Morales-Sarriera, Salvucci 1

Transcription

1 Morales-Sarriera, Salvucci 1 THE RISING COSTS OF TRANSIT AND BAUMOL S COST DISEASE Javier Morales-Sarriera (corresponding author) Massachusetts Institute of Technology 77 Massachusetts Avenue, Building 1-235, Cambridge MA Tel: javierms@mit.edu Frederick P. Salvucci Massachusetts Institute of Technology 77 Massachusetts Avenue, Building 1-270, Cambridge MA Tel: salvucci@mit.edu Word Count: 5,999 words + 6 tables/figures = 7,449 words Submission Date: November 15 th, 2015

2 Morales-Sarriera, Salvucci 2 ABSTRACT This paper analyzes whether the public transit sector suffers from Baumol s cost disease by assessing the evolution of labor productivity and average labor costs across transit agencies in the United States, as compared with other industries. We found that (i) labor productivity in the transit sector has been mostly stagnant over , even more so in bus operations than rail operations (0.0% and 0.7% average labor productivity growth rates, respectively), and even more so measuring output as vehicle revenue miles rather than passenger miles traveled; (ii) the transit sector is highly labor intensive as it represents on average 64% of total costs (operating and capital) for bus and 40% for rail; (iii) compensation per employee has risen at a faster pace than inflation in 85% of the agencies we analyzed; and (iv) compensation per employee has risen at a faster pace than the average local wage rate in 65% of the agencies we analyzed. These findings support the hypothesis that not only does the transit sector suffer from Baumol s cost disease, but also that additional factors contribute to spiraling labor costs. Although there is no clear antidote to the disease, policymakers should recognize that as the economy becomes more productive overall, it can continue to support growing levels of transit service in recognition of its growing external benefits, despite its inherent nature of stagnant productivity growth. Keywords: Baumol s cost disease, transit, productivity, labor costs

3 Morales-Sarriera, Salvucci 3 INTRODUCTION Baumol s cost disease is a theory that has preoccupied economists in the last four decades because of its bleak predictions for industries with little productivity growth over time. The seminal paper by Baumol and Bowen (1) outlining the theory focused on the performing arts industry and argued that its unit cost must continually rise faster than the rate of inflation. Baumol (2) revisited the original paper 30 years later, finding that the cost predictions were confirmed not only in the performing arts, but also in other stagnant industries, such as education, health care, and legal services. In one the most comprehensive empirical analysis of Baumol s cost disease in the United States (US), Nordhaus (3) finds compelling evidence that technological stagnant industries have increasing unit costs and declining outputs relative to the rest of the economy. Moreover, other research, such as Zureiqat (4), Evangelinos et al. (5), and Gordon (6), has argued that Baumol s cost disease also affects the transit sector. The theory by Baumol and Bowen (1) claims that as wages rise in jobs with growing productivity, wages in jobs with stagnant productivity must increase at the same pace, so that all jobs continue to attract workers. As a result, wage growth rates would equalize between industries, despite uneven growth rates in productivity. In the long term, unit costs in stagnant industries are deemed to rise faster than productivity and the inflation rate. In the classic illustration of Baumol s cost disease, the economy may be divided into two industries. On the one hand, Industry A is more progressive and often adopts technology advances that streamline its production process. Industry B, on the other hand, is inherently more labor intensive and has been long operating with the same technology. Let s suppose there is an improvement in the production process in Industry A that results in an increase of 5% in the industry s output per worker. In industry B, output per worker remains constant. The economic theory predicts that wages in Industry A will rise at 5%, at the same level of its labor productivity, maintaining labor cost per unit unchanged. Wages in Industry B will also increase at a similar 5% regardless of its null productivity growth, in order to match the wage increase in Industry A and continue to attract workers. Therefore, the direct labor cost per unit in Industry B will also rise 5%. As a result, Baumol and Bowen (1) argue that the unit cost of output in Industry B must rise boundlessly with respect to the unit cost of output in Industry A. The essence of Baumol s cost disease lies on the different pace of productivity growth between industries. While the overall productivity of the economy grows, some industries inherently perform better than others. According to Nordhaus (3) productivity is directly correlated with technological innovation, therefore sectors such as transit, in which the labor-to-output ratio is fairly constant over time because of the fixed requirement of one operator per vehicle, will experience a faster increase in labor costs than in productivity. In transit, technological progress is available, such as automatic train operations, advanced signaling systems, articulated buses, dedicated bus lanes, or advanced labor scheduling methods that may enhance productivity via speed improvements, expanded vehicle capacity, or automation. However, either because these tools haven t been widely adopted or because they haven t had a remarkable impact on productivity, the ratio of output per worker in transit has tended to lag behind other industries (as evidence in Evangelinos et al. [5] has shown).

