ABSTRACT 

  

 

 

 

Title of Dissertation: EXAMINATION OF US 
TRANSPORTATION PUBLIC-PRIVATE 
PARTNERSHIP EXPERIENCE: 
PERFORMANCE AND MARKET 

  

 Kunqi Zhang, Doctor of Philosophy, 2024 

  

Dissertation directed by: Professor Qingbin Cui, Civil & Environmental 
Engineering 

 

 

Worldwide, public-private partnership (P3) project performance and benefits accrued 

to market participants are understudied. Focusing on the US, this dissertation 

examines the country’s transportation P3 experience through three empirical studies 

comparing P3 to design-bid-build (DBB), the traditional delivery method. 

Throughout, the Information Source for Major Projects database, built by a 

University of Maryland team in which the author led the data collection effort, served 

as the data source. In the first study, the researchers examined P3 cost and time 

performance using piecewise linear growth curve modeling, recognizing that past 

cross-sectional studies had produced mixed results. With 133 major transportation 

projects, the longitudinal analysis confirmed P3’s time performance advantage and 

efficiency diffusion effecting cost savings in DBB, where efficiency diffusion was a 



  

new term describing the spillover and internalization of technical and managerial 

innovations inducing an efficient outcome. The second study used social network 

analysis to investigate collaboration patterns among different types of players in the 

P3 market (i.e., public sponsors, special purpose vehicles, investors, lenders, advisors, 

contractors, and professional service firms). With 135 projects and 1009 

organizations, data found that both P3 and DBB networks are small worlds. 

Exponential random graph modeling revealed that practicing in the DBB market 

helps firms participate in P3 projects and that large firms (vis-à-vis small/medium-

sized firms) are not privileged. The third study, further exploring the P3 market, 

focused on the Disadvantaged Business Enterprise (DBE) program. Administered by 

the US Department of Transportation, the program promotes the participation of 

small, disadvantaged firms in federal-aid projects. Linear regressions on 134 contracts 

showed that P3 associates with higher DBE goals in terms of percentage of dollars to 

be awarded to DBEs, whereas the delivery method does not affect the actual 

attainment. Overall, the findings justify continued policy support towards P3 

implementation. 

  



  

 

EXAMINATION OF US TRANSPORTATION PUBLIC-PRIVATE 
PARTNERSHIP EXPERIENCE: PERFORMANCE AND MARKET 

 

 

 

by 

 

 

Kunqi Zhang 

 

 

 

 

 

Dissertation submitted to the Faculty of the Graduate School of the  

University of Maryland, College Park, in partial fulfillment 

of the requirements for the degree of 

Doctor of Philosophy 

2024 

 

 

 

Advisory Committee: 

Dr. Qingbin Cui, Chair 

Dr. Gregory B. Baecher 

Dr. Yueming Qiu, Dean’s Representative 

Dr. Paul M. Schonfeld 

Dr. Mirosław J. Skibniewski 

 



  

 

 

 

 

 

 

 

 

 

 

 

 

 

 

© Copyright by 

Kunqi Zhang 

2024 

 

 

 

 

 

 

 

 

 

 

 

 

 



 

 

ii 

 

Preface 

“It’s a very sobering feeling to be up in space and realize that one’s safety factor was 

determined by the lowest bidder on a government contract.” 

– Alan Shepard, US Astronaut (1923-1998) 

Besides light laughter, this quote may make us worry about the bleak prospect of low-

bid procurement and, by extension, traditional public procurement. But a project 

manager or public administrator may have a different take. We know that low-bid 

procurement, and relying on it, the traditional design-bid-build project delivery 

method in the construction world, takes care of fairness and perceived accountability. 

That is, it is conceivably reasonable to break up a large construction project into 

smaller pieces and separate the design process from the construction process to ensure 

that more private companies get a piece of the project. Low bids and prescriptive 

design are meant to control the risk of price manipulation in the spirit of “laying it out 

to the public.” Still, we see cases of market monopoly (Erfani et al. 2021). The 

bottom line is, as flawed as it may be, design-bid-build is not going away any time 

soon. Still, nothing stops us from experimenting with alternative project delivery, 

most notably, public-private partnership (P3 or PPP).1 Recent examples include the 

Maryland Op Lanes project, a so-called progressive P3 (Casady and Garvin 2022). 

The project has a tortuous past, with the developer having walked out and its fate 

 

1 Refer to Section 1.2 for more explanation of P3 necessary for navigating this dissertation. 



 

 

iii 

 

uncertain. Another project also in the Washington metropolitan area, I-66 Outside the 

Beltway, a 22.5-mile-long, $3.9-billion reconstruction of urban highway, has been 

acclaimed for its success (Cho 2022); according to our data, its completion was two 

months ahead of what was required at contract award. 

Now, more than three decades after Virginia and California allowed modern P3s in 

the country (Gómez-Ibáñez and Meyer 1991), we are hard-pressed to answer this: 

“With P3, is the public better off?” A major hindrance to empirically addressing this 

question is data, whereas we see more inquiries into design-build performance simply 

because there are more projects and more accessible data. Hence the challenge to 

establish a proper counterfactual for comparative analyses (Verweij et al. 2022). With 

a sufficiently large data set, we might be able to establish a proper benchmark. 

Thanks to a one-of-a-kind, comprehensive database of US major transportation 

projects, the researchers were able to establish associations comparing P3 with other 

delivery methods (i.e., design-build and design-bid-build). Notably, the researchers 

used novel statistical methods, which were rarely seen in this field of research but 

were not new in other social sciences. The overall investigation is committed to one 

theme: to understand what the US P3 experience has to teach us! 

Finally, resonating with Shepard, ignorance is not bliss when it comes to government 

contracts because our lives and livelihood depend on them! 



 

 

iv 

 

Acknowledgements 

I owe a lot of people a great many thanks for completing this dissertation work. First, 

my thanks go to my advisor, Dr. Qingbin Cui, for his continued support, guidance, 

and patience. Second, I thank my committee members for their insightful critiques 

and suggestions. Third, I thank my colleagues and friends at the University of 

Maryland who inspired or edited the work or helped with data collection—

AbdolMajid (Mazi) Erfani, Yu Wang, Paul Hickey, and Sirish Suwal, to name a few. 

Last but not least, I thank my family—my wife, two children, parents, and in-laws for 

their encouragement and sacrifice. 



 

 

v 

 

Table of Contents 

Preface ........................................................................................................................... ii 

Acknowledgements ...................................................................................................... iv 

Table of Contents .......................................................................................................... v 

List of Tables ............................................................................................................... ix 

List of Figures ............................................................................................................... x 

List of Abbreviations ................................................................................................... xi 

Chapter 1: Introduction ................................................................................................. 1 

1.1 US Transportation Infrastructure ........................................................................ 1 

1.2 Public-Private Partnership .................................................................................. 3 

1.2.1 Definition ..................................................................................................... 3 

1.2.2 P3 Implementation in the US ....................................................................... 5 

1.3 Research Needs, Aims, and Purpose ................................................................ 10 

1.4 ISMP Database ................................................................................................. 11 

1.5 Dissertation Structure ....................................................................................... 11 

Chapter 2: Performance and Efficiency Diffusion (Paper #1) .................................... 13 

2.1 Abstract ............................................................................................................. 13 

2.2 Introduction ....................................................................................................... 14 



 

 

vi 

 

2.3 Literature Review ............................................................................................. 16 

2.4 Method .............................................................................................................. 25 

2.4.1 Hypotheses ................................................................................................. 25 

2.4.2 Data ............................................................................................................ 28 

2.4.3 Preliminary Analysis .................................................................................. 36 

2.4.4 Model ......................................................................................................... 39 

2.5 Results ............................................................................................................... 41 

2.5.1 Cost Performance ....................................................................................... 42 

2.5.2 Time Performance ...................................................................................... 46 

2.6 Discussion ......................................................................................................... 50 

2.7 Conclusion ........................................................................................................ 55 

Chapter 3: Interorganizational Collaboration (Paper #2) ............................................ 56 

3.1 Abstract ............................................................................................................. 56 

3.2 Introduction ....................................................................................................... 56 

3.3 Literature Review ............................................................................................. 57 

3.3.1 Social Network Analysis in Construction Literature ................................. 57 

3.3.2 P3 Market Network .................................................................................... 59 

3.4 Method .............................................................................................................. 60 



 

 

vii 

 

3.4.1 Hypotheses ................................................................................................. 61 

3.4.2 Small World ............................................................................................... 62 

3.4.3 Exponential Random Graph Models .......................................................... 63 

3.4.4 Data ............................................................................................................ 63 

3.5 Results and Discussion ..................................................................................... 72 

3.5.1 Small-world Assessment ............................................................................ 72 

3.5.2 Entry Barrier .............................................................................................. 75 

3.5.3 Limitations ................................................................................................. 79 

3.6 Conclusion ........................................................................................................ 79 

Chapter 4: Market Participation of Disadvantaged Business Enterprises (Paper #3) . 81 

4.1 Abstract ............................................................................................................. 81 

4.2 Introduction ....................................................................................................... 82 

4.3 DBE Program .................................................................................................... 83 

4.3.1 DBE Overall Goal ...................................................................................... 84 

4.3.2 DBE Contract Goal .................................................................................... 86 

4.3.3 DBE Commitment ..................................................................................... 86 

4.3.4 DBE Attainment ......................................................................................... 88 

4.4 Method .............................................................................................................. 88 



 

 

viii 

 

4.4.1 Hypotheses ................................................................................................. 89 

4.4.2 Model and Data .......................................................................................... 91 

4.5 Results and Discussion ..................................................................................... 95 

4.6 Conclusion ...................................................................................................... 100 

Chapter 5: Conclusion and Next Steps ..................................................................... 103 

5.1 Summary of Findings and Significance .......................................................... 103 

5.2 Limitations and Future Work .......................................................................... 105 

5.3 Final Remarks ................................................................................................. 106 

Appendices ................................................................................................................ 108 

Bibliography ............................................................................................................. 118 

 

 



 

 

ix 

 

List of Tables 

Table 2.1 Studies of Cost and Time Performance in Relation to Delivery Method ... 21 

Table 2.2 Project Distribution ..................................................................................... 32 

Table 2.3 Cost and Duration Statistics ........................................................................ 32 

Table 2.4 Direct Comparisons of Cost Growth Ratios by Delivery Method .............. 37 

Table 2.5 Direct Comparisons of Time Growth Ratios by Delivery Method ............. 37 

Table 2.6 Longitudinal Analysis Result for Cost Performance .................................. 44 

Table 2.7 Longitudinal Analysis Result for Time Performance ................................. 48 

Table 3.1 Select Terms in Social Network Analysis .................................................. 58 

Table 3.2 Project Profile ............................................................................................. 65 

Table 3.3 Organization Profile at the Parent Level ..................................................... 68 

Table 3.4 Organization Profile at the Root Level ....................................................... 69 

Table 3.5 Small-world Assessment Result ................................................................. 74 

Table 3.6 Exponential Random Graph Model Result ................................................. 78 

Table 4.1 Regression Results ...................................................................................... 97 

 



 

 

x 

 

List of Figures 

Figure 1.1 Selected Transportation P3 Projects in the US ............................................ 9 

Figure 1.2 Dissertation Structure ................................................................................ 12 

Figure 2.1 Distribution of Project Delivery Durations ............................................... 33 

Figure 2.2 P3 Market Annual Growth for US Major Transportation Projects ........... 35 

Figure 2.3 Individual Cost Trajectories ...................................................................... 46 

Figure 2.4 Individual Duration Trajectories ............................................................... 50 

Figure 3.1 Data Preparation and Analysis Workflow ................................................. 67 

Figure 3.2 Market Networks by Delivery Method for US Major Transportation 

Projects ................................................................................................................ 70 

Figure 3.3 P3 Equity Market Network for US Major Transportation Projects ........... 71 

Figure 4.1 Project Distribution by Contract Size and Delivery Method ..................... 94 

Figure 4.2 Distributions of DBE Goal and DBE Attainment by Delivery Method .... 95 

Figure 4.3 Project Distribution by Contract Size and Delivery Method ................... 100 

 

  



 

 

xi 

 

