Three Faces of Eve: How Engineers, Economists, and Planners Variously View Congestion Control, Demand Management, and Mobility Enhancement Strategies

Erik Ferguson*
ETF Associates

Abstract

The political acceptability (A) of public policy measures correlates positively with program effectiveness (E) and negatively with program cost (C) and other obstacles to implementation (I) under normal circumstances. Ferguson (1991) observed that the political acceptability of many demand management strategies seemed to correlate negatively with implied program effectiveness. Engineers, economists, and planners each have their own unique professional standards. Increased effectiveness is the primary goal of engineering. Improved efficiency is the generally accepted standard in economics. Process issues are of vital concern in planning. A review of the literature indicates few studies that rate demand management strategies in terms of all four variables of interest (A, E, C, and I) simultaneously. Three relevant studies were identified: one each by an engineer, an economist and a planner. Raw data, regression results, bivariate correlations, and model output reveal that two of the three studies support the Ferguson hypothesis. The other supports a more traditional public policy model. E is the most influential variable in the engineer's data. C is the most influential variable in the economist's data, while I is the most influential variable in the planner's data. These revealing results suggest the subtle manner in which professional training and experience may alter perceptions of transportation policies and programs in professional practice.

Introduction

Transportation planning is a capital-intensive process. The Federal-Aid Highway Act of 1956 removed gasoline tax revenues from the federal budget and set up a dedicated Highway Trust Fund. The Act further specified that all local return monies from this source must be allocated to the construction of new highway facilities (Weiner 1999). Project evaluation of highway capital investments was relatively simple in those days. Capital costs were fixed and low, at least by later standards. Operating costs were negligible from the government's perspective. Benefits included reduced out-of-pocket travel costs and time. Reduced costs were sufficient to justify most proposed projects, making concerns about the imputed value of travel time irrelevant. In the 1960s, rising land acquisition, highway construction costs, and a diminished potential to shorten travel distances made saving time more important in justifying new highway investments (Walters 1961).

In the 1970s, transportation system management (TSM) came into vogue. Even with savings in travel time properly accounted for, new highway construction became more expensive and difficult to justify, particularly in the face of local citizen opposition. The purpose of TSM is to soften the blow of transportation planning by making it more short-range, user-friendly, and demand-oriented. TSM actions include improved vehicular flow, preferential treatments for high occupancy vehicles, reduced peak period travel, parking management, promotion of alternative modes of transportation, and transit and paratransit service improvements (UMTA 1977). TSM's most ardent admirers predicted its possible demise from political opposition, the projected outcome of institutional inertia, and professional apathy (Gakenheimer and Meyer 1979). However, many TSM strategies have done much better than expected.

In the 1980s, travel demand management (TDM) became the watchword of the day. TDM is the demand side of TSM, making it more suitable for private sector participation. TDM operates even closer to end users such as individuals, households, and firms. TDM strategies include alternative modes and hours of travel, alternative locations for specific activities, as well as economic incentives and institutional arrangements that may be required (Orski 1987).

Due to the wide range of strategies available to deal with traffic problems, the complexity of decisions associated with congestion management has increased in recent years. Sorting through the bewildering array of alternative policies, programs, and projects can be a daunting task. Word of mouth, hearsay evidence, and the occasional case study can only go so far. The more comprehensive the evaluative outlook, the better the implied advisement should be. Comprehensive comparative assessments of traditional planning, TSM, and TDM techniques, however, remain relatively few and far between (Ferguson 2000).

Previous Studies

One study by the Urban Mass Transit Association (UMTA) (1977) categorizes TSM measures without identifying any of the potential impacts. Wagner and Gilbert (1978) evaluates TSM measures in terms of effectiveness and cost but omits any discussion of implementation issues. Misch and Margolin (1981) categorizes the "feasibility or prospects" of TSM actions taken in support of ridesharing in a limited fashion. Schonfeld and Chadda (1985) focuses on the effectiveness of travel reduction options with particular emphasis on parking management strategies. Levinson et al. (1987) evaluates the effectiveness of TSM strategies in terms of connected "impact chains." Bhatt and Higgins (1989) focuses mainly on results, measured in terms of mode split (table 1).