4 Morales-Sarriera, Salvucci 4 Baumol (1) also indicates that more labor intensive industries are more likely to suffer from the disease. This happens in part due to low labor mobility that prevents access to lower labor costs, so in a metropolitan area with a strong local economy and rising wages, the transit agency cannot avail itself of labor in lower cost areas, as other industries may do with capital and intermediate goods. Data from the National Transit Database (NTD) shows that the transit sector is indeed highly labor intensive. In 2013, labor costs, including wages, salaries, and fringe benefits, represented on average 73% of bus operating expenses and 66% of rail operating expenses (for the sample of transit agencies used in this paper). After accounting for capital costs, labor represents roughly 64% of total bus expenses (operating and capital), and 40% of total rail expenses. Baumol s cost disease may help explain in part the financial struggle of many transit agencies and the increasing amount of required government support. In fact, transit agencies in the US have an average farebox recovery ratio of about 25-30% in bus and light rail operations, and 40-45% in heavy and commuter rail operations (as a ratio to operating expenditures), an indicator that has been stationary from 2001 to 2013 (based on the NTD). If transit indeed suffers from Baumol s cost disease, its poor financial situation will likely deteriorate even further over time. Increasing service efficiency, improving managerial controls, actively managing collective bargaining, or introducing technology innovations may alleviate the impact of the disease, and the importance of these type of efforts must be underscored in stagnant industries, however they may not reverse its inherent productivity disadvantage. However, because of the overall productivity improvement in the economy, society can afford to continue to support growing levels of transit service, in recognition of growing external benefits of transit. This paper focuses on the analysis of costs and productivity of major US transit agencies in light of Baumol s cost disease. We conduct an analysis of labor productivity trends, followed by an assessment of the evolution of labor costs in the transit industry. Finally, we discuss the results and policy implications. LABOR PRODUCTIVITY IN THE TRANSIT SECTOR In order to test the hypothesis that Baumol s cost disease affects transit, we start by estimating productivity growth. We first follow Evangelinos et al. (5) in assessing the evolution of the productivity (measured as the ratio of gross output to workers) in the transit and ground passenger transportation sector, compared with other industries. We also analyze the evolution of labor productivity in transit agencies across the US based on operational indicators discussed in Fielding (7) and in Meyer and Gómez-Ibáñez (8), based on service and ridership metrics. We then estimate a labor productivity trend using a simple fixed effects panel regression model.

5 Morales-Sarriera, Salvucci 5 Industry-Wide Productivity An analysis of aggregated data, following Evangelinos et al. (5), reveals that US labor productivity (measured as the inflation-adjusted gross domestic output per worker) has grown at an annual compound growth rate of 1.1% over the period On the other hand, labor productivity in the transit and ground passenger transportation industry has declined. These results can be observed in Figure 1, which shows the evolution of an index with the ratio of real domestic industry output to total jobs (part- and full-time), using annual industry data from the Bureau of Economic Analysis (BEA). Figure 1 also shows that among the selected industries, labor productivity growth in the transit and ground passenger transportation industry is only superior to that in education. Among the 195 industries in the US economy (using the official BEA classification), the transit and ground passenger transportation industry has the 9th smallest output per worker growth rates. As expected, more technology driven industries, such as telecommunications and financial intermediation, experienced faster productivity improvements. This aggregated data analysis, nevertheless, has significant limitations because it relies on industry gross domestic output (composed of an average between gross domestic product and gross domestic income). This measure may not be directly relatable to the transit sector, in which output is more often evaluated with indicators of service or ridership. Moreover, the BEA classification of transit and ground passenger transportation includes taxi, charter bus, school bus, employee bus, and interurban bus, which may seriously bias the results of a transit productivity analysis. Unfortunately, BEA data for the transit sector separately is not available. In the next section, we use other measures of labor productivity at the transit agency level in order to overcome the limitations of data aggregation. Productivity in Transit Agencies A measure of labor productivity in transit agencies can be defined in different ways. First, from the supply side, productivity can be calculated as a function of vehicle revenue miles (VRM) or vehicle revenue hours (VRH). Second, from the demand side, productivity may be measured in terms of passenger miles traveled (PMT) or ridership. Third, service quality may also be considered a measure of productivity, taking into account factors such as service frequency and service reliability. For the current analysis we will focus on the first two approaches. Firstly, VRM is a good measure of transit output because it is the decision variable of the agency and is directly impacted by the choice of inputs in the production function (number of employees, vehicles, etc.). Fielding (7) argues that VRH may be a better measure of service supply than VRM, since it may also reflect the expansion of routes into areas with different levels of traffic congestion and speeds. Nevertheless, data on VRH is hardly reliable since transit agencies usually assume a constant average speed for each mode (that can vary widely per year or per agency). We also calculate demand-side labor productivity based on PMT. This is a measure of consumption rather than production, and it is dependent on ridership levels that are not entirely

