The High Cost of the Purple Line Light Rail Transit Project: Free Bus Service Would Be Cheaper For Everyone, and Provide a Better Service

Purple Line Costs vs BRT

A.  Introduction

The Purple Line is a proposed light rail transit project that would thread itself through suburban neighborhoods over 16 miles in an arc from the east of Washington, DC, to its north.  It is a controversial project, but with strong political pressure to sign soon a contract with a private concessionaire who would construct and then operate the rail line over a 30 year life.  The aim is to begin construction in 2015, complete construction by late 2020, and open the line to ridership by early 2021.

The project is controversial for several reasons.  There are environmental and noise concerns, as a portion of the line will be routed over what is now a park (on an old, abandoned, rail line) with a walking and biking trail that is the most popular in Maryland in terms of usage.  Two parallel rail lines would be built on this trail, with a new trail then built alongside the tracks, necessitating the clear cutting of the mature trees along the trail to allow for the much wider right of way.  There will also be major noise issues, as frequent trains (every 10 minutes during the off-peak hours, and every 6 minutes during the peak) will go by, until 3:00 am on weekends and starting at 5:00 am on week-day mornings.  Homes now backing on to a quiet park will instead have to contend with the noise of the frequent passing trains.  No compensation will be provided to those adversely impacted, and it should not be surprising that they, as well as others, are opposed.

The line is also expensive.  The most recent estimate, from July 2014, puts the capital cost alone at $2.4 billion, with annual operating costs then of $58 million.  But the Purple Line will only serve suburban neighborhoods of medium to low density, so ridership will not be high.  The cost estimates are of course only estimates, and the final costs will not be known until the work is completed (when it is too late to do anything).  Based on past experience with such projects, one should expect that the final costs will be substantially higher than these estimates.  And as will be discussed below, the published cost estimates do not even cover all of the costs that will be incurred for the Purple Line.  Finally, even these estimates have increased substantially from what they were initially.  As late at June 2007, with initial design work well under way and alternatives being considered, the estimated capital cost was only $1 billion.  Subsequent estimates were $1.5 billion (in August 2009), $1.9 billion (in September 2011), and $2.2 billion (in September 2012).  The most recent estimate is $2.4 billion.  Few will be surprised if this goes higher, and perhaps much higher.

These cost totals by themselves do not tell us much, however, unless they are put in the context of how many riders will use the system.  While thousands of pages of documents have been posted on the web on the proposed project, with the Final Environmental Impact Statement (FEIS, August 2013) the most comprehensive review, I have not been able to find any serious economic analysis of the project, nor of the alternatives to provide such transit services.  The FEIS does describe in great detail a set of alternatives it states they considered, and I am sure such work was done.  There are full chapters in the FEIS on the alternatives (see in particular Chapter 2 and Chapter 9).  But figures are not presented which would allow one to compare one alternative to another.

Evaluating major projects such as this is something I did during my career at the World Bank.  This blog post will summarize estimates I have made of what the full costs of the Purple Line will be, and will compare these to some alternatives.

B.  The Cost of the Purple Line

A transit project such as the Purple Line will incur both upfront capital costs to build the system, and then annual operations and maintenance (O&M) costs to operate it.  Ridership will start only once the system is built, and then should grow over time.  Determining the full cost of the system per boarding (one rider getting on board for one trip) is therefore complex.  While it would be easy to determine the annual O&M costs per boarding once the system is up and running, one should not ignore the up-front capital costs that are incurred.  And since the capital costs are incurred up-front, there will be interest costs, either explicit (for what the private operator borrows) or implicit (if government grants are used –  but such funds will still need either to be borrowed or to come from some other use, so there will be an opportunity cost in such usage of the limited funds available).  One cannot simply ignore the costs of these funds, and yet the published analysis appears to do just that.

One therefore needs to use a spreadsheet which separates out by year when the costs are incurred (both capital and O&M costs), and when the ridership occurs.  One can then calculate what the cost would be per boarding which, over some given lifetime, would cover the full costs incurred by building and then operating the Purple Line.  If riders are charged this cost per boarding (and assuming the projected ridership would still be the same, even though such a fare was charged), the system would cover its costs from the ridership.  While transit systems rarely cover their full costs from the fare box, one will still need to know what this cost will be to judge whether the system is worthwhile, as well as to judge whether some alternative would be a better use of the funds.

The Technical Note at the bottom of this post describes in some detail the methodology followed, the sources for the data used, and the assumptions then made.  The end result is that the estimated full cost for the Purple Line comes out to be $10.42 per boarding, in terms of constant dollars of 2012.  This is a lot.  The riders on the Purple Line will mostly be making only short trips of just a few miles, connecting to Metrorail lines and/or traditional bus routes to get to and from work.  At $10.42, private taxi service would likely normally to be cheaper.

The busiest portion of the route is expected to be between Silver Spring and Bethesda, connecting two business centers each on two effectively separate Metrorail lines (although in fact they are the same line, after looping through downtown Washington, DC).  This is the portion of the route that would destroy the existing park.  It is only 4.3 miles long, and the time savings would be small.  Existing local bus service between these two points only requires 17 minutes, and that is during rush hour.  The Purple Line light rail service would require 9 minutes, producing a savings of only 8 minutes.

It is expected that few if any travelers would ride the full 16 miles of the line.  Traveling that route on the Purple Line would take an estimated 63 minutes based on the current design.  But one could travel between the same two points on the existing Metrorail service in 51 minutes now, during rush hour.  The Purple Line is designed for local service.

Riders would of course not pay that $10.42.  If they were charged such fares for the short trips being taken, very few would take the Purple Line (as noted, taxis would likely be cheaper).  The FEIS (Chapter 3, page 3-8) estimates that the additional fare box revenue in 2040 (but in 2012 dollars) would be $9,615,564 (which is more precise than one would think they intend).  Based on the FEIS ridership projections, this comes to just 38 cents per boarding.  It is so low because most of the riders would be transfers to and from Metrorail and traditional bus services, or would displace ridership on existing services.  Transfers pay zero or small additional fares.

The cost per boarding of $10.42 and the fare per boarding of $0.38 implies that the subsidy that would be provided to those riding the Purple Line would be $10.04 per boarding.  These figures are shown in the chart at the top of this post.  A subsidy of over $10 per ride is huge.

C.  Comparison to a Bus Rapid Transit System for Montgomery County

To put the $10.42 per boarding cost of the Purple Line in perspective, one needs to look at alternative forms of transit.  Montgomery County, Maryland (through which roughly half of the Purple Line will run) is also looking closely at use of Bus Rapid Transit (BRT) systems for certain of its public transit routes.  A consultant’s report completed in 2011 commissioned by the county provides figures that can be used to provide perspective on the Purple Line costs.

A Bus Rapid Transit system provides high-capacity and streamlined bus services along selected routes.  By use of larger buses, dedicated stations where one will pay the fares before boarding (thus streamlining boarding), various road improvements and perhaps dedicated bus lanes, one can provide transit services that are significantly faster than, and more comfortable than, traditional bus services.

The Montgomery County BRT study looked at a system whose capital cost came to an estimated $2.4 to $2.6 billion (in 2012 dollars).  This was roughly the same, coincidently, as the current estimated cost of the Purple Line Light Rail project.  But what one would obtain for that similar investment would be far more:

Comparison of Purple Line to BRT BRT Purple Line Difference
Capital Cost $2.4 to $2.6b $2.43b similar
Number of routes 16 1 16 times
Number of miles covered 150 16 9.4 times
Daily boardings, 2040 (mid-point) 186,300 59,130 3.2 times
O&M cost per boarding (mid-point) $2.424 $2.688 10% less
Total cost per boarding $4.16 $10.42 60% less

The Montgomery County BRT system would cover 16 routes, versus only one for the Purple Line.  It would cover 150 miles, versus only 16 for the Purple Line.  The projected daily boardings in 2040 of 186,300 (based on the mid-point of the range projected) would be over three times the 59,130 projected for the Purple Line.  And the operational and maintenance (O&M) costs per boarding (again based on the mid-point of the range in the BRT study) would be 10% less.  Normally one justifies the higher capital expenditures per mile of a rail system by its then lower O&M costs.  But the O&M costs of the Purple Line would be higher.

The full cost (including capital costs) per boarding of the BRT system is then far below the cost of the Purple Line.  As discussed above, the estimated full cost of the Purple Line would be $10.42 (in 2012 dollars).  Using a similar methodology, but with the BRT cost and ridership estimates, the full cost of the BRT system would be $4.16 per boarding, or 60% less.

The BRT system would be a far better investment, then, of the scarce transit dollars available.  Many more people would be served, at a far lower cost.  For the Purple Line corridor itself, various BRT systems (as well as alternative light rail systems and other options) were examined by the Purple Line consultants, but rejected in favor of the light rail system selected.  However, I cannot find in any of the thousands of pages of documentation now posted any presentation of figures on the total cost per boarding of a light rail system versus a BRT for the selected route.  It is not clear if this was ever examined.  And some have argued that the BRT alternative was never seriously considered as an option, but rather that the light rail approach was chosen early, with the analysis then done by the hired consultants directed at justifying this choice.

