The Census Bureau provided customized mailing and reporting arrangements for owners of large fleets of trucks selected in the sample.3 Based on the owners' preference, questionnaires were boxed by "panel"-each panel composed of truck registrations in selected states-and mailed to the corporate headquarters or other designated location for completion.

Each mailout package consisted of the appropriate questionnaire, cover letter, and a return envelope. The Census Bureau sent questionnaire TC-9501 to owners of trucks in the pickup and van strata, and questionnaire TC-9502 to owners of all other truck types (see Table 10-1). Census Bureau staff identified vehicles on the questionnaire prior to mailing: they included-in the registration information section-the vehicle make, model year, license number, and vehicle identification number. Owners were asked to reply only for the identified truck, regardless of other vehicles they might own or operate. They were to return their questionnaires to the NPC within 30 days of receipt.

Four days prior to each panel's mail follow-up, the file of nonrespondents was sent to the NPC. Then, a second form TC-9501 or TC-9502 questionnaire was printed, with the registration information, and mailed with the proper follow-up letter. Recipients of questionnaires TC-9501 or TC-9502 were mailed letter TC-9500-F during the first follow-up. During the second follow-up, letter TC-9500-CF was mailed with the questionnaire.

The Census Bureau used certified mail for the second follow-up mailing of the questionnaires. Then, staff conducted a telephone follow-up, by strata, within states whose response rate was less than 75 percent. In some cases, the phone call served as a reminder, and resulted in a mail response. If questionnaires were not returned to the NPC, analysts phoned the company's contact person to request a response. This continued until the strata were closed out for each panel. The data collection process was completed in September 1998.

The response rate for certified mail was 38.6 percent. The Census Bureau received responses from 105,054 (or about 80.4 percent) of companies in the sample. The final response rate for customized reports was 88.7 percent.4

3While the Census Bureau forms typically were mailed to the registered owner of the vehicle, the Census Bureau contacted owners of large fleets and asked if they would prefer another arrangement. Some owners wanted all forms mailed to their corporate headquarters to a particular individual who became the Census Bureau contact for follow-ups. Other owners preferred that the Census Bureau sort the forms by a given characteristic (such as state of registration or subsidiary) and mail each to a particular address or individual.

"The response rate for customized reports was higher because the Census Bureau was dealing at the company level for the completion of many forms-in some cases, hundreds. The Census Bureau offered companies options on completing one form to represent others with "like" characteristics.

100 Chapter 10

Data Processing

Receipt and check-in. Clerks at the NPC checked responses in as the questionnaires arrived. The clerks used laser scanners to read the barcoded Census File Numbers (CFNs-Chapters 7, "Preparations for Census Mailout," and 8, "Data Collection and Processing," provide more detail on CFNs) on each mail address label. This procedure checked each report form into the Transportation Automated Control System (TRACS), a system that tracked questionnaires through all stages of data processing. The NPC received two types of survey-related mail-questionnaires and correspondence. Clerks assigned a check-in (source) code for the type of receipt: mail, telephone, fax, undeliverable as addressed, received in Washington DC, delinquent, telephone follow-up, congressional correspondence, and other correspondence. After checkin, clerks entered the appropriate status code for each questionnaire in the TRACS, and referred the questionnaires to the appropriate processing unit. When more than one questionnaire was received in one package, the clerks kept the questionnaires together through all stages of processing so that one contact could be made, if required, for all questions.

Data screening and entry. Clerks screened all questionnaires for completeness and to ensure that the information provided by respondents could be keyed. They also (1) looked for "must items" that needed to be completed, (2) converted responses (e.g. pounds to tons), (3) transcribed respondent remarks, (4) clarified illegible responses, and (5) rounded figures. If information for any of the "must items" was missing, the questionnaire was routed to the telephone unit, where Census Bureau staff called the truck owner to obtain complete information. All questionnaire items that could not be completed satisfactorily through screening or by phone were forwarded to a statistical assistant at the NPC for resolution.

Once the questionnaires were screened, and all problems had been resolved, they were forwarded to the microfilming unit. Microfilming equipment scanned the bar-code on each address label as each questionnaire was filmed, and included it in a file sent to Census Bureau headquarters with the accompanying reel and frame number. The original microfilm was stored at the NPC and a copy of each reel was sent to headquarters to be used in data editing and data analysis. After microfilming, the questionnaires were routed for data keying.

The data entry unit keyed (1) corrections made to the registration information, (2) all data responses provided on the questionnaires, and (3) comments written by respondents. The NPC transmitted the information to the Census Bureau's computer center in Bowie, MD, via dedicated telephone lines. Data tapes were retained for possible retransmission until the Census Bureau's computer center notified the NPC that the data had been received.