6 Morales-Sarriera, Salvucci 6 in the control of the agency. Measuring labor productivity via PMT may capture an increase in productivity from exogenous factors, such as population growth, densification, land use changes, or preference changes, which the agency is not directly responsible for. Moreover, it also does not represent a fair assessment of quality, since higher PMT may also denote overcrowding in vehicles. However, it is a good proxy of the transit benefit to society, in other words, the total ridership that can be produced with a unit of labor. Data In order to estimate labor productivity we used data from the NTD over the period in which information was available, between fiscal years 1997 and Since most agencies have multimodal transit systems, we analyzed bus, light rail, heavy rail, and commuter rail separately (we do not consider boat, trolley, paratransit, or other smaller modes). This is necessary because VRM and PMT represent a different scale of output for each mode of transportation. For example, the total VRM that a train operator can sustain is clearly higher than that of a bus operator because, in the former, average speeds are higher, and stations are more widely spaced. Moreover, the NTD calculates revenue miles for a passenger car rather than for the train. A similar logic can be applied for a comparison between light rail, heavy rail and commuter rail. As a measure of employee headcount, we used total full-time employees and part-time employees (combined with a weighted average in which we assume that part-timers work half the hours of full-timers). The fact that the NTD apportions employees among different modes is useful in our mode-specific analysis. We highlight that since we are mostly concerned with the evolution of productivity and not with levels of productivity, different criteria in apportioning administrative personnel between modes would not impact the analysis. Moreover, since only directly operated transit service provides information about employee headcount to the NTD, we were not able to include private operations in the sample. In bus operations, we only consider agencies that have bus revenue service over 10 million miles in any given year between 1997 and This avoids the inclusion of very small transit systems that provide service with significantly different characteristics than the larger agencies in the analysis. Overall, we included 42 agencies with bus service, which represents 51% of the total nationwide bus revenue miles in Another 30% is supplied by 439 small bus transit agencies, and the remaining 19% by privately operated systems for which labor data is not available (of which only 6 agencies would be above the 10 million VRM threshold). For rail operations, we included 16 light rail, 14 heavy rail and 9 commuter rail agencies, which combined represented 93% of nationwide rail transit service in The remaining 7% are supplied by privately operated systems. A full list of the transit agencies and modes included (and the time series available for each) is detailed in Table 1.

7 Morales-Sarriera, Salvucci 7 Productivity: Vehicle Revenue Miles per Employee The descriptive statistics shown in Table 2 reveal that average annual VRM per employee (VRM productivity, henceforth) ranges from 9,600 miles per worker in light rail to 12,300 in heavy rail between 1997 and There is significant heterogeneity across agencies, which is confirmed by the large coefficients of variation (ratio between the standard deviation and the average), from 0.18 in commuter rail to 0.42 in light rail. Light rail has both the lowest levels and highest dispersion of VRM productivity across agencies, which may be explained by several new light rail lines inaugurated during the last two decades, which contrasts with the labor productivity levels of older light rail systems. With regards to bus operations, as general rule, lower levels of VRM productivity are associated with cities with more traffic congestion and less spacing between bus stations. Table 2 also shows descriptive statistics of the average annual compound VRM productivity growth by mode, which ranges between 0.0% in bus operations and 1.0% in commuter rail operations. VRM productivity is stagnant for buses, with only minor fluctuations around a mean of roughly 11,000 miles per employee. Moreover, 7% of the bus transit agencies experienced alarming annualized declines of under 1.0%. On the other hand, only 5% of bus transit systems had growth rates above the estimated 1.1% productivity growth in the US economy as a whole. The results are less discouraging for VRM productivity in rail operations. The highest gains were observed in light rail and commuter rail (annual average growth of 0.7% and 1.0%, respectively). Finally, 18% of the rail systems experienced a decline in VRM productivity over , and 46% experienced an increase in productivity above the estimated nation-wide productivity rate. Finally, heterogeneity across agencies is more apparent in rail than in bus, as shown by the larger coefficients of variation. The findings reveal that the transit industry as a whole is fairly stagnant in terms of VRM productivity. Moreover, its average growth rate in any of mode lags behind the overall economy s productivity growth. The results are consistent with the findings of Evangelinos et al. (4), who also finds that buses tend to have relatively lower productivity growth than trains. Productivity: Passenger Miles Traveled per Employee Descriptive statistics for the ratio between passenger miles traveled to employees (PMT productivity, henceforth) are also shown in Table 2. PMT productivity levels are the highest for commuter rail, followed by heavy rail, light rail and bus, as expected, due to the average capacity in each vehicle type. Once again, we observe that light rail has the highest dispersion (with a coefficient of variation of 0.52), which can also be attributed to new lines that introduced heterogeneity with respect to older light rail systems. The data reveals higher average PMT productivity growth in heavy rail (1.4%) and commuter rail (1.6%) compared to other modes (-0.5% and 0.2% in light rail and bus, respectively) and also compared with their own VRM productivity growth rates. Light rail once again stands out as the outlier, explained in part by the fact that ridership has changed at a slower pace than service supply. The opposite can be observed for heavy and commuter rail, for which ridership on average increases faster than service supply.