It is possible that the BRT alternative was rejected for the Purple Line corridor due to the nature of the streets it would pass through, in particular on the Prince George’s County portion of it.  However, a BRT would likely work quite well for the section between Silver Spring and Bethesda, where there is a four-lane major road connecting the two centers.  A BRT could simply run along that.  A BRT would also provide an option to loop up to another major employment center just north of Bethesda, where the Naval Medical Center and headquarters (and main labs) of the National Institutes of Health is located.  The proposed light rail system would not do that.

Use of a BRT line between Silver Spring and Bethesda would also mean that the linear park between the two would not be destroyed.  A hybrid system of light rail up to Silver Spring, and then BRT between Silver Spring and Bethesda, would be a possible compromise.  The BRT could then join up with north-south BRT lines being planned separately for Bethesda, as well as BRT lines being planned for Silver Spring.

D.  A Cheaper and Better Alternative:  Free Bus Service

As noted above, the subsidy of over $10.00 per boarding for the Purple Line is huge.  The cost will be borne in one form or another (either capital subsidies or operational payments) by the government, and hence ultimately by the taxpayer.  Recognizing that government would be providing a subsidy of $10.00 per boarding to transit users in this corridor, provides a new and better perspective on how best to provide transit services.  Instead of asking the question of how much will it cost to build and then operate a light rail transit line, the question shifts to how best to use the funds that would be made available for transit in this corridor.

When one looks at the issue this way, one alternative stands out:  Why not simply charge a zero fare for bus service along the Purple Line corridor (and perhaps more broadly)?  While I was not able to find figures to allow a calculation of the full cost of operating a traditional bus system in an area of similar density as the Purple Line corridor, the cost should be expected to be less than the cost of a BRT system in Montgomery County.  That is, the cost will likely be less than $4.16 per boarding.

And note that with no fare being collected, there will be at least two additional advantages gained over current bus service.  First, the new bus system will have a similar advantage in terms of speed as a BRT system.  BRT buses are able to move more quickly on regular roads primarily because they can load passengers quickly, since fares have already been paid at the special bus stations built at each stop along a BRT line.  But if no fares are being collected, one can simply get on a traditional bus quickly, with no delays due to people lining up to pay their fare.  Over time, one could also replace current buses with ones with multiple entrances and exits, since everyone would not need anymore to pass through the front door by the driver, to ensure fares were being paid.  This would allow even speedier boarding.

Second, collecting individual fares is costly in itself.  Cash fares need to be kept secure and later counted and deposited, and one needs special equipment and technology to keep track of fares paid by those using electronic smart cards or similar devices.  In addition, speedier bus trips mean that the number of driver-hours one needs to pay for (the most significant expense in operating a bus system) will be reduced in per rider terms.  Both of these factors reduce costs, and significantly so.

But even assuming the traditional bus system will have full costs of $4.16 per boarding (the same as the BRT), one could still carry 2.4 times as many passengers as the Purple Line would carry, for the same net cost (of $10.04).  With a likely cost of well less than $4.16 per boarding, one could carry even more.  And with a larger number of riders, a higher frequency of bus service on each route (say every five minutes instead of every 15 minutes) could then be supported.  Free fares for riders coupled with more frequent service would then be expected to attract even more riders, and possibly many more.  The main concern public officials should probably have is that such bus service would become so popular that many more than 2.4 times as many riders would want to ride the system.  While economies of scale (more riders on each bus, on average) will reduce costs per rider to even less, a large number of new riders eager to take buses is a “problem” that most public officials would welcome.

One would then also expect that such ridership shifts to public transit would start to have a significant impact on car usage and hence road congestion, even with additional bus service.  An individual bus with reasonable ridership levels displaces many cars from the roads along the corridor.

Even if it were argued that such a shift to free and frequent bus service were not possible for much of the Purple Line, it is clear that it would work well for at least the Bethesda to Silver Spring section.  As noted above, there is an existing four lane road, and even during congested rush hour traffic, the current traditional bus line (with its frequent stops, and passengers lining up at each stop to step aboard and pay their fare) only requires 17 minutes.  This could be sped up significantly with a shuttle service where no fare is paid (so need to line up to pay it) and perhaps a limited number of stops.  Such a service would likely match or almost match the 9 minutes the Purple Line light rail system would require for this 4.3 mile segment.  Furthermore, one could start to offer this free shuttle service immediately.  There is no need to wait until 2021 for the Purple Line to be built.  This alternative would also save the park that the Purple Line would destroy, and the residents whose land now backs on to this park would not need to contend with the noise of rail cars passing their windows every 10 minutes until midnight, and until 3:00 am on weekends.

E.  Reality Check:  Why the Better Solution is Unlikely to be Followed

So far, the analysis above has kept to what would make most sense to provide transit services along the corridor the Purple Line would serve.  But just because a simpler, cheaper, and better service might be available, does not mean that it is likely to be done.  There are at least three reasons in this case:

a)  Bureaucratic rules:  Government support for transit projects is biased to providing capital support to build things, rather than operational support to run things.  State and especially federal government support is biased in this way.  This creates distortions when decisions are made, as an option requiring much up-front capital will be favored over a solution which instead has primarily on-going operational expenses.  Funds for the capital investment may be available as a grant, while operational expenses are not covered (or are not covered to the same degree).

There would likely be such an issue here, as the state and federal funding is focussed on providing grants for construction.  Those advocating the expensive light rail system will argue that while they can get these funds for construction, they could not obtain such funds to operate improved bus services along this corridor.

But these are bureaucratic rules.  Such rules can be changed.  If a cheaper option than a light rail system (such as free and frequent bus service) provides a better solution, then elected politicians should be able to find a way to make this possible.

b)  Some parties will gain by an expensive light rail system:  Even though transit users as well as taxpayers might lose by building the expensive option, there are some groups that may gain.  Two in particular should be noted.  One is developers who own land parcels close to the proposed stations of the Purple Line.  These parcels will gain significantly in value as transit users are channeled to those locations (and not to others), with land values that may well rise by hundreds of millions of dollars.  Someone else will be paying the $2.4 billion construction cost.

The second is the group of private construction and engineering companies that will participate in the construction, as well as the ultimate concessionaire.  Profits on a $2.4 billion project are substantial.

c)  The embarrassment factor from admitting your choice was wrong:  Finally, one should not neglect that politicians and others will be extremely reluctant to admit that they made a mistake on a project they had previously supported and indeed championed.  But they should not be criticized if they recognize that the information they had before was perhaps insufficient, or that conditions have changed as more information has been gathered.

The Governor of Maryland announced in August 2009 that a light rail line would be the “locally preferred alternative” for the corridor the Purple Line would serve.  At that time, the capital cost was estimated to total just $1.5 billion, with construction that could start in 2013 and be competed by 2016, and with projected daily boardings of 64,800 by 2030.  But the current estimates are that the capital cost will come to $2.4 billion (60% more), construction will not begin until 2015 and only be completed in 2020 (four years later), and that daily boardings now projected for 2030 are only 53,000 (18% less).

Estimates are of course only estimates, and one cannot know for certain beforehand what the costs and ridership will be, nor how long it will take to build such a system.  But how high do the costs need to go before one agrees that earlier decisions need to be reconsidered?  A 60% increase is not small.

One way to resolve this:  Why not hold a vote?  Arrange for a ballot referendum in the areas impacted, where the population would be allowed to vote on whether they prefer the Purple Line light rail system (to be built as currently proposed, and with regular fares then to be paid to ride it), or the alternative of using the funds to provide free bus service along this corridor, starting immediately.  Since the issue is one of service preferences, as the costs would be similar, the general population should be given a say in how the funds are utilized.

 

———————————————————————————

Technical Note on Methodology, Data, and Assumptions Used

This technical note presents in some detail the methodology, sources of data, and assumptions made, to come up with an estimate of the full cost per boarding of the proposed Purple Line Light Rail transit project.  The basic approach is to develop a spreadsheet which estimates the full costs (for each year over the lifetime of the project) of building and then operating the rail line.  One then subtracts from these costs what would need to be “charged” per boarding, so that the “revenues” thus generated (given the ridership estimates) will suffice so that the project will have paid for itself in full by the end of the time horizon chosen.  The “shadow fare” thus computed is not the fare that would actually be charged, but rather the cost per boarding that would need to be covered for the full cost of the project to be covered by the end of the time horizon.  Riders are not in fact charged this fare, but rather something far less.  The purpose of the exercise is to calculate what the full cost per boarding will be.

The spreadsheet needs to break out the costs by year since, like any project, capital costs are incurred up-front, ridership starts only when the project is completed, and ridership generally will grow over time as the region grows and develops.  Annual operations and maintenance budgets will also grow over time to cover the costs incurred from carrying more riders (with more frequent train service, for example).

Importantly, because major capital expenses are incurred up front, there will be a cost from providing the necessary funds up front, to be repaid only later.  These will be interest costs.  These interest costs will be incurred whether the project itself borrows directly the funds necessary for the construction, or if some level of government (federal, state, or local) provides the funds as a grant.  The grant funds need to come from somewhere, and governments need to borrow.  Even if the governments were currently running a budget surplus, they could have used the funds being provided to the transit project instead to pay down some of the government’s existing outstanding debt, or for some other use.  Economists call this the opportunity cost of capital, and it exists even when the transit project itself is receiving the funds as a grant.  This cost cannot be ignored, even though it often is.