Two studies by the Institute of Transportation Engineers (ITE) (1989 and 1997) present information on impacts, costs, and obstacles to implementation associated with a wide range of congestion control and mobility enhancement tools but discuss political acceptability only tangentially. Ferguson (1991) observes that the public acceptability of many TDM strategies correlates negatively with their implied effectiveness. Road and parking pricing are more effective but less popular strategies. However, voluntary efforts to promote alternative modes and hours of travel are more popular but less effective.

Downs (1992), Zupan (1992), and Arnold (1993) evaluate various mixes of congestion control and demand management strategies in terms of four generalized performance measures: effectiveness, cost, ease (or difficulty) of implementation, and political acceptability. Dueker et al. (1998) perform a similar analysis on a smaller number of parking management strategies. Comsis (1993) and the Organisation for Economic Cooperation and Development (OECD) (1994) revert to an earlier emphasis on program effectiveness. McBryan et al. (1996) balances effectiveness, cost, and implementation issues but deliberately excludes political acceptability as "too arbitrary" (Shadoff 1997).

The remainder of this paper focuses on an analysis and evaluation of data derived from Arnold (1993), Downs (1992), and Zupan (1992), hereinafter referenced to as such.

Hypotheses

There are two major hypotheses tested here, one rather precise in nature, the other far less so.

1. Do Arnold, Downs, and Zupan confirm or deny Ferguson's (1991) assertion that TDM program effectiveness and political acceptability are negatively rather than positively correlated? An objective comparison of the signs and significance of relevant parameters determines the outcome of this test.
2. Are there any other differences in data or results among the three studies examined? How do such differences relate to the professional outlook of the respective authors? This is a much more subjective test but intriguing nonetheless.

Authors

In order to evaluate the possible contribution of professional perspectives to the understanding of complex policy issues such as congestion control or demand management, it is important to know with whom one is dealing. The three studies in question were each authored by a single individual (table 2).

Eugene Arnold is a Senior Research Scientist at the Virginia Transportation Research Council, Virginia Department of Transportation (VDOT) in Charlottesville, Virginia. The Commonwealth of Virginia is one of the most conservative states in the Union. VDOT ranks among the more innovative state transportation agencies, thanks in no small part to its proximity to the nation's capital. Arnold is an active member and national leader in the Institute of Transportation Engineers (ITE). For example, he chaired the committee that prepared the most recent update of Trip Generation (ITE 1997). It should come as no surprise to find that Arnold's structure of the research problem closely parallels that identified in ITE (1989).

Anthony Downs is a Senior Fellow in Economic Studies at the Brookings Institution in Washington, DC. His expertise extends to topics as diverse as democracy, demographics, housing, metropolitan policy, real estate, real estate finance, smart growth, suburban sprawl, and urban policy. This is his first foray into transportation planning in more than a quarter of a century. Downs (1992) expands on the ideas originally set forth in Downs (1962).

Jeffrey Zupan is Senior Fellow for Transportation at the Regional Plan Association in New York City. His research on the relationship between land use and public transportation is well known and respected. Pushkarev and Zupan (1977) showed 1) what type of transit works best, 2) where it works best, 3) which support policies are most effective, and 4) how to estimate transit demand and costs (see Lee 1978). The connection between Zupan (1992) and Pushkarev and Zupan (1977) is clear.

Data

The data used in this analysis are widely available and based on previously published research results. They are reproduced here mainly for the reader's convenience (see tables A-1 to A-3). However, Arnold's original survey data were obtained directly from Arnold for this study. Table A-1 includes more information than was published in the original paper.

Arnold

Arnold surveyed 85 local, regional, and state transportation agencies in Virginia regarding 53 congestion-reducing measures. The ratings and rankings shown in table A-1 reflect the collective judgment of traffic engineers and transportation planners in Virginia, not necessarily Arnold's personal opinions.

