The purpose of the Import and Export Price survey and sample design is to provide a representative and unbiased measure of price change in each published index. For the majority of product categories, the Bureau of Labor Statistics (BLS) conducts the U.S. Import and Export Price survey to collect prices of individual items that represent detailed product categories. For a limited number of product categories, price data come from alternative data sources. (See data sources). Merchandise goods product categories and services product categories are sampled separately.
The universe of all merchandise goods trade is available from transaction data collected by the U.S. government from importers and exporters. Importers are required to file an entry summary for shipments with the U.S. Customs and Border Protection (CBP) using the Automated Commercial Environment, which is an electronic data collection system. Exporters report merchandise that is exported to countries other than Canada to the Census Bureau using the Automated Export System platform. The Canadian Border Services Agency collects data for export shipments to Canada and provides them to the U.S. government. The trade records provide information on the company, the dollar value and volume of transactions, and the classification of the goods being traded based on the harmonized system (HS). The Bureau of Labor Statistics (BLS) receives the import and export records from CBP and Census Bureau for sampling purposes.
The sample frame is divided into two non-overlapping product categories, called panels A and B, for imports and exports separately. The panels remain relatively consistent over time and are roughly equal in dollar value. To ensure the most up-to-date sets of sampled items, BLS draws an import sample and an export sample annually from alternating panels. The balance of the dollar value for panels is kept roughly equal by revising on a periodic basis when major revisions to classification systems or trade value occur. (See exhibit 2.)
|Import panel A||Export panel A|
Food and beverages
|Food and beverages|
|Minerals and chemicals|
Import panel B
|Export panel B|
Minerals and chemicals
Source: U.S. Bureau of Labor Statistics, International Price Program.
Each panel comprises all records for import or export trade data for a 12-month period. The panels are prepared on a 6-month rolling basis, alternating between imports and exports to ensure both imports and exports are fully resampled every 2 years. At any given time, two to three panels are contributing to the market basket. When items from previous panels are phased out, the process is spread over 12 months to avoid undue loss of coverage to any published indexes. At the same time, the initial collection of data for a new panel is completed. As a result, the new panel reaches peak initiation at about the same time the earlier panel begins to be eliminated.
For all imports and for exports to countries other than Canada, an establishment is defined by the establishment identifier number (EIN), an 11-digit number assigned by the Internal Revenue Service and recorded on the frame transactions for those shipments. In some instances, multiple establishments may be located at the same address. For Canada, export establishments are defined by the combination of the name and zip code of the exporter because data received from the Canadian Border Services Agency do not include U.S. EINs. A company may comprise one or many establishments.
A sampling stratum is typically the four-digit level of product detail of the HS and this is the level of publication of price indexes for the HS product classification system. Stratifying by these commodity codes allows the U.S. Import and Export Price Indexes survey to more accurately represent important subsections of trade than if a sample were selected from a non-stratified frame. The U.S. Import and Export Price Indexes survey samples independently within strata, which tends to reduce the sampling error of the estimates derived from the sample.
Determining if a four-digit HS area is a publishable stratum is based on whether the dollar value of trade in that area exceeds a predetermined threshold. The threshold is calculated based on the total number of budgeted field initiation visits, the average number of quotes per establishment that is sent to the field, the total dollar value of the panel being sampled, and the minimum number of quotes to be priced per stratum in order to maintain publishability. The total number of budgeted field initiation visits divided by the average number of quotes per establishment determines the expected number of quotes to be fielded. The total dollar value of the panel is then divided by the expected number of quotes to be fielded and multiplied by the minimum number of quotes to be priced per stratum to maintain publishability.
Product categories are created from the 10-digit HS detail to provide consistency across time and to establish concordances with other product and industry classifications. These product categories are called classification groups. In most cases, classification groups are the 10-digit HS level of aggregation. There are two exceptions. The first is for areas where a series of 10-digit HS level groupings are merged together, generally for cases with small and relatively homogeneous groupings. The second is where a 10-digit harmonized level is split, generally for cases with large groupings that carry a diverse set of items.
The sample selection process is carried out in three stages to efficiently identify large traders and ensure that as many indexes can be published at as great a detail as possible by identifying specific import and export items to price over time. The following subsections discuss the three stages in detail so the reader will better understand what takes place at each stage.
The first stage of sampling begins with setting the sample frame. The sample frame comprises 12 months of import or export transactions for all sampled strata in each panel using their respective frame sources. The first stage of the sampling process relies on three constraints:
an upper bound of item burden for an establishment within a sampling stratum,
an upper bound of establishments to be fielded for the panel being sampled based on the total number of budgeted field initiation visits, and
the expected number of items per establishment to be collected.
Using the aforementioned constraints, BLS begins allocating the sample. First, BLS establishes some upper and lower boundaries. BLS also establishes an upper bound of the total number of items to be collected for the sample. This total can be found by multiplying the expected number of items per establishment by the total number of establishments in the sampling frame. Next, BLS allocates the total number of items across all sampled strata on the basis of each stratum’s dollar value of trade. The lower boundaries ensure that there is sufficient item coverage at the classification group-level for the three classification systems.
BLS distributes the number of establishments by strata based on the distribution of allocated items and subject to the upper bound of item burden for an establishment. Readjustment of item and establishment allocation occurs throughout the process to meet the minimum number of items and establishments required to satisfy publication criteria.