Thus the basic structure of the spreadsheet starts by accounting for the capital costs during the construction period, by year, and including the interest costs incurred (implicit or explicit) to cover those capital costs (and after the first period, also the costs of covering the accumulated interest itself).  The construction period is primarily 2015 to 2020 according to the current planned schedule.  Operation then begins in early 2021, with annual operations and maintenance costs starting then and ridership beginning.  Since the current plan is to provide a concession to a private firm to build and operate the system, with the operations concession lasting for 30 years from the end of the construction period, the spreadsheet was used to determine what “shadow fare” would be necessary so that at the end of this 30 year concession, the “revenues” thus generated (given the ridership projections) less the annual operations and maintenance expenditures, would have covered the up-front capital costs incurred (along with accrued interest on the outstanding annual balances).  An iterative process was used to arrive at that shadow fare.  That shadow fare will be the full cost incurred, per boarding, of this light rail line.

The calculations were done all in current dollar terms.  That is, certain inflation rates were assumed and the implicit interest rate on the capital costs was defined in nominal terms.  However, all the figures reported here on cost per boarding are expressed in terms of prices of 2012.  One could have set up the spreadsheet to do all the calculations in real, inflation-adjusted, terms, but the results (if everything was done correctly) would be the same.  For the purposes here, working in current price (or nominal) terms, was simpler.

Data were taken from the documents posted on the internet for this project.  Most important were the most recently updated summary sheet from the US Federal Transportation Agency (FTA) of July 2014; the Final Environmental Impact Statement (FEIS) of August 2013, in particular its Chapters Two, Three, and Nine, plus its Volume III Technical Report on Capital Costs; and the “Request for Proposals (RFP) to Design, Build, Finance, Operate, and Maintain the Purple Line Project”, issued by the State of Maryland in July 2014.

While one would have expected that with all these reports, totaling thousands of pages, the project designers would have made available a spreadsheet of their own with the expected costs by year as well as ridership.  But the information from such a spreadsheet does not appear to have been posted.  I am sure they would have themselves made such calculations, but they evidently chose not to make them available to the public.  I therefore had to make various estimates of my own, drawing on the figures they did make available and anchoring the projections in the figures they provided for only certain of the outlying years (most commonly 2035 or 2040).

Due to the inherent uncertainties in all this, I erred on the side of conservatism whenever assumptions needed to be made.  That is, I aimed to err on the side of keeping estimated costs low, so that the estimated cost per boarding (in 2012 dollars) of $10.42 in the base case is probably low.  The true figure will probably be higher.  But I have some confidence it will be at least this high.

Specific figures used included:

1)  Estimated capital costs (construction costs) was taken from the FTA summary sheet.  The figure reported there of $2,427.97 million includes, however, $126.0 million in “finance charges”.  These finance charges appear to include the financing costs that will be incurred only on the private borrowing portion of the total costs (estimated to cover $800 million of the overall $2.4 billion cost) and only during the construction period.  Since the total financing cost (including on government borrowed funds) will be accounted for separately, the capital cost figure used for construction expenses only was $2,302 million ($2,428 million less $126 million).  Like all the cost figures presented in the FEIS and RFP, it is assumed these are expressed in prices of 2012.  They were then spread evenly (in real terms) over the construction period of 2015 to 2020.

2)  While this capital cost figure of $2,302 million was used, it should be noted that all of the capital costs of the project have not been accounted for in this widely reported figure.  In particular, it does not include the cost of perhaps the most complex and difficult light rail station to construct, at the western end of the line (Bethesda).  This will be fitted into an existing underground tunnel under a building (where the old train line had run), with underground connections made there to link it to an existing subway line station.  Consideration was given to tearing down the existing building above the lines to allow the construction, but a recent decision was made not to, as the costs would be even higher.  The capital cost figure also does not include the cost of re-building the existing walking/biking path that the new rail line will take over, as this cost will be covered by Montgomery County.  However, it is still a cost, and should have been included.  Finally and perhaps most importantly, the capital cost figure of $2,302 million does not include anything for the significant costs incurred (mostly by the State of Maryland) for the design work, environmental impact and other assessments, and all else that has been done to bring the project to this point.  As has been noted, thousands of pages of analysis have been posted on the internet, consultants were hired to produce these reports, and public officials have devoted a good deal of time to organizing and overseeing this work.  These costs should not be ignored.  While it can be argued that these costs are already incurred and hence should not be a factor in what to do now, one should then not present the capital cost estimate (of $2,302 million currently) as the total capital cost of the project.  Rather, it is an estimate of the additional capital cost now needed to complete the project.  But in any case, since I do not have figures on the costs already incurred, I have had to leave them out.  The true total capital costs are higher.

3)  Also left out is any valuation for the cost of the public lands taken (including public park lands) for the rail line.  The public park and other public lands taken have been treated as if they were free, with zero value.  In particular, the western section of the line, from Silver Spring to Bethesda, will be built over an existing walking/biking path, and will need to clear-cut the existing trees on both sides to allow for the two new parallel rail lines plus a re-built path adjacent to it.  The park will be effectively destroyed.  Instead of a walk through the woods, one will have a utilitarian paved path next to a busy rail line.  If this project were being financed by the World Bank in a developing country, the World Bank would have required (by its environmental standards) that a new similarly sized park be created near-by, as an environmental offset to the land taken for the transit project.  The cost of acquiring this new park land would then be reflected in the project cost.  The cost would not be small, which is probably why it was never seriously considered, but that high cost (reflecting the high value of such land) is precisely the point.  And while poor countries are expected to follow such measures to protect the environment, there is no such plan here, even though Montgomery County (where this section of the line will run) is one of the richest counties in one of the richest countries in the world.

4)  The implicit interest rate used (the opportunity cost of capital) to cover the cost of the up-front capital expenditures will also be important.  The project documents appear to have all left this out (except for the relatively minor $126 million finance charge included in the most recent FTA summary sheet, discussed above).  The current financing plan is for two-thirds of the cost to be covered by government grants (federal and state) and one-third by private borrowing by the project concessionaire.  The private borrower will of course need to cover its interest costs.  While interest rates are currently low, and have been since the Lehman Brothers collapse in September 2008 (as the Fed has kept rates low to spur the recovery), it is expected that interest costs will return to normal once full employment is recovered.  Over the ten year period leading up to September 2008, the average corporate bond borrowing rate for a AAA borrower averaged 6.2%, while it averaged 7.1% for a BBB borrower over this same period.  To be conservative, I assumed the borrowing rate would be 6.0% for this project, even though this is likely to be low.  Note that this is a nominal, not real, interest rate.

5)  More importantly, one also needs to include a cost for the government funds being provided.  It is certainly not zero, even if the project itself receives the funds as a grant.  The government has to obtain the funds from somewhere.  And while the government can borrow, in this case it is choosing to have the private concessionaire borrow funds for a substantial share of the project, rather than provide additional government borrowed funds.  This implies that the government would rather have the private entity borrow funds for the project, and that it views this cost (assumed to be 6.0%) as preferable to whatever it would pay for directly borrowed funds.  Therefore, the spreadsheet calculations were done based on a 6.0% interest cost, implicit or explicit, for the full project cost.

6)  Finally, all the calculations were undertaken in nominal terms, and hence one needed to make certain inflation assumptions.  Based on figures from the RFP and the FEIS, I assumed inflation rates of 3.1% for the construction costs, 2.5% for operations and maintenance costs, and 2.0% for general consumer prices (reflected in the shadow fare rates).

7)  Ridership forecasts were taken from the most recent FTA summary sheet, which shows figures for 2014 (which I interpret reflect what ridership would be today, if the system were operational today) and for 2035.  It was assumed ridership between these dates would grow at a steady growth rate.  This worked out to 1.113% a year, which is reasonable for the already developed region the rail line would go through.

Based on these cost and ridership assumptions, the cost per boarding for the proposed Purple Line comes to $10.42.  This is a lot, for what is designed to be basically a local service (providing connections to and from Metrorail lines and traditional bus services).

There is of course uncertainty in this single point estimate.  It depends on the accuracy of the underlying cost and other estimates used.  One needs to know the sensitivity of this point estimate to the data assumptions made, in order to judge how meaningful the point estimate is.  Several different scenarios were therefore examined to test the sensitivity.  Most of the scenarios tested looked at changes that would lead to higher costs, but the impacts would be similar going in the opposite direction:

Purple Line Scenarios Cost per Boarding
         (prices of 2012)
Base Case $10.42
Interest rate 6.0% → 7.0% $11.62
Time horizon 30 → 40 years $9.28
Ridership 20% less $13.02
Capital Cost + 20% $11.92
Construction Period + 2 years $11.02
Capital Cost + 20%, and also
    Construction Period + 2 years $12.64

The base case assumptions, as noted, lead to an estimated break-even cost per boarding of $10.42.  If the borrowing costs (implicit or explicit) were 7.0% rather than 6.0%, then the cost per boarding would rise to $11.62.  Some would argue that a 7% borrowing rate over the long term would likely be a better estimate of what it will be for such a project entity in coming decades than 6% (the BBB borrowing rate averaged 7.1% over the decade before Lehman Brothers collapsed), but the Base Case was deliberately conservative.

