Skip to contentU.S. Department of Transportation/Federal Highway Administration

Description of the National Highway Construction Cost Index

The National Highway Construction Cost Index (NHCCI) is intended as a price index that can be used both to track pure price-changes associated with highway construction costs and to convert current-dollar expenditures on highway construction to real- or constant-dollar expenditures. The NHCCI is intended to replace the Federal Highway Administration's (FHWA) Bid-Price Index (BPI) in the future, but also to be compared with the BPI for historical purposes. This report presents a description of the research into developing a new price index using Oman Systems, Inc. (OSI) Bid-Tabs data and the methodology used to produce the NHCCI on a quarterly basis. A companion document describes the mathematics of the indexing used in NHCCI. The conclusion is that a Fisher Index based on the OSI Bid-Tabs data gives results that are both representative of construction costs as a whole and are consistent with related price indexes, such as the PPI for Highway and Street Construction and the historical BPI.

Background

Since 1922, the Federal Highway Administration (FHWA) has compiled a continuing record of variations in contract bid prices on Federal-aid highway construction. This computation gives useful information on price changes that affect highway projects. The FHWA published this information quarterly.

Until April 2007, FHWA compiled the index using data on the award for contracts by the State highway agencies for the Federal-aid highway projects, except for those on the Federal-aid secondary system. Since 1977, FHWA counted only contracts greater than $500,000 in an attempt to reduce the paperwork burden. Until April 2007, the data was collected on Forms FHWA-45, entitled Bid Price Data, and FHWA-47, entitled Statement of Materials and Labor. Form FHWA-45 recorded information on bid prices for major work items on Federal-aid highway construction contracts. Form FHWA-47 recorded data on labor and materials usage for Federal-aid highway construction contracts. The FHWA used the data provided on the forms as input to create the Bid-Price Index (BPI), published quarterly in the Price Trends for Federal-Aid Highway Construction1. The development of the BPI is discussed in some detail in "Public Roads" Magazines Volumes 31, 36, and 45.

The BPI has many uses including: the Biennial Report to Congress, Status of the Nation's Highways, Bridges, and Transit: Conditions and Performance; the annual Highway Statistics and Our Nation's Highways publications; special studies of various material prices; and policy issues related to employment and materials usage. It is also used as a price deflator by the Bureau of Economic Analysis (BEA) in the production of the National Income and Product Accounts. However, the BPI has undergone criticism from the General Accountability Office (GAO) in their Report GAO-04-113R Comparison of States' Highway Construction Costs.2

That report reinforced the FHWA's belief that few States rely upon the data and that the quality of the BPI was below Department standards.3 Reviews of data reported on the Forms FHWA-45 and FHWA-47 show uneven data quality. In April 2007, FHWA discontinued the collection of Forms FHWA-45 and FHWA-47 because of data quality issues and to reduce the paperwork burden on States. The final quarterly report using the FHWA-45 data is for the fourth quarter of calendar year 2006.

Research into a replacement for the BPI began in 2000. One of two research reports documenting this research is Alternatives to the FHWA-47 and FHWA-45: Data Forecasting Future Inflation Rates for Highway Construction. This report presents an evaluation of 14 databases but found none that could be used to create the BPI. The second report, Development of a New Highway Cost Index, recommended a detailed data collection effort with broader scope than the current Forms FHWA-45 and FHWA-47 but with fewer data providers. The FHWA rejected this proposal as being both too costly to implement and presenting a small likelihood of success given the regulatory requirements to issue surveys.

The FHWA decided to purchase the Bid-Tabs database from Oman Systems, Inc. (OSI) to provide construction cost data for the NHCCI. This data captures the State web-posting of bids submitted on highway construction contracts. The OSI database contains historical State bid tabulations data. The amount of historical OSI bid-tabs data varies from State to State. The OSI database contains all States except for Alaska and Hawaii (as of September, 2009); some States with data back to the mid-1990's, but other States with data starting in the 2000's. As new data becomes available from the States, OSI processes the data and updates are downloaded into the Oman-developed Bid-Tabs propriety software. The Oman Bid-Tabs software includes an analytical system geared for State and contractor analysis, but not for national analysis. The FHWA developed its own software application (the NHCCI) to use Oman collected and compiled State Bid-tab data.