Arnold treats the ordinal scales he uses to measure effectiveness, cost, and implementation as interval scales in his analysis: a testable proposition. Arnold's data are purely quantitative in presentation but largely qualitative in nature. This may reflect a preference among engineers for data and methods that are more objective. Arnold's data include 53 observations and only 4 variables, a case of problem overidentification. This is positive, of course. Engineers prefer larger margins of error.

Arnold's principal research findings include the following.

1. TSM measures have almost caught up with supply-side measures in terms of actual use. Supply-side measures are more effective but extremely costly and face many obstacles to implementation.
2. TDM measures are much less effective and slightly more costly than TSM measures. With obstacles to implementation about the same, it is no surprise that TSM is much more popular than TDM in the Commonwealth of Virginia.

Downs

Downs evaluates 23 congestion-reducing policies using 7 items corresponding to 4 variables (table A-2). Whereas the Arnold data appear purely quantitative, the Downs data appear purely qualitative in nature. Downs uses different semantic scales to describe most of his items. The two cost items share an identical scale.

All of the scales constructed by Downs are ordinal in nature, with the possible exception of the institution required for implementation, which may be categorical. The Downs data are overidentified but not nearly as much as the Arnold data. Downs assigns a non-ambiguous descriptive adjective to each attribute for every policy he considers.

Downs' principal research findings include the following.

1. Supply-side policies are more expensive but require less institutional change than demand-side policies.
2. Demand-side policies have broader effects but are less politically acceptable than supply-side policies.

Zupan

Zupan evaluates 22 TDM solutions using 12 items corresponding to 4 variables (table A-3). Zupan is the everyman of performance measurement. He incorporates a little bit of everything in his evaluation matrix, including both words and numbers, scales both absolute and relative, as well as suitable descriptors of measurement variability, non-applicability, and the unknown.

Zupan's 12 items are just barely identified by the 22 observations in his matrix, providing another indication of his tolerance for uncertainty. There is not one item in Zupan's matrix that is ranked consistently using a single semantic scale. The only possible exception is ease of implementation, which still includes two "unknowns" and one "not applicable" rating. Planners may be less technical than engineers and less consistent than economists, but one thing stands clear: they are much more comfortable with uncertainty.

Zupan's principal research findings include the following.

1. Strategies that reduce traffic congestion, have no negative effect on transit, and are politically popular should be pursued.
2. Strategies that have little effect on traffic congestion, negative effects on transit, or are politically unpopular should be avoided.
3. Strategies for which too little is currently known should be studied more carefully.

Comparing the methodological results of these three studies would be intriguing but difficult to accomplish. Observed differences between the three databases seem much greater on the surface than any prospective similarities might be. The problem of compatibility must be resolved in order to make any more meaningful comparisons between these studies and their results.

Semantic Scales

The next objective is to evaluate a model of the following general form:

A= f (E, C, I)

where

A = (political) acceptability

E = (program) effectiveness

C = (program) costs

I = (obstacles to) implementation

It is hypothesized that (political) acceptability should be

1. Either positively or negatively correlated with program effectiveness, depending on whether a traditional public policy model applies, or Ferguson's 1991 conjecture is correct.
2. Negatively correlated with program costs and obstacles to implementation.

In order to test these hypotheses, it is necessary to convert the data found in tables A-1 to A-3 to standard form. Certain assumptions are required; the most important ones follow.

1. Observed semantic scales correspond to unobserved, logically consistent, and mathematically comparable probability surfaces that can be mapped.
2. Observed variable components correspond to unobserved or latent variables that may be combined.

Given these assumptions, composite variables representing A, E, C, and I may be computed and subsequently analyzed using regression. The initial assignment of values to each semantic scale in the data is straightforward (table 3).