History-1997 Economic Census

Data edits. At Census Bureau headquarters, each data record underwent a detailed computer edit for inconsistencies and reasonableness. During this stage, the Census Bureau sometimes inserted data-based on administrative information supplied by R.L. Polk & Co. or from other data-item responses-into the data file to complete blank, or obviously incorrect, responses. The computer flagged items that failed edits, and displayed each failed record for review by an analyst. Once the analyst made corrections, the record was re-edited and displayed again if it still failed the edit. In addition to making corrections, analysts also deleted records-determined by pre-set criteria-to be out of scope of the survey.

Census Bureau staff developed procedures to impute annual miles and lifetime miles. Averages were calculated based on responses to the five characteristics most closely correlated with annual and lifetime miles-model year, vehicle type, acquisition, disposal, and weeks operated. Cubic-inch displacement and number of cylinders were derived from administrative records from each state. Length and average weight (if data were missing) were imputed for each truck. These imputations were based upon responses from records with similar characteristics. Once data records from the 1997 VIUS had been edited and tabulated, the NPC stored the paper questionnaires until January 1, 2000, after which they were destroyed. After data tabulation and publication for the 1997 VIUS, the microfilm of the questionnaires was sent to the Washington Federal Records Center in Suitland, MD, for archiving.

Data Tabulation and Release

For each state, data were tabulated using unweighted and weighted estimates of the number of trucks and their characteristics. Headquarters analysts computed corresponding sampling variability, checked tabulations for consistency and logic, and made comparisons between 1992 and 1997 data. Prior to the preparation of Portable Document Format (pdf) files for release on the Internet, certain data were automatically suppressed. Such data were not published if the estimates did not meet publication standards on either of two bases: the response rate, or the associated or relative standard error.

In addition to the respective printed and Internet releases of the United States Summary—as well as Internet releases of state level data—the 1997 VIUS produced microdata information for each truck in the sample. These records were modified to avoid identification of individual trucks or operating establishments, and subsequently were made available on a CDROM. The CD-ROM included state level and United States' data tables, tabulating software, and documentation.

1997 COMMODITY FLOW SURVEY

Precursors to the 1997 Commodity Flow Survey (CFS)

From 1963 through 1977, the Census Bureau produced measures on the flows of goods and materials according to mode of transportation. These data were gathered in

History-1997 Economic Census

the Commodity Transportation Survey (CTS). An evaluation of the 1977 CTS showed that major problems were caused by the methodology used to collect the shipment samples, as well as the small sample size used. In 1983, the Census Bureau conducted an abbreviated CTS that required summary data from those surveyed. Because an evaluation of the 1983 survey found that the summary data were of poor quality, the Census Bureau did not publish 1983 survey results.

In 1990, the Department of Transportation and the Census Bureau began exploring the possibility of restoring a commodity-based survey. The need for this information was recognized in the "Statement of National Transportation Policy" issued by the Secretary of Transportation in 1990. This was reinforced by the Intermodal Surface Transportation Efficiency Act that Congress enacted in 1991. Thus, the Census Bureau designed the Commodity Flow Survey (CFS) to collect and publish statistics that had previously been produced by the CTS. The new survey was first conducted for 1993; the data were collected by the Census Bureau with additional funding and support from the Bureau of Transportation Statistics.

Scope of the 1997 CFS

The 1997 CFS collected data on commodities shipped, their value, weight, and mode of transportation, as well as the destination of shipments. The 1997 survey involved a sample of approximately 100,000 domestic manufacturing, mining, wholesale, and selected retail establishments. Changes Made From Previous Commodity Flow Surveys

Using the results of the 1993 CFS, the Census Bureau developed a more efficient sample design. Whereas the sample size in 1993 was 200,000 establishments, the 1997 survey sampled only 100,000 establishments. In addition, the sampling period for each questionnaire was reduced from 2 weeks for the 1993 survey to 1 week for 1997.

Other changes for 1997 included

Uses of the Data

Data from the CFS were used by various Federal agencies for policy and program purposes, as well as by private companies for research:

The U.S. Department of Transportation used these data for policy development, assessing infrastructure needs, and in evaluating transportation safety issues.

• State Departments of Transportation used the CFS results in combination with other data sources to evaluate existing uses/needs of their transportation infrastructure.

Sample Design

The 1997 CFS used an approximate 1-in-8 sample drawn from a universe of 800,000 establishments in the mining, manufacturing, wholesale trade, and selected retail trade industries, including auxiliary establishments of multiestablishment companies. The sample was selected using a stratified three-stage design in which the first-stage sampling units were establishments, the second-stage sampling units were groups of four 1-week periods (reporting weeks) within the survey year, and the third-stage sampling units were shipments.

Data Collection

Establishments were mailed four questionnaires—one for each of the reporting weeks. Establishments were to provide data on approximately 25 of their outbound shipments for a 1week period in each of the four calendar quarters of 1997. Establishments with 40 shipments or fewer for the 1-week period were to report all shipments.