Extending the time horizon would also affect the break-even cost.  The private concession is planned to extend for thirty years of operation following completion of construction, so determining the break-even cost per boarding at that point is of interest.  But some of the assets would likely last longer.  Offsetting this, however, is that there will also be major rehabilitation costs periodically, and I was not able to find any estimates for what those would be.  They were therefore implicitly set at zero.  But even assuming rehabilitation costs were zero, and that assets were all able to last for 40 years rather than 30, the break-even cost per boarding would still be high at $9.28.

Ridership is also difficult to predict with great confidence.  Ridership that turns out to be 20% less than projected would raise the break-even cost per boarding to $13.02.

Construction costs (capital costs) also often turn out to be higher than projected, and/or completion takes longer than planned, and these often come together (delays in completion lead to higher costs).  If the capital cost turns out to be 20% higher, then the break-even cost per boarding rises to $11.92.  If completion is delayed by two years (but with no additional capital cost), the cost per boarding would be $11.02.  And if both the capital cost turns out to be 20% higher and completion is delayed by two years, the break-even cost per boarding rises to $12.64.

Finally, one could have (and indeed generally will have) a combination of differences.  Some might be offsetting, but one could also have some combination of lower ridership, higher construction costs, delays in completion, and higher borrowing costs.  But the degree of difference in each case might well be less than those tested here.

Based on the sensitivities in these scenarios, the estimated cost per boarding of $10.42 in the base case is probably accurate within a dollar or perhaps two.  Given past experience with such projects, there is a greater likelihood that costs will turn out to be higher than expected rather than lower.  I would therefore doubt that the final cost per boarding turns out to be less than the base scenario estimate of $10.42, while there is a significant risk that it could be $12 or even more.

===============================================

October 1, 2014:  Update

The Washington Post reported (in its print edition today, and in an on-line note yesterday) that the official estimate of the capital cost of the Purple Line has increased again, by $21 million this time from the estimate published in July.  The total is now $2.45 billion.  While the $21 million increase should perhaps not be considered large in itself, it comes as the most recent such increase that has steadily raised the estimated cost of the Purple Line from just $1 billion in 2007, to the estimated $2.45 billion now.

I have not changed any of the text above.  With this new capital cost estimate and assuming nothing else has also been changed, the cost per boarding would now work out to $10.48, a bit more than the $10.42 estimated before.

At One Time, You Could Work Your Way Through College – But Not Any More.

Earnings from Min Wage vs. University Costs, 1963-2013

 

At one time, not that long ago, a student could work at a minimum wage job over the summers and during holidays, and be able to cover the total cost (including room and board) of attending a four-year state university.  That is now far from possible.

With students now returning to school, it is perhaps a good time to look at what has happened to the affordability of college in recent decades for middle class families.  The chart above provides one indicator.  It compares what a student could earn in a summer job at the minimum wage, or in year-round work at the minimum wage while attending school (i.e. during summers, holidays, and part time during the academic term), as a ratio to what it would cost to attend a four-year state university.

The state university costs are for in-state tuition and required fees, plus the cost of on-campus room and board.  The figures are from the National Center for Education Statistics of the US Department of Education (with figures for 2013 calculated based on the 2012 to 2013 growth in the College Board estimates).  The university cost figures are for four-year, degree granting, state colleges and universities (i.e. they do not include two-year community colleges), and cover all such state schools.  The cost of attending the elite state schools (such as Berkeley, UVA, or the University of Michigan) would be more.  The years shown on the chart are for the beginning of the respective academic years (i.e. 2013 is for the 2013/14 academic year), and the minimum wage rate used is that which was in effect in July of that year.

The chart indicates that one could have covered the cost of attending a state university in the 1960s and 70s solely through minimum wage work.  Based on just a 17 week summer break, one would have earned enough to cover an average of 82% of the full cost of attending school.  An industrious student working full time over the summer and during vacation breaks (such as Christmas), plus 10 hours per week during the academic term, would have been able to cover the full cost and more – an average of 143% of the cost of school.  Hence summer work plus a bit more during vacations would have sufficed to cover the full cost of college.  In terms of dollar figures, the full cost of attending a state university in 1963/64 would have been $929, in the then current dollars.  A student could have earned $782 just from working at minimum wage over the summer, or $1,357 by working at minimum wage over the summer, during vacations, and 10 hours per week during the academic term.

These are, of course, just simple indicators.  One might have been able to earn more than the minimum wage, and/or worked a different number of hours.  But the point is that in the 1960s and 70s, when baby boomers such as myself were going to college, it was possible for the student alone, simply by working at the minimum wage, to have paid for the full cost of attending a four-year state university.

That began to change in the 1980s, as Reagan took office.  The change is indeed striking.  Affordability then began to fall, and it has fallen steadily since, as seen in the chart above.  By 1986, a student working even full time over the summer and during vacation breaks, and 10 hours a week during the academic term, no longer would have been able to cover the full cost of attending school.

The share of schooling costs that could be covered by work then continued to decline (with some bumps up when the minimum wage was sporadically changed) until the present day.  By 2013, summer work would only cover a quarter of the cost of schooling, while more comprehensive work over the entire year would only cover less than half.  In dollar terms, the average cost of attending a state university (for tuition, room, and board) was $18,037 per year in 2013.  But a student working over the summer at the minimum wage would have only been able to earn $4,930, or only a bit over a quarter of the cost of attending school.  Working full time over the summer and during vacations, plus 10 hours per week during the academic term, the student could have only earned $8,555, or less than half the cost of attending school.

As a consequence, students must now rely on their parents (when their parents can afford it), or a scarce number of scholarships (highly limited, especially for state schools), or on student loans.  Otherwise, they must give up on attending university.

The result has been an explosion in student loan debt outstanding.  As of June 30, 2014, student loan debt totaled an estimated $1,275 billion (based on Federal Reserve Board estimates), or five times the level outstanding in 2003 of $250 billion (the earliest figures I could find on a comparable basis; the amounts were so small earlier, that the Fed did not separately break them out).  Student loans have long been common in the US (I had them when I went to school in the early 1970s).  But the amounts outstanding then were relatively small, were at low interest rates, and were for most of us easily manageable.  It is different now.  Student loan debts have exploded in recent years, with a five-fold increase over just the past decade.

The declining affordability of college by this measure is of course a consequence of what has been happening to the two components of the measure.  One has been the unwillingness of Congress to allow the minimum wage to keep up with inflation.  As noted in an earlier post on this blog, the minimum wage in the US has stagnated over the last half century, and is indeed lower now (in real terms) than it was in 1950, when Harry Truman was president.  Real GDP per capita is 3.5 times higher now than it was in 1950, and real labor productivity has increased similarly.  These are not small increases.  I find it amazing (and shameful) that the real minimum wage is lower now than it was then.

For the period since 1963 (the earliest date in the chart), real GDP per capita and real labor productivity are both now 2.7 times higher than what they were then.  But the real minimum wage is close to 20% less now than it was in 1963.

The fall in the real minimum wage fall since the 1960s is half the story.  Note that the inflation measure used for determining the real minimum wage is the general consumer price index (the CPI).  This is the price index for the overall basket of goods and services a US household will purchase.  But the price index is an average over all the goods and services that households buy, and individual items can have price increases that are more than, or less than, this overall average.

In particular, the cost of attending a state university has increased by a good deal more than the overall CPI.  Based on the overall CPI, the real cost of attending a state university (for tuition, room, and board) is now 2.5 times what it was in 1963.  The cost of the tuition component alone is now 4.5 times higher.  The basic cause has been the cutbacks in state budgetary support for their colleges and universities, with tuition and other charges then increased to make up for it.

As a result, the minimum wage has fallen in real terms (based on the overall CPI) since the 1960s, at the same time that the real cost of attending school (relative to the overall CPI) has increased sharply.  The two factors together account for the steep fall in the share of state university costs that one can pay for by working at the minimum wage.  The curves in the chart at the top of this post show that path.

It is important to recognize that this declining affordability of attending state schools was not inevitable, but rather the result of policy choices.  The minimum wage has not been adjusted to reflect general inflation, even though real GDP per capita and labor productivity have both grown substantially.  And as was discussed in another post on this blog, there is no evidence that raising the minimum wage by the modest amounts now being discussed would lead to adverse effects on employment.

Government support for state colleges and universities has also been scaled back, leading to tuition and other cost increases substantially higher than that reflected in the general price index.  This has also been a policy choice.  And it is a policy choice that has prioritized the present generation (with tax cuts a prime example) over the coming generation, that is denying many of the coming generation the educational opportunities we ourselves had.

Transparency of Quality is Essential for a Well-Functioning Health Care System

New York State CABG Mortality, with distribution, 1989-2011

A.  Introduction

Prospective patients will try to assess the quality of the medical care provided by the doctors or hospitals where they might go, when deciding where to seek treatment.  They seek out recommendations from friends and family, they look at publicly available rankings such as those of US News and World Report, and they have their own past experience with some doctor or hospital.  More recently, more information has become available on the internet, allowing prospective patients to look up personal histories on medical providers (where they went to medical school, their age, what languages they speak), as well as to view consumer comments and ratings on dedicated medical websites as well as websites such as Yelp.  There may also be reputational ratings (where doctors are asked what other doctors they would recommend), such as those conducted by the Washingtonian magazine in the Washington, DC, area.