After the allocation process is completed, establishments are selected independently within sampling strata, which represent product areas at the most detailed level of publishable harmonized indexes. Because the size of trade for establishments varies greatly, selection is based on probability proportionate to size (PPS), where size is defined as dollar value traded. This approach reduces sampling error compared with equal probabilities of selection. Some establishments are selected with 100 percent probability; these are the large, frequent traders in a stratum. The rest of the establishments have probabilities of selection that are proportional to the value of their trade in the stratum. These comparatively smaller traders are designated as uncertainties. An establishment can be selected in more than one stratum and with different probabilities of selection; that is, each stratum is independently sampled.
Some sampling strata have low dollar values of trade, and the corresponding price index may not be publishable because the stratum does not meet the dollar threshold for publication. Even so, sampled establishments in these strata are weighted accordingly and aggregated to a higher level to ensure representativeness of overall trade for all-goods import and export price indexes.
Upon selection of the first-stage sample, the sampled establishments are refined by survey economists. The economists use historical survey data and other reference materials to validate the establishment’s name and address. In addition, economists determine if an establishment should be included in the survey based on one of two criteria: the company historically provides price data or is new to the survey. During the first-stage sampling process, multiple establishments with the same company name and address are combined into one entity representing the unique name and address for data collection, known as the collection unit. Also, establishments that are part of the same company but located at different addresses are combined into a collection unit if survey history or reference-material research suggest data collection should be at one centralized location for the entire company.
The second stage of sampling is carried out in three steps: 1) capping burdens for collection units when necessary; 2) allocating items to selected establishments; and 3) selecting classification groups per establishment. The sampling unit is the classification group within the establishment. The first stage of sampling is determining which establishments BLS will attempt to request data from. The second stage is to determine what groupings of items by establishment BLS will ask about.
A cap is put on burden because BLS does not wish to overwhelm data providers. This is primarily done to control an establishment’s item burden for those situations in which it was sampled in multiple strata, in addition to those cases in which multiple establishments were combined into one collection unit during refinement. If either or both of these instances occur and result in the collection unit burden exceeding the upper limit, the item burden within each stratum of each establishment constituting the collection group is proportionately reduced. This prevents the overall burden for the collection unit from exceeding the upper limit.
After capping collection-unit burdens, second-stage sample allocation initially distributes the item burden assigned to a sampled establishment within a stratum to the corresponding classification groups by proportional dollar value. The classification group item allocations then undergo a series of readjustments across and within all sampled establishments that trade in these classification groups through an iterative process known as raking. The raking process involves redistributing items across classification groups in order to meet minimum item requirements. This process also ensures each classification group is allocated enough items across establishments to support publishability across the three primary classification-system strata (Bureau of Economic Analysis end use, harmonized, and the North American Industry Classification System) to which they contribute. At the same time, this maintains the item burden that was assigned to each of the sampled establishments in the stratum.
In the second stage of sampling, classification groups are selected with replacement within the sampled establishments. The process uses PPS methodology in which the measure of size is the expected number of items to be selected within each classification group for the establishment. The PPS method may result in a given establishment’s classification group being assigned a number of items, as items are selected multiple times based upon the relative proportion of the classification group’s value within the establishment.
Both raking and selection with replacement are performed in order to control respondent burden while maximizing the quality of price indexes by ensuring sufficient item coverage for the publication of indexes in the three classification systems.
Further sample refinement occurs when survey economists conduct a final review of the sampled classification groups and counts of items per classification group within the collection units, making any adjustments necessary. The collection units are then assigned to the regions where the data collection occurs and the field economist contacts the company to carry out the initial interview and conduct the price survey.
The third stage of sample selection occurs at the first interview with the company respondent. During that interview, the respondent provides unique items within each sampled classification group. The respondent indicates a dollar share, weight, or importance to establish how representative an item is of a company’s trade. The respondent-identified weights are then used to draw a weighted random selection of items. In cases where the company is unable to identify importance, items are selected randomly.
Once the items are established, the same unique items are subsequently priced over time by the respondent. Sampled items are priced for approximately 5 years until the items are replaced by a fresh sample. Generally, each index spans two to three panels. See the calculation section for specifics on how indexes are calculated.
For more information on the methods used to collect prices for sampled items, see data sources.
U.S. Import and Export Price Indexes are calculated for air passenger travel and air freight transportation.
For the air passenger fares indexes, there is no need to sample. The full universe of data is drawn from administrative data from the Airlines Reporting Company. The air passenger fares universe consists of fares to and from the United States and U.S. territories, and to and from roughly 230 foreign countries serviced by more than 200 U.S. and foreign airlines. In addition, the Export Air Passenger Fares Index includes foreign-to-foreign fares when foreign residents are flying on U.S. airline carriers.
The universe for the air freight price indexes is aggregated transaction data published by the Bureau of Transportation Statistics, U.S. Department of Transportation (DOT). Freight is tendered by the company that needs items shipped to an airline for transportation, excluding mail and personal baggage; the service measured includes shipment from airport to airport only, omitting any ground transport or port service costs, which are classified as different types of services. Data cover both U.S. and foreign air carriers, with at least one point of service in the United States or one of its territories. The records are reported in the DOT T-100 International Market file and include route-specific information: origin and destination airports, air carrier names and nationalities, and the amount of cargo transported.
Air freight services are fully resampled approximately every 5 years. For the air freight price indexes sample, BLS uses probability sampling methods to select a sample of company-routes from the T-100 file that is representative of international air freight transportation. A company route is composed of an air carrier transporting freight internationally between a specific origin and destination.