8 Morales-Sarriera, Salvucci 8 Overall, the results show relatively higher growth rates of PMT productivity than VRM productivity, expect for light rail. The best results are in commuter rail (67% of the systems had a higher increase of PMT productivity than the US economy), followed by heavy rail (43%), light rail (38%) and bus (22%). On the other hand, roughly 20% to 40% of the agencies in each mode have experienced declines in PMT productivity. Productivity Trends in a Fixed Effect Panel Regression While descriptive statistics reveal the average and variability of labor productivity and of its compound growth rates, a panel regression analysis is better suited to understand the overall trends in productivity by mode. A fixed effects panel regression model allows to control for agency-specific characteristics, to disentangle the variance of the data within and across agencies and, therefore, to better explore the relationship between a time trend and labor productivity. When using a fixed effects model we assume that something within the agency (observable or unobservable) impacts its level of productivity, which needs to be controlled for in order to estimate the impact of the time trend. The equation we estimate is given by: ln(productivity it ) = α 0 + α i + T t β 1 + T t 2 β 2 + u it for i = 1,, n and t = 1,, τ. where the dependent variable is VRM productivity or PMT productivity for agency i in period t, α 0 is a constant, α i is a vector with individual intercepts for each agency i, T t is a time trend for time period t, β 1 and β 2 are unknown parameters, and u it are random errors. We estimated this same equation for each of the four modes. Since we use a quadratic trend to fit productivity growth, we calculate the impact of the time trend at means, that is, the partial derivative of the trend with respect to productivity evaluated at the average year. We then transform this result in terms of the level of productivity (instead of its log-transformed value) in order to facilitate the interpretation of the coefficient. The estimation results in Table 3 show that the average impact of the trend on VRM productivity is slightly smaller (but similar) than the results in the previous descriptive analysis. Furthermore, only the trends associated with light rail and commuter rail are positive and significantly different from zero. For the bus mode, the estimated average annual trend is slightly negative (-0.04%), followed by heavy rail (0.53%), light rail (0.66%) and commuter rail (1.02%). We also highlight that the quadratic component for the trend is negative and significant for bus and heavy rail, which means that VRM productivity growth rates have tended to decline. Table 3 also shows the model results for a specification with PMT productivity as the dependent variable, revealing a positive impact of the trend across modes. The trend for PMT productivity is higher than that for VMT productivity for bus and heavy rail (more significantly for heavy rail, which increases from 0.5% to 1.8%, and it is the only statistically significant trend across modes). Light rail and commuter rail have a lower PMT productivity growth rate, which may be explained by longer trips, more vacant seats, or a mix of both. The quadratic components for the trend in these estimations are not statistically significant and have small values, which reveals no changing PMT productivity trends over time.

9 Morales-Sarriera, Salvucci 9 These results help support the hypothesis that labor productivity growth rates in the public transportation sector are smaller than productivity growth rates in the economy overall, with the exception of heavy rail after considering PMT as the output variable. Moreover, it confirms that the labor productivity growth of bus operations has been mostly stagnant for the period of analysis, while the rail sector has experienced positive but small rates of productivity growth. In conclusion, the low labor productivity growth, especially in the case of VRM productivity, but also in the case of PTM productivity, can constitute a serious threat to the long term financial sustainability of the transit industry, particularly in bus operations. According to Baumol s cost disease, an industry may have bleak cost prospects if it experiences a higher growth in average labor costs than productivity. As a result, transit agencies may face a spiraling of costs with respect to other goods and services in the broader economy. LABOR COSTS IN THE TRANSIT SECTOR We expand the analysis by estimating the impact of Baumol s cost disease on operating costs, moving the focus to the financial accounts of the same agencies previously analyzed. We use agency operating expenses (from the NTD) apportioned by mode in order to assess the level and evolution of labor costs per employee, calculated as the ratio of wages, salaries and benefits (henceforth compensation) to the number of employees (as defined in the previous section). Its evolution can be directly compared with the rate of inflation and with the evolution of average labor costs in the national economy and in each metropolitan area (Quarterly Census of Employment and Wages and Consumer Price Index from the Bureau of Labor Statistics [BLS]). As we have highlighted, the evidence that transit is among stagnant industries indicates that labor costs might face spiraling growth rates above inflation due to the implications of Baumol s cost disease. Table 4 reveals statistics of average compensation in public transportation agencies in It varies from $89,000 for workers apportioned to bus operations to $108,000 for workers apportioned to rail operations, a difference may reflect the premium for a more specialized labor in the latter. The agency that pays the highest average compensation per mode spends approximately $140,000 per worker, while the lowest average compensation is roughly $50,000 for bus transit operations. The dispersion among agencies is similar across modes as the coefficients of variation are roughly Differentials in compensation across agencies may be explained by several factors, including different cost of living in metropolitan areas, higher seniority of workers in older transit systems, different union bargaining power, among others, which results in agencies with average compensation costs two to three times higher than the agency that pays the minimum average compensation. It is important to highlight that the analysis includes all larger and smaller agencies shown in Table 1. In light of Baumol and Bowen s theory (1), the major threat to the sustainability of the sector is not the level of costs, but their growth rates, which are predicted to increase faster than the general rate of inflation. Table 4 reveals that the average compound annual growth rate in nominal labor costs per employee is 3.5% over in bus operations, and between 2.9% and 3.6% in rail operations. The agency with the smallest average labor cost increase had an

