ECONOMIC STATISTICS: WHY THEY OFTEN LIE Those government-issued numbers on which so many business judgments, plans, and forecasts depend are often seriously defective. Blame bureaucratic inertia as well as budget cuts.
By Irwin Ross RESEARCH ASSOCIATE Lorraine Carson

(FORTUNE Magazine) – TO ANYONE who believes economic statistics, the announcement from the U.S. Bureau of the Census last September was a bit of a shocker: its monthly figures on imports were way off. The snafu developed because the Customs Service, which collects the raw data, could not get all its paper to the Census Bureau on time. It was still operating a cumbersome manual system of data collection. As a result, each month's totals included imports from prior months. Delayed reports for August, say, swelled the import figures for September and October. Over a period of two years, anybody trying to track trade performance on a monthly or even quarterly basis was badly misled. Among the deluded was the Commerce Department's Bureau of Economic Analysis, whose initial reports on GNP growth were far too high for the last quarter of 1984 and far too low for the first quarter of 1985, largely because an enormous volume of December imports was erroneously credited to January. After the bureau made the adjustment, GNP growth in the fourth quarter shrank from 4.3% to 0.6%, while growth in the first quarter of 1985 swelled from 0.3% to 3.7%. People at the Bureau of Economic Analysis must have been relieved to learn that at last the Customs Service is acquiring an automated system to process information on imports. That little horror story is perhaps the most dramatic example of the eroding reliability of U.S. economic statistics. The subject is finally getting some national attention, in part because of the Census Bureau's candor in confessing its goof. In March the Joint Economic Committee of Congress held a widely publicized hearing on the problem. The basic trouble is that the world has changed and statistics have not all kept up. ''Too many statistical series are outmoded, and there are too many data gaps,'' says Courtenay Slater, former chief economist of the Commerce Department, in a report prepared for the Joint Economic Committee. ''Information about new industries and rapidly growing economic sectors is often scanty and sometimes misleading.'' To a considerable extent, the statistical lags result from budgetary constraints. Cutting began in the Carter Administration and went deeper under Reagan. The Gramm-Rudman-Hollings act is forcing an across-the-board 4.3% cut on all agencies. Some deficiencies, however, such as the failure of price indexes to deal adequately with changes in quality, long antedated the financial pinch. In any account of obsolescent statistical systems, that vast compendium known as the Standard Industrial Classification (SIC) must head the list. A compilation of 1,005 numerical industry codes, the widely used SIC was designed to classify business establishments by their economic activity, thereby facilitating the collection, tabulation, and analysis of all sorts of data. Obviously any classification system has to be reasonably current to be of value, but the last major revision of the SIC code was published in 1972. A revision was to have been undertaken in 1982, but the Administration declined to put up the money. The present SIC manual is in many ways only of antiquarian interest. It has a separate code number for ''extraction of pine gum'' but not for manufacture of computers. At long last the Administration has decided to undertake a revision. Proposed changes were published in the Federal Register for comment. A total of 78 industries are to be eliminated as stand-alone categories with their own code numbers, and 79 new industries are to be recognized as distinct for the first time. Out as stand-alones are ''rock salt,'' ''artificial flowers,'' and ''billiard and pool establishments.'' In are ''travel agencies,'' ''videotape rentals,'' and ''animal aquaculture'' (mostly fish farming). ''Management, consulting, and public relations'' will be divided into four separate industries and ''engineering, architectural, and surveying services'' into three. If all goes well, the new manual will be out next year, 15 years after the old one. The consumer price index is also out of date -- a matter of moment not just to economists and statisticians. Some 37 million recipients of Social Security have their payments adjusted annually by the CPI, as do military and federal civil service retirees. Some 20 million food stamp recipients see their allotments shift with the CPI. A rise of a single point in the index means a $4.6-billion increase in the federal deficit, according to the Office of Management and Budget. As newspaper readers are reminded every month when the new CPI number comes out, the index is based on a ''market basket'' of the goods and services purchased by a typical urban dweller -- a basket that was first put together in 1972-73. Here we are 13 years later with the same shopping list and the same weighted averages, even though that typical urban consumer is likely to have changed his spending habits considerably. For one thing, stores now offer a host of new electronic wares to spend money on. New consumer products not in the CPI include videocassette recorders, personal computers, and compact-disk players. Since the index does not reflect the decline in per capita energy use since 1973, it gives too much weight to energy prices. Retreats in prices of gasoline and fuel oil largely accounted for the CPI's 0.4% dip in February, the first downtick since December 1982. The