Standard Classification of Transported Goods (SCTG).

For the 1997 CFS, commodities shipped were classified using the SCTG. The SCTG is a four-level hierarchical system, created jointly by agencies of the United States and Canada to produce better measures of transportation activity from the Commodity Flow Survey. The SCTG is based on building blocks provided by the Harmonized Commodity Description and Code System and Canada's Standard Classification of Goods coding system. (The Harmonized Commodity Description and Code System is a six-digit coding system used in the United States to valuate exports, imports, and customs.)

Mileage calculations. The mileage estimate component of the 1997 CFS was produced by the Oak Ridge National Laboratory's (ORNL) Center for Transportation in Oak Ridge, TN. The Center used a series of transportation networks developed and maintained at the ORNL to create estimates of miles traveled by mode(s) of transportation on a shipment by shipment basis.

102 Chapter 10

Mailout. Mailout and data collection were conducted by the NPC and took place on a flow basis during each of the four calendar quarters of 1997. Initial mailout began in December 1996, and follow-up concluded in February 1998. During mailout, establishments were sent a questionnaire, an instruction guide, and a Commodity Coding Manual.

The Census Bureau used two versions of the questionnaire-form CFS-1000 and form CFS-2000. Each establishment received form CFS-1000 in each of its first 3 reporting weeks. For the fourth reporting week of each quarter, a subsample (approximately 25,000) received form CFS-2000, while the remaining establishments received form CFS-1000.

The Census Bureau requested that each establishment report the following information:

• Domestic-destination city, state, and ZIP Code.

• Whether the shipment was an export and, if so, the mode of export and the city and country of destination. In addition to the above items, recipients of the form CFS-2000 were asked to furnish additional data on the availability and use of transportation facilities and equipment as follows:

• Were the following on-site facilities available, and were they used: rail siding, waterway docks, airport/landing strips, or pipeline terminals?

Did you use the following off-site facilities and, if so, what was the distance of the facility from your establishment and mode of transport used to reach it: rail siding, waterway dock, airport/landing strip, and pipeline terminal?

History—1997 Economic Census

■ Did you use rail cars that (1) your company owned or leased, (2) a common carrier owned or leased, or (3) another party owned or leased? Also, what percentage of your total rail shipments were each of these three methods?

■ Did you use trucks (with six or more tires) or truck tractors that your company (1) owned, (2) leased with a driver, or (3) leased without a driver?

Did you use the following equipment that your company owned or leased: truck trailers, aircraft, barges, and other equipment (specify)?

• Who generally decided on the mode of transportation used for your shipments—your company, receiver of the shipment, or other?

Data Processing

Mailout, check-in, screening, and data capture for the 1997 CFS were conducted by the NPC. Additional automated edits were conducted at Census Bureau headquarters. Edit problems were resolved at both the NPC and Census Bureau headquarters. The NPC also was responsible for the follow-up, which consisted of a reminder mailed to nonrespondents. A telephone follow-up was conducted, during which the largest establishments were asked to return their completed questionnaires. Census staff did not attempt to collect data by phone because of the large amount of data requested by the questionnaires. The overall response rate for the 1997 CFS was approximately 75 percent.

Editing. The Census Bureau headquarters staff created a flow-processing edit program to identify inconsistencies, range errors, and critical omissions. The edit program was run on a computer system, following the transmission of the data entry files from the NPC to Census Bureau headquarters. The edit program assigned a series of flags for missing or incorrect data.

Imputation of missing data. Item nonresponse occurred when a data variable (either value or weight) either was not reported or was determined to be invalid by the edit process. Establishment nonresponse, in turn, took place when an establishment failed to report shipment data for at least one quarter. Under certain conditions, item nonresponse was handled by imputation procedures. Establishment nonresponse was handled primarily by an adjustment of statistical weights.

Disclosure review. To make certain that information about individual companies would not be disclosed, the Census Bureau subjected the data to disclosure analysis. In conformance with the Census Bureau's confidentiality rules, a computerized disclosure analysis system identified and withheld from publication-data that might have revealed the identities of, and information about, individual companies. In those instances, the Census Bureau substituted a reference symbol "D" for the data in the published tabulations.

Publication Program

Data from the 1997 CFS were made available in both printed and electronic formats. Tables provided information on commodities shipped, their value, weight, and mode of transportation, as well as the origin and destination of shipments. The survey also provided selected data on hazardous material shipments and exports. Commodity detail for national tables was at the four-digit SCTG level. Data were at the two-, and where possible, three-digit SCTG level for state reports. Data from the 1997 CFS were released for the following geographic levels: United States, census regions and divisions, states and selected metropolitan areas and remainder of states.

History-1997 Economic Census

Chapter 10 103