10 Morales-Sarriera, Salvucci 10 annual growth rate of only 1.3%, while the agency with largest had a whopping 5.8%, both observed in bus service. The growth rates by mode are clearly above the average US inflation rate of 2.3% and the average US wage growth rate of 2.8%. The data also reveals that, on average, compensation per employee in transit agencies has increased 42% above the inflation rate in rail operations and 52% in bus operations. In fact, 90% of the 42 bus operating agencies had compensation per employee growing above the US inflation rate (or 86% if we consider the metropolitan area inflation for each agency). In the rail mode, the percentage of agencies that have experienced labor cost increases above US or local inflation rates is also high (79%). Table 4 also shows annual growth rates for three sub-periods: , , and The results reveal that the periods between and had the largest increases in compensation per employee (on average 4.2% per year), almost twice as much as the following period (on average 2.0% per year). These variations between periods may in part be attributed to economic cycles. We can also observe that the differential between the inflation rate and the transit labor costs increase was higher in the first sub-period, followed by the second and third. In all three sub-periods, however, average compensation per employee has increased at rates above inflation. Finally, by comparing compensation growth rates in transit and in the overall economy, we find that, on average, it has increased faster in transit agencies (the average compound growth rates are 3.4% for transit and 2.8% for the wider economy). Moreover, from Table 4, we can also observe that about 77% of the agencies in our sample increased compensation above the national wage growth (86% for bus operations only), and about 65% increased compensation above local metropolitan wage growth (71% when considering bus operators only). Finally, the evolution of inflation and labor costs in the transit industry and in the overall economy are shown in Figure 2. The graph shows that average labor costs in transit agencies have consistently increased above the local inflation rate, confirming the predicates of Baumol s cost disease. Moreover, often these compensation growth rates are above the average wage rate growth in the economy. Another important observation is that bus operations seem to have experienced higher growth in compensation compared to the rail mode, although having lower productivity growth rates. This may actually indicate that other institutional factors beyond Baumol s cost disease add to the effects of the disease. DISCUSSION AND POLICY IMPLICATIONS We may draw a few conclusions about transit and Baumol s cost disease. First, labor productivity in the transit sector has been mostly stagnant, especially in bus operations, since agencies have not been able to adopt new disruptive technologies or operating efficiencies that reduce the labor required to supply a fixed capacity, despite heterogeneity across agencies and modes. Second, compensation (wages, salaries and benefits) make up for the majority of the costs in transit agencies, even after accounting for capital costs. Third, labor costs have steadily increased at a faster pace than the inflation rate, oftentimes twice as fast. Fourth, average transit labor costs have grown on average faster than wages in the local economies. All these factors

