0%); among nonwhite subscribers (31.7%); and among males aged 18-34 (24.5%) and females aged 18-34 (27.0%).

Since transportation research tends to focus on large metropolitan areas, the high incidence of non-listings in such areas should be cause for serious concern among those who generate their phone or address samples from telephone directories. And since public transportation tends to attract a higher proportion of people in lower income groups and from non-white ethnic groups, there is also a significant risk of bias from that standpoint. By contrast, those in the lowest income brackets (under $5,000) have a relatively low incidence of non-listing (17.6%) as do those earning $15,000 and over (17.3%), men aged 50 and over (8.1%) and women aged 50 and over (11.8%).

Glasser and Metzger (1972) suggest random digit dialing as a method of telephone sampling which avoids the non-listing problem. However, as emphasized earlier, transportation study samples should be geographically specific when studying modal choice behavior, since proximity of access to a transit route may be a key determinant. Unfortunately, in most locations the preliminary two or three digits of a number may delineate telephone subscriptions within quite large neighborhoods or even entire cities and are thus unsuitable for pinpointing sample locations with sufficient precision.

A second, and equally important problem with this medium is that the amount of information which may be obtained over the phone is typically limited. Although Payne (1974) has indicated some subtle ways of overcoming the telephone's shortcomings, it remains difficult to administer more than a limited number of questions employing psychological scaling methods such as paired comparisons, Likert scales or semantic differential scales through this medium.

Personal Interviews

Personal interviews, by contrast, are an appropriate and frequently used way of administering detailed attitudinal surveys, providing skilled personnel are available to minimize interviewer-induced response bias. However, they can prove extremely expensive to implement. Figures of $20 to $50 per completion are common where such interviews are conducted on an individualized basis. Where large sample sizes are desired for statistical purposes, this method of obtaining information may quickly become infeasible for all but the most generously funded projects, unless some way can be found to bring respondents together in a central location.

There are many works detailing the advantages and disadvantages of personal interviewing in survey research. [Ferber (1974) is a good sourcebook.] So I will not reiterate these here, except as in they relate specifically to transportation studies. Selection of households in personal interviews conducted for this purpose can be made on the basis of housing unit location relative to transit routes, major highways and distance from major employment centers, in consultation with demographic census data (which is available on a block by block basis). Additionally, the interviewer may be able to observe certain characteristics, such as physical condition of respondents, model and condition of household automobiles, etc., which might be related to modal choice behavior.

One problem with personal interviews nowadays, however, is concern for the safety of the interviewer, especially in neighborhoods with an above average crime rate. Additionally, many people are increasingly unwilling to open their doors to a stranger. These two factors may make it-particularly difficult to interview respondents in the lower income brackets.

Never the less, certain techniques, such as prompted recall, paired comparisons and the priority evaluation methods employed by Hoinville (1973) and Hatch and Flack (1974) effectively require the presence of a personal interviewer. Direct questioning by an interviewer also has advantages when measuring consumer awareness since it does not permit time for extensive reflection by the subject nor provide an opportunity for cheating.

Self-Completion Questionnaires

If the telephone is deemed inappropriate and personal interviews considered too expensive, then the researcher is forced to use questionnaires designed for self completion by respondents. Again, I will refrain from discussing the general pros and cons of mail questionnaires, since these are treated in detail elsewhere in a variety of different articles and reference works. Suffice to say that such questionnaires have proven to be an effective medium for often quite extensive attitudinal surveys on a wide variety of topics.

As in personal interviews, the sample may be defined with reference to transit route and stop locations, in conjunction with census data. Questionnaires may then be mailed directly to addresses on appropriate streets.

An alternative delivery procedure calls for personal drop-off and pick up of questionnaires by lightly trained survey takers, following preassigned routes and visiting a given proportion of household units on that route. If desired, the number of questionnaires distributed to each household can be varied, to reflect the number of eligible subjects resident therein. This approach also allows for selective replacement of subjects falling outside the sample frame (e.g., non-English speaking persons). Survey takers may be required to keep log sheets, allowing for identification of reasons for non-response and thus a better understanding of the nature of non-response bias than is possible in mail surveys. I have used this drop-off and pick up procedure successfully myself and it is described in detail in Lovelock (1973).

CONCLUSION

This paper has reviewed a number of methodological issues which I consider particularly significant for transportation researchers. I have deliberately chosen to focus on methods relating to research design and data collection rather than data analysis, since I believe that it is in the former categories that improvements are most needed and will yield the most significant advances. The analytical techniques employed in transportation research are, in many instances, highly sophisticated. However, the usefulness of any resulting findings is conditioned by the quality of data collected in the first instance, and I perceive many opportunities for improvement in sample design and measurement techniques. Researchers also need to exercise care in their choice of methods for communicating with subjects; in particular, they should recognize that telephone interviews may be inappropriate for certain types of transportation research.