List of Abbreviations 

3R Resurfacing, Restoration, and Rehabilitation 
AIAI Association for the Improvement of American Infrastructure 
ANOVA Analysis of Variance 
ACC Average Clustering Coefficient 
ACG Award Cost Growth 
APL Average Path Length 
ATC Alternate Technical Concept 
ATG Award Time Growth 
CCG Contract Cost Growth 
CFR Code of Federal Regulations 
CTG Contract Time Growth 
BIL Bipartisan Infrastructure Law 
BF Build-Finance 
CMAR Construction Manager at Risk 
DB Design-Build 
DBB Design-Bid-Build 
DBE Disadvantaged Business Enterprise 
DBF Design-Build-Finance 
DBFOM Design-Build-Finance-Operate-Maintain 
DBM Design-Build-Maintain 
DBIA Design-Build Institute of America 
DOT Department of Transportation 
ENR Engineering News-Record 
ERGM Exponential Random Graph Model 
FHWA Federal Highway Administration 
FOIS Federal Highway Administration Organization and Information 

System 
EPC Engineering, Procurement, and Construction 
ERGM Exponential Random Graph Model 
GMP Guaranteed Maximum Price 
HOT High Occupancy Toll 
ICC Intra-Class Correlation 
ISMP Information Source for Major Projects 
ISTEA Intermodal Surface Transportation Efficiency Act 
ITS Intelligent Transportation System 



 

 

xii 

 

MBE Minority-Owned Business Enterprise 
NEPA National Environmental Protection Act 
O&M Operation and Maintenance 
P3 Public-Private Partnership 
PAB Private Activity Bond 
PDM Project Delivery Method 
REML Restricted Maximum Likelihood 
SEP-14 Special Experimental Project Number 14 
SEP-15 Special Experimental Project Number 15 
SNA Social Network Analysis 
SPV Special Purpose Vehicle 
STA State Transportation Agency 
TCG Total Cost Growth 
TIFIA Transportation Infrastructure Finance and Innovation Act 
TTG Total Time Growth 
USDOT US Department of Transportation 

 

 



 

 
 

1 

Chapter 1: Introduction 

Hailed as an alternative of and an antithesis to design-bid-build (DBB) with 

integrated delivery and private financing, public-private partnership (P3) has played 

an integral part in the development of US transportation infrastructure. 

Incommensurate with P3’s sprawling growth is empirical research, inadequate in at 

least three areas: 1) whether P3 has better cost and time certainty; 2) collaboration 

pattern among the market participants; and 3) how small, disadvantaged firms have 

fared with P3 projects. This dissertation attempts to bridge these knowledge gaps by 

examining the US experience using a comprehensive database and robust methods. 

Sections 1.1 and 1.2 provide the context of US experience. 

1.1 US Transportation Infrastructure 

The provision, upkeep, and improvement of US transportation infrastructure are no 

mean feat, with an existing stock of more than 4.1 million route miles of public roads 

and more than 600 thousand bridges (USDOT 2021). The US Department of 

Transportation (USDOT) budgeted $145.3 billion for its operation in FY 2024, of 

which $70.2 billion belonged to the Federal Highway Administration (FHWA) 

(USDOT 2023). Meanwhile, the primary source of federal appropriations for 

federally aided projects, the Highway Trust Fund, has seen and will continue to see a 

dwindling balance—barring any changes in the revenue structure, with a projected 

negative balance in both highway and transit accounts as early as 2028 (CBO 2023). 

To ease the pressure on funding and advance project delivery schedules, private 

financing, road usage charging, tolling, and public-private partnership (P3) have been 



 

 
 

2 

advanced as solutions (Geddes 2011; Poole 2018). Private financing allows a private 

firm to borrow against its own credit or the project’s toll/farebox revenues and to hold 

an equity stake in the project, thereby financing the project to “get the ball rolling.” 

Usually, public funding/financing or some form of government subsidy/guarantee 

needs to supplement private financing, resulting in a hybrid model known as P3. 

Road usage charging follows the “users-pay/users-benefit” principle and is a direct 

response to the dwindling gas-tax-based revenue for transportation infrastructure 

funding (e.g., the federal Highway Trust Fund). With the increasing use of fuel-

efficient and electric vehicles, the traditional gas tax falls short of reflecting actual 

road usage and, therefore, inadequately funds transportation. While road usage 

charging (a.k.a. mileage-based user fees) charges motorists registered in a jurisdiction 

for using the road system based on the actual/estimated mileage traveled, tolling is 

asset-specific and charges the user for traveling on the facility. For better or worse, 

tolled or partially tolled facilities have been delivered and/or operated under a P3 

model. Tolling entangles with the P3 model so much that opposition to tolling could 

embroil the use of P3. For example, with the Pennsylvania court declaring the Major 

Bridges P3 Initiative void ab initio over the legitimacy of imposing tolls (Ceisler 

2022), the state legislature revamped its P3 authorization to limit the state’s ability to 

levy user fees in a P3 (Wolf 2022). 

Yet, P3, as a contracting vehicle to mobilize projects and complement the 

aforementioned strategies while retaining public ownership and keeping liability “off 

the balance sheet,” has appealed to the palate of many governments (Murphy 2008). 

Section 1.2 formally defines P3, so what the researchers mean by P3 is clear. 



 

 
 

3 

1.2 Public-Private Partnership 

1.2.1 Definition 

P3 carries many meanings, including a political-cultural phenomenon, a policy, a 

governance tool, a delivery method, and a project (Greve and Hodge 2013). For the 

purposes of this dissertation, the researchers reference two definitions. 1) The World 

Bank (2017) defined P3 as “a long-term contract between a private party and a 

government entity, … in which the private party bears significant risk and 

management responsibility, and remuneration is linked to performance.” 2) Eggers 

and Startup (2006) described it as “a contractual agreement formed between a 

government agency and a private sector entity that allows for greater private sector 

participation.”  

A helpful way of characterizing P3 is through typology. Within contracts franchising 

long-term operation and/or maintenance among other services, three main types of P3 

have emerged in the US: asset recycling (a.k.a. long-term lease), revenue-risk (a.k.a. 

toll-concession), and availability-payment. If the roadway asset is already operating 

and revenue-positive (or potentially revenue-positive), a private party takes over to 

operate, maintain, and improve it for a specific period; the government, in turn, takes 

a sizable upfront fee—this is usually what the government is after in this type of deal. 

This arrangement is asset recycling (e.g., Chicago Skyway). Or, the facility may not 

have been built, and the government solicits a developer to design, construct, operate 

and/or maintain it; in this case, the private party may or may not take the revenue risk, 

i.e., a revenue-risk deal (e.g., I-77 Express Lanes) or an availability-payment deal 

(Presidio Parkway Phase II), respectively. In the latter model, the public sponsor pays 



 

 
 

4 

the developer milestone payments at construction milestones and regular, unitary 

availability payments adjusted for noncompliance penalties over the lifetime of the 

concession agreement for keeping the lanes “available.” 

Another way of enumerating P3 types is by specifying services delivered: build-

finance (BF), design-build-finance (DBF), design-build-finance-maintain (DBFM), 

design-build-finance-operate-maintain (DBFOM), design-build-maintain (DBM), and 

design-build-operate-maintain (DBOM). Essentially, any combination that exists in 

practice with an “F” (private financing), “O” (operation), or “M” (maintenance) 

affixed to a “B” (build) is a P3. 

Progressive P3s (a.k.a. pre-development agreements), a variant of traditional P3 

(typically a DBFOM type), have gained traction. A progressive P3 brings in a 

developer, based on qualifications (and possibly a fee structure), to help scope and 

shape the project. Governed under a pre-development agreement, the developer works 

with the public sponsor to advance project design, allocate risks, and refine the 

financing structure (Casady and Garvin 2022). A separate agreement may be executed 

following sufficient design development—an example being North Tarrant Express 

Segments 3A & 3B (TxDOT 2013, 2009). Or, a single development agreement can 

authorize both pre-development work and development work, as in the case of the 

Maryland Op Lanes (MDOT and MDTA 2021). 

Various risk-sharing and incentive mechanisms could be incorporated. For example, 

in what Gross and Garvin (2011) called upside revenue sharing and Rouhani et al. 

(2018) called the collar type of minimum revenue guarantee, the Capital Beltway 

High Occupancy Toll (HOT) Lanes project specified three tiers of revenue sharing 



 

 
 

5 

percentages based on the concessionaire’s internal rate of return (Titus-Glover et al. 

2016). With an 85-year concession including 80 years of operation and maintenance 

(O&M), the $2.1-billion revenue-risk DBFOM added two 14-mile HOT lanes in each 

direction in the median of I-495 in Fairfax County, Virginia. Another example is the 

I-595 Corridor Roadway Improvements project. The $1.7-billion 35-year-concession 

(30-year-O&M) availability-payment DBFOM widened 11 miles of I-595 mainline in 

Broward County, Miami, adding three reversible express lanes with variable tolls. 

The developer could earn a $50-million bonus for meeting all eight milestones, 

subject to a 1% deduction per day of delay off the respective milestone bonus (FDOT 

2009). 

1.2.2 P3 Implementation in the US 

Private toll roads and bridges in the US date back to the late 1700s, e.g., the 

Philadelphia to Lancaster Turnpike in Pennsylvania and a private toll bridge 

connecting Boston with Charlestown in Massachusetts (Buxbaum and Ortiz 2009; 

Poole 2018). Developments before World War II relied on both public 

funding/financing and private financing. In an analysis of federal legislation 

authorizing infrastructure procurements between 1789 and 1933, Miller (2000) 

showed that 62% of laws belonged in Quadrant II (combined delivery with private 

financing) and 31% in Quadrant I (combined delivery with public appropriations). 

The early 1900s saw more private toll roads and bridges with a rapid growth of motor 

vehicles—notably, the Long Island Motor Parkway in Nassau and Suffolk counties, 

New York, and the Ambassador Bridge linking Detroit, Michigan, with Windsor, 

Ontario. Development of public authority toll roads surged from the end of the War to 



 

 
 

6 

the 1960s (Poole 2018). From the 1956 launch of the interstate highway program to 

the early 1990s, segmented and direct procurement was in full swing. According to 

Miller (2000), federal laws in this period dominated Quadrant IV (segmented delivery 

with public appropriations). “A combined or systems delivery approach [had] not 

been generally available,” and “indirect [(private)] finance [had] not been part of 

America’s public infrastructure development policy” (Miller 2000). However, the 

state and local governments had extensively used municipal bonds to finance their 

capital assets (Marlowe 2015). In the late 1980s, influenced by neoliberalism, which 

characterizes the privatization of national assets, Virginia and California enacted their 

P3 laws. Soon after, the two states developed Dulles Greenway and State Route 91 

Express Lanes, respectively, both of which opened to traffic in 1995.2 Also 

facilitating the onset of P3 in the US was New Public Management, promoting a 

performance-driven approach to administration and an increased private role, an idea 

developed in the 1980s and popularized in the 1990s (Bovaird 2010; Casady et al. 

2020). Signaling a switch of mindset at the federal level was FHWA’s institution of 

Special Experimental Project Number 14 (SEP-14) in 1990, authorizing state 

recipients of federal grants to use innovative contracting techniques (e.g., design-

build and cost-plus-time-bidding). Congress further opened the door. The Intermodal 

Surface Transportation Efficiency Act of 1991 (ISTEA) permitted tolls to a much 

greater degree than in the past on federal-aid facilities, allowing for various P3 

possibilities (Walcoff & Associates 1991). “For the first time, private entities were 

 

2 Teodoro Moscoso Bridge in Puerto Rico was the first executed P3 contract, signed in December 
1991. However, Puerto Rico enacted its P3 enabling law after Virginia and California did theirs. 



 

 
 

7 

allowed to own [federal-aid] toll facilities, and states were allowed to loan the federal 

share of a project cost to another public agency or private entity constructing the 

project” (USDOT 2004). 

The US was among the last developed countries to adopt the P3 model, partially due 

to its functioning municipal bond system (Yescombe 2007). Unsurprisingly, overseas 

developers and toll operators became the early leaders—and some still are today—in 

the US private sector. For example, the Spanish developer Abertis owned 75% equity 

of Teodoro Moscoso Bridge, the first greenfield transportation P3 in the US; the 

French toll operator Cofiroute (VINCI) owned and operated State Route 91 Express 

Lanes. A few more P3 deals closed through the early 2000s; among them were 

Virginia’s Pocahontas Parkway and South Carolina’s Southern Connector. Both 

projects used 63-20 non-profit corporations to float tax-exempt toll revenue bonds 

under Internal Revenue Service Rule 63-20.3 While these projects leveraged little 

public funding/financing, the federal TIFIA program—created in 1998 under the 

Transportation Infrastructure Finance and Innovation Act and reauthorized in 

subsequent transportation acts—unlocked the potential for P3s to scale and multiply. 