11 Morales-Sarriera, Salvucci 11 combined indicate a persistent spiraling pattern of costs compared to the costs of other goods and services in the economy. These findings support the hypothesis that transit suffers from Baumol s cost disease. One caveat is that that the transit sector has a different market structure than the industries to which Baumol s (1) model was originally conceived. Transit agencies are, for the most part, monopolistic, and do not face perfectly competitive markets as the traditional examples. Economic theory indicates that monopolies slow down the adoption rate of new technologies, reducing even further a sluggish rate of productivity growth. In addition, the labor supply for transit agencies is also characterized by a monopolistic market, since labor unions often hold significant bargaining power over wages and hiring, which may add another layer to the spiraling costs. The monopolistic nature of transit can only add to its spiraling cost trend. An indication of the additional effect is that the growth in compensation per employee in transit agencies has often exceeded the wage growth in the local economy. In this sense, Baumol s cost disease may only predict labor costs growing at the average economy levels, therefore additional effects related to the monopolistic characteristics of the sector must be driving further the increase in labor costs. On a different note, we can estimate the impact of the spiraling labor costs on the financial sustainability of transit agencies. If compensation per employee had risen at the same pace as local inflation since 1997 (expected from a stagnant industry in the absence of Baumol s cost disease), labor costs would be about 19% lower than its current value, on average. This total represents roughly $3 billion in our sampled agencies and modes for fiscal year 2013 (it might reach $3.8 billion after considering all US transit agencies and modes). This amount could have been used to produce more service, which could improve frequency and comfort, and even more employment that transit labor unions might be interested in. Stimulating productivity appears to be the main mitigation strategy for Baumol s cost disease. For example, adopting automation, updating of vehicle fleets to avoid excessive maintenance costs, creating exclusive bus lanes to increase the speeds, or using more efficient scheduling and routing tools. Nevertheless, we must recognize that transit is, essentially, labor intensive, and thus no disrupting technology is likely to significantly alter this characteristic. The spiraling trend of transit costs imposes a serious threat for the quality and service that transit agencies can provide. On the one hand, raising fares to support higher levels of cost recovery may significantly impact ridership levels and make transit unaffordable for many (currently, fares usually increase roughly at the inflation rate, according to NTD data). On the other hand, in a scenario where neither fares nor subsidies increase above inflation, agencies will have to continuously cut back on service quality, i.e., frequency, reliability, and capacity. Therefore, there are at least two challenges in transit finance that policymakers must overcome. First, understanding that the costs of providing transit are rising, partially explained by Baumol s cost disease, and, as a result, that the budgets for transit agencies often grow above the rate of inflation in the absence high productivity growth. A second challenge is avoiding

12 Morales-Sarriera, Salvucci 12 cyclical government financial support, since transit agencies are hardly likely to adapt and downsize during economic cycles that drive tax revenue or political will to fund transit. Karner et al. (9) found that economic downturns do have a direct impact on transit productivity, a result of budget cuts and shortfalls. Therefore, maintaining service capacity during busts is important to avoid long term loss in productivity. In spite of these gloomy prospects, the bigger picture behind Baumol s cost disease is that although transit becomes more expensive relative to other goods and services, productivity in the overall economy grows, and society becomes richer as a whole. In other words, as the economy grows, society can continue to afford relatively more expensive goods and services since consumption of other goods becomes relatively cheaper. Baumol (2) argues that, in fact, while the unitary price of lagging industries continue to rise (and their exchange rate against industries with increasing productivity), the cost it takes society (in terms of hours of labor) to purchase these services is steady over the long run. It is therefore conceivable that policymakers choose to prioritize transit funding, coupled with measures that may help alleviate the effects of Baumol s cost disease. First, adopting new cost-effective technologies that help increase productivity, reducing the implications of the cost disease. Second, increasing fares at the same pace as inflation something that most agencies do, according to NTD data. Third, active administration of labor-management relations, with potential adjustments on work rules and labor practices. Finally, we should point out that Baumol s cost disease may only describe the spiraling costs of transit, disregarding its benefits in the form of externalities in the urban environment. Over time, investment in transit will continue to produce a stream of benefits to society, such as urban densification, displacement of auto ownership, alleviation of traffic congestion, beyond other additional externalities such as reduced pollution emissions, and improved road safety. Although these external benefits can hardly be captured by transit agencies, they clearly produce an additional value which is increasing over time and captured by commuters, drivers, firms, landowners, among other groups. Although these externalities have been documented in theoretical models, they are not easily quantifiable. Among the few studies that have attempted to do so, Antos (10), for example, has quantified transit externalities for the Chicago region. Pushkarev and Zupan (11) indicated that a city can only grow to a fixed urban development density based on auto, and at the margin, all additional density has to be supported by public transportation. In addition, Graham (12) has led the path in the quantification of agglomeration benefits related to these urban densities which are supported by transit investment. Since most transit benefits are external to the agencies, the sector may require added taxpayer support to continue and to expand service. But the tendency for subsidy levels to grow at a higher rate than inflation undermines the credibility of the transit agency, and makes it more difficult to generate the political will to support the growing level of transit subsidy required. A better recognition of the Baumol effect, and clear strategies to contain it, may help to increase the public understanding and support for the need for growth in transit subsides.