But such information is limited, possibly biased, and superficial.  Recommendations of friends and family, your own experience, and comments and ratings on sites such as Yelp, are really just anecdotal, based on a very limited number of cases.  Individuals will also not always know whether the care they received was in fact high quality or not (there may have been complications, but they will normally not know if they were avoidable).  Rankings in reports such as that of US News and World Report have been criticized for being based on a small set of statistics (limited to those that the publication can obtain) which might have limited relevance.  And reputational ratings can be self-reinforcing, as those being surveyed rate some doctor or hospital highly simply because they have been highly rated in the past.  They may well have no real basis for making an assessment.

Most fundamentally, this information does not focus on what one really wants to know:  Does the doctor or hospital provide good quality care that will cure the patient?  Information such as that above has little on whether the doctors or hospitals are in fact any good at what they do.  Rather, the information is mostly on inputs (where did the doctor go to medical school, for example), or on superficial factors (was the receptionist pleasant when one checked in).

As a result, one can find out more on the quality of a $500 television that one is looking to buy, than on the quality of a doctor who will perform a coronary artery bypass surgery on you.

But information on actual results of doctors and hospitals, in terms of success rates (was the condition cured) and mortality rates, the frequency of medical complications, and other such measures, in fact exist.  The problem is that most of this information, with some exceptions noted below, is kept secret from the public.  Especially limited is information on the performance of specific doctors.  But the information is collected.  There are mandatory reports filed with government and regulatory authorities (both at the federal and state levels in the US).  Insurance companies (including Medicare) will know for the population they cover whether the treatment actually worked or required additional attempts or changes in approach.  Insurance will also know whether there were complications that then had to be treated (with the resulting expenses then filed).  And they will know all this at the level of the individual doctor and medical facility, and for the well defined specific medical procedures which were performed.

The information therefore exists.  The problem is that it is not made publicly available.  The normal rationale provided for this secrecy is that the information is complex and can be difficult to interpret by someone other than a medical professional.  But that is a lame excuse.  The information could be released in a form which adjusts for such factors as the underlying riskiness of the particular cases a doctor has dealt with (there are standard statistical ways to do this), and with accompanying information on the degree of uncertainty (derived statistically) in the information being provided.  One would also expect that if such information were made publicly available, then specialized firms would develop who would take such information and assess it.  Based on their technical analysis, they would sell their findings to insurance companies and firms, as well as interested individuals, on which doctors and facilities performed the best for specific medical procedures.  Government entities interested in good quality care (such as Medicare, in the public interest and also because good quality care costs less in the end) could also assess and make such information available, for free.

The real reason such information on outcomes is in general not made publicly available is rather that the results can be embarrassing for the doctors and hospitals.  And more than simply embarrassing, there could be huge financial implications as well.  Patients would avoid the doctors and hospitals who had poor medical outcomes.  With close to $3 trillion now being spent each year on medical care in the US, this means there are huge vested interests in keeping this information secret from the public.

This is starting to change, however.  As noted above, there are exceptions as well as experiments underway to provide such information to the public.  But it has been fragmented, partial, and highly limited.  The limited information that has been provided so far has been primarily at the level of hospitals, although there have been some experiments with data also being provided on the performance of individual doctors in certain specialties.

From these trials and experiments, we know that widespread availability of such information in an easily accessible form could have profound impacts on the practice of American medicine.

B.  The Impact of Transparency – A New York Experiment

The oldest and longest lasting experiment has been in New York.  Starting with data from 1989 (made publicly available in 1990), the New York State Public Health Commissioner has released the risk-adjusted 30-day in-hospital mortality rates of those undergoing coronary artery bypass graft (CABG, or simply heart bypass) surgery, by specific hospital.  They started to release physician specific mortality rates (on a three-year rolling basis) from December 1992.  There have been a number of good descriptions of, and analyses of the impacts of, the New York program.  Sources I have used include the articles here, here, here, and here.  In addition, a good description is provided as the third chapter in the excellent book by Dr. Marty Makary, Unaccountable, a source I will make further use of below.  Dr. Marty Makary is a physician at The Johns Hopkins Hospital, specializing in pancreatic surgery.  In addition to his many medical research publications, Dr. Makary has undertaken research on how to improve the quality of medical care delivery.

The chart at the top of this post shows what happened to 30-day in-hospital mortality rates following heart bypass surgery since 1989, across hospitals in New York State performing this procedure.  Only hospitals doing 70 or more such surgeries in any given year are included in the chart.  This was to reduce the statistical noise arising from small samples (and there were only a few exclusions:  two hospitals were excluded in two of the 23 years of data, and only one or zero in all of the other years).  A total of 28 hospitals were covered in the 1989 set, with the number rising over time to 38 in 2011.

The data were drawn from the annual reports issued by the New York State Department of Health.  Reports for 1994 to 2011 (the most recent report issued) are available on their web site.  Reports for earlier years were provided to me by a helpful staff member (whom I would like to thank), and the figures for the first half of 1989 were published in a December 1990 article in the Journal of the American Medical Association.  All the mortality rates shown are risk-adjusted rates, as estimated by the New York Department of Health, which controls for the relative riskiness of the patients (compared to the others in New York State that year) that were treated in the facility.

The chart depicts a remarkable improvement in mortality rates once it became known that the figures would be gathered and made publicly available, with individual hospitals named.  The chart shows the fall over time of the average rate across the state (note this is not the median rate, but rather the mean), as well as the minimum and maximum rates across all hospitals with 70 or more CABG procedures in the year.  The ranges at the 90th and 10th percentiles are also shown.  Among the points to note:

1)  The average risk-adjusted mortality rate fell sharply in the early years, and since then has continued to improve.  Furthermore, the underlying improvement was in fact greater than what it appears to be in these figures.  The average mortality rates shown in the chart are for the mix of patients (by riskiness of their health status) in each given year.  But especially in the early years, when angioplasty and coronary stent procedures were developing and found to be suitable for lower risk patients, the pool of patients for whom coronary bypass surgery was needed became a riskier mix.  Taking this into account, while the overall average mortality rate fell by a very significant 21% between 1989 and 1992, once one accounts for the higher risk of the patients operated on in 1992, the fall in the cross year risk-adjusted mortality rate was an even larger 41% over just this three year period.  Technology for CABG procedures did not change over this period.  Transparency did.

2)  The improvement in the coronary artery bypass surgery mortality rate in New York is especially impressive as New York was starting from a rate which was already in 1989 better than the average across all US states.  And by 1992, the rate in New York was the best across all US states.

3)  What is perhaps even more interesting and important, not only did the average rate in New York improve, but also the dispersion in mortality rates across hospitals was dramatically reduced.  The maximum (worst) mortality rate dropped from almost 18% in the first half of 1989 to under 6% by 1992.  The minimum rate was 2.1% in 1989H1, and fell to zero in 9 of the 12 most recent years.  One sees this narrowing in dispersion also in the range between the 90th and 10th percentile bands.

Publication of the mortality results got a good deal of media attention in the early years, and led to pressure, especially on the poor performers, to improve.  Note that the information being gathered was not anything new.  State health authorities long had reports on death rates by hospitals.  What was new was to make this information publicly available, with hospitals named.

Hospitals with poor records then scrambled to improve.  A range of actions were taken.  Some might have seemed obvious, but even so, were not undertaken until the mortality rates by hospital were made publicly available.  For example, hospitals with poor records began to create cardiac specific teams of nurses and other staff, rather than draw on staff from a pool who could be assigned to a wide range of different medical conditions.  Such specialization allowed them to learn better what was needed in cardiac surgery, and to work better as teams.  Such a reorganization at Winthrop Hospital, which included bringing in a new Chief of Cardiac Surgery who led the effort, led to a drop in its mortality rate from 9.2% in 1989 (close to the worst in the state in that year) to 4.6% in 1990 and to 2.3% in 1991 (better than the state wide average that year of 3.1%).

Other issues were highly hospital specific.  For example, one hospital (St. Peter’s in Albany) saw that its mortality rates for pre-scheduled elective and even urgent CABG surgery cases were similar to those elsewhere in New York.  But it had especially poor rates for emergency cases, which raised its overall average.  After reviewing the data, its doctors concluded that they were not stabilizing sufficiently the emergency patients before the surgery.   After it corrected this, its mortality rates fell sharply.  They were among the highest in New York in 1991 and 1992 (at 6.6% and 5.8%), but the rates then fell to 2.5% in 1993 and 1.4% in 1994 (when the New York average rate was 2.5%).  Mortality in emergency cases fell from 26% in 1992 (11 of 42 cases) to 0% in 1993 (zero in 54 emergency cases).

Another hospital (Strong Memorial) also found that its mortality rates for routine elective cases were similar to the New York average, but very high for the emergency cases, bringing up its overall average.  The problem was that while they had a good adult cardiac surgeon, he was always fully booked with routine cases, and hence was not available when an emergency case came in.  They then used one of two doctors who were not trained in adult cardiac surgery to handle the emergencies (one was a vascular surgeon, and the other a specialist in pediatric cardiac surgery).  By hiring a new adult cardiac surgeon and then better balancing the schedule, the rates soon dropped to normal.

