Cletus C. Coughlin andThomas B. Mandelbaum

Cletus C. Coughlin is a research officer and Thomas B.Mandelbaum is an economist at the Federal Reserve Bank ofSt Louis. Thomas A. Pollmann provided research assistance.

THE RISING LEVEL OF U.S. EXPORTS in re-cent years has caused jobs and incomes inmany states to become more closely tied to ex-ports. To assess the economic effects of stateexports, it is essential to have reliable informa-tion on the level of export activity by firms with-in the individual states. Such information isessential for numerous other purposes as well.For example, policymakers and others interestedin state economic development require exportdata to assess the effectiveness of programs de-signed to stimulate export activity; they also re-quire such data to assess the effects of tradepolicy changes, such as the proposed free trade

agreement with Mexico.’ Unfortunately, no idealmeasure of state export activity currently exists.

This article describes the two available stateexport series and compares their estimates ofmanufactured exports. Such a comparison was

‘As reported in Business America (1991), state governmentsengage in a wide variety of activities to promote exports.These activities include overseas trade missions, technicalassistance (such as seminars on the legal and financialaspects of trade), and the dissemination of trade leads.Seven states have export finance programs: 41 statesmaintain offices in 24 countries to promote trade. Thesepromotional activities raise the issue of whether interna-tional exports by firms within a state generate differenteconomic results than domestic exports (or exports toother states by firms within a state). Empirical evidence to

not possible until recently because the two se-ries were not available for the same year. Ourcomparison for 1987 reveals that the two seriesprovide conflicting information about export ac-tivity in many states.

The most prominent deficiency of both mea-sures is that they are based on the value of ex-port shipments by firms within a state ratherthan on the value of goods produced within astate that ace exported. While this distinctionmay sound arcane, the discussion below indicatesthat it is not. Moreover, income and employ-ment in a state are dependent on the lattermeasure, not on the value of export shipments.To address this deficiency, a third estimate ofstate manufactured exports is developed in thisarticle. Comparisons show the differences be-tween this new measure and one of the existingmeasures of export activity. Such a comparisonfurther illuminates the shortcomings of the two

assess whether such a distinction is meaningful in aneconomic sense is scarce. See Webster et al. (1990) forevidence that the employment effects of international ex-ports exceed those of domestic exports for manyindustries.

I 65

Measuring State Exports: IsThere a Better Way?

IIIIIIIII’IIIIIIIIa

66

available series and the advantages of a serieslike that developed here.

EXISTING MEASURES OF

STATE EXPORTS

Historically, the focus of U.S. trade data hasbeen on country-to-country trade flows (thatis, U.S. exports to and imports from individualcountries). Recently, increasing attention hasbeen focused on trade flows involving individualstates. Exports of Boeing aircraft from Wash-ington and imports of foreign cars by Missouriresidents are just two examples of traded goodsthat have attracted attention to the fact thatstate jobs and incomes are related to the inter-national economy. Our focus is restricted to ex-port activity at the state level. To date, those in-terested in the magnitude of these flows haverelied on two data sources published by the U.S.Census Bureau: Exports From ManufacturingEstablishments (EME) and the Origin of Move-ment of Commodities (OMC).

Exports From ManufacturingEstablishments

Approximately 56,000 of 220,000 manufactur-ing establishments are asked in the Annual Sur-vey of Manufactures to report the total value ofproducts shipped for export.’ Since many estab-lishments do not know the final destination oftheir products, the reported exports understatethe value of all manufacturing export shipments.i’o compensate, the total amounts reported areadjusted to include estimates of exports by otherdistributors, such as wholesalers.

Differences between the directly reportedvalues and the national total derived from Ship-per’s Export Declarations are allocated to states.3

A Shipper’s Export Declaration is a documentthat exporters must file which includes thevalue of each export shipment.’ The allocationprocedure is complicated slightly because the in-dustry classification scheme used in Shipper’sExport Declarations differs from that used in theAnnual Survey of Manufactures. An additionalcomplication is that the value of export ship-ments in Shipper’s Export Declarations includesfreight and wholesale margins. Since the valueof export shipments in the EME is reported asshipments leave the plant, the costs associatedwith transportation and wholesaler activity musthe removed from the values reported in Ship-per’s Export Declarations.

EME was first produced in 1960 as the Originof Exports of Manufactured Products. It was pro-duced at varying intervals until it became anannual report in 1983. This series possessessome significant shortcomings. First, the seriesis restricted to manufactured exports. It pro-vides no information for establishments engagedin exporting services or unprocessed commodi-ties produced by the agricultural, mining, fores-try and fishing sectors.’

Second, this series is available with a two-yearor more delay. For example, data for 1985 and1986 became available in early 1989 and datafor 1987 became available in 1991. Many ana-lysts view these data as having only historicalvalue because information on recent activity isnot available for use in current decisions, suchas those involving targeting export promotionexpenditures.6

2At five-year intervals, a more comprehensive coverage ofmanufacturing establishments occurs with the Census ofManufactures. See appendix A of U.S. Census Bureau(1991) for details on the 1987 Census of Manufactures andthe 1986 Annual Survey of Manufactures.

‘For details, see appendix C of the U.S. Census Bureau(1991).

‘A Shipper’s Export Declaration must be filed for all exportshipments except for those going to Canada. EffectiveNovember 30, 1990, this document was no longer requiredfor Canadian shipments because of a decision tosubstitute Canadian import statistics for U.S. exportstatistics. See Ott (1988) for an explanation why Canadianimport data are considered more accurate than U.S. ex-port data.

‘Processed food, forestry, petroleum and coal products thatoriginate in these primary sectors are included asmanufactured exports.