13 Morales-Sarriera, Salvucci 13 Therefore, comparing the evolution of transit costs and of transit benefits, after considering Baumol s cost disease and a stream of transit externalities, may an interesting field for further research. Furthermore, also as future research, the evolution of transit costs in other agencies outside of the US remains to be analyzed, since other countries may have been more successful in containing higher costs when facing low productivity growth. Moreover, it also important to further understand and compare the evolution of costs between transit agencies with unionized and non-unionized labor, and explore whether services contracted out have different cost growth experience than directly operated service. Finally, another area of further research may use service quality as a dimension of productivity, for example, an increase in service frequency would most likely hurt productivity as measured from the supply or demand side, however, a service quality improvement such as increased frequency would directly affect the value of the transit system to its users. REFERENCES 1. Baumol, W., and W. Bowen. Performing Arts, the Economic Dilemma: a Study of Problems Common to Theatre, Opera, Music, and Dance. Twentieth Century Fund, New York, Baumol, W. J. Children of Performing Arts, The Economic Dilemma: The climbing Costs of Health Care and Education. Journal of Cultural Economics, Vol. 20, 1996, p Nordhaus, W. D. Baumol s Diseases: A Macroeconomic Perspective. The B.E. Journal of Macroeconomics, Vol. 8, Zureiqat, H. Baumol s Cost Disease in Public Transit: Historical Evidence and Future Implications. Presented at Kuhmo Nectar Conference and Summer School, University of Urbino, Evangelinos, C., B. Wieland, and T. Khunhausen. Baumol s Cost Disease in the Local Transit Sector: a Comparative Analysis for Germany and the USA. International Journal of Transport Economics, Vol. 31, 2012, p Gordon, M. Developing Strategies for Resource-Constrained Transit Growth through Increased Private Sector Involvement. Master Thesis, Massachusetts Institute of Technology Fielding, G. Managing Public Transit Strategically. Jossey-Bass, San Francisco, Meyer, J. R., and J. A. Gómez-Ibáñez. Measurement and Analysis of Productivity in Transportation Industries. In New Developments in Productivity Measurement (J. W. Vaccara and B. N. Kendrick, eds.), University of Chicago Press, Chicago, 1980, pp Karner, A., A. Urrutia, and D. Niemeier. US Public Transit Fantasies: Performance and Economic Stimulus. International Journal of Transport Economics. Vol. 39, 2012, p Antos, J. Paying for Public Transportation: the Optimal, the Actual, and the Possible. Master's Thesis, Massachusetts Institute of Technology, Pushkarev, B., and J. M. Zupan. Public Transportation and Land Use Policy. Indiana University Press, Bloomington, Graham, D. Agglomeration, Productivity, and Transport Investment. Journal of Transport Economics and Policy, Vol. 41, 2007, p

14 Morales-Sarriera, Salvucci 14 LIST OF TABLES AND FIGURES Figures FIGURE 1 Industry Gross Domestic Output per Worker..15 FIGURE 2 Evolution of Average Compensation in Transit Agencies Vis-à-Vis Inflation and Average Compensation in the Overall Economy..20 Tables TABLE 1 Description of the agencies, modes, and time periods used in the analysis..16 TABLE 2 Summary Statistics of Labor Productivity in US Transit Agencies ( ).17 TABLE 3 Regression Results for Labor Productivity (Fixed Effects Model)...18 TABLE 4 Summary Statistics of Compensation per employee in US Transit Agencies..19

15 Morales-Sarriera, Salvucci Colleges, universities, professional schools Construction Retail trade Telecommunications Source: Own calculations, based on Bureau of Labor Statistics Transit + ground passenger transportation Real estate Monetary authorities, credit intermediation Total Economy FIGURE 1 Industry Gross Domestic Output per Worker (constant dollars, index 100=1990)

16 Morales-Sarriera, Salvucci 16 TABLE 2 Description of the agencies, modes, and time periods used in the analysis Agency Name State Bus LR HR CR Los Angeles Metropolitan Transportation Authority (LAMTA) CA Sacramento Regional Transit District CA San Diego Trolley CA San Diego Metropolitan Transit System (MTS) CA Alameda-Contra Costa Transit District (AC Transit) CA San Francisco Bay Area Rapid Transit District (BART) CA San Francisco Municipal Railway (MUNI) CA Santa Clara Valley Transportation Authority CA Orange County Transportation Authority (OCTA) CA Southern California Regional Rail Authority (Metrolink) CA Denver Regional Transportation District CO Washington Metropolitan Area Transit Authority (WMATA) DC Broward County Transit FL Central Florida Regional Transportation Authority (LYNX) FL Miami-Dade Transit FL Metropolitan Atlanta Rapid Transit Authority (MARTA) GA Honolulu Department of Transportation Services HI Chicago Transit Authority IL Pace - Suburban Bus Division (PACE) IL Northeast Illinois Regional Commuter Railroad Corporation (Metra) IL Northern Indiana Commuter Transportation District (NICTD) IN New Orleans Regional Transit Authority LA Massachusetts Bay Transportation Authority (MBTA) MA Maryland Transit Administration (MTA) MD Ride-On Montgomery County Transit MD Suburban Mobility Authority for Regional Transportation (SMART) MI City of Detroit Department of Transportation (DDOT) MI Metro Transit - Minneapolis/St. Paul MN Metro Transit - St. Louis MO Charlotte Area Transit System NC New Jersey Transit Corporation (NJ TRANSIT) NJ Transit Alliance (GTJC) NY Metro-North Commuter Railroad Company (MTA-MNCR) NY Metropolitan Suburban Bus Authority (MTA Long Island Bus) NY MTA Bus Company (MTA BUS) NY MTA Long Island Rail Road (MTA LIRR) NY MTA New York City Transit (NYCT) NY Niagara Frontier Transportation Authority (NFT Metro) NY Port Authority Trans-Hudson Corporation (PATH) NY Port Authority Transit Corporation (PATCO) NY Staten Island Rapid Transit Authority (MTA Staten Island Railway) NY Southwest Ohio Regional Transit Authority (SORTA) OH The Greater Cleveland Regional Transit Authority (GCRTA) OH Tri-County Metropolitan Transportation District of Oregon OR Port Authority of Allegheny County PA Southeastern Pennsylvania Transportation Authority (SEPTA) PA Dallas Area Rapid Transit (DART) TX Metropolitan Transit Authority of Harris County (Metro) TX VIA Metropolitan Transit (VIA) TX Utah Transit Authority (UTA) UT Hampton Roads Transit VA King County Metro Transit WA Milwaukee County Transit System (MCTS) WI