1. Arnold's use of semantics requires little comment. He assigned values of 0, 1, and 2 to his 3 ordinal scales. This has been changed to 1, 2, and 3 for consistency with the Downs and Zupan data but otherwise should have no effect.
2. Downs employs a wide variety of different labels. By inspection, all of his labels correspond to a single four-point scale of zero to three, with relatively minor variations as noted.
3. Zupan employs a bewildering array of positive and difference scales, often simultaneously. He occasionally uses different labels that mean the same thing (for example, he uses "same" and "none" on a difference scale). He uses three different labels to represent aspects of uncertainty ("varies," "unknown," and "not applicable"). Despite these obstacles, it remains possible to assign plausible initial values to all of his labeling conventions.

Regression Results

With an initial assignment of semantic values in hand, it is possible to calculate regression equations for each dataset. This does not imply that the results will maximize the potential of any available information, of course. For that to happen, several more questions must be answered.

1. Semantic values: the initial assignment of semantic values is unambiguous in the case of Arnold and Downs. The accuracy of the semantic values assigned to Zupan's mixed scales and uncertainty measures is open to question.
2. Semantic differences: Arnold and Downs employ a series of ordinal scales. Semantic differences between values on an ordinal scale may be linear or non-linear: a testable proposition. It is difficult, if not impossible, to evaluate semantic differences where the semantic values themselves are open to question, leaving Zupan out of this discussion.
3. Component weights: where more than one item describes a variable, these must be combined to provide a unitary performance measure. For simplicity, it is assumed here that each item bears the same weight: a testable proposition.
4. Variable transformations: none of the variables used in this analysis, with the possible exception of Arnold's measure of political acceptability, is anything more or less than a dimensionless performance indicator. All variables were standardized under an assumption of normality in recognition of this fact and to facilitate comparisons. The limited sample sizes involved do not allow for any other assumption.

Arnold

The Arnold data generally support the traditional public policy model (table B-1):

1. Arnold's measure of political acceptability is positively correlated with program effectiveness and negatively correlated with program cost and obstacles to implementation. The cost variable is not statistically significant in any of the equations although it does approach significance as the model's assumptions are relaxed.
2. Supply-side measures are most politically acceptable, followed by TSM and growth control, with TDM last. The only statistically significant difference is between supply-side and TDM measures.
3. Semantic differences between values on Arnold's ordinal scales are clearly nonlinear.
4. The extent of nonlinearity varies. Implementation is least, and cost is most nonlinear. Low and medium costs are not semantically distinguishable in Arnold's data.

Downs

The Downs data generally support the Ferguson hypothesis (table B-2).

1. Political acceptability is negatively correlated with program effectiveness, cost, and obstacles to implementation. There is a statistically significant bias toward supply-side policies.
2. Most of the semantic differences in the Downs data are highly correlated with their respective component weights. Given the limited sample size, it is not possible to test simultaneously for both semantic differences and component weights under these conditions. The sole exception to this rule is the difference between no cost and minor costs, which seems to operate somewhat more independently.
3. Component weights are of somewhat greater interest than semantic differences, at least from a public policy perspective. Component weights that maximize overall model goodness-of-fit include the following:
   1. Effectiveness: impact is more important than extent, but the difference is fairly slight.
   2. Cost: social costs are twice as important as commuting costs.
   3. Implementation: required institution (which operates efficiently as a linear variable) weighs in at an order of magnitude greater than ease of administration.
4. Simultaneous estimation of low-cost semantic differences and variable component weights produce results roughly equivalent to those found separately, indicating a stable solution to the model.

Zupan

The Zupan data generally support the Ferguson hypothesis (table B-3).