Administered by USDOT, TIFIA provides direct loans, loan guarantees, and standby 

lines of credit with favorable terms to finance surface transportation projects.4 

 

3 For this to work, a state or political subdivision must have a “beneficial interest” in the non-profit 
during the bond term. The title to the property must revert to the public sponsor when the bonds are 
retired without consideration. Another type of 63-20 financing is through a lease-back arrangement, 
where the bonds are secured by future lease payments from the public sponsor (Smith 2009). 
4 USDOT generally limits loans and loan guarantees to no more than 33% of the eligible project cost, 
whereas lines of credit can take up a maximum of 49% share. The interest rate is the Treasury rate 
(one-half the Treasury rate for rural infrastructure projects) for a similar maturity at closing. Maturity 
on loans can be up to 35 years from Substantial Completion, with up to five years of deferred 
repayments (USDOT 2015). 



 

 
 

8 

Another bonanza was the 2004 Special Experimental Project Number 15 (SEP-15). 

Designed to encourage P3 approaches to project delivery, SEP-15 allows for 

experiments deviating from traditional project approval procedures and policies under 

Title 23 while ensuring FHWA’s stewardship to protect the public interest and the 

environment (FHWA 2006, 2015). For example, in the I-595 Corridor Roadway 

Improvements project, the Florida Department of Transportation requested—and the 

request was approved—experimental features allowing each pre-qualified proposer to 

separately negotiate a conditional TIFIA term sheet without delaying the procurement 

process, which would have otherwise waited for a successful proposer to come out 

before letting the developer engage with the TIFIA Joint Program Office (Ray 2008). 

The market saw a wave of asset-recycling P3s in the mid-to-late 2000’s, featuring a 

99-year lease of Chicago Skyway to Cintra/Macquarie, a Spanish/Australian 

consortium, and Transurban’s (an Australian toll operator) taking over the operation 

of Pocahontas Parkway from the 63-20 corporation, also for 99 years. P3 Investment 

has since taken off. 

While the federal government facilitated P3 uptake and best practices, state 

authorization was vital in driving P3 adoption. By October 2023, 41 states plus Puerto 

Rico and District of Columbia had authorized P3 in one form or another (AIAI 2023). 

Major transportation projects5 that have been awarded the contract as of May 1, 2022, 

totaled over $56 billion, according to the Information Source for Major Projects 

 

5 Major transportation projects are federal-aid projects with a total capital cost of over $500 million, 
including all cost elements incurred from environmental study and planning to final completion. 



 

 
 

9 

(ISMP) database.6 The FHWA Center for Innovative Finance Support, an 

authoritative knowledge hub on P3, keeps a list of representative P3 projects on their 

website. Figure 1.1 shows the location and type of those projects. 

 

Figure 1.1 Selected Transportation P3 Projects in the US 
Note: P3 = public-private partnership. Those were the same highway and transit P3 projects listed by 
Federal Highway Administration, Center for Innovative Finance Support 
(https://www.fhwa.dot.gov/ipd/p3/p3_projects/) as of October 2023. 
 

Another indicator of the scale of P3 implementation is private activity bonds (PABs). 

Intended to attract private investment, PABs are municipal bonds issued by and on 

behalf of state or local governments with zero tax on interest for “qualified projects” 

where more than 10% of the proceeds from bond issue are used for private business. 

A 2005 amendment to Section 142 of the Internal Revenue Code added highway and 

 

6 Refer to Section 1.4 for more about the ISMP database. 



 

 
 

10 

surface freight transfer facilities as “qualified projects” and capped USDOT’s PABs 

authority at $15 billion. This amount was nearly depleted by 2021 when Congress 

passed the Bipartisan Infrastructure Law (BIL), which expanded PABs’ authority to 

$30 billion By August 2023, USDOT had approved $17 billion worth of PABs on P3s 

and privately-owned projects (e.g., Brightline), although not all of them were 

highway and transit projects, the focus of this dissertation (Build America Bureau 

2022). Cross-referencing those projects with ISMP revealed over $6-billion PABs 

issued for 14 of the major transportation P3 projects in the database. 

1.3 Research Needs, Aims, and Purpose 

Despite wide implementation of the delivery method, empirical research is lacking in 

three areas. First, while P3 bolsters efficiency gains (GAO 2008), performance 

assessments reported conflicting results (Deep et al. 2022; Raisbeck et al. 2010), 

likely due to the cross-sectional study design and the legacy problem of data quality. 

Performing longitudinal analysis on a robust data set could reconcile the mixed 

results and instill more confidence in the furtherance of nationwide policy support of 

P3. Second, understanding of the relations among P3 players is far from adequate. 

Notably, evidence has yet to demystify the claim that P3 is a game of a few with a 

high entry barrier and is preferential to sophisticated, large firms (Hodge and Greve 

2005). This calls for quantitative approaches befitting the dependency structure of 

relational data and a sufficiently large data set. Third, the myth that P3 market is not 

equitable also haunts socially and economically disadvantaged firms. This warrants 

checking if those firms have received differential treatment in the execution of 

affirmative action programs. Accordingly, this investigation aims to assess the cost 



 

 
 

11 

and time performance, interorganizational collaboration, and disadvantaged firms’ 

participation in transportation P3 projects in the US. The purpose of this investigation 

was to assess the P3’s public benefits through examining the US transportation P3 

experience. 

1.4 ISMP Database 

Empirical studies are as good as the data. Using a one-of-a-kind, comprehensive 

database for over 140 major transportation projects, this investigation ensures high 

data quality and reliable results. The open-access ISMP database 

(http://www.transportationproject.org) for which the researchers led the data 

collection efforts covers various data types, including yearly cost and funding 

breakdowns, estimated and actual milestones, stakeholders, and O&M data. The 

University of Maryland team also collected over 3TB of project documents spanning 

the planning (e.g., environmental documents), procurement (e.g., contracts), 

construction (e.g., construction progress reports), and O&M phases (e.g., traffic and 

revenue reports). 

1.5 Dissertation Structure 

The dissertation adopts the three-paper model (Figure 1.2), with each paper 

constituting a self-contained, comparative study between P3 and DBB. Chapter 2 uses 

piecewise linear growth curve modeling to assess P3’s cost and time performance 

while also checking if a learning process that diffuses efficiency to the DBB process 

exists. Chapter 3 checks collaboration patterns in P3 and DBB market networks 

inclusive of different types of participants. Exponential random graph models probe 



 

 
 

12 

the myths of a high entry barrier and preference for large firms. Chapter 4 examines 

the USDOT’s Disadvantage Business Enterprise (DBE) program. Linear regressions 

compare the difference between P3 and DBB in terms of DBE goal and actual 

attainment. Chapter 5 summarizes the three studies and suggests future avenues of 

research to continue to benefit the public. 

 

Figure 1.2 Dissertation Structure 

 



 

 
 

13 

Chapter 2: Performance and Efficiency Diffusion (Paper #1) 

This chapter is based on the paper submitted to the Journal of Construction 

Engineering and Management (Zhang, K., Q. Cui, and P. DeCorla-Souza (Submitted 

2023). “Performance and Efficiency Diffusion of US Surface Transportation Public-

Private Partnerships”. J. Constr. Eng. Manage., TBD.) 

2.1 Abstract 

With successes and attritions, the US transportation industry has implemented public-

private partnership (P3) projects for over three decades. However, performance 

inquiries of P3 have not kept up with its development, showing mixed results. The 

predominantly cross-sectional approach may mask a certain within-project effect on 

cost and time certainty. Using 133 major transportation projects from the Information 

Source for Major Projects database, the authors performed piecewise linear growth 

curve modeling to parse out variation attributable to delivery method and variation 

attributable to that within-project effect, conveniently termed “efficiency diffusion.” 

Results revealed that P3 is on par with DBB in terms of cost certainty but performs 

better in time certainty. Contrastingly, efficiency diffusion from P3 is evident in cost 

but not in time. In other words, P3 makes for a spillover and internalization of 

technical and managerial innovations inducing cost savings. The first longitudinal 

study to reconcile previous cross-sectional efforts, this study showed P3’s better time 

certainty over DBB using project-level data in the US while confirming the delivery 

method’s cost-saving spillover effect, effectively “raising the bar.” 



 

 
 

14 

2.2 Introduction 

Public-private partnerships (P3) have been widely implemented by state 

transportation agencies across the US. A major argument for using P3 is to leverage 

private sector’s efficiency in project delivery and incentivize private partners for on-

budget and on-time project development and execution. This is especially crucial for 

large transportation projects where cost overruns and delays are not uncommon and 

have raised considerable public concerns (Flyvbjerg et al. 2009). The current P3 

practice needs to be examined for its performance results. In their monograph on the 

subject, Casady et al. (2022), recognizing the dissonant results, called for more 

country-specific studies. The current transportation authorization law, the Bipartisan 

Infrastructure Law (BIL), both supports and exercises caution over the use of P3. The 

BIL requires that all projects seeking federal credit assistance with a total cost of 

$750 million or more must perform a value-for-money analysis, where the P3 option 

is compared against the publicly funded and less integrated alternatives (Muriello 

2022). It is critical and urgent to understand P3 performance using empirical 

evidence, so new investments under the BIL can be efficiently spent for public 

interest. 

Despite a pressing need for P3 performance assessment, not many studies used 

objective performance measures (from hard project data vis-à-vis subjective, 

surveyed opinions of project performance) comparing P3 to non-P3, e.g., design-bid-

build (DBB) and design-build (DB), especially in the transportation sector. Yet, the 

few that used objective performance measures and conducted statistical analyses 

reported mixed results. Some studies found P3’s performance advantage over non-P3. 



 

 
 

15 

Those claiming a cost advantage include Raisbeck et al. (2010), Ramsey and El 

Asmar (2020), and Verwij and van Meerkerk (2021). Raisbeck et al. (2010) studied 

the social, transport, water, and information technology sectors in Australia and 

asserted that P3 had superior cost performance over traditional delivery methods; 

Ramsey and El Asmar (2020) held that P3 associated with a smaller cost growth than 

DB in transportation projects in the US; Verwij and van Meerkerk (2021) concluded 

the same based on Dutch transportation projects. In support of P3’s time performance 

advantage, Atmos et al. (2017) examined Indonesian power plant projects and 

contended that P3 had a smaller time growth than engineering, procurement, and 

construction (EPC) contracts. Of note, the same studies found no difference in time 

performance (Raisbeck et al. 2010; Ramsey and El Asmar 2020; Verweij and van 

Meerkerk 2021) or in cost performance (Atmo et al. 2017). Fathi and Shrestha (2022) 

found that neither cost performance nor time performance significantly differed 

between P3 and DB highway projects in the US. 

The differing results across studies are perplexing and perhaps an artifact of the cross-

sectional study design. This study used growth curve modeling on 133 projects from 

34 states to capture the time dimension while examining the presence/absence of P3 

performance advantage. A concept called “efficiency diffusion” was introduced and 

incorporated in cost and time performance assessment. To afford this analysis, this 

study capitalized on the Information Source for Major Projects (ISMP) database 

(http://www.transportationproject.org) which the authors took leading roles in 

building (Zhang et al. 2022). The most comprehensive database covering the 

lifecycles of major transportation projects in the US, ISMP adopts a standard process 



 

 
 

16 

and uniform terminology and only reports cost and schedule data from documents 

prepared for the Federal Highway Administration (FHWA). 

2.3 Literature Review 

Despite a focus on P3, it is helpful to review the literature on DB performance as 

well, since P3 is in many ways a scaled-up and more integrated version of DB. There 

exist four main research directions (objectives or subobjectives). The first direction is 

a direct comparison of performance by project delivery method. Analysis techniques 

include t-test, Welch’s t-test (Tran et al. 2018), and one-way analysis of variance 

(ANOVA) for normal data; and Mann-Whitney U test (Verweij and van Meerkerk 

2021) and Kruskal-Wallis test (Choi et al. 2020) for nonnormal data. The purpose of 

this direction is to find a superior delivery method based on a specific measure. The 

second direction is constructing performance measures. For example, using principal 

component analysis, Lam et al. (2007) developed an index for evaluating DB project 

success. Considering performance objective indicators of different stakeholders and 

using system dynamics, Xiong et al. (2015) proposed a satisfaction adjustment model 

to measure project performance. Liang et al. (2020) performed k-means clustering on 

ratio-scale cost and time performance measures for DB projects; their purpose of 

doing scale transformation was to assess differences in numerical project attributes 

among the cluster levels. The third direction is determining factors affecting project 

performance. A major approach is factor analysis, with which Chan et al. (2001) 

derived six factors based on 31 project success factors for DB projects. Regressing a 

performance variable on those six factors identified three critical success factors. The 

fourth direction is investigating the relationships between factors and performance 



 

 
 

17 

measures or predicting project performance. Methods include Pearson and Spearman 

correlation tests (Shrestha and Maharjan 2018), t-test (Fathi et al. 2020), one-way and 

factorial ANOVA (Liang et al. 2020), fuzzy cluster analysis (Nguyen et al. 2020), 

linear discriminant analysis (Kim et al. 2008), linear regression (Papajohn and El 

Asmar 2020; Zheng et al. 2018), logistic regression (Lu et al. 2017), discrete choice 

modeling (Park and Kwak 2017), and structural equation modeling (Mathew et al. 