American health care has traditionally relied on state regulators, armed with reports on hospital and indeed surgeon specific practices and outcomes, to impose safety and good practice measures.  But there is no way a central regulator can know all that might be underlying the causes of poor outcomes, or what actions should be taken to remedy the problem.  They also will not focus on hospitals with relatively good, or even average, mortality rates, even though such institutions could often still improve.  By releasing the data to the public, hospitals with poor records will be under great pressure to improve, while even those with relatively good records will see the need to get better if they are to stay competitive.  And the actions taken will often be actions that no central regulator would have been able to see, much less require.

C.  Staff Surveys as Another Indicator of Quality

Outcome indicators, up to and including mortality rates, are one set of measures which could have a profound impact on the quality of health care delivery if made publicly available.  An additional type of measure has been developed by Dr. Marty Makary, tested with a number of hospitals, and is now routinely used in hospitals across the US.  But the results are then typically kept secret from the public.

Specifically, Dr. Makary developed a simple staff survey (see here and here, in addition to his book Unaccountable referenced above) with some key questions.  The survey goes to all staff in a hospital, and asks questions such as whether the respondent would feel comfortable having their own care performed in the hospital unit in which they work.

In the original test, the surveys were sent to all staff at 60 hospitals across the US.  They got a 77% response rate, which is quite good.  What is most interesting was the wide range they found in the results across the hospitals.  For example, on the question of whether the staff member would want their own care performed at the hospital unit in which they work, there were two hospitals where close to 100% of the staff said they would, but also one hospital in which only 16% said they would.  There was a fairly even spread between these two extremes, and in about half of the hospitals surveyed, less than half of the staff said they would want their own care performed there:

Makary Hospital Staff Survey - Care in Own Unit.003

This would be powerful information to have as a patient.  The insiders are really the ones who know best what quality of care is being provided.  If even they would not want their health care needs met at their hospital, one knows where one would want to avoid.

It is recognized that the original Makary survey was done with the promise that the identities of the individual hospitals would not be revealed.  Should such surveys be made publicly available, the staff responding might well be less negative.  But the identities of the individual staff members would still be kept confidential (with the data gathered by an independent third party, and anonymously over the web).  There would certainly still be some dispersion in results across hospitals, and one could take into account the possible biases when judging the results.  And if a hospital is rated poorly by its staff even when they know the results will be made public, one knows which hospitals to avoid.  One would expect such hospitals then to scramble to improve the quality of the care they provide.

D.  While a Number of Transparency Initiatives Are Underway, They Remain Fragmented and Partial

Patients have always sought information on the quality of the care they will need, and have made decisions on where to go based on what they can find out.  But the information that they have been able to obtain has been only partial, highly fragmented, and far from what they really need to know to make a wise decision.

People will also find measures that are easily observed, but not necessarily terribly important to the quality of the care they will receive.  For example, they may find out whether parking is free and convenient, but this should not normally be a driving factor for their decision.  More relevant, and obviously something they will know, will be geographic location:  Is the facility close to them, or further away?  But they will normally have little basis for determining whether it is worthwhile to go a facility that is further away.

There has been a substantial expansion in recent years in the amount of information one can find on providers.  While still limited, one can find out more now than before.  There is the New York experiment described above, which New York soon extended from hospitals to individual surgeons, and also to angioplasty and cardiac stent procedures.  New York has also brought together on one web site easy access to a wide range of health topic data sets.  These include data sets on outcomes and quality of care indicators (such as the most recent CABG mortality rates by hospital and by surgeon, for example) but also many others (such as the most common baby names chosen).

The Obama administration has also expanded substantially the public availability of information on hospital quality measures.  The Centers for Medicare and Medicaid Services (CMS) now makes available at its Medicare Hospital Compare site results at the hospital level, drawn primarily from the data they have for Medicare patients, on such outcome measures as mortality rates, complications, hospital readmission rates, and other indicators.  However, they are still partial, and instead of showing, for example, actual and historical figures by hospital for indicators such as the rate of complications or mortality, they simply show whether the rates are similar to the national norm, or better or worse by a statistically significant margin (at the 95% significance level).

With the clear positive impact of the New York experiment, other states have also begun to implement similar programs.  But they remain partial and fragmented, and do not provide the comprehensive picture a patient really needs if they are to make a wise choice.

In addition, many professional medical societies have begun to collect similar data from their members, and then calculate risk-adjusted measures.  However, they have then kept the individual results secret, with identifying information by hospital or physician not made available.  Individual hospitals and physicians could release them if they so chose, and some have.  But one can safely assume that only those with good results will release the information, while those with poor results will not.

The same is true for hospital staff surveys, such as the one described above pioneered by Dr. Makary.  Such surveys are now widely used.  Dr. Makary reports in Unaccountable (published in 2012) that approximately 1,500 hospitals were then undertaking such surveys.  The number is certainly higher now.  But the results are in general kept secret.  Some hospitals make them publicly available, but one can again safely assume that these will be the ones with the better results.  Without the others for comparison, it is difficult to judge how meaningful the individual figures are.

So the relevant data are often collected already.  It is only a matter of making them public.  There is not a question of feasibility in collecting such data, but rather a question of willingness to make them public.

E.  What a Transparent System of Information on Quality Should Include

As noted above, people will gather what information they can.  But what they can gather now is limited.  What is needed is hard data on actual outcomes, identified by hospital and by individual doctor.  As the New York experiment discussed above indicates, the result could have a profound impact on quality of care.

Specifically, there should be easy access to the following specific measures:

a)  Volume:  While not directly an outcome measure, it is now well established in the literature that a higher frequency of a doctor undertaking some specific medical procedure, or that is done by all the doctors at some hospital or medical facility, is positively associated with better outcomes.  A doctor that undertakes a procedure a hundred times a year, or more, will on average have better outcomes than one who does the procedure only a dozen times a year (i.e. once a month).  And volume can be easily measured.  The problem is in obtaining easy access to the information, and at the relevant level of detail (i.e. by individual doctor, and for the procedure actually being considered for the patient, not just of some standard benchmark procedure).

b)  Success rates:  While many of the outcome measures being used in various trials and experiments are negative measures (mortality rates; complication rates), a more useful starting point would be risk-adjusted success rates.  What percentage of the procedures undertaken by the individual doctor or at the medical facility for some condition actually leads to a cure of the condition?  How success is defined will vary by the medical issue, but standard ones are available.  If the risk-adjusted success rate is 80% for one doctor and 99% for another, the choice should be clear.  Yet I have never seen a trial or experiment where such success rates by medical facility, much less at the level of individual doctors, were made publicly available.

c)  Success rates without complications:  A more stringent measure would be not only that the procedure was a success, but that it was achieved without a noteworthy complication such as an infection.

d)  Complication rates:  Moving to negative measures, one wants to see minimized the complications associated with some procedure.  The medical profession has identified the complications often found as a result of some medical procedure, and significant complications will be reported.  They can also normally be identified from medical insurance records, as they require treatment.  As with mortality rates, these should be published on a risk-adjusted basis.

e)  Mortality rates:  The ultimate “complication” is mortality.  As discussed extensively above, these should be made available by medical procedure and by individual doctor on a risk-adjusted basis.  The 30 day mortality might be appropriate for most medical procedures, but for others the 60 day or 90 day rates might be more appropriate.  Medical societies can work out what makes most sense for a given procedure.  But everyone should then be required to use the same measure, to allow comparability.

f)  Bounceback rates:  Bounceback rates are the percentage of patients undergoing some procedure, who then need to be readmitted back to a hospital (the original one or some other) within some period following release, usually 90 days.  Readmission rates are regularly collected by hospitals, and they can also be risk adjusted when made publicly available.  They are a good indication that some problem developed.  Some rate of readmission might well be expected for certain procedures.  They are not risk free.  But one wants to see if the bounceback rates are especially high, or low, for the physician or medical facility being considered.

g)  Never events:  Never events are events that should never occur.  While a certain rate of complications will normally be expected, one should never see an operation done on the wrong side of the body, or sponges or medical instruments left in the body after the surgeon has sewn up.  Hospitals know these and keep track of them (as such never events often lead to expensive lawsuits), but not surprisingly want to keep them secret.

h)  Hospital Staff Surveys:  As discussed above, Dr. Marty Makary developed a survey that would go to all hospital staff, which asks a series of questions on the quality of care being provided at the facility.  While approximately 1,500 hospitals were already administering the survey in 2012 (when his book Unaccountable was published), they are voluntary and in general not made publicly available.  They should be.

While the surveys can cover a long list of questions, Dr. Makary recommends (Unaccountable, page 216) that the percentage of hospital staff responding “yes” to the following three questions, at least, should be made public:

-  “Would you have your operation at the hospital in which you work?”

-  “Do you feel comfortable speaking up when you have a safety concern?”

-  “Does the teamwork here promote doing what’s right for the patient?”

F.  Conclusion

There are of course many other measures of quality one could examine, and there has been some movement in recent years to making more available.  These include results from patient surveys (“were you content with your experience at the hospital?”, “were the rooms kept clean?”), as well as the percentage of cases where certain established medical best practices were followed (“was aspirin given within 24 hours of a suspected heart attack?”).

Such additional measures might well be useful in particular cases.  It will depend on the individual, their particular condition, and what specifically is important to them.  People should have a choice, and do the research they personally wish to do.

But until hard measures on actual outcomes, such as those described above, are made widely available, and on a comprehensive rather than partial and fragmented basis, it will not be possible to make a well informed and wise choice on which doctor and medical facility to go to.  Without this, there can be no effective competition across providers.  There will be little pressure on the poor quality providers either to improve their performance, or drop out and let providers who can deliver better quality care treat the patients.