1. Political acceptability is negatively correlated with program effectiveness and obstacles to implementation and is, paradoxically, positively correlated with program cost.
2. "Varies" is an unknown quantity, normally assigned to the midpoint of any given range of values. Zupan uses two distinct scales to describe key elements of program effectiveness and political acceptability, making it unclear where "varies" belongs on a uniform or combined scale. The regression results clearly support the notion that "varies" refers to a positive (low-medium-high) rather than a difference (negative-none-positive) scale in this analysis.
3. Two other semantic-value assumptions bear additional scrutiny.
   1. "Unknown" and "not applicable" belong at the midpoint of their respective ranges, rather than at one or the other of the two extremes; this is true in all cases.
   2. The difference and positive scales are separate and distinct for program effectiveness but are semantically indistinguishable in the case of political acceptability. This is an unusual finding to say the least.
4. Best fitting component weights include the following:
   1. Transit impact is the most important component of program effectiveness, followed by vehicle-miles traveled (VMT) reduction and peak trip reduction. Single-occupancy-vehicle (SOV) reduction has no weight.
   2. Employer cost is the most important component of program cost, followed by public capital cost. Employee cost has no weight.
   3. Municipal acceptance is the most important component of political acceptability, followed by public acceptance and employee acceptance. Employer acceptance has no weight.
   4. The unexpected positive correlation between program cost and political acceptability may be explained largely as a function of component weights. The relationship is significant only in the last equation, where employer costs dominate the cost variable, while employer acceptance is omitted from the political acceptability variable. Economists refer to this as a transfer payment (Small 1999). Others might call it a tax or even an unfunded mandate.

Dueker et al. (1998) evaluate 10 parking management strategies in terms of 3 scopes or applications (temporal, functional, and spatial), 2 benefits or impacts (effectiveness and efficiency), ease of administration and political feasibility. A simple regression reveals that political feasibility is negatively correlated with effectiveness, lending support to the Ferguson hypothesis. No measure of cost was provided, and ease of administration was not correlated with political feasibility among these data.

Booz-Allen & Hamilton (2000) evaluate 32 TDM strategies loosely based on OECD (1994) in terms of 9 direct travel effects, 12 indirect policy effects or implications, practical feasibility (implementation), and political acceptability. A simple regression reveals that obstacles to implementation are strongly negatively associated, direct travel effects marginally negatively associated, and indirect effects not associated with political acceptability, lending support to the Ferguson hypothesis.

It would seem that the Ferguson hypothesis holds sway over congestion control and demand management strategies in many instances. Program effectiveness is negatively associated with political acceptability according to data derived from four out of five independent studies (table 4). This is a most unfortunate result, at least from a public policy perspective.

Variable Correlations

For those who prefer simpler explanations, bivariate correlations based on nave assumptions are shown in table 5.

These correlations illustrate that

1. Arnold's independent variables are highly correlated. This is an unanticipated response to the original survey design. Although not very desirable, these associations do not seem to have affected model estimation in any negative way. In fact, the estimated equations in table B-1 are, if anything, superior to the direct correlations shown in table 5. Program effectiveness exhibits the highest direct correlation with political acceptability in Arnold's data.
2. Downs' independent variables are largely independent of one another. Program effectiveness and cost are marginally correlated with each other, in a positive direction. Program cost exhibits the highest direct correlation with political acceptability in Downs' data.
3. Zupan's variables are more like Downs' than Arnold's. Program effectiveness and cost are marginally correlated, in a negative direction. This is yet another odd finding. Obstacles to implementation exhibit the highest direct correlation with political acceptability in Zupan's data.

Engineers seek technical solutions to the problems they face. Economists search for low-cost solutions. Planners search for institutional answers (Mandelbaum 1996; Marshall 1997). The independent variable most highly correlated with political acceptability (A) in this analysis is

1. Program effectiveness (E)-Arnold, the engineer
2. Program cost (C)-Downs, the economist
3. Obstacles to implementation (I)-Zupan, the planner

These simple bivariate correlations would seem to support the idea that political acceptability is in fact defined to some limited extent with particular professional perspectives kept in mind.

Outliers

One advantage of the equations shown in tables B-1 to B-3 is that values can be calculated for each of the four variables (A, E, C, and I) and compared across each data set.