2021). This study pursues the first and fourth lines of inquiry using growth curve 

modeling, an approach yet to be applied in this field, to the authors’ best knowledge. 

The authors summarized the findings of relevant studies in Table 2.1 (P3 studies are 

in bold letters). Different performance measures were used and called different 

names. To better compare the results, the authors defined the following performance 

measures. 

 Award	Cost	Growth	(ACG) =
Award	Cost − Planned	Cost

Planned	Cost  (2.1) 

 Contract	Cost	Growth	(CCG) =
Actual	Cost − Award	Cost	

Award	Cost 	 (2.2) 

 Total	Cost	Growth	(TCG) =
Actual	Cost − Planned	Cost	

Planned	Cost  (2.3) 

 Award	Time	Growth	(ATG) =
Award	Duration − Planned	Duration

Planned	Duration  (2.4) 

 Contract	Time	Growth	(CTG) =
Actual	Duration − Award	Duration	

Award	Duration 	 (2.5) 

 Total	Time	Growth	(TTG) =
Actual	Duration − Planned	Duration	

Planned	Duration 	 (2.6) 

where Actual Cost = final cost; Actual Duration = final construction duration; Award 

Cost = cost estimate at contract award; Award Duration = construction duration 



 

 
 

18 

estimate at contract award; Planned Cost = engineer’s estimated cost before contract 

award; Planned Duration = engineer’s estimated construction duration before contract 

award. These performance measures can extend to the project level as well. Whereas 

a P3 or a DB project usually comprises a single P3 or DB contract, a DBB project 

may consist of multiple DBB contracts. For example, the Californian DBB project I-5 

HOV (High Occupancy Vehicle) North [State Route 134 to State Route 118] 

consisted of four segments. Segment 1 #07-1219U4 was awarded in February 2010 

for $109.3 million, whereas the last contract awarded was Segment 3 #07-1218W4 in 

November 2012 at $147.3 million. In such case, “contract” means the first major 

DBB contract, and “award” means the award of the first major DBB contract. As 

such, Contract Cost Growth of a DBB project refers to the relative increase of project 

cost from the award of the first major DBB contract to the final project cost. 

As mentioned earlier, studies produced conflicting P3 performance results. On the 

cost side, some studies showed that CCG was similar between P3 and EPC (Atmo et 

al. 2017) and that TCG was similar between P3 and DBB (Deep et al. 2022 p. 7) and 

between P3 and DB (Fathi and Shrestha 2022). Notwithstanding, P3 was also 

reported to associate with a smaller ACG and TCG than non-P3 (Raisbeck et al. 

2010) and a smaller CCG than DB (Ramsey and El Asmar 2020). On the schedule 

side, similar ATGs, CTGs, and TTGs were observed between P3 and non-P3 

(Raisbeck et al. 2010), and similar CTGs (Ramsey and El Asmar 2020) and TTGs 

(Fathi and Shrestha 2022) were found between P3 and DB. Contrastingly, Atmo et al. 

(2017) revealed that P3 was associated with a smaller CTG than EPC. 



 

 
 

19 

Also not reassuring are conflicting DB performance results. For cost performance, 

some studies reported DB and DBB had similar ACGs (Minchin et al. 2013; Shrestha 

et al. 2012), CCGs (Choi et al. 2020; Park and Kwak 2017; Shrestha et al. 2012), and 

TCGs (Minchin et al. 2013). In other studies, DB was associated with a shorter CCG 

(Hale et al. 2009; Konchar and Sanvido 1998; Nguyen et al. 2020) and TCG 

(Shrestha et al. 2007) than DBB. For time performance, DB and DBB were found to 

have no significant difference in terms of ATG, CTG (Minchin et al. 2013; Shrestha 

et al. 2012), or TTG (Shrestha et al. 2007). In comparison, other studies showed that 

DB performed better on ATG (Park and Kwak 2017) and CTG (Konchar and Sanvido 

1998; Hale et al. 2009; Park and Kwak 2017; Choi et al. 2020). Results could differ 

depending on the project size (Tran et al. 2016) and project type (Tran et al. 2018). 

Reflecting on the result inconsistency, the authors identified four challenges facing 

existing studies. First, they predominantly focused on contract-level performance 

such that the cost excluded elements such as right of way and utilities, and the 

construction duration was only specific to the contract, not the larger project. This 

might have missed (in)efficiencies underlying the “project” delivery processes. For 

example, a DBB project arguably consists of a more linear sequence of work 

packages than P3/DB, which, by design, has more design-construction-interface 

efficiency and efficiency arising from increased overlapping of work packages. 

Performance at the project level deserves more attention, in that usually at the 

completion of the project (vis-à-vis the contract), the whole or part of the facility is 

operation-ready, which appeals more to the public. Second, data quality may be a 

concern. The surveyed individuals could have different interpretations of the 



 

 
 

20 

requested data, and the consulted databases could have relied on secondary sources. 

Third, the time points of cost and duration estimates were sometimes unclear, 

compromising the comparability across studies (Bain 2010). Fourth, the cross-

sectional study design hinders uncovering a certain within-project effect and 

accommodates as far as a growth ratio between two time points, not a growth rate per 

unit time, whereas the time dimension may contain valuable information. For 

example, a $5-million cost increase in two years connotes different cost performance 

than if the time span was ten years. In response, this study prepares a panel data and 

conducts a longitudinal analysis in hopes of reconciling the discordant results 

afflicting the subject. 

 



 

 
 

21 

Table 2.1 Studies of Cost and Time Performance in Relation to Delivery Method 

Study Data Sample Size Data Period Project Type Method Finding 
      Cost Time 

Konchar and Sanvido 
(1998) Survey 

351 (44% 
DB, 23% 
CMAR, 33% 
DBB 

NA Building 

Univariate 
analysis and 
linear 
regression 

DB < DBB for CCG DB < DBB for CTG 

Ling et al. (2004) 
Singapore Building 
and Construction 
Authority 

87 (54 DBB, 
33 DB) 

Completed 
after 1992 Building Linear 

regression 

Contractor’s paid-up capital 
is negatively associated 
with CCG in DB 

Importance for project to be 
delivered is positively 
associated with CTG in DB 

El Wardani et al. 
(2006) Survey 76 (DB) Completed 

1984-1997 Building Mood's 
median test 

Qualification-based 
selection < other 
procurement methods for 
CCG in DB 

No significant difference 
among procurement methods 
in DB for CTG 

Shrestha et al. (2007) Texas DOT 
database 

15 (4 DB, 11 
DBB) 

DB 
(completed 
2001-2006), 
DBB 
(completed 
2000-2004) 

Highway One-way 
ANOVA DB < DBB for TCG 

No significant difference 
between DB and DBB for 
TTG 

Hale et al. (2009) 
Naval Facilities 
Engineering 
Command 

77 (38 DB, 
39 DBB) 

Authorized 
1995-2004 Building  One-way 

ANOVA DB < DBB for CCG DB < DBB for unnormalized 
CTG 

Migliaccio et al. 
(2010) 

Survey, published 
project information, 
previous research 
data, and state DOT 
documents and 
websites 

146 (DB) NA Highway and 
bridge 

Pearson 
correlation 

Procurement duration has 
little relationship with CCG 
in DB 

Procurement duration is 
negatively associated with 
CTG in DB 

Raisbeck et al. 
(2010) 

Australian 
parliamentary and 
governmental 
documents and 
websites 

54 (21 P3, 
33 non-P3) 

Commenced 
after 2000 Mixed 

Mann-
Whitney U 
test 

P3 < non-P3 for ACG 
(from original 
announcement to award) 
and TCG (from budget 
approval to final) 

No significant difference 
between P3 and non-P3 for 
ATG (from original 
announcement to award), 
CTG, or TTG (from original 
announcement to final and 
from budget approval to 
final) 

Chasey et al. (2012) Various databases 
and interview 12 (P3) Completed 

1990-2010 
Highway and 
bridge 

No statistical 
test NA NA 



 

 
 

22 

Shrestha et al. (2012) Texas and other 
state DOTs 

22 (6 DB, 16 
DBB) 

Completed 
2000-2009 Highway 

One-way 
ANOVA, 
Welch’s t-
test, and 
Pearson and 
Spearman’s 
correlation 
tests 

No significant difference 
between DB and DBB for 
ACG or CCG 

No significant difference 
between DB and DBB for 
CTG 

Minchin et al. (2013) Florida DOT 60 (30 DB, 
30 DBB) 

Completed 
2002-2010 Highway 

t-test and 
Mann-
Whitney U 
test 

No significant difference 
between DB and DBB for 
ACG and TCG 

No significant difference 
between DB and DBB for 
CTG 

Ramsey and El 
Asmar (2015) 

3 databases and 
interview 25 (P3) Completed 

1995-2013 

Transportation 
(road, bridge, 
tunnel, rail, and 
multimodal) 

No statistical 
test NA NA 

Chen et al. (2016a) DBIA database 418 (DB) 
Data 
collection by 
Jul 2014 

Mixed One-way 
ANOVA 

GMP < other contract 
methods for CCG in DB 

Qualification-oriented 
selection < cost-oriented 
selection for CTG in DB 

Chen et al. (2016b) DBIA database 418 (DB) 
Data 
collection by 
Jul 2014 

Mixed 
One-way 
and two-way 
ANOVA 

GMP < lump sum for CCG 
in DB 

No significant difference 
between GMP and lump sum 
for CTG in DB 

Tran et al. (2016) 

5 State DOTs 
(Florida, Indiana, 
Ohio, Oregon, and 
Utah) 

2976 (210 
DB, 2766 
DBB) 

NA Highway 

t-test, 
Welch's t 
test, and 
Mann-
Whitney U 
test 

DB < DBB in projects $10 
million - $50 million for 
CCG; no significant 
difference between DB and 
DBB in projects $2 million 
- $10 million and over $50 
million for CCG 

NA 

Atmo et al. (2017) 

Published project 
information and 
commercial 
documents 

56 (28 P3, 
28 EPC) 

Completed 
2000-2014 Power plant t-test 

No significant difference 
between P3 and EPC for 
CCG 

P3 < EPC for CTG 

Lu et al. (2017) 
Survey of projects 
in China and other 
Asian countries 

144 (71 DB, 
73 DBB) NA Building 

Factor 
analysis and 
logistic 
regression 

Cost overrun is sensitive to 
contractor abilities in DB NA 

Park and Kwak 
(2017) Florida DOT 

1512 (255 
DB, 1257 
DBB) 

Let from 
2001 and 
completed 
by 2010 

Transportation 
(road, bridge, 
ITS, etc.) 