The new NHCCI quarterly construction cost index will replace the BPI with data starting in March of 2003. At that point in time, OSI was collecting data from 45 States. FHWA believes this is sufficient coverage to initiate the NHCCI with some data overlap with the BPI. Overlap coverage between the FHWA NHCCI and the FHWA Bid-Price data therefore is from the beginning of 2003 to the end of the 2006 calendar year.

Report Overview

The first section of this report describes the data and index in terms of basic characteristics and coverage. The second section describes how the data was analyzed for inclusion in the NHCCI. The third section presents results for the nation as a whole and some States and compares these results to related measures. The final section is a summary of findings and suggestions for future research and development. A separate report presents the chained Fisher Index methodology used to construct the index.

NHCCI as a Construction (Output) Cost Index

State highway agencies typically prepare a set of construction engineering design plans (called Plans, Specifications and Estimates, or PS&E) which include the construction plan, estimates of the type and quantity of materials, and specifications (grade, quality, etc.). These materials, along with other goods and services become State-defined bid items. States maintain long lists of these specific bid items, and the lists can often be found on State websites. Construction companies interested in the work then submit proposals based on the PS&E package. In these proposals, the price associated with each item includes the material itself, and all other associated costs for moving, placing, and installing the material. The price also includes a component of profit and overhead associated with each item. Clearly in this usage, bid-items include more than just the direct price of the item.

Price indexes in general combine prices of individual goods and quantity weights to track the percentage change in prices over time for a particular basket of goods. Implicitly, the quality of goods represented in a given time frame is assumed to be constant. For the NHCCI, individual 'goods' on the Bid-Tabs data are represented by 'pay-items' for successfully bid contracts. Pay-items are defined at the State level and so cannot be combined across States. During data preparation, each State is processed individually before the data is used to create the national index, so substitution across State lines is not an issue. For our purposes the relevant information which is included with each pay-item is: State, price, unit of measure, general expenditure category and the date the contract was awarded.

The NHCCI procedures are, therefore, in contrast to the procedures used by other price indexing agencies such as the U.S. Bureau of Labor Statistics who calculate the Producer Price Index for Highways and Streets. It is proper to think of the NHCCI as a construction (or output) cost index as opposed to an input price index.

The choice of the Fischer Ideal index embodies the idea that for a fixed market basket, changing relative prices will lead to changes in the relative quantities being purchased in the basket as entities make substitutions within an item category. Over time then, the market basket is changing, and the Fischer Ideal Index recognizes this explicitly. FHWA is adopting the view that since the government is purchasing the entire project, rising input costs could lead to a changed mix or timing of projects.

Both FHWA's BPI and the NHCCI indexes reflect the changing mix of inputs over time. Since the final product that the NHCCI measures is highway construction, then it is appropriate that the index reflect the mix of inputs. FHWA intends to monitor how the mix of goods in the NHCCI changes over time. The NHCCI is intended to cover the universe of highway projects and therefore arrive at an average cost index for all highway construction. One of the advantages of the Fischer index is that the market basket is updated throughout the index. This removes the bias that would otherwise occur if the mix of goods changes over time.

Data Preparation

The second major step in constructing the NHCCI is data preparation. Data preparation involves producing a 'clean' dataset of useable variables (goods identifiers, prices, and quantities) at a set time-frequency (e.g., monthly, quarterly, or annual). Cleaning the data involves identifying likely errors in the variables, problems with defining a consistent unit of measure for a particular good, and determining an appropriate time-frequency. Likely errors in the variables will be identified by looking for extreme values and analyzing other statistical measures. Problems with identifying consistent measurement units for a pay item may appear because the unit for the item is not precisely defined (e.g. lump-sum), or because goods with different characteristics are reported under a single pay-item.