Arnold

Arnold's model output (table C-1) does not vary much from the original data (table A-1). The values in table C-1 are directly comparable with those in tables C-2 and C-3, however, while those in table A-1 are not. Residuals are calculated using the following formula:

where

SRSD = studentized residual
A_P = political acceptability (predicted)
A_O = political acceptability (observed)
= standard error of A_P

A negative SRSD implies a positive bias in favor of that measure. In Arnold's case, survey respondents reported more examples of new and reconstructed highways in Virginia than the model predicts. Toll roads and high-occupancy vehicles (HOV) lanes produced fewer examples than the model predicts, serving as a confirmation of the model.

TSM measures show greater variability in Arnold's data.

1. Intersection improvements, restricted on-street parking, and five other traffic flow enhancements are favored.
2. Intelligent transportation system improvements (still relatively new in 1992) are not favored. This may have been a short-term effect, since remedied during the intervening time period.

Paratransit services and growth management round out the list of strategies that apparently receive a limited form of preferential treatment in Virginia (Cervero 1997).

Downs

Downs finds that demand-side policies are slightly more effective than supply-side policies but more difficult to implement (table C-2). This opposes Arnold's findings.

Downs' data show the following supply-side biases.

1. For: incident management and HOV construction
2. Against: new roads and transit service improvements

Down's data show the following demand-side biases.

1. For: growth management
2. Against: commuting allowances, staggered hours, and automobile license fees

Downs is the principal author of a widely read Real Estate Research Corporation report on the costs of sprawl (Altshuler 1977). Downs presumably should favor growth management beyond any limited ability it might have to deal with traffic congestion as an urban problem, based on this previous work.

Zupan

Zupan's data exhibit the following model biases (table C-3).

1. For: tax deductible transit vouchers, transit-friendly design, carpools, staggered shifts, and preferential parking
2. Against: trip reduction ordinances, employer subsidized transit, parking pricing, and growth management

None of the biases reported here are particularly consistent, other than that growth management shows up as an outlier in all three models. This identifies growth management as an issue of greater than average political controversy. Few of these biases are statistically significant, and none is influential enough to bias parameter estimates.

In summary (table 6):

1. Arnold reveals a slight bias in favor of certain types of highway projects.
2. Zupan reveals a slight bias in favor of certain types of transit projects.
3. Most of the residuals in Downs are random artifacts of model construction.

Other comparisons are best left to readers to explore on their own. Consider the relatively straightforward issue of parking pricing.

1. Arnold has something he calls differential parking rates, a limited variation on the general theme of parking pricing.
2. Downs provides two examples, a parking tax on peak-hour arrivals and the elimination of incentives to provide free employee parking. Both of these are indirect forms of parking pricing.
3. Zupan lists two aspects of parking pricing, including parking pricing and parking ratios. Parking ratios affect parking pricing indirectly, similar to federal tax incentives.

These three authors do not necessarily agree on the individual ratings of parking-pricing approaches any more than they do on how parking pricing should be labeled as a congestion control or demand management strategy.

Conclusions

How are program effectiveness, cost, obstacles to implementation, and political acceptability defined?

1. Effectiveness: effectiveness may be defined in terms of applications or impacts and is usually represented as some combination of both. The overwhelming emphasis in the literature on effectiveness may be overdone, however, especially if effectiveness is negatively correlated with political acceptability.
2. Cost: cost seems to be a somewhat neglected issue, at least in relation to effectiveness. Given that TSM and TDM programs often are intended to serve as low-cost solutions, this may be less of an impediment than it seems. Equity impacts may be more important than efficiency concerns in political terms and deserve greater scrutiny as a result.
3. Implementation: obstacles to implementation remain nebulous, at least in terms of definition. More work certainly could be done to identify the nature of and means to overcome those obstacles to existing implementation.
4. Acceptability: political acceptability is, if anything, the least understood aspect of the congestion control, demand management, and mobility enhancement problem.

How do professional perspectives influence these definitions?