The impact on the quality of health care services provided would be profound.

More on the Widely Varying Charges for Common Health Procedures: Price Variation for Standard Blood Tests

Blood Test Prices in California - Lipid Panel

A.  Widely Varying Prices Charged Even for Standard Blood Tests

This post is an addition to an earlier post on this blog that looked at the widely varying prices being charged in the US for common health procedures.  As that post noted, such differences in prices for what are fundamentally the same services are a clear indication that the market is not working.  The prices would be similar if the market was working, with differences that are relatively small and explainable by factors such as geography.  But that is not the case.

That post looked at data from a number of studies (including my own simple research on the prices that I would be charged in the Washington, DC, area, for a common surgical procedure).  Prices could vary by a factor of 10, and indeed often even more.  And as that post showed in a series of charts, the prices actually paid in the US (at the rates negotiated by insurers) are not only widely varying, but also consistently far higher than the prices paid for the same procedures in other countries.

A criticism of studies that examine the prices being charged for health care procedures is that individual cases can differ, with some more complex than others.  Thus prices might vary for that reason.  Even though it is difficult to see how costs can vary by a factor of ten or more even with differing levels of complexity for some standard procedure (such as a hip or knee replacement, for example), one can recognize that differing degrees of complexity might explain at least some of the price differences.

Thus a study published last week in the BMJOpen, an open-access on-line journal affiliated with the British Medical Journal, is of interest as it addresses the question of whether such price variation is found also for procedures where case complexity does not enter.  The lead author is Dr. Renee Hsia, of the Department of Emergency Medicine at the University of California – San Francisco.  In an earlier study, summarized in the blog post cited above on health care price variation, Dr. Hsia had looked at the prices charged by hospitals in California for an uncomplicated but urgent appendectomy.  She found that the prices varied by a factor of 120, between the lowest rate charged and the highest.

In the current study, Dr. Hsia with her colleagues looked at the prices charged by California hospitals for ten common blood tests.  The prices reviewed are the so-called “chargemaster” rates, or the list prices at the hospitals for the tests.  The actual price paid will then normally be a lower rate negotiated with the hospital by your insurer (if you have insurance), but the chargemaster rate is the starting point.  Why this matters will be discussed below.

Dr. Hsia was able to obtain the data for California because hospitals there are required to report to state authorities the average prices they charged for a number of common procedures.  Since routine blood tests are standard, and are not more or less complicated for one patient vs. another (the blood is drawn, brought to a standard machine, and the results then read), one cannot argue that the price variation observed might be a consequence of different degrees of case complexity.

The results from one of the blood tests examined, that of a standard lipid test (which measures blood cholesterol levels), is shown graphically at the top of this post.  Data was available from 178 hospitals, and each hospital reported the average price it charged for this test over the course of 2011.  The price charged at one hospital was only $10 per test.  The average price charged at a different hospital, for the exact same blood test, was $10,169 per test, or over 1,000 times as much.  Such variation is absurd.

These are, of course, the extremes.  But even if one focusses on observations in the middle of the distribution, it is impossible to see how such variation in prices charged can be justified.  The price at the 5th percentile (meaning 5% of the hospitals charged this price or less) was $76.  The price charged at the 95th percentile (meaning 5% charged this price or more) was $602.  This is almost 8 times higher than the price at the 5th percentile.  The results for the other nine blood tests examined were broadly similar (with ratios between the prices at the 95th and 5th percentiles varying from a high of 12 times and a low of 6.8 times).

B.  Chargemaster Rates Matter

What can justify such a spread?  Nothing that I can see.  The tests are standard, use standard machines, and all use similarly drawn blood.  The response of a spokeswoman for the California Hospital Association was that the prices reviewed in the study are “meaningless”, since virtually no one (she states) pay these rates.  As noted above, the rates reviewed in the study, as in the earlier study of the prices charged for appendectomies, are the chargemaster rates of the hospitals.  These are the regular list prices for the procedures, which are then typically discounted in negotiations with individual insurers.

But there are still several problems with this, including:

1)  How much the prices are negotiated down will vary according to the bargaining strength of the patient’s individual insurer in the region.  In the bargaining process discussed in an earlier post in this series on health reform, insurers will bargain with hospitals on what the rates will be.  Their relative bargaining strength will depend on how concentrated the local market is in terms of hospitals (if there is only one hospital, or one chain of hospitals all owned by the same entity, but a number of insurers, the bargaining power of the hospital will be great) versus insurers (in one insurer dominates in the market, while there are many hospitals, that insurer will have great bargaining power).  If you have insurance with an insurer who does not command great market share in the region, the price you will have to pay may be close to the chargemaster rate.

2)  If you do not have insurance (and many could not get health insurance, prior to the reforms of Obamacare), you will be charged the chargemaster rate.  You might then try to bargain individually with the hospital, but the starting point will be the chargemaster rate.  And many hospitals will insist, unless you are poor, that you have to pay that chargemaster rate.

3)  You may well have insurance, but if the particular hospital you are in is not in your insurance network (perhaps because you were brought by an ambulance to the nearest hospital in an emergency), you will be charged the chargemaster rate.  Your insurance company might pay a portion of this at what they consider to be a “reasonable rate”, but this is likely to be close to what your insurer has negotiated with others, and as we have discussed in the earlier blog posts cited above, this might be only one-tenth of the chargemaster rate.  You will then still be responsible for the other 90%.  This can be a lot, if you are at the hospital where a simple lipid panel blood test is charged at over $10,000.

4)  You may well again have insurance, and be in a facility that is in-network for your insurer, but your insurer might disagree on whether some standard blood or other test ordered by your doctor was really needed.  Your insurer will then refuse to cover the cost of that test, and you will be charged the chargemaster rate.

I am personally facing a case of that right now.  While the amounts are small in absolute terms, the issue is the same.  My doctor ordered a set of routine blood tests for me earlier this year, and my insurer covered all except one.  For that one, the insurer asserted that there had not been a need (even though both my doctor, and research I found on the web, made clear that the test was in fact needed).  The lab therefore charged me the full chargemaster rate (which in this case was $213.98), even though the negotiated rate Aetna would have paid, had they agreed it should be covered, was only $16.23.  That is, the full billed rate was 13.2 times the negotiated rate.  I would have been glad to pay the negotiated rate in full, and the $16.23 the lab has negotiated with Aetna is evidently a rate sufficient to provide an adequate profit to the lab.  But find it absurd that I should have to pay over 13 times more.  I am appealing, but do not know yet the outcome.

5)  Finally, it is worth noting that the chargemaster rates can matter for other issues as well. For example, hospitals are typically required to provide a certain amount of “charity care” (care provided to the poor without health insurance for free or at discounted rates) in order to benefit from certain tax breaks.  This is especially important and valuable for private, profit-making, hospitals.  Valuing such services at the chargemaster rates, when these rates are 1000 times higher than what someone else might charge, will make it look as if the hospital is providing a good deal of charity care.

C.  Conclusion

This new study should put to rest the argument that price variation in health care services is principally due to variation in the degree of complexity of individual cases.  Common blood tests are standard, and they show price variation which is huge as well as similar in degree to that seen for standard health care procedures (see the review in the earlier post).  The prices vary not principally due to case complexity, but rather due to a grossly misfunctioning market for health care services, where there are strong forces keeping out effective competition and any pressure to converge on low prices from efficient providers.

The (Lack of) Recovery in the Employment to Population Ratio: Not the Concern It Might Appear to Be

Employment to Population Ratios, Jan 2007 to July 2014

Unemployment Rates, Ages 25 to 54, Jan 2007 to July 2014A.  Introduction

A critically important policy question is how close the US economy now is to full employment.  The unemployment rate has been falling, albeit slowly, from a peak of 10.0% in October 2009, to a current 6.2% as of mid-July (ticking up from 6.1% in June, but a 0.1% change is not statistically significant).  That is, the unemployment rate has come down by a bit less than 4% points from its peak.

However, some have noted that one does not see such a recovery if one focusses on the employment to population ratio.  Excellent analysts, such as Paul Krugman and Brad DeLong, have argued that one should.  If the unemployment rate has come down by close to 4% points, then the employment to population ratio should have gone by almost the same in percentage points unless people are dropping out of the labor force.  [It will not go up by exactly the same amount in percentage points since the base for the employment to population ratio is population while the unemployment rate is expressed as a share of the labor force.  But, all else equal, they will be close.  One could make the relationship exact by expressing the unemployment rate in terms of the share of population rather than share of the labor force, but this is not how the unemployment rate is normally reported.]

If the employment to population rate has not recovered by the same amount (in percentage points) as the unemployment rate has, then by arithmetic this is only possible if the labor force participation rate has come down.  The concern is that the pool of unemployed is coming down not because people are finding jobs (which would then be seen in a rising employment to population ratio), but rather because they are dropping out of the labor force after trying, but failing, to find a decent job (thus lowering the labor force participation rate).

There are of course demographic factors as well to take into account to explain what might be happening to the labor force participation rate, in particular the increasing share of the baby boom generation that is reaching normal retirement age.  One way to do this is to focus the analysis on the prime working age group of those aged 25 to 54 only.  All the charts in this post therefore do this.  But even with this refinement, the apparent concern remains:  The employment to population ratio does not show the same recovery that one sees in the falling unemployment rate.  What is going on?