Determining an appropriate time frequency involves balancing three characteristics: our desire for the highest frequency index possible (monthly or quarterly), the availability of complete or nearly complete series of observations for the included pay items, and the inclusion of enough pay items that the resultant index is representative of a sufficiently large subset of all highway construction cost components. These characteristics interact because pay items which occur infrequently are more likely to have a sufficient number of consecutive useable observations when the index is calculated at a lower frequency (e.g. annual). Possible reasons pay-items may occur infrequently are because they are specialized inputs to highway construction or because some items are seasonal. The overall question of seasonality in bid prices will not be addressed in this or subsequent steps, but could be taken up in further research.

Analysis of the Oman Systems Bid-Tabs with Respect to Constructing the NHCCI

The data analysis presented in this section of the report evaluates FHWA's use of Oman Systems, Inc.'s (OSI) Bid-Tabs data to construct a highway construction cost index. FHWA's objective in constructing a highway cost index is to provide a measure that is useful not only as a pure cost index for constructing federally subsidized highways, but also an index that the States can use to gauge differences in their relative cost experiences for highway construction. Two advantages of reliable, State-specific cost indexes are:

  • To help States monitor how quickly unit costs are rising to better match budget appropriations with actual expenditures; and
  • To provide States with a more informed bargaining position in writing contracts.

Either of these results would help the States to better meet the real transportation needs and desires of their citizens. Given the relevant indexing and economic theory, the requirements of the data are:

  1. Consistent and specific product characteristics so that price comparisons over time compare apples-to-apples rather than excavation in corn-fields to excavation in the Rockies.
  2. Continuity in the time series for the individual pay-items such that there are few breaks in the series. Mathematically, the index methodology requires prices and quantities to be available in consecutive periods.
  3. A sufficiently large or representative set of goods so that the resultant index is relevant for those that make the real budget decisions.

The first requirement above is fundamental to creating a well-defined cost index and involves taking care in specifying what constitutes an individual product. We use guidelines developed by the Bureau of Labor Statistics (BLS) to further inform our data analysis. BLS produces a wide variety of price/cost indexes, such as the Consumer Price Index for All Urban Consumers (CPI) and the PPI for various industries or stages of manufacturing. BLS' indexes are widely accepted and used. BLS outlines the criteria for determining what constitutes an individual good for the PPI in Chapter 14 of their Handbook of Methods—the overriding philosophy of which is that they treat any variation in the product, or in the delivery of the product, that affects the price as a price-determinant. The following excerpt is from the PPI Frequently Asked Questions (FAQ) at http://www.bls.gov/ppi/ppifaq.htm:

"How are producers and products selected for the PPI survey?"

PPIs are published for the output of virtually all U.S. mining and manufacturing industries and are gradually being introduced for the output of industries in other sectors of the economy. For any given industry, producers are selected for the survey via a systematic sampling from a listing of all firms that file with the Unemployment Insurance System. Typically, a firm's probability of selection is based on its employment size. After a firm is selected and agrees to participate in the survey, a probability sampling technique called disaggregation is used to determine which specific products or services will be in the PPI. Disaggregation is a process in which iterative steps are taken to select items based on their proportionate value to the manufacturer's overall revenue. First a reporter breaks down the type of items shipped into categories. Next, these categories are broken down further by price determining characteristics, for example, options, color, size. Further break downs may be necessary to differentiate between types of buyers or discounts. Disaggregation continues until a specific product sold to a specific buyer is selected."

Products which are characterized by different price-determinants should be treated as distinct products. BLS has a large staff of industry experts and statisticians to help them design survey questionnaires to isolate different products and to evaluate the responses.