1. Engineers: the dominance of engineering in the debate over congestion control, demand management, and mobility enhancement is clear in the definition of the problem to be solved. The emphasis on effectiveness to the practical exclusion of all else is a good indication of the engineering mindset at work. Transportation engineers must look beyond their own interests to see other aspects of the problem.
2. Economists: low-cost solutions to traffic congestion and air pollution may reduce the urgent necessity to think in economic terms, except for the considerable efficiency and equity impacts that might otherwise accrue (Garrett and Wachs 1996; Howitt and Altshuler 1999). Cost estimation seems undervalued in the current literature, with the identification of distributional effects even less in evidence. There is clearly work to be done by transportation economists in this area.
3. Planners: implementation issues are discussed slightly more often than cost issues in the literature, but the emphasis remains more anecdotal than evidentiary in terms of presentation. A practical definition of implementation issues and how these may be categorized from a strictly analytical point of view remains to be done.

How well is the decisionmaking process actually understood?

1. Policymakers: most of the transportation decisions are made by bureaucrats or politicians-hence the frequent confusion between "bureaucratic" implementation issues and "political" acceptability. A clearer separation of roles and responsibilities might help eliminate this problem, at least partially.
2. Stakeholders: policymakers are influenced by a variety of considerations in addition to their own personal opinions and moral judgments. Developers, employers, neighborhood associations, environmentalists, and a wide variety of other advocacy groups are involved in decisions that affect local communities. The perspectives of each of these groups may color perceptions of political acceptability and ultimately influence the outcome of transportation policy decisions.

Arnold (1993) is an excellent example of how to go about evaluating the political acceptability of transportation policies, programs, or projects. Downs (1992), Zupan (1992), and McBryan et al. (1996) provide useful guidance on additional aspects of program cost, obstacles to implementation, and political acceptability. Many others have contributed to the identification of congestion management strategies and measures of their effectiveness.

Confirmation of the Ferguson (1991) hypothesis leaves one important question unanswered. If the traditional public policy model holds true under most normal circumstances, either the model itself or one of its variable components must be in error in the present case. The model itself is too straightforward to be associated with any kind of specification error. The acceptability variable operates much as expected in relationship to the other variables in the model, and is therefore free from further suspicion as well. The cost variable is moderately suspect, due to its omission in some cases and its unusual performance in yet another.

Effectiveness behaves more like a cost than a benefit in most of the models tested here. One must conclude that the beneficial effects thus measured are associated with some additional indirect costs. Such hidden costs must supercede and override the direct benefits included in the measure's original, more limited definition. Further research should confirm or deny the validity of this conjecture.

References

Altshuler, A. 1977. Book Review. Journal of the American Institute of Planners 43:207-9.

Arnold, E.D. Jr. 1993. Evaluation of Congestion-Reducing Measures Used in Virginia. Transportation Research Record 1404:4-7.

Bhatt, K. and T. Higgins. 1989. An Assessment of Travel Demand Management Approaches at Suburban Activity Centers. U.S. Department of Transportation, Washington, DC.

Booz-Allen & Hamilton. 2000. Draft Travel Demand Management Strategy. Auckland, NZ: Auckland Regional Council.

Cervero, R. 1997. Paratransit in America: Redefining Mass Transportation. Westport, CT: Praeger.

Comsis Corporation. 1993. Implementing Effective Travel Demand Management Measures. U.S. Department of Transportation, Washington, DC.

Downs, A. 1962. The Law of Peak-Hour Expressway Congestion. Traffic Quarterly 16:393-409.

____. 1992. Stuck in Traffic: Coping with Peak-Hour Traffic Congestion. Washington, DC: Brookings.

Dueker, K.J., J.G. Strathman, and M.J. Bianco. 1998. Strategies to Attract Auto Users to Public Transportation. TCRP Report 40. Transportation Research Board, Washington, DC.

Ferguson, E. 1991. Transportation Management Actions: What Price Traffic Congestion? Transportation Planning 18, no. 2:8-10.

