General Description of the ACES Survey
The Annual Capital Expenditures Survey (ACES) is designed to provide detailed and timely information on capital investment in structures and equipment by nonfarm businesses. The data are used to improve the quality of current economic indicators of business investments, as well as the quarterly estimates of gross domestic product by industry. The data also reveal trends in capital expenditures useful for product development, business planning, and identifying business opportunities. The Capital Spending Report presents select information from recent consecutive survey cycles to illustrate changes over time.
Background of the ACES Survey
The scope of the survey includes all private, nonfarm, domestic companies. Major exclusions are government-owned operations, including the U.S. Postal Service; foreign-owned operations of domestic companies; establishments located in the U.S. territories; establishments engaged in agricultural production (agricultural services are not excluded); and private households. The survey collects information from the company as a whole, and asks companies to report how they distribute their expenditures across different economic activity codes.
Congress first provided funding for ACES to the U.S. Census Bureau in fiscal year 1991. The first cycle was a relatively small feasibility study, whose main goal was to test the quality of questions and collection instruments, and to determine the ability of companies to report the requested data. A larger but still relatively small study the next year used the results of the feasibility survey in collecting 1992 data. The goal in that second cycle was to further evaluate what the survey content should be, assess burden to respondents, refine the survey forms and instructions, and test how large a sample might be required to obtain reasonably precise estimates.
To limit burden, beginning in 1993, the number of questions in most survey years was reduced, with more detailed questions asked every five years. That more detailed information is not part of this longitudinal report. This report focuses on total capital expenditures for new and used structures and equipment for all U.S. nonfarm companies with and without employees, and expenditures by economic sector for companies with employees.
Initially, the Standard Industrial Classification (SIC) system was the basis for the industry categories used in ACES. The North American Industry Classification System (NAICS) became the new basis beginning in 1999. Comparisons across classification systems are problematic. In addition, the sample methodology has changed over time, including expanding the scope to include companies without employees, although data collection from these firms is not as detailed as data collection for firms with employees.
ACES uses various industry combinations developed through consultation with data users. NAICS has gone through several revisions since its inception, and has resulted in changes in the ACES industry categories. These industry categories are used to publish capital expenditures data at various industry or sector levels for companies with employees.
Investment estimates from ACES may not be directly comparable to investment data from other sources due to differences in the scope of the survey, definitions of concepts, the company based collection, ACES specific industry codes, and other sources.
The Current ACES Survey
ACES publications of annual data come out about 13 months after the close of each reference year. The 2010 ACES report was published in February 2012. The 2012 Capital Spending Report continues the Census Bureau's effort to present annual ACES data longitudinally under NAICS, and covers the 2001 through the 2010 collection cycles. The series first report was published in 2007, presenting information from the 1999 to 2005 collection cycles. The series presented may not always be internally consistent due to changes in sample methodology, in NAICS categories, and in company organization and classification over time. Previous sets of ACES data, either annual reports or prior longitudinal reports, are available at https://www.census.gov/econ/aces/historic_releases.html
As of the 2010 data collection cycle, the survey has a sample size of about 45,000 companies with employees from a frame of about 5.7 million. These companies receive a more detailed ACE-1 form, of which there are three variations. The survey form asks for capital expenditures data for each industry in which respondents had activity, and asks them to classify these expenditures as new and used structures and equipment based on the definitions provided.
The sample also includes about 30,000 businesses without employees from a frame of about 28 million. These nonemployer businesses receive an ACE-2 form. The ACE-2 form asks for capital expenditures data separately for new and used structures and equipment, but not for any industry level information.
ABBREVIATIONS AND SYMBOLS
The publication uses the following abbreviations and symbols:
- Represents zero in the table.
(Z) The value is greater than zero, but would round to zero in the table.
ACES estimates are based on annual probability samples. Samples are drawn from a database containing records for each physical business establishment, which are consolidated prior to sampling to create company level records. The sampling begins with automatically taking the largest companies in the frame. It continues with random sampling of the rest of the population, after stratifying by factors that include size, economic activity, and whether or not the company has employees. In that random sampling stage, larger companies are still more likely to be included in the sample. Each selected company has a sampling weight, reflecting both their own and other similar but unselected companies' investments. Larger companies will have weights near one, while smaller companies could have weights above one hundred. A company's impact on the estimates will vary with their sampling weight and their reported data. Sampled companies in the same substratum have identical weights. An adjustment to account for sampled companies that did not respond may increase sampling weights further.
Like all probability samples, the results are subject to both sampling and nonsampling errors. Sampling error is the uncertainty due to seeking to measure only a subset of the target population, while nonsampling error is a general term for all other sources of error. Types of nonsampling error include the inability to provide the requested information, difficulties in interpreting the question, mistakes in retrieving, entering, or recording the information, and issues concerning the sampling frame or how to process inconsistent or missing information. While a census seeks to obtain measurements from an entire population, a probability sample reduces the resources required by reducing the number of measurements required.
Sampling error may be expressed in different ways. The standard error (SE) is expressed in the same units as the estimate itself; a relative standard error (RSE) is often a percentage; while a margin of error (MOE) is useful for creating confidence intervals. The SE is the dispersion of the distribution created from the set of all possible estimates corresponding to all the possible samples. The RSE is the ratio of the SE to the estimate itself. Tables usually denote RSEs, but the SE for estimates that are themselves percentages instead of absolute values is registered as an SE. This is to avoid misinterpretations of a ratio of two percentages. In either case, the RSE and the SE are simply estimates derived from the sample data, just as the estimates themselves are. Therefore, these measures of sampling error are also subject to sampling and nonsampling errors.
An MOE is the product of the SE and a multiplier and is used to create confidence intervals (CI). A 90-percent CI uses an MOE that is approximately 1.6 times the SE. The upper bound is the addition of the MOE to the estimate, while the lower bound is the subtraction of the MOE from the estimate. A 90-percent CI contains the true value 90-percent of the time over repeated applications of the sampling methodology. Although in reality only a single sample is drawn, CIs are still useful in assessing relative quality of the estimate, or for making comparisons between estimates. CIs only interpret sampling error. Any bias from nonsampling error that impacts the CI is not considered.
To create a CI using an RSE, first divide the RSE by 100 to make it a proportion instead of a percentage. Then multiply this by the estimate to obtain the SE. As an example, here is a 90-percent CI for the total capital expenditures for companies with employees invested in the mining sector in 2001: from table 2a, the estimate is $51,278 million. From table 2b, the RSE is 3.6. Converting the RSE into a SE means converting to a proportion, 0.036, and multiplying by the estimate to get $1,846.0 million. To get the MOE, multiply this by 1.6 to get $2,953.6 million. Adding and subtracting this to the estimate gives us a CI of $48,324.4 million to $54,231.6 million. Intervals created in the same way over repeated samples would contain the true value of total capital expenditures invested in the mining sector in 2001 90 percent of the time.
A complete description of the ACES survey design is at