The OSI Bid-Tabs data does have an advantage with respect to the third requirement above, in that it represents a virtual universe of our items of interest – the components of federally-subsidized highway costs. The wealth of Bid-Tabs data allows us to use alternative methods to arrive at the same objective: a reliable indicator of highway construction costs that can be used for both general cost comparisons and for States to gauge changes in their costs against those in a similar situation. Our approach is to eliminate pay-items that may be defined too broadly to hold price-determining characteristics constant or have statistical properties that imply variable price determining characteristics. In the next section, FHWA uses the three criteria above as the basis of our data analysis and to develop the data cleaning rules.

Edits Used to Improve Data Quality

FHWA's approach to creating the NHCCI attempts to reliably reflect changes in the prices of the underlying goods. To achieve this goal, FHWA procedures first eliminate three categories of data:

  1. Non-standard pay items – These are pay items that have the same pay item number but have different pay item descriptions (or units of measure) from project to project. For example, on one project in a State an item had a description of "Control Survey," and on another project in the same State the item description was "Cleaning Existing Paved Ditch". These types of items, even if they fit all the statistical criteria listed below, cannot be included due to the differing types of work from project to project.
  2. Unit of measure problems – There are some pay items where the unit of measure makes it difficult to track prices changes. Many of these items are Lump Sum items where the quantity of the item is "1." The prices bid on these types of items are generally not related to any specific price trend but are more due to many other factors such as project type, duration, location, size, traffic patterns, etc.
  3. Suspect categories – All the pay items in the historical databases have been categorized into 31 predefined work categories. Some of these categories relate to aspects of a contract such as start-up costs, incentives, etc. Some of these categories generally relate to groups of pay items that are generally not related to any specific price trend but are more due to the project type, location, size, etc. (just like the units problems listed above). These categories are: Mobilization, and Alternates/Bonuses/Time.

Additional Edits to Improve Data Quality

Statistical edits are used to eliminate pay-items that are unlikely to have constant price-determining characteristics with the objective of improving the quality of the price index data. The statistical edits used for the NHCCI data are applied sequentially, and are:

  1. An observation must have a lagged observation in order to mathematically construct the index, so observations that do not have a lagged value will be eliminated from the analysis.
  2. A pay-item must have at least 8 quarters worth of data to be included. This is to reduce the influence of items that have low statistical validity.
  3. Outlier observations, defined as being at least two standard deviations from the mean, are set to the average change in logged price for non-outlier observations for the State in the same period. This particular threshold represents the 95th percentile of pay-items, so that only 5 percent of all pay-items had a value exceeding the cut-off.
  4. Pay-items for which the adjusted R-squared is greater than 0.60 from a regression of the log change in price on the log change in quantity are eliminated. Pay-items meeting this criterion represent a break in the price-quantity relationship required by an index. The BLS considers transaction terms, such as quantity discounts, as a price determining factor. Pay-items for which the price is highly related to quantity are likely to be subject to quantity discounts or volume penalties and are therefore eliminated.
  5. Pay-items for which the maximum observed price is more than 16 times the minimum observed price are eliminated. Casual observation of the highway goods and services bought suggests that prices for a single constant-quality good/service rarely change by the very large amounts observed in the Bid-Tabs data. As an example, we use data from the Energy Information Agency (EIA) on the weekly price for a gallon of regular gasoline for the most recent five years available (1/2004-1/2009). The data shows a maximum price of $4.11 for the week of July 7, 2008 and a minimum price of $1.61 for the week of December 29, 2008 for ratio of the maximum to minimum price of 1.6. Extending the time period back ten years, the minimum price was $0.91, for a ratio of 4.5. Given that the price of gasoline is considered highly variable; our cut-off of 16 should be considered a conservative choice.
  6. Pay-items for which the coefficient of variation of 100 times the log change in price is greater than 42 are eliminated. The coefficient of variation is the standard deviation divided by the absolute value of the mean. We use the log change in price to control for trends that would make the standard deviation ill-defined. We divide by the mean of the log-change in price for the usual reasons that the coefficient of variation is used rather than the standard deviation itself—prices that have a high average change are also likely to have a higher standard deviation (the standard deviation is used directly in cases where the absolute value of the mean is less than one.) The justification for this edit is that pay-items having prices which are extremely variable are unlikely to represent goods/services with constant price-determining factors.