Discrete 
choice 
modeling 
and linear 
regression 

No significant difference 
between DB and DBB for 
CCG 

DB < DBB for CTG 



 

 
 

23 

Fathi and Shrestha 
(2018) DBIA database 57 (DB) 

Data 
collection by 
July 2017 

Highway and 
building 

Mann-
Whitney U 
test 

No significant difference 
between highway and 
building projects for CCG 
in DB 

No significant difference 
between highway and 
building projects for CTG in 
DB; CTG < CCG for both 
highway and building 
projects in DB 

Hasanzadeh et al. 
(2018) 22 State DOTs 

56 (23 DB, 6 
CMAR, 27 
DBB) 

NA Highway 

Kruskal-
Wallis test 
and Mann-
Whitney U 
test 

No significant difference 
between DB and DBB for 
cost overrun 

DB < DBB for time overrun 

Tran et al. (2018) Florida DOT 
278 (139 
DB, 139 
DBB) 

Completed 
1999-2018 

Transportation 
(road, bridge, 
3R, ITS, etc.) 

t-test, 
Welch's t 
test, and 
Mann-
Whitney U 
test 

DB < DBB in 
miscellaneous construction 
projects for CCG 

DB < DBB in 3R projects for 
TTG 

Choi et al. (2020) Florida DOT 610 (59 DB, 
551 DBB) 

Completed 
2002-2011 Highway (3R) 

Tukey HSD 
pairwise test 
and Kruskal-
Wallis test 

No significant difference 
between DB and DBB for 
CCG 

DB < DBB for CTG 

Fathi et al. (2020) DBIA database and 
state DOTs 97 (DB) NA 

Highway and 
water and 
wastewater 

t-test 

Water and wastewater < 
highway for change order 
cost as a percentage of 
award cost in DB 

No significant difference 
between DB highway and 
DB water and wastewater for 
CTG 

Liang et al. (2020) DBIA database 167 (DB) Completed 
2008-2019 Mixed Factorial 

ANOVA 

Progressive DB < DB w/ 
other procurement methods 
for CCG 

No significant difference 
among procurement methods 
for CTG in DB 

Nguyen et al. (2020) 28 state DOTs 254 Completed 
2004-2015 

Transportation 
(road, bridge, 
ITS, etc.) 

Fuzzy 
cluster 
analysis 

DB < DBB in new, 
complex, and highly risky 
projects for CCG; DB > 
DBB in certain 
reconstruction projects for 
CCG 

NA 

Papajohn and El 
Asmar (2020) 7 state DOTs 31 (DB) Completed 

2008-2018 
Highway and 
bridge 

Linear 
regression 

Agency years of experience 
with DB is positively 
associated with ACG 

NA 

Ramsey and El 
Asmar (2020) 

3 databases and 
interview 

75 (27 P3, 
48 DB) 

Completed 
1995-2015 

Transportation 
(road, bridge, 
tunnel, rail, and 
multimodal) 

Mann-
Whitney U 
test 

P3 < DB for CCG No significant difference 
between P3 and DB for CTG 



 

 
 

24 

Mathew et al. (2021) 28 state DOTs 118 (DB) Completed 
2004-2015 

Transportation 
(road/drainage, 
bridge, etc.) 

Factor 
analysis and 
structural 
equation 
modeling 

Project complexity has a 
direct positive effect on 
CCG in DB 

NA 

Verweij and van 
Meerkerk (2021) 

Dutch 
Rijkswaterstaat 
database 

65 (9 P3, 56 
DB) 

Commenced 
2008-2017 

Transportation 
(road, bridge, 
and tunnel) 

Mann-
Whitney U 
test 

P3 < DB for CCG No significant difference 
between P3 and DB for CTG 

Deep et al. (2022) 

Integrated Database 
on Indian 
Infrastructure 
Projects 

997 (516 P3, 
481 DBB) 

Completed 
1996-2017 Road Paired t-test 

No significant difference 
between P3 and DBB for 
TCG 

P3 < DBB for TTG 

Fathi and Shrestha 
(2022) 

Online research and 
state DOT 
documents 

144 (22 P3, 
122 DB) 

Completed 
1995-2020 Highway 

Mann-
Whitney U 
test 

No significant difference 
between P3 and DB for 
TCG 

No significant difference 
between P3 and DB for TTG 

Koppenjan et al. 
(2022) 

Dutch 
Rijkswaterstaat 
database, survey, 
and interview 

30 from 
database (12 
P3, 18 DB); 
161 from 
survey (110 
P3, 51 DB) 

Database 
data 
collection: 
Apr 2018-
Nov 2019; 
survey data 
collection: 
Oct 2019-
Feb 2020 

Transportation 
(road and 
waterway) 

Mann-
Whitney U 
test and 
linear 
regression 

Database: P3 < DB for 
CCG; survey: no significant 
difference between P3 and 
DB for CCG from survey 

Database: P3 < DB for CTG; 
survey: no significant 
difference between P3 and 
DB for CTG from survey 

Note: 3R = resurfacing, restoration, and rehabilitation; ACG = Award Cost Growth; ATG = Award Time Growth; ANOVA = analysis of variance; DB = design-
build; DBB = design-bid-build; DBIA = Design-Build Institute of America; DOT = Department of Transportation; EPC = engineering, procurement, and 
construction; CCG = Contract Cost Growth; CTG = Contract Time Growth; GMP = guaranteed maximum price; ITS = intelligent transportation system; NA = 
not applicable or not available; P3 = public-private partnership; TCG = Total Cost Growth; TTG = Total Time Growth. The studies in bold involved public-
private partnerships. 
 



 

 
 

25 

2.4 Method 

This study used piecewise linear growth curve modeling under the multilevel 

modeling framework. This approach is appropriate for four reasons. First, it 

operationalizes the notion of cost and time growth rates. Second, it distinguishes 

within-project and between-project sources of variation. Level 1 models the time 

dimension for each project (within-project), and level 2 models the project-level 

characteristics (between-project). This setup assumes that the repeated measures have 

a correlational structure due to unobserved variables, and that correlational structure 

is captured in what is called random effects (Hothorn and Everitt 2009). Random 

effects are project-specific, whereas fixed effects reflect the population characteristics 

(Fitzmaurice et al. 2011). In the growth curve modeling context, random effects 

represent variance of the individual trajectories (i.e., within-project patterns of 

change) around the mean trajectory pooling all projects in the sample (Curran et al. 

2010). Third, the multilevel modeling framework is preferred over the other 

approach, the structural equation modeling framework, when assessment schedules 

are highly variable (Grimm et al. 2017). In other words, it is better suited to model a 

continuous time scale. Fourth, “piecewise linear” allows for modeling differential 

growth rates pre-award and post-award. 

2.4.1 Hypotheses 

2.4.1.1 Cost and Time Certainty 

In favor of P3’s cost and schedule advantages are five explanations. First, with 

bundling or vertical integration of design, construction, finance, operation, and 

maintenance components, the P3 developer is incentivized to better control cost and 



 

 
 

26 

schedule in order to maximize its gains (Iossa and Martimort 2015; Koppenjan et al. 

2022). Bundling also leads to economies of scope and increased efficiency 

(Koppenjan et al. 2022). Second, private financing and performance-dependent 

payments spur on-time delivery since a delay will cost the developer additional 

interest and trigger a deduction or withholding of payments. Also, financiers’ 

involvement brings additional monitoring and precipitates more effective risk 

management (van Meerkerk et al. 2022). Third, milestone payments in P3 (vis-à-vis 

invoiced payments in DBB) reduces front-loading (i.e., the most profitable work is 

completed first), allowing for more sensible costing and scheduling and better cost 

and schedule risk control. Fourth, a less-opportunistic bidding behavior in P3 helps 

reduce chances of surprise elements later. Because developers take more risk than 

DBB contractors, they are less likely to submit a lowball bid; instead, they add a risk 

premium. So, it is less likely to see a drastic increase in project cost after award 

because much of the risk has been priced in the bid. Fifth, cross-sector collaboration 

such as P3 drives value creation and improves project outcomes through 

complementarity in capabilities (Mahoney et al. 2009; Roberts and Siemiatycki 

2015). Therefore, the authors propose the following hypotheses: 

H2.1a: P3 projects have better cost certainty than DBB projects. 

H2.1b: P3 projects have better time certainty than DBB projects. 

2.4.1.2 Efficiency Diffusion 

In a broader view, cost certainty and time certainty are but two aspects of project 

delivery efficiency (Hodge and Greve 2017; Sharma et al. 2019). Other aspects 



 

 
 

27 

include value for money and technical efficiency. The former includes traditional 

value for money focusing on project financials and value for money further 

incorporating the social impact such as emissions and fuel consumption (DeCorla-

Souza and Lee 2017). Technical efficiency gauges the relationship between resource 

inputs and outputs. Examples include project-focused technical efficiency considering 

lanes and miles as outputs (Daito and Gifford 2014) and infrastructure input-output 

efficiency for economic development (Chen 2021). While a high-level definition of 

efficiency is the ability to produce the same product with minimal resources, this 

study focuses on how well the project meets its on-budget and on-time goals. 

Following the previous argument for P3’s cost and schedule advantages, with 

increasing investment in P3 transportation projects, could DBB projects have 

benefited from the P3 experience and, therefore, have improved their cost and time 

performance? After comparing the public-private hybrid arrangement for prison 

operation (i.e., private operator with public agent supervision) to the state-run model 

in Brazil, Cabral et al. (2013) claimed that “some routines and organizational 

procedures developed in the hybrid mode will spill over to other similar activities in 

the public sector.” Domingos et al. (2020) examined cross-sector collaboration in the 

Brazilian education sector and contended that the private partner’s novel practices 

had transferred not only to the schools they collaborated with (targeted units) but also 

to other schools managed under the same district managers but did not partake in the 

program (non-targeted units). Consequently, performance of both targeted and non-

targeted units improved. The boundary spanners (district managers) played an 

important role in the performance improvement of the non-targeted units: the higher 



 

 
 

28 

the internalization of practices by the boundary spanner, the more the performance 

improvement. In light of these advancements and to better facilitate discussion, the 

authors coined the term “efficiency diffusion” to describe the spillover of technical 

and managerial innovations underlying an efficient outcome from an integrated 

process such as P3 to a less integrated, public process such as DBB, leading to 

performance improvement of the latter process. The learning process may happen 

within the same delivery method; in this case, the innovations are internalized. This 

positive externality (i.e., beneficial side effect of a process) is in line with strategic 

management theorists’ belief that cross-sector collaboration induces sustainable 

development value creation (Mahoney et al. 2009). Also in concert with this view 

were investigations into infrastructure P3s’ spillover on sustainable economic growth 

(Chen 2021) and a multi-actor innovation ecosystem approach to space technology 

procurement (Mazzucato and Robinson 2018). Yet, efficiency diffusion so defined 

lacks empirical evidence in the transportation context. To this end, the authors make 

the following hypotheses. 

H2.2a: P3’s cost efficiency diffuses to DBB projects such that DBB’s cost 

performance improves. 

H2.2b: P3’s time efficiency diffuses to DBB projects such that DBB’s time 

performance improves. 

2.4.2 Data 

The analysis capitalized on ISMP, the most comprehensive database of US 

transportation major projects by far, featuring project life-cycle data ranging from 



 

 
 

29 

yearly cost and funding breakdowns to estimated and actual key milestone dates. All 

projects were FHWA major projects, meaning they each 1) cost more than $500 

million to build (although a phased project may cost less than that for the funded 

phase as seen in the prepared data set); and 2) had received or were due to receive 

some form of grant or credit assistant from FHWA. Data quality was assured because 

of the following qualifications. 1) ISMP follows standard processes and uniform 

terminology (Zhang et al. 2022); 2) data were sourced from primary sources such as 

Cost Estimate Reviews and Financial Plans, which were prepared by the sponsoring 

agency and FHWA in accordance with FHWA’s risk-based estimate guidelines; 3) 

data were additionally cross-checked with FHWA’s FOIS (FHWA Organization and 

Information System) major projects website, which publishes approved budgets and 

estimated completion dates every year, albeit not as detailed as ISMP; (e.g., ISMP 

reports the scope of work of estimates, allowing for parallel comparisons.) 4) having 

built the ISMP and involved in every stage of the data collection process, the authors 

conducted another round of data verification with the original documents to ensure 

data accuracy for the intended analysis. 

2.4.2.1 Project Distribution 

Updated through May 1, 2022, the data set contained 133 projects representing 34 

states, of which 95 had been completed. DBB projects, usually comprising multiple 

DBB contracts, were 64 in number, whereas the numbers of P3 and DB projects, each 

comprising a single P3 and DB contract, were 38 and 32, respectively. Contract 

award—of the first major DBB contract in a DBB project or the P3/DB contract in a 

P3/DB project—ranges from September 1991 to March 2021. To be included, a 



 

 
 

30 

project must 1) have a defined set of constituent segments and geographic limits; 2) 

have identified funding sources in the Financial Plans; and 3) have available at least 

two risk-based cost estimates or two schedule estimates made at two separate 

milestones as defined in the next subsection. Project type includes roadway (including 

new construction and widening and reconstruction), bridge/tunnel (with a bridge or 

tunnel comprising a main component of the project), and other (e.g., a combination of 

highway and rail). All projects totaled $185.5 billion in terms of total capital cost. In 

year-of-expenditure dollars, total capital cost includes preliminary engineering, right 

of way, utilities, construction, construction administration, risks, public outreach, 

finance charge, developer expense, concession fee, management reserve, and others. 