17 Morales-Sarriera, Salvucci 17 TABLE 2 Summary Statistics of Labor Productivity in US Transit Agencies ( ) Measure Statistic Bus Light Rail Heavy Rail Commuter Rail Average 11,319 9,619 12,333 11,408 Vehicle Revenue Miles Median 11,206 8,801 11,825 10,916 (VRM) per Employee Coefficient of variation Annual VRM Productivity Compounded Growth Rates, by agency Passenger Miles Traveled (PMT) per Employee Annual PMT Productivity Compounded Growth Rates, by agency Source: Own calculations based on the National Transit Database Average 0.0% 0.7% 0.5% 1.0% Median 0.0% 0.9% 0.9% 1.0% Std. Deviation 0.7% 1.6% 2.2% 0.9% Average 123, , , ,318 Median 120, , , ,345 Coefficient of variation Average 0.2% -0.5% 1.4% 1.6% Median 0.6% 0.6% 0.7% 1.5% Std. Deviation 1.3% 3.3% 3.3% 1.8%

18 Morales-Sarriera, Salvucci 18 TABLE 3 Regression Results for Labor Productivity (Fixed Effects Model) Dependent Variable: ln(vrm/employees). Years: Bus Light Rail Heavy Rail Commuter Rail T t 0.011** * (0.003) (0.012) (0.014) (0.007) 2 T t ** * (0.0002) (0.0007) (0.0007) (0.0004) α ** 9.06** 9.26** 9.23* (0.01) (0.05) (0.07) (0.03) ln ( VRM Labor ) t ( VRM Labor ) t * ** (0.001) (0.004) (0.005) (0.003) -0.04% 0.66% 0.53% 1.02% Observations Groups ρ F-Statistic Dependent Variable: ln(pmt/employees). Years: Bus Light Rail Heavy Rail Commuter Rail T t (0.006) (0.020) (0.026) (0.019) 2 T t (0.0004) (0.0011) (0.0012) (0.0010) α 0 ln ( PMT Labor ) t ( PMT Labor ) 11.71** 12.26** 12.28* 12.77** (0.02) (0.08) (0.12) (0.08) ** (0.002) (0.007) (0.008) (0.006) 0.25% 0.21% 1.83% 0.89% t Observations Groups ρ F-statistic Own calculations with data from National Transit Database. ρ is the ratio of the variance across panels to total variance. Standard errors in parentheses. * Significant at 90% confidence level ** Significant at 95% confidence level

19 Morales-Sarriera, Salvucci 19 TABLE 4 Summary Statistics of Compensation per employee in US Transit Agencies Average compensation per employee in 2013 (US$) Average annual compounded growth rate of compensation per employee Percentage of agencies with growth in compensation per employee above Bus Light Heavy Commuter Wage Inflation Rail Rail Rail rate Average 89,295 95, , , Maximum 142, , , , Minimum 52,236 59,236 68,943 76, Coeff. of variation % 3.6% 2.9% 3.4% 2.3% 2.8% % 4.9% 3.2% 4.7% 2.3% 2.9% % 3.5% 4.8% 4.0% 3.0% 3.6% % 2.5% 1.2% 1.3% 1.7% 2.0% Inflation (USA) 90% 81% 71% 88% - - Local inflation 86% 81% 71% 88% - - Wage growth (USA) 86% 69% 64% 75% - - Local wage growth 71% 63% 50% 63% - - Source: Own calculations with data from the National Transit Database, and the Quarterly Census of Employment and Wages and Urban Consumer Price Index from Bureau of Labor Statistics.

20 Morales-Sarriera, Salvucci Average Compensation (Transit Agencies) Average Compensation (Civilian workers) Inflation Rate Source: Own calculations, based on National Transit Database and Bureau of Labor Statistics FIGURE 2 Evolution of Average Compensation in Selected Transit Agencies Vis-à-Vis Inflation and Average Compensation in the Overall Economy (indices, base year=1997)