Bureau of Labor Statistics freely admits that the 1972-73 market basket is out of date. Indeed, the bureau has been engaged in a five-year project to design a new basket to reflect consumption patterns in 1982-84. The job required a comprehensive household survey of consumer expenditures to discover what people were spending their money on. The revised CPI will be out in February 1987. The other major drawback in the CPI is that changes in quality largely elude it, a problem as old as the index. If the quality of a product improves as its price rises, the true rise is obviously less than the nominal rise. It is generally agreed that the BLS does an excellent job of adjusting for quality in automobiles through a complex process of pricing components; the auto companies cooperate by furnishing a lot of data. But with other consumer durables, the same painstaking analysis is only infrequently applied -- for reasons of methodological difficulty as well as cost. For services, the BLS makes only scattered efforts to assess quality. How would the BLS go about tracking quality in dry cleaning, for example? Obviously by close scrutiny over a period of time in many places. That would be fearfully expensive. THE PRODUCER PRICE index (PPI), which these days gets a lot of attention as an indicator of ebbing inflation, has a major deficiency of its own, apart from the fact that it also reflects 1972 weights. The index covers only commodities, yet the service sector, including trade, communications, transportation, and finance, accounts for almost half of GNP and more than 70% of business employment. The BLS has long been aware of the slighting of services, and in 1984 Congress appropriated $750,000 as seed money toward adding coverage of services in various statistical series, including the PPI. Since then the BLS has received a trickle of additional funds, totaling $2.3 million. The money is largely for research on how to do it rather than for doing it. Incorporating the service sector into the PPI would probably cost as much as the present index -- $14 million a year, according to Kenneth Dalton, associate commissioner of labor statistics in charge of price indexes. In the present budgetary atmosphere, nobody is proposing that. The employment statistics do not suffer from underfunding, but they are far from problem free. They often give off confusing signals. Over the past four years, and particularly since mid-1984, the two BLS measurements of employment -- the household survey and the business establishment survey -- have been seriously out of sync. The establishment survey estimates the number of payroll jobs through a questionnaire sent to a sample of 260,000 companies. At the same time, interviewers question a sample of 60,000 households to determine how many members are working or have actively sought work in the preceding four weeks. The establishment survey picks up only wage and salary workers and thus is expected to report a lower figure than the household survey, which includes the self-employed, farm workers, and domestics. When these groups are deducted, the household total should logically show about the same number of payroll jobs as the establishment total. Yet it does not. Adjusted household employment is substantially lower, and the gap is widening. In November 1982 households reported 3,142,000 fewer wage and salary jobs than establishments did. By January 1986 the gap had widened to 4,230,000, a 35% increase. BLS experts speculate that a household member who is reported as employed may actually hold two jobs and that the number of moonlighters has been increasing. But there is no question on the interview sheet that would elicit information about dual jobholders. Perhaps a question or two should be added. By all accounts, the index of leading indicators is in especially bad shape. Produced monthly by the Bureau of Economic Analysis, it combines 12 series -- among them new orders, net business formation, and building permits -- that are supposed to herald up or down movements in the economy. The index does so only on occasion and with such variable lead times as to be of little value. The main problem, according to Feliks Tamm, the official who gets out the index, is that it has not been revised since 1975. ''There is hardly one series that we are completely satisfied with,'' he says. Improvements do occur in the statistics the government puts out, but slowly. Late last year the Bureau of Economic Analysis introduced an index for computer prices. In the past it had no method for evaluating quality changes, so it simply held prices constant to compensate for improvements. When the bureau finally developed a sophisticated index, it discovered that with computing power taken into account, computer prices had dropped at a rate of 18% a year between 1972 and 1984. This revelation had a pronounced effect on the widely followed price index for producers' durable goods, reducing its rate of increase over that period from 6.9% to a mere 1.8%. A widely welcomed change was the recent demise of the ''flash'' estimate of the current quarter's GNP. Released in the final month of the quarter, the flash was partly based on projections from the first month's data and partly on the second. It was almost always grievously off the mark, but not predictably in one direction or the other. The Bureau of Economic Analysis began publishing the flash in 1983; before that it had been furnished to other ) government economists and had invariably leaked. Last January the Commerce Department mercifully killed it. But that solution can hardly be invoked for most of our statistical problems.