Effect of the Edits on the Data

Most of the change in both value and the number of observations comes from the first edit in which observations with no lagged value are excluded. The effects of these observations are, however, only lost in the case in which they are isolated points. A major advantage of using a chained-index is that pay-items with breaks are still included in the index because the market basket is continuously updated. Also note that the total value and number of observations will be lower than is reported because the first observation in the time-series is always lost.

The data tests show that $38 billion is lost in the first three edits, and another $9 billion from the other edits, leaving about 60% of the value of all good pay-items. Another feature to note is that shares of the various categories in total value do not change much from the initial share for all good pay-items. Only 3 categories change by more than one percent, leaving the final basket of pay-items with a similar make-up to the basket made up of all good pay-items.

The edits eliminate almost $50 billion of the $122 billion that remained after inconsistent pay-items and units were eliminated. Over half of the lost value was due to the elimination of values that were missing a lagged value. These observations may represent pay-items for which there was no expenditure and therefore may not truly be lost. Another quarter of the lost value is due to the elimination of pay-items that had less than eight quarters worth of data and were therefore infrequent components of expenditure. Outlier adjustments actually contributed to reducing lost value as extremely low prices were on average adjusted upward.

The number of observations falls by almost 400 thousand in the first three edits, and then another 30 thousand in the other edits. The relative contribution of the remaining statistical edits consists of about $6 billion due to adjusted R-squared, about another $5 billion for the ratio of the maximum price to the minimum price, and a bit more than $1 billion for the coefficient of variation for the log change in price. The last two are likely to be highly related as pay-items with large differences in the maximum and minimum price are likely to also have large standard deviations. Note that the number of observations declines by a much larger percentage than value, which indicates the pay-items lost through the edits tend to be the low-priced items.

Overall, the effect of the edits on the data, although large, still leaves a useful dataset for constructing the NHCCI. The remaining value is representative of all Bid-Tabs expenditures, with almost $70 billion of the initial $170 billion still available. Further, the distribution over the expenditure categories, which will determine the weights in the NHCCI, is largely unaffected by the edits. This statistical analysis highlights another advantage the NHCCI has over the discontinued BPI, such statistical analysis could not be performed the data provided by the earlier method.

Future Work

While the NHCCI offers many improvements in FHWA's indexing efforts, further improvements are desirable. Additional research should be conducted and should focus on the following areas:

  1. Monitoring how the NHCCI performs with updated data, particularly related to input substitution over time.
  2. Further develop computer programs to provide indexes by individual States and analyze on how the individual State series perform.
  3. Develop methodology and software application to derive sub-indexes based on category groupings such as excavation, reinforcing steel, structural steel, structural concrete, structures, asphalt, etc.
  4. Develop computer programs to provide indexes by project types such as capital improvements, maintenance, etc, if possible to identity in the Oman dataset, and analyze on how different project types perform.
  5. Investigate whether pre-engineering costs should be included in the NHCCI. Normally, these pre-engineering costs are covered in a separate contract, and are not reflected in the Oman dataset.
  6. Further investigate the randomness of the excluded items from the Oman dataset.
  7. Investigate combining the Oman and the Recovery Act datasets to identify and analyze selected characteristics.
  8. Ask States to review and possibly better define some items within the Oman dataset to reduce data edits.

1 http://www.fhwa.dot.gov/programadmin/pricetrends.htm

2 http://www.gao.gov/new.items/d04113r.pdf

3 http://www.bts.gov/publications/guide_to_good_statistical_practice_in_the_transportation_field/