Of note, preliminary engineering as a cost element is differentiated from preliminary 

design as a stage in project development. The former refers to pre-construction 

activities such as testing, environmental documentation, and agency’s design, 

whereas the latter refers to additional design beyond conceptual design to ~30% plan. 

Tables Table 2.2 and Table 2.3 present more information on project characteristics. 

Figure 2.1 plots two distributions by delivery method: duration from NEPA Start to 

Substantial Completion and duration from NEPA Completion to Substantial 

Completion, where NEPA stands for the National Environmental Protection Act and 

is a required environmental review procedure for major federal actions. NEPA Start 

represents the start of the respective NEPA class of action, be it Environmental 

Impact Statement, Environmental Assessment, or Categorical Exclusion, in 

descending order of environmental impact (Zhang et al. 2022). NEPA Completion 

refers to the last major reevaluation (covering major design modifications or phasing 



 

 
 

31 

for construction advertisement) of the respective decision document. For both 

durations, P3 and DB associated with a shorter duration than DBB. 

 



 

 
 

32 

Table 2.2 Project Distribution 

N 
Total Capital Cost  Project Type 

≤ 
$500M 

$500M-
$1000M 

$1000M-
$2000M 

$2000M-
$3000M 

> 
$3000M  Roadway Bridge/Tunnel Other 

DBB 5 35 16 4 3  52 10 1 

DB 4 11 12 3 2  21 10 1 

P3 2 13 14 7 2  30 8 0 
All 11 59 42 14 7  103 28 2 

Note: DB = design-build; DBB = design-bid-build; NA = not applicable; P3 = public-private partnership. 
 

Table 2.3 Cost and Duration Statistics 

 

N 
(Completed) 

N 
(All) Total Capital Cost (M$) 

 
Construction Duration (Month) 

 
  

(Mean, Median, SD, Min, Max)  (Mean, Median, SD, Min, Max) 

DBB 38 63 (1366.6, 793.8, 2094.9, 116.1, 14808.0)  (129.9, 106.5, 81.4, 26.6, 417.0) 

DB 26 32 (1293.8, 1098.9, 958.2, 130.6, 4825.0)  (63.4, 58.1, 24.8, 11.3, 124.6) 

P3 31 38 (1472.7, 1312.9, 842.0, 307.3, 3862.8)  (56.0, 54.9, 14.2, 28.8, 96.3) 

All 95 133 (1379.4, 921.0, 1574.8, 116.1, 14808.0)  (92.8, 68.8, 67.6, 11.3, 417.0) 

Note: DB = design-build; DBB = design-bid-build; P3 = public-private partnership. Statistics are based on the last 
available update. 
 

 



 

 
 

33 

 
Figure 2.1 Distribution of Project Delivery Durations 
Note: DB = design-build; DBB = design-bid-build; NEPA = National Environmental Protection Act; 
P3 = public-private partnership. The crosses and bars in the boxes represent the means and medians, 
respectively. 
 

2.4.2.2 Cost and Schedule Estimates 

For each project, data preparation allowed up to four cost estimates and four schedule 

estimates, made at the following milestones: 

• Milestone 1: conceptual design completion 

• Milestone 2: preliminary design completion 

• Milestone 3: Notice to Proceed 2 for P3/DB and Notice to Proceed 1 for DBB 

• Milestone 4: final/last available update 

 



 

 
 

34 

Data preparation also followed the following rules. To ensure parallel comparison, 

costs in this study refer to estimates of the total capital cost spanning from the 

planning phase to the construction phase; durations refer to construction durations. In 

line with FHWA’s practice (FHWA 2014), when the source is Cost Estimate 

Reviews, the 70th percentile cost and completion date were taken as the estimated 

values. These cost estimates typically exclude financing costs, and adjustments were 

made to the Financial Plan cost estimates to ensure parallel comparison. Construction 

duration was calculated as months between the estimated/actual construction start 

date and the respective estimated/actual Substantial Completion date. Substantial 

Completion marks when the facility is available for use and is arguably more relevant 

than Final Completion for the purposes of this study. 

2.4.2.3 P3 Market Annual Growth 

To capture efficiency diffusion, assumed to be proportional to the cumulative relative 

P3 experience in the US market, and to indicate its direction, the authors created a 

variable called P3	Market	Share. It is defined as the ratio of cumulative award cost 

of P3 projects and cumulative award cost of all projects in the sample. Formally, 

P3	Market	Share in Year	k is represented as: 

 P3	Market	Share! =
[P3"]#×%AAward	Cost"&B%×'AYear&B'×#
[1]#×%AAward	Cost"&B%×'AYear&B'×#

 (2.7) 

 

Similarly, P3	&	DB	Market	Share in Year	k is: 



 

 
 

35 

 P3	&	DB	Market	Share! =
([P3"]#×% + [DB"]#×%)AAward	Cost"&B%×'AYear&B'×#

[1]#×%AAward	Cost"&B%×'AYear&B'×#
 (2.8) 

where, 

Award	Cost!" 

= ,Cost	Estimate	of	Project	i	at	Milestone	3, if	j = Year	of	Milestone	3	of	Project	i
0, otherwise ; 

DB! = 11, if	Project	i	is	a	DB0, otherwise ; P3! = 11, if	Project	i	is	a	P30, otherwise ; 

Year" = 11, j ≤ k
0, k < j ≤ l; l = total number of years; n = total number of projects. Figure 

2.2 shows the market share and cumulative number of projects undertaken for P3 as 

well as P3 and DB. 

 

Figure 2.2 P3 Market Annual Growth for US Major Transportation Projects 
Note: DB = design-build; P3 = public-private partnership. 



 

 
 

36 

2.4.3 Preliminary Analysis 

The authors obtained cost and time growth ratios as per Equations (2.1-(2.6 and 

performed direct comparisons by delivery method. The two pre-award milestones 

(i.e., Milestone 1 and Milestone 2) allowed for additional growth ratios. For example, 

long-term ACG and short-term ACG still invoked Equation (2.1, but referenced 

Milestones Milestone 1 and Milestone 2, respectively. The same applies to the other 

performance measures in Equations (2.2-(2.6. As such, a total of five cost growth 

ratios and five time growth ratios were compared. Involving three delivery methods—

P3, DB, and DBB, comparisons used one-way ANOVA when the normality of 

residuals and homoscedasticity conditions were met and Kruskal Wallis test when 

normality of residuals was violated. Post-hoc tests used Tukey HSD and Mann-

Whitney U tests, respectively. Tables Table 2.4 and Table 2.5 summarize the results. 

No significant difference was found in any of the five cost growth ratios (Table 2.4). 

But in four of the five time growth ratios (except for long-term ATG), P3 performed 

better than DBB (Table 2.5). 

 



 

 
 

37 

Table 2.4 Direct Comparisons of Cost Growth Ratios by Delivery Method 

 Long-term ACG Short-term ACG CCG Long-term TCG Short-term TCG 

 

(N, Mean, Median, 
SD, Min, Max) 

(N, Mean, Median, 
SD, Min, Max) 

(N, Mean, Median, 
SD, Min, Max) 

(N, Mean, Median, 
SD, Min, Max) 

(N, Mean, Median, 
SD, Min, Max) 

DBB 
(21, 12.1%, 3.1%, 
47.2%, -52.4%, 

181.0%) 

(56, 3.1%, 0.2%, 
14.3%,  

-22.7%, 50.6%) 

(61, 6.6%, -0.2%, 
30.9%,  

-25.7%, 156.0%) 

(20, 12.4%, 3.4%, 
50.0%,  

-55.3%, 181.0%) 

(55, 13.0%, 1.3%, 
46.4%,  

-30.7%, 234.0%) 

DB 
(15, -1.7%, 0.3%, 
23.1%, -51.6%, 

42.6%) 

(29, 1.0%, 1.5%, 
18.6%,  

-49.6%, 45.9%) 

(30, 8.6%, 0.8%, 
19.4%,  

-7.9%, 77.1%) 

(15, 14.3%, 0.9%, 
49.3%,  

-52%, 152.0%) 

(27, 11.0%, 1.0%, 
34.0%, -50.1%, 

102.0%) 

P3 
(9, -10.9%, -11.2%, 

15.4%, -38.3%, 
17.2%) 

(32, -3.9%, -2.0%, 
12.7%,  

-34.5%, 23.7%) 

(37, 4.2%, 0.8%, 
11.5%,  

-12.4%, 59.6%) 

(8, -13.5%, -11.6%, 
13.8%,  

-40.2%, 1.7%) 

(31, 0.2%, -0.1%, 
13.3%,  

-32.1%, 36.4%) 

Kruskal 
Wallis 
(χ2, df, 
p-value) 

(3.638, 2, 0.162) (4.504, 2, 0.105) (3.353, 2, 0.187) (2.858, 2, 0.240) (0.410, 2, 0.815) 

 
Note: CCG = Contract Cost Growth; DB = design-build; DBB = design-bid-build; Long-term ACG = Award Cost 
Growth based on Milestone 1 estimate; Long-term TCG = Total Cost Growth based on Milestone 1 estimate; P3 = 
public-private partnership; Short-term ACG = Award Cost Growth based on Milestone 2 estimate; Short-term TCG 
= Total Cost Growth based on Milestone 2 estimate. 
 

Table 2.5 Direct Comparisons of Time Growth Ratios by Delivery Method 

 Long-term ATG Short-term ATG CTG Long-term TTG Short-term TTG 

 (N, Mean, Median, 
SD, Min, Max) 

(N, Mean, 
Median, SD, Min, 

Max) 

(N, Mean, Median, 
SD, Min, Max) 

(N, Mean, Median, 
SD, Min, Max) 

(N, Mean, 
Median, SD, Min, 

Max) 

DBB 
(11, 27.0%, 17.8%, 

32.6%, -20.8%, 
85.8%) 

(52, 9.8%, 4.1%, 
29.5%, -75.4%, 

135.0%) 

(61, 33.8%, 25.2%, 
38.6%, -23.1%, 

123.0%) 

(11, 66.6%, 37.9%, 
73.7%, -12.7%, 

202.0%) 

(52, 45.7%, 
35.5%, 58.1%, -
65.6%, 255.0%) 

DB 
(9, -11.0%, -8.9%, 

29.8%, -73.2%, 
31.6%) 

(27, -3.8%, -
3.4%, 19.1%, -
61.4%, 46.0%) 

(30, 20.1%, 9.5%, 
28.7%, -21.9%, 

99.9%) 

(9, 9.9%, 5.4%, 
43.6%, -73.7%, 

73.5%) 

(28, 15.6%, 6.2%, 
36.7%, -62.0%, 

82.8%) 

P3 
(9, -10.1%, -

16.0%, 53.3%, -
71.8%, 72.6%) 

(27, -3.5%, -
1.0%, 24.0%, -
64.9%, 72.6%) 

(37, 9.4%, 2.8%, 
20.9%, -16.3%, 

87.8%) 

(8, -15.5%, -
15.5%, 48.9%, -
69.9%, 63.7%) 

(26, 8.4%, 8.5%, 
28.0%, -62.5%, 

60.4%) 

One-way 
ANOVA (F, 
numerator df, 
denominator df, 
p-value) 

(3.091, 2, 26, 
0.063) NA NA (4.964, 2, 25, 

0.015) NA 



 

 
 

38 

Tukey HSD NA NA NA 

P3 < DBB (0.016); 
no significant 
difference between 
P3 and DB (0.651) 
or between DB and 
DBB (0.100) 

NA 

Kruskal Wallis 
(χ2, df, p-value) NA (12.176, 2, 0.002) (12.374, 2, 0.002) NA (10.242, 2, 0.006) 

Mann-Whitney 
U (p-value) NA 

P3 < DBB 
(0.013); DB < 
DBB (0.007); no 
significant 
difference 
between P3 and 
DB (0.829) 

P3 < DBB (0.002); 
no significant 
difference between 
P3 and DB (0.078) 
or between DB and 
DBB (0.195) 

NA 

P3 < DBB 
(0.012); DB < 
DBB (0.032); no 
significant 
difference 
between P3 and 
DB (0.649) 

Note: CTG = Contract Time Growth; DB = design-build; DBB = design-bid-build; Long-term ATG = Award Time 
Growth based on Milestone 1 estimate; Long-term TTG = Total Time Growth based on Milestone 1 estimate; NA = 
not applicable; P3 = public-private partnership; Short-term ATG = Award Time Growth based on Milestone 2 
estimate; Short-term TTG = Total Time Growth based on Milestone 2 estimate. 
 