B.  Recent Years

The chart at the top of this post shows the employment to population ratios from January 2007 to July 2014, for those aged 25 to 54, and for everyone together as well as for males and females separately.  The chart below it shows the unemployment rates for these same groups.  The data all come from the Bureau of Labor Statistics.  The peak unemployment rate was hit in October 2009, after which there was a fairly steady recovery.  [The month to month fluctuations mostly reflect statistical noise.  The employment, unemployment, and labor force participation figures are all based on surveys of households, and there will be statistical noise in any such surveys.]

For the group as a whole (male and female), the unemployment rate for those aged 25 to 54 rose by about 5% points between late 2007 / early 2008 and its peak in October 2009.  Over this period the employment to population ratio fell by a similar 5% points.

But this relationship then broke down going forward.  Over the two years between October 2009 and October 2011, for example, the unemployment rate for those aged 25 to 54 fell by 1.1 percentage points, dropping to 7.9% from 9.0% at the peak (for this age group).  But the employment to population ratio hardly moved.  And between October 2009 and the most recent figures (for July 2014), the unemployment rate came down 3.8% points, while the employment to population ratio rose by only 1.6% points.

The question for policy makers is whether the 3.8% fall in the unemployment rate is a reasonable measure of how far the economy has recovered from the 2008 collapse, or the 1.6% recovery in the employment to population ratio is.  As noted above, both the unemployment rate and the employment to population ratio deteriorated by 5% points during the 2008 collapse and follow-on into 2009.  If the 3.8% recovery in the unemployment rate is the right indicator, then we would have retraced about three-quarters of the fall (3.8/5.0 = 0.76).  But if the 1.6% recovery in the employment to population ratio is the right indicator, then we are less than one-third of the way (1.6/5.0 = .32) back.  This is a huge difference.

Since the difference between the two measures must be reflected, by arithmetic, in a declining labor force participation rate, one needs to look there to see what is going on.  For the January 2007 to July 2014 period, the picture is:

Labor Force Participation Rates, Jan 2007 to July 2014

The rates are all falling after October 2009, for males and females, and hence for the two combined.  What is interesting is that they appear to be falling at a fairly steady pace throughout the period (aside from the month to month squiggles that are mostly statistical noise).  And for males, the rate appears to be falling at a broadly similar pace before October 2009.  The trend is not so clear for females before October 2009, whose rate may have been rising until a few months before October 2009.  This then leads to little change in the overall rate for males and females combined, but the period is so short that the trends are not clear.

C.  A Longer Term Perspective

When one then takes a longer view, the trends do become clear:

Labor Force Participation Rates, Jan 1948 to July 2014

Going back to 1948 (the first year in the BLS series for all these labor market indicators), one sees a pretty steady fall in the labor force participation rate for males from around the mid-1950s (with the squiggles in the curves due to statistical noise), and a strong rise in the female labor force participation rate from the initial year with data (1948) to around 2000.  There was some acceleration in the rise for females in the 1970s, and then a deceleration from the early 1990s, leading to a leveling off around 2000.  Since then, the labor force participation rate for females has fallen, on a path that appears to parallel the similar fall in the rate for males, but at 14 to 15% points lower.

The data are consistent with the broader socio-economic story we have of the labor market in the post-World War II period.  Male labor force participation rates are quite high, but have fallen some over time.  Female rates started very low but then grew, and grew at an especially rapid rate starting in the 1970s.  Female labor market participation rates then reached maturity and leveled off around 2000, after which the female rates paralleled the downward path of the male rates, but at a certain distance below.

In this longer term perspective, the decline in the labor force participation rates since 2009 therefore does not appear to be unusual, but rather a continuation of the longer term trend.  There have been some small fluctuations around the long term trends in recent years that appear to coincide with the business cycle (in particular for the female rates), but they are small and dominated over time by the long term trends.  There have also been similar fluctuations in the participation rates in the past (such as in the mid-1990s) that did not coincide in the same way with the business cycle, as well as large business cycle changes in the past that did not show such fluctuations (such as during the big downturn in the early 1980s at the start of the Reagan presidency, that did not lead to such fluctuations in the labor force participation rates).

The implication of this analysis is that the reported unemployment rates are a better indicator of the state of the labor market than the employment to population ratio is.  The fall in the labor market participation rates in recent years has not been something new, driven by the 2008 economic downturn, but rather a continuation of the trend seen in these rates over the longer term.

Looking at unemployment rates for this age group going back to 1948 provides a useful perspective on what to expect for it:

Unemployment Rates, Jan 1948 to July 2014

Unemployment rates continue to be high in mid-2014.  Even though they have retraced about three-quarters of the deterioration in 2008/2009 (more for males, less for females), they are, at 5.2% currently (for males and females together) still well above the unemployment rates for this group of about 4% in late 2007 /early 2008, and of only 3 1/2% in late 2006 / early 2007.  And the unemployment rate for this group was only 3.0% in late 2000, at the end of the Clinton years.

There is therefore still a significant distance to go before the economy will have returned to full employment.  But the improvement since October 2009 is substantial, and is real.

D.  Implications of the Long Term Trends for Aggregate GDP

Finally, while the employment to population ratio might not be a good indicator of how much slack there is in the labor market in the short run, there are long term implications of the trends noted above.  Specifically, while the overall labor force participation rate rose steadily from 1948 (the earliest year for which we have this data) to about 2000, this was entirely due to the strong rise in the female rate over this period.  The male rate was falling, steadily but slowly.  Once the female rate peaked in the year 2000 and then began to fall at a rate similar to that for males, the overall rate began to fall.  There is no indication this will be reversed any time soon.  Indeed, the degree to which the female rate is now paralleling the male rate suggests that this really is a “new normal”.

A falling labor force participation rate is not necessarily an indication of something bad in itself.  It might reflect increased prosperity, which is being enjoyed by choosing not to work but to retire early, or to attend university or post-graduate education programs in your 20s, or to stay at home and raise a family.  But to the extent it reflects lack of free choice, such as being fired in your 40s or 50s and then not being able to find a job, or to remain a perpetual student due to lack of job opportunities, or to stay at home due to the unavailability of affordable child care, the implications are different.  But it is well beyond the scope of this blog post to dig into this deeper.

But there will be important long term implications of declining labor force participation rates on long term GDP growth.  With fewer in the labor force, aggregate GDP growth will be less.  Note that this does not imply growth in GDP per capita (or more precisely, GDP per worker) will be less.  GDP per worker is a function of productivity growth.  But with fewer workers than otherwise, aggregate GDP growth will be less.

Two final charts, then, to close this blog post.  The first shows the absolute number of people in the ages 25 to 54 population cohort, who are not in the labor force:

Population Not in Labor Force, Jan 1948 to July 2014

The number of males in this age group not in the labor force has been growing steadily since the late 1960s.  The number of females not in the labor force fell until around 1990, was then flat for a decade, and then began to grow.  Overall, the number aged 25 to 54 not in the labor force started to grow around 1990, and has continued to grow since.

Looking at the numbers of those in the 25 to 54 age group in the labor force:

Labor Force Number, Jan 1948 to July 2014

Due to a growing population in this age group (baby boomers, for example, but others as well), and the growing labor force participation rates of females until 2000, the total labor force in this group rose from the starting year (1948) until 2008.  It grew especially fast in the 1970s, 80s, and 90s.  But the absolute size of the labor force (in the 25 to 54 age group) then started to fall from 2008.  This is a historic change for the US, and based on the fall in labor force participation rates discussed above, as well as slowing population growth, should be expected to continue.  While GDP growth per capita (or per worker) might continue to grow as it has in the past (and it has grown at a remarkably consistent 1.9% a year since 1870 in the US, as discussed in this earlier blog post), one should expect aggregate GDP growth to slow.

E.  Summary and Conclusion

The unemployment rate has fallen substantially since hitting its peak in October 2009, but one does not see a similar recovery in the employment to population ratio.  The labor force participation rate therefore has to have fallen.  However, it does not appear that this fall in the labor force participation rate has been driven by the economic downturn, where high unemployment and poor job prospects led workers to drop out of the labor force on a widespread basis.  Rather it appears largely to be a continuation of longer term trends, that become clear when one separates out the paths for male and female labor force participation rates.

The implication is that the unemployment rate is probably a good indicator of how much slack there is in the labor force.  The unemployment rate has retraced about three-quarters of the rise during the 2008/2009 downturn, but is still high.  And it is substantially higher than what was seen as possible in late 2006 / early 2007, and especially the rate achieved in late 2000.

But there are longer term implications.  The analysis suggests that we should not expect much of a recovery in the labor force participation rate when the economy finally returns to full employment.  Rather, the labor force participation rate is on a downward slope, and has been since the year 2000 (when the female rates reached maturity).  This is likely to continue.  The result is that the absolute size of the labor force in the prime working age years of 25 to 54 should be expected to continue to fall for the foreseeable future.  Japan and most of the European economies have already been facing this.  While GDP per worker, which is driven by productivity change, need not necessarily slow, one should expect growth in aggregate GDP to be less than what one saw in the past.  The ability to adapt to, and manage in, this new economic environment remains to be seen.