____. 2000. Travel Demand Management and Public Policy. Aldershot, England: Ashgate Publishing.

Gakenheimer, R. and M. Meyer. 1979. Urban Transportation Planning in Transition: The Sources and Prospects of TSM. Journal of the American Planning Association 45:28-35.

Garrett, M. and M. Wachs. 1996. Transportation Planning on Trial: The Clean Air Act and Travel Forecasting. Thousand Oaks, CA: Sage.

Howitt, A.M. and A. Altshuler. 1999. The Politics of Controlling Auto Air Pollution. Essays in Transportation Economics and Policy. Washington, DC: Brookings.

Institute of Transportation Engineers (ITE). 1989. A Toolbox for Alleviating Traffic Congestion. Washington, DC: ITE.

____. 1997. Trip Generation. Washington, DC: ITE.

____. 1997. A Toolbox for Alleviating Traffic Congestion and Enhancing Mobility. Washington, DC: ITE.

Lee, D.B. Jr. 1978. Book Review. Journal of the American Institute of Planners 44:358-61.

Levinson, H.S., M. Golenberg, and K. Zografos. 1987. Transportation System Management-How Effective? Some Perspectives on Benefits and Impacts. Transportation Research Record 1142:22-32.

Mandelbaum, S.J. 1996. Making and Breaking Planning Tools. Computers, Environment and Urban Systems 20:71-84.

Marshall, T. 1997. Futures, Foresight and Forward Looks: Reflections on the Use of Prospective Thinking for Transport and Planning Strategies. Town Planning Review 68:31-53.

McBryan, B., J. Shadoff, D. Pike, and D. Loseff. 1996. Transportation Demand Management: A Guide for Including TDM Strategies in Major Investment Studies and Planning for Other Transportation Projects. Seattle: Washington State Department of Transportation.

Misch, M.R. and J.B. Margolin. 1981. Guidelines for Using Vanpools and Carpools as a TSM Technique. Washington, DC: Transportation Research Board.

Organisation for Economic Cooperation and Development (OECD). 1994. Congestion Control and Demand Management. Paris: OECD.

Orski, C.K. 1987. Can We Manage Our Way Out of Traffic Congestion? Transportation Quarterly 41:457-76.

Pushkarev, B.S. and J.M. Zupan. 1977. Public Transportation and Land Use Policy. Bloomington, IN: Indiana University Press.

Schonfeld, P.M. and H.S. Chadda. 1985. An Assessment of Urban Travel Reduction Options. Transportation Quarterly 39:391-406.

Shadoff, J. TDM Program Manager, Office of Urban Mobility, Washington State Department of Transportation. 1997. Personal Communication. 15 November.

Small, K.A. 1999. Project Evaluation. Essays in Transportation Economics and Policy. Washington, DC: Brookings.

Urban Mass Transportation Administration (UMTA). 1977. Transportation System Management: State of the Art. U.S. Department of Transportation, Washington, DC.

Wagner, F.A. and K. Gilbert. 1978. Transportation System Management: An Assessment of Impacts. U.S. Department of Transportation, Washington, DC.

Walters, A.A. 1961. The Theory and Measurement of Private and Social Cost of Highway Congestion. Econometrica 29:676-99.

Weiner, E. 1999. Urban Transportation Planning in the United States: A Historical Overview. Westport, CT: Praeger.

Zupan, J.M. 1992. Transportation Demand Management: A Cautious Look. Transportation Research Record 1346:1-9.

Address for Correspondence and End Notes

Erik Ferguson, ETF Associates, P.O. Box 888729, Dunwoody, GA 30356. Email: etfassoc@bellsouth.net.

* significant at 0.01 level
** significant at 0.05 level
*** 1.00 < t < 2.00
d.f. = degrees of freedom
HOV = high-occupancy vechicle
NA = not applicable
SOV = single-occupancy vehicle
SRSD = studentized residual
TDM = travel demand management
TMA = transportation management association
TSM = transportation system management
U = data are unavailable
VHT = vehicle hours of travel
VMT = vehicle-miles traveled
* Significant at 95% confidence interval.