By way of conclusion, I should like to make a plea for more detailed reporting in transportation studies of what I consider to be essential background information. Since the value of research findings is so closely tied to the nature of the methodologies employed, readers have a right to expect reports to provide information on such basics as the characteristics of the sample (as well as just its size), how subjects were selected, and the percentage response rate; it is also desirable to include an identification and evaluation of any possible biases. This information can generally be provided quite concisely but some significant background details are missing all too often.

One explanation for this situation may perhaps be found in the fact that in transportation studies, as in other fields, the tasks of data collection and data analysis are often entrusted to two different researchers. However, if this paper achieves no more than to lead the analysts to ask more pointed questions about where their data came from and how it was collected, it will have served a useful purpose.

REFERENCES

Alpert, Mark I., "Identification of Determinant Attributes: A Comparison of Methods," Journal of Marketing Research, 8 (May 1974), 184-191.

Beckwith, Nell E. and Donald R. Lehmann, "The Importance of Halo Effects in Multi-Attribute Attitude Models," Journal of Marketing Research, 12 (August, 1975) 265-275.

Blumer, Herbert, Symbolic Interactionism: Perspective and Method (Englewood Cliffs, N. J.: Prentice-Hall 1969).

Bonsall, J. A., Dial-A-Bus: The Bay Ridges Experiment (Toronto, Ont.: Ontario Dept. of Transportation and Communications, 1971).

Day, George S., "Theories of Attitude Structure and Change," In S. Ward and T. S. Robertson (eds.), Consumer Behavior: Theoretical Sources, (Englewood Cliffs, NJ: Prentice-Hall, 1973).

Ferber, Robert, Handbook of Marketing Research, (New York: McGraw-Hill, 1974).

Glasser, Gerald J., and Dale D. Metzger, "Random Digit Dialing as a Method of Telephone Sampling," Journal of Marketing Research, 9 (February, 1972), 59-64.

Glasser, Gerald J. and Gale D. Metzger, "National Estimates of Nonlisted Telephone Households and Their Characteristics," Journal of Marketing Research, 12 (August, 1975), 359-361.

Golob, Thomas F., Eugene T. Canty, Richard L. Gustafson, and Joseph E. Vitt, "An Analysis of Consumer Preferences for a Public Transportation System," Transportation Research, 6 (March, 1972), 81-102.

Golob, Thomas F., Ricardo Dobson, Jagdish N. Sheth, 'Perceived Attribute Importance in Public and Private Transportation,"(Warren, Mich.: General Motors Research Publication GMR-1465, July, 1974).

Hatch, Donald and Mervyn Flack, "Priority Research Techniques Applied to Public Transport Investment," Journal of the Market Research Society, 16:1 (1974).

Hoinville, G., "Evaluating Community Preferences," Journal of the Market Research Society, 15:1 (1973), 1-23.

Lovelock, Christopher H., Consumer Oriented Approaches to Marketing Urban Transit, Ph.D. dissertation, Stanford University (Springfield, VA.: National Technical Information Service, #PB. 220-781, 1973).

Lovelock, Christopher H., "Researching and Modeling Consumer Choice Behavior in Urban Transportation," in M. J. Schlinger (ed.), Advances in Consumer Research, Vol. II, 1974 Proceedings, Association for Consumer Research.

Lovelock, Christopher H., "A Market Segmentation Approach to Transit Planning, Modeling and Management," Proceedings, Transportation Research Forum, 1975.

McMillan, Robert K. and Henry Assael, National Survey of Transportation Attitudes and Behavior (Washington, D.C.: National Cooperative Highway Research Program Reports #49, 1968; #82, 1969).

Metropolitan Toronto and Region Transportation Study (MTARTS), GO Transit Commuter Rail Project, Special Reports #4A and #9 (Toronto, Ont., 1968).

Nicolaidis, Gregory C., "Quantification of the Comfort Variable," Transportation Research, 9 (1975), 55-66.

Paine, Frank T., et al., Consumer Conceived Attributes of Transportation: An Attitude Study (Springfield, Va.: National Technical Information Service, #PB 176-485, 1967).

Payne, Stanley L., "Data Collection Methods: Telephone Surveys," in R. Ferber (ed.) Handbook of Marketing Research (New York: McGraw Hill, 1974).

Ratchford, Brian T., "The New Economic Theory of Consumer Behavior: An Interpretive Essay," Journal of Consumer Research, 2 (Sept., 1975), 65-75.

Solomon, Kathleen M., Richard J. Solomon and Joseph S. Silien, Passenger Psychological Dynamics, Sources of Information on Urban Transportation, Report #3 (New York: American Society of Civil Engineers, 1968).

Stearns, Mary Donahue, Proceedings: TSC Workshop on Attitudinal Studies for Transportation Planning and Evaluation (Cambridge, Mass.: Transportation Systems Center, 1975 [forthcoming]).

United States Department of Transportation, Proceedings, National Transit Marketing Conference (Washington, D.C.: Office of Transit Management, 1975).

Wilkie, William L. and Edgar A. Pessemier, "Issues in Marketing's Use of Multi-Attribute Attitude Models," Journal of Marketing Research, 10 (November, 1973). 428-441.

---------------------------------------