 

 
 

39 

2.4.4 Model 

The authors performed piecewise linear growth curve modeling contrasting different 

groups of projects: P3 vs. DB vs. DBB with DBB as the reference group (Models 

2.1a-2.1d and 2.5a-2.5d); P3 & DB vs. DBB with DBB as reference (Models 2.2a-

2.2d and 2.6a-2.6b); P3 vs. DBB with DBB as reference (Models 2.3a-2.3b and 2.7a-

2.7d); and P3 vs. DB with DB as reference (Models 2.4a-2.4b and 2.8a-2.8b), as 

shown in Tables Table 2.6 and Table 2.7. All models incorporated a two-stage growth 

profile specifying separate slopes joined at Milestone 2. Understandably, from 

Milestone 2 onward, the mean trajectory may exhibit a different pattern than before 

the developer’s principal involvement. Slope coding used an incremental scheme 

such that the slope coefficient of the second time piece @t!"A#represents the deviation 

from the slope of the first time piece t!", where t!" and @t!"A# are defined in the 

example below. The a- and b-models and the c- and d-models included 

P3	Market	Share (Equation (2.7) and P3	&	DB	Market	Share (Equation (2.8) as a 

time-varying covariate, respectively. As such, efficiency diffusion was modeled as 

the within-project effect of market shares on the dependent variables: log(Cost) and 

log(Durations). Two random effects covariance matrices were used: the maximal 

structure (Equation (2.14) and a restricted structure assuming independence between 

the @t!"A# random effect and other random effects (Equation (2.16). Although Barr et 

al. (2013) recommended the maximal random effects structure which the a- and c-

models used, the b- and d-models adopted the restricted structure to show sensitivity 

of model specification. The models did not explicitly control for inflation and 

material cost escalation since cost estimates had: 1) been escalated to the proposed 



 

 
 

40 

midpoint of construction, by virtue of year-of-expenditure dollars (FHWA 2007); 2) 

considered uncertainty in escalation factor due to bidding competition and the 

economic condition. Equations (2.9-(2.15 exemplify Model 2.1a specification. By 

default, model estimation used the lmer function with the restricted maximum 

likelihood (REML) method in the lme4 package in R. Where lmer resulted in singular 

fits, the Bayesian method (the brm function) from the blme package was used instead 

(Chung et al. 2013). 

Level 1: 

 logICost"&J = β(" + β#"t"& + β)"It"&J* + β+&IP3	Market	Share&J + ε"&	 (2.9) 

 

Level 2: 

 β(" = γ(( + γ(#PDM" + γ()log(Duration") + ζ("	 (2.10) 

 β#" = γ#( + γ##PDM" + ζ#"	 (2.11) 

 β)" = γ)( + γ)#PDM" + ζ)"	 (2.12) 

 β+& = γ+(	 (2.13) 

 O
ζ("
ζ#"
ζ)"
P~MVNTO

0
0
0
P , W

τ() τ(# τ()
τ(# τ#) τ#)
τ() τ#) τ))

YZ	 (2.14) 

 ε"&~N(0, σ))	 (2.15) 

 



 

 
 

41 

where β$! = level-1 intercept; β%! = level-1 coefficient of t!"; β&! = level-1 coefficient 

of @t!"A#; β'" = level-1 coefficient of P3	Market	Share"; γ$$ = fixed effect intercept 

for β$!; γ$% = fixed effect coefficient of PDM! for β$!; γ$& = fixed effect coefficient of 

log(Duration!) for β$!;	γ%$ = fixed effect intercept for β%!; γ%% = fixed effect 

coefficient of PDM! for β%!; γ&$ = fixed effect intercept for β&!; γ&% = fixed effect 

coefficient of PDM! for β&!; γ'$ = fixed effect intercept for β'"; ε!" = time-specific 

residual error; ζ$! = random error for β$!; ζ%! = random error for β%!; ζ&! = random 

error for β&!; σ& = variance of 𝜀(); M
τ$& τ$% τ$&
τ$% τ%& τ%&
τ$& τ%& τ&&

O = covariance matrix of P
ζ$!
ζ%!
ζ&!
Q; 

Cost!" = cost estimate of Project	i in Year	j; Duration! = final/last available duration 

estimate of Project	i; P3	Market	Share": see Equation (2.7; PDM! = project delivery 

method of Project	i; t!" = number of years from Milestone 2 of Project	i to Year	j; 

@t!"A# = R
t!", if	t!" > 0
0, if	t!" ≤ 0. 

All else the same, Model 1b specified a restricted random effects structure: 

 O
ζ("
ζ#"
ζ)"
P~NTO

0
0
0
P , W

τ() τ(# 0
τ(# τ#) 0
0 0 τ))

YZ (2.16) 

 

2.5 Results 

Residual diagnostics indicated a violation of normality of residuals (Equation (2.15). 

However, considering mixed effects models were robust to this violation (Schielzeth 

et al. 2020), the authors proceeded with the present analysis. Tables Table 2.6 and 



 

 
 

42 

Table 2.7 present the analysis result. The intra-class correlations (ICCs) were high, 

indicating a nested data structure (i.e., repeated measurements of the same project 

were more similar than they were between projects) (Chen and Chen 2021). The 

marginal 𝑅& quantifies the proportion of variance explained by the fixed effects, 

whereas the conditional 𝑅& concerns both fixed and random effects. Figures Figure 

2.3 and Figure 2.4 show the predicted individual cost and duration trajectories for 

projects with all four estimates, respectively. 

2.5.1 Cost Performance 

Table 2.6 showed similar fixed effects between the a- and b models and between the 

c- and d-models. Positive coefficients of log(Duration) suggested that projects with a 

longer duration had a higher cost. Nonsignificant coefficients of t!" and @t!"A# 

suggested that neither pre-award slope nor post-award slope deviation (hereafter as 

slopes) attributable to the project being a DBB differentiated from zero. Compared to 

DBB, P3 and DB projects had a higher cost; but cost estimates were similar between 

P3 and DB. For any given project, P3 market growth and P3 and DB market growth 

negatively associated with its cost estimates. Take Model 2.1a for example, a 1% 

increase of P3	Market	Share associated with a -0.48% (i.e., *
,-./01+%
%$$

) cost cut across 

the board, i.e., the efficiency diffusion effect. Figure 2.3 illustrates this effect as the 

deviation from the dotted lines to the solid lines of the same color, where the dashed 

lines represent the counterfactual that efficiency diffusion from P3 was absent. 

Neither P3 nor DB had an effect on the slopes, albeit Models 2.1b, 2.3b, and 2.3d 

indicated a smaller pre-award slope. For P3 vs. DB, P3’s cost efficiency diffused to 



 

 
 

43 

DB, and P3 slopes did not differentiate from those of DB. Models 2.4c and 2.4d 

failed to detect an effect of P3	&	DB	Market	Share, possibly because of a smaller 

sample size, therefore less power, compared to Models 2.1c and 2.1d. Taken together, 

1) cost certainty did not differentiate between P3 and DBB or between P3 and DB; 

and 2) cost efficiency diffusion was evident. 

 



 

 
 

44 

Table 2.6 Longitudinal Analysis Result for Cost Performance 

 Model 
2.1a 

Model 
2.1b 

Model 
2.1c 

Model 
2.1d 

 Model 
2.2a 

Model 
2.2b 

Model 
2.2c 

Model 
2.2d 

 Model 
2.3a 

Model 
2.3b 

Model 
2.3c 

Model 
2.3d 

 Model 
2.4a 

Model 
2.4b 

Model 
2.4c 

Model 
2.4d 

Data subset P3 vs. DB vs. DBB (ref.)  P3 & DB vs. DBB (ref.)  P3 vs. DBB (ref.)  P3 vs. DB (ref.) 
N of 
observations / N 
of projects 

382 / 115 
 

382 / 115 
 

283 / 86 
 

200 / 60 

Dependent 
variable log(Cost) 

Fixed effect Coefficient (SE) 

  Intercept 4.321 
(0.511) 

4.363 
(0.528) 

4.446 
(0.516) 

4.482 
(0.546) 

 4.339 
(0.512) 

4.390 
(0.535) 

4.451 
(0.517) 

4.498 
(0.530) 

 4.371 
(0.564) 

4.428 
(0.569) 

4.627 
(0.558) 

4.721 
(0.568) 

 4.199 
(0.863) 

4.202 
(0.855) 

4.335 
(0.883) 

4.327 
(0.850) 

  log (Duration) 0.553 
(0.107) 

0.549 
(0.111) 

0.553 
(0.107) 

0.548 
(0.113) 

 0.549 
(0.107) 

0.542 
(0.112) 

0.550 
(0.107) 

0.544 
(0.111) 

 0.532 
(0.118) 

0.523 
(0.118) 

0.530 
(0.116) 

0.511 
(0.117) 

 0.674 
(0.212) 

0.672 
(0.210) 

0.648 
(0.215) 

0.652 
(0.208) 

  t 0.017 
(0.022) 

0.029 
(0.017) 

0.012 
(0.022) 

0.024 
(0.016) 

 0.016 
(0.022) 

0.028 
(0.017) 

0.012 
(0.022) 

0.023 
(0.016) 

 0.004 
(0.028) 

0.023 
(0.017) 

0.015 
(0.027) 

0.028 
(0.016) 

 0.020 
(0.027) 

0.019 
(0.026) 

0.006 
(0.025) 

0.005 
(0.025) 

  (t)+ -0.004 
(0.021) 

-0.015 
(0.016) 

0.002 
(0.021) 

-0.009 
(0.016) 

 -0.004 
(0.021) 

-0.014 
(0.016) 

0.002 
(0.021) 

-0.009 
(0.016) 

 0.005 
(0.027) 

-0.012 
(0.016) 

0.003 
(0.027) 

-0.011 
(0.016) 

 0.002 
(0.026) 

0.003 
(0.025) 

0.015 
(0.025) 

0.016 
(0.025) 

  DB 0.332 
(0.152) 

0.331 
(0.157) 

0.329 
(0.152) 

0.321 
(0.157) 

               

  P3 0.549 
(0.153) 

0.545 
(0.155) 

0.547 
(0.152) 

0.542 
(0.159) 

      0.523 
(0.152) 

0.521 
(0.149) 

0.540 
(0.150) 

0.525 
(0.151) 

 0.234 
(0.160) 

0.233 
(0.153) 

0.224 
(0.163) 

0.217 
(0.156) 

  P3 & DB      0.441 
(0.132) 

0.432 
(0.141) 

0.439 
(0.132) 

0.436 
(0.133) 

          

  P3 Market 
Share 

-0.647 
(0.252) 

-0.751 
(0.252) 

   -0.637 
(0.251) 

-0.713 
(0.258) 

   -0.360 
(0.268) 

-0.470 
(0.277) 

   -0.860 
(0.382) 

-0.840 
(0.370) 

  

  P3 & DB 
Market Share   -0.601 

(0.181) 
-0.638 
(0.185) 

   -0.581 
(0.181) 

-0.615 
(0.188) 

   -0.799 
(0.207) 

-0.798 
(0.219) 

   -0.470 
(0.277) 

-0.482 
(0.277) 

  t * DB 0.009 
(0.036) 

-0.013 
(0.028) 

0.012 
(0.037) 

-0.014 
(0.027) 

               

  t * P3 -0.031 
(0.039) 

-0.058 
(0.029) 

-0.030 
(0.040) 

-0.053 
(0.030) 

      -0.023 
(0.050) 

-0.062 
(0.030) 

-0.031 
(0.049) 

-0.060 
(0.029) 

 -0.043 
(0.036) 

-0.047 
(0.033) 

-0.034 
(0.035) 

-0.037 
(0.033) 

  t * (P3 & DB)      -0.008 
(0.030) 

-0.034 
(0.023) 

-0.007 
(0.030) 

-0.032 
(0.023) 

          

  (t)+ * DB -0.003 
(0.036) 

0.019 
(0.029) 

-0.004 
(0.036) 

0.023 
(0.028) 

               

  (t)+ * P3 0.026 
(0.039) 

0.054 
(0.030) 

0.025 
(0.039) 

0.048 
(0.031) 

      0.018 
(0.050) 

0.057 
(0.031) 

0.026 
(0.049) 

0.055 
(0.030) 

 0.032 
(0.037) 

0.036 
(0.035) 

0.021 
(0.036) 

0.023 
(0.035) 

  (t)+ * (P3 & 
DB) 

     0.009 
(0.030) 

0.034 
(0.024) 

0.008 
(0.030) 

0.034 
(0.024) 

          

Random effect                    



 

 
 

45 

  SD (Intercept) 0.585 0.592 0.584 0.594  0.588 0.598 0.588 0.599  0.560 0.569 0.555 0.569  0.597 0.594 0.603 0.603 

  SD (t) 0.069 0.029 0.071 0.029  0.068 0.030 0.070 0.029  0.101 0.022 0.101 0.022  0.034 0.028 0.034 0.029 

  SD ((t)+) 0.048 0.009 0.051 0.009  0.046 0.008 0.049 0.008  0.087 0.009 0.088 0.008  0.020 0.015 0.018 0.015 
  Cor (Intercept, 
t) -0.174 -0.270 -0.264 -0.253  -0.158 -0.275 -0.254 -0.272  -0.174 -0.101 -0.292 -0.142  -0.140 -0.370 -0.186 -0.344 

  Cor (Intercept, 
(t)+) 0.103  0.242   0.073  0.225   0.185  0.311   -0.270  -0.162  

  Cor (t, (t)+) -0.949  -0.946   -0.957  -0.953   -0.983  -0.981   -0.226  -0.180  

ICC 0.975 0.973 0.976 0.974  0.975 0.974 0.976 0.974  0.976 0.970 0.977 0.971  0.975 0.975 0.976 0.976 
Marginal 𝑅!/ 
conditional 𝑅! 

0.192 / 
0.979 

0.210 / 
0.977 

0.191 / 
0.980 

0.207 / 
0.978 

 0.185 / 
0.979 

0.195 / 
0.977 

0.184 / 
0.980 

0.193 / 
0.977 

 0.199 / 
0.981 

0.212 / 
0.975 

0.213 / 
0.982 

0.217 / 
0.976 

 0.163 / 
0.976 

0.161 / 
0.976 

0.154 / 
0.976 

0.153 / 
0.976 

Note: DB = design-build; DBB = design-bid-build; ICC = intra-class correlation; P3 = public-private partnership. The log(Duration) predictor took the final/last available duration 
estimate. Models 2.1a, 2.1c, 2.2a, 2.2c, 2.3a, 2.3c, 2.4a, and 2.4c used the maximal random effects structure, whereas Models 2.1b, 2.1d, 2.2b, 2.2d, 2.3b, 2.3d, 2.4b, and 2.4d used 
the restricted structure. Significant fixed effects at the 95% confident level are highlighted in bold. 
 



 

 
 

46 

 

Figure 2.3 Individual Cost Trajectories 
Note: DB = design-build; DBB = design-bid-build; P3 = public-private partnership. Only projects with 
all four cost estimates are plotted (N = 40). Trajectory prediction was based on Model 2.1a. Efficiency 
diffusion refers to the learning process whereby technical and managerial innovations are 1) spilled 
over from an integrated process (e.g., P3) to a less integrated process (e.g., DBB) and 2) internalized 
within the integrated process, resulting in efficiency gains. Model 2.1a modeled efficiency diffusion as 
the effect of P3	Market	Share on log(Cost). 
 

2.5.2 Time Performance 

The a- and c- models and b- and d- models showed similar fixed effects results (Table 

2.7). A higher cost associated with a higher duration. DBB had a positive pre-award 

slope in the b- and d-models, i.e., conditional on DBB, duration estimate increased 

before Milestone 2. P3 and DB associated with a shorter duration than DBB. No time 

efficiency diffusion was found. Despite sensitivity to the random effects structure, 

evidence existed that P3 negatively associated with duration growth rate, indicating 

better time performance than DBB. For example, Model 2.5b showed that, compared 



 

 
 

47 

to DBB, P3 projects had a pre-award duration drop of -15.3%/year (i.e., e+$.%-- − 1) 

and a post-award rate of change of -2.7%/year (i.e., e+$.%--#$.%'. − 1). Moreover, P3 

and DB had similar time performance (Models 2.8a-2.8d). 

 



 

 
 

48 

Table 2.7 Longitudinal Analysis Result for Time Performance 

 Model 
2.5a 

Model 
2.5b 

Model 
2.5c 

Model 
2.5d 

 Model 
2.6a 

Model 
2.6b 

Model 
2.6c 

Model 
2.6d 

 Model 
2.7a 

Model 
2.7b 

Model 
2.7c 

Model 
2.7d 

 Model 
2.8a 

Model 
2.8b 

Model 
2.8c 

Model 
2.8d 

Data subset P3 vs. DB vs. DBB (ref.)  P3 & DB vs. DBB (ref.)  P3 vs. DBB (ref.)  P3 vs. DB (ref.) 
N of 
observations / N 
of projects 

346 / 107 
 

346 / 107 
 

256 / 80 
 

179 / 54 

Dependent 
variable log(Duration) 

Fixed effect Coefficient (SE) 

  Intercept 2.471 
(0.404) 

2.451 
(0.419) 

2.594 
(0.416) 

2.572 
(0.436) 

 2.533 
(0.402) 

2.515 
(0.405) 

2.648 
(0.414) 

2.618 
(0.414) 

 2.195 
(0.516) 

2.174 
(0.514) 

2.240 
(0.530) 

2.216 
(0.538) 

 2.795 
(0.351) 

2.839 
(0.369) 

2.943 
(0.374) 

2.955 
(0.393) 

  log (Cost) 0.287 
(0.057) 

0.290 
(0.059) 

0.285 
(0.057) 

0.288 
(0.060) 

 0.278 
(0.057) 

0.281 
(0.058) 

0.278 
(0.057) 

0.281 
(0.056) 

 0.328 
(0.074) 

0.329 
(0.073) 

0.329 
(0.074) 

0.329 
(0.076) 

 0.163 
(0.048) 

0.160 
(0.051) 

0.164 
(0.050) 

0.162 
(0.051) 

  t 0.066 
(0.047) 

0.076 
(0.032) 

0.070 
(0.047) 

0.079 
(0.031) 

 0.067 
().047) 

0.078 
(0.032) 

0.070 
(0.047) 

0.081 
(0.031) 

 0.063 
(0.052) 

0.075 
(0.033) 

0.063 
(0.052) 

0.074 
(0.034) 

 -0.015 
(0.067) 

-0.049 
(0.053) 

-0.008 
(0.067) 

-0.039 
(0.053) 

  (t)+ -0.023 
(0.046) 

-0.034 
(0.031) 

-0.024 
(0.047) 

-0.034 
(0.031) 

 -0.024 
(0.046) 

-0.036 
(0.032) 

-0.024 
(0.046) 

-0.036 
(0.031) 

 -0.020 
(0.051) 

-0.033 
(0.032) 

-0.019 
(0.051) 

-0.032 
(0.033) 

 0.038 
(0.066) 

0.073 
(0.054) 

0.037 
(0.066) 

0.068 
(0.054) 

  DB -0.398 
(0.093) 

-0.399 
(0.098) 

-0.394 
(0.093) 

-0.400 
(0.093) 

               

  P3 -0.509 
(0.095) 

-0.512 
(0.096) 

-0.502 
(0.095) 

-0.510 
(0.097) 

      -0.519 
(0.105) 

-0.527 
(0.109) 

-0.521 
(0.104) 

-0.528 
(0.103) 

 -0.090 
(0.075) 

-0.086 
(0.082) 

-0.084 
(0.076) 

-0.084 
(0.085) 

  P3 & DB      -0.451 
(0.078) 

-0.450 
(0.079) 

-0.447 
(0.077) 

-0.449 
(0.080) 

          

  P3 Market 
Share 

-0.230 
(0.374) 

-0.234 
(0.384) 

   -0.271 
(0.372) 

-0.266 
(0.369) 

   -0.249 
(0.439) 

-0.197 
(0.441) 

   0.328 
(0.473) 

0.210 
(0.497) 

  

  P3 & DB 
Market Share   -0.396 

(0.297) 
-0.366 
(0.306) 

   -0.408 
(0.296) 

-0.383 
(0.294) 

   -0.247 
(0.373) 

-0.181 
(0.383) 

   -0.186 
(0.367) 

-0.187 
(0.415) 

  t * DB -0.083 
(0.077) 

-0.119 
(0.058) 

-0.084 
(0.077) 

-0.119 
(0.057) 

               

  t * P3 -0.137 
(0.070) 

-0.166 
(0.049) 

-0.138 
(0.070) 

-0.166 
(0.048) 

      -0.127 
(0.076) 

-0.170 
(0.050) 

-0.128 
(0.077) 

-0.168 
(0.049) 

 -0.052 
(0.088) 

-0.044 
(0.064) 

-0.052 
(0.087) 

-0.049 
(0.064) 

  t * (P3 & DB)      -0.114 
(0.061) 

-0.150 
(0.042) 

-0.115 
(0.061) 

-0.151 
(0.043) 

          

  (t)+ * DB 0.066 
(0.078) 

0.103 
(0.059) 

0.068 
(0.078) 

0.105 
(0.058) 

               

  (t)+ * P3 0.109 
(0.070) 

0.139 
(0.050) 

0.111 
(0.070) 

0.140 
(0.050) 

      0.100 
(0.076) 

0.143 
(0.052) 

0.100 
(0.076) 

0.142 
(0.052) 

 0.038 
(0.087) 

0.030 
(0.067) 

0.038 
(0.087) 

0.034 
(0.066) 

  (t)+ * (P3 & 
DB) 

     0.093 
(0.060) 

0.129 
(0.043) 

0.094 
(0.061) 

0.130 
(0.044) 

          

Random effect                    



 

 
 

49 

  SD (Intercept) 0.369 0.380 0.372 0.381  0.369 0.376 0.373 0.381  0.403 0.416 0.404 0.416  0.245 0.272 0.250 0.274 

  SD (t) 0.114 0.048 0.115 0.047  0.113 0.047 0.113 0.047  0.130 0.045 0.131 0.046  0.134 0.056 0.134 0.055 

  SD ((t)+) 0.089 0.011 0.091 0.011  0.086 0.012 0.088 0.010  0.102 0.013 0.104 0.012  0.102 0.016 0.104 0.016 
  Cor (Intercept, 
t) -0.219 -0.277 -0.202 -0.284  -0.204 -0.264 -0.188 -0.270  -0.214 -0.273 -0.200 -0.271  -0.417 -0.642 -0.387 -0.631 

  Cor (Intercept, 
(t)+) 0.167  0.137   0.154  0.126   0.177  0.156   0.312  0.272  

  Cor (t, (t)+) -0.941  -0.940   -0.941  -0.939   -0.968  -0.966   -0.962  -0.959  

ICC 0.883 0.883 0.886 0.883  0.884 0.881 0.887 0.884  0.892 0.889 0.893 0.889  0.756 0.760 0.767 0.764 
Marginal 𝑅!/ 
conditional 𝑅! 

0.389 / 
0.929 

0.408 / 
0.923 

0.390 / 
0.931 

0.409 / 
0.924 

 0.382 / 
0.928 

0.396 / 
0.922 

0.384 / 
0.930 

0.398 / 
0.923 

 0.387 / 
0.934 

0.403 / 
0.926 

0.388 / 
0.934 

0.401 / 
0.926 

 0.152 / 
0.793 

0.172 / 
0.797 

0.149 / 
0.802 

0.174 / 
0.798 

Note: DB = design-build; DBB = design-bid-build; ICC = intra-class correlation; P3 = public-private partnership. The log(Cost) predictor took the final/last available cost 
estimate. Models 2.5a, 2.5c, 2.6a, 2.6c, 2.7a, 2.7c, 2.8a, and 2.8c used the maximal random effects structure, whereas Models 2.5b, 2.5d, 2.6b, 2.6d, 2.7b, 2.7d, 2.8b, and 2.8d used 
the restricted structure. Significant fixed effects at the 95% confident level are highlighted in bold. 
 



 

 
 

50 

 

Figure 2.4 Individual Duration Trajectories 
Note: DB = design-build; DBB = design-bid-build; P3 = public-private partnership. Only projects with 
all four duration estimates are plotted (N = 27). Trajectory prediction was based on Model 2.5b. 
 

2.6 Discussion 

Contrary to H2.1a, P3’s cost certainty is similar to DBB. This is consistent with the 

cross-sectional result in Table 2.4. Regarding time certainty, H2.1b proved true that 

P3 holds a time performance advantage. Both results concur with Deep et al. (2022