6Such decisions are also complicated by the fact that, until1987, export data were generally unavailable at the three-

digit Standard Industrial Classification (SIC) level. The SICis the standard by which establishment-based U.S. govern-ment economic statistics are classified by industry. Fordetails, see U.S. Office of Management and Budget(1987). For manufacturing, 20 industries are identified atthe two-digit SIC level. The industry becomes more nar-rowly defined as the number of digits for an SIC level in-creases. Prior to 1987, the export data were presented atthe two-digit SIC level, or only for broad industries. An ex-ample of the disaggregation offered by the use of three-digit SIC codes is chemicals and allied products (SIC 28)which has eight industry groups: industrial inorganicchemicals (SIC 281); plastics materials and synthetics (SIC282); drugs (SIC 283); soaps, cleaners and toilet goods(SIC 284); paints and allied products (SIC 285); industrialorganic chemicals (SIC 286); agricultural chemicals (SIC287) and miscellaneous chemical products (SIC 289).

67

The final and most important shortcoming isthat this series reports the value of shipmentsinstead of what is termed “value added.” Valueadded is the value of a firm’s sales minus thevalue of the goods and services it purchasesfrom other firms to make its products. As theterm implies, value added measures the dollarvalue a firm adds to the value of purchased in-puts in its production process.

One way to calculate the market value offinal goods and services produced during a yearis to sum the value added at each stage of pro-duction by the firms in an economy.7 To illus-trate, assume an automobile producer had totalsales of $18 billion, of which $10 billion reflectthe value of steel, tires, plastics, electricity andother inputs used by the producer to make auto-mobiles. The cost of these intermediate inputs issubtracted from the producer’s revenue to cal-culate value added, so the automobile pro-ducer’s value added was $8 billion. This pro-cedure is repeated for each firm in the econ-omy; the sum of all firms’ value added equalsthe total value of production within aneconomy.

Using the value of export shipments ratherthan the value added related to exports mightbe a misleading indicator of export activity in astate. Some manufactured products are not ex-ported directly, but are combined as inputs withother resources to produce an export. If theseinputs were produced in one state and trans-ported to another for final processing, the valueof export shipments from the latter manufactur-ing establishment in the exporting state wouldoverstate the value added that actually occurredin that state.8 The value of export shipments in-cludes value added in both states.

A state’s value of export shipments will exceedits export value added if its exporting firms relyheavily on inputs produced elsewhere or if its

7The value added approach is one of three standardmethods for calculating the market value of production.The other methods focus on income and expenditures.The income approach sums the incomes derived fromeconomic activity, which are primarily wage, profit and in-terest incomes from employment of labor and capitalresources. The value added in an establishment is the in-come generated by the establishment’s activity. The ex-penditure approach sums four general categories of spen-ding on goods and services: consumption, investment,government and net exports (that is, exports minus im-ports). The income and expenditure approaches are usedmore extensively in the United States than the value add-ed approach. In the European Community, however, thevalue added approach is used extensively in the ad-ministration of taxes.

firms produce relatively few inputs used by ex-porting firms in other states. On the other hand,the value of export shipments from a state willfall short of its export value added if its export-ing firms produce more inputs that are used byexporting firms in other states than its firmspurchase from elsewhere. Export value addedand the value of export shipments will only bethe same if the value of shipments used to pro-duce exports in other states exactly offsets thevalue of inputs from other states that are usedto produce exports. Overall, a state’s value ofexport shipments may overstate, understate orequal the value added that actually occurred inthe state. The empirical importance of this dif-ference is examined below.

Origin of Movement qfCommodities

Prompted by a request from the transporta-tion industry, a second export series, the 0MG,began in 1987. The goal of this series is to iden-tify where merchandise begins its export jour-ney so that it can be tracked to its port. In thecase of a manufactured good, the so-called “pointof origin” does not require that the location ofproduction of all component parts be identified,but rather where a completed manufacturedgood began its export journey. According to theinstructions that accompany the Shipper’s Ex-port Declaration, the point of origin could beany of the following: 1) the state in which themerchandise actually began its journey to theport of export (indicated by the two-digit U.S.Postal Service abbreviation); 2) the state oforigin of the commodity with the greatest shareof value in a bundle of exports; or 3) the stateof consolidation (the state where goods are con-solidated by an intermediary for overseas ship-ment). In practice, the ports from which goodsare shipped overseas are frequently used to iden-tify the point of origin. This discretion in identi-

8Although such intermediate products are identified by thestate of production as “supporting exports” in the EME,the state from which they are ultimately exported is not in-dicated. In addition, adding a state’s supporting exports toits final shipments would result in some “double counting”of exports and overstate the value added associated withmanufactured exports. Note that the national value of ex-port shipments is a theoretically appropriate measure ofvalue added because the sum of export shipments acrossall establishments does measure the market value of thesemanufactured exports. At the national level, there is nodouble counting. The shipments of intermediate inputs usedfor the exports are already included in the value of exportshipments and are not added again in the calculation ofmanufactured exports.

IIIIIIIIIIIIIIIIIa

68

fying the point of origin reflects the fact thatdetermining the location of production is not aprimary objective of this data set-ies.°

Origin of movement totals are determined bysorting Shipper’s Export Declarations by thestate ~vhere a commodity became an export. Aproblem, however, is that Declarations for manyshipments contain no point of origin. For exam-ple, in 1987 about 25 percent of the 9.7 millionDeclarations for shipments contained no statecode. To make the data more useful, the CensusBureau contracted with the Massachusetts In-stitute for Social and Economic Research to de-velop estimates for the origin of shipments lack-ing state codes.’°This expanded series is usedin the following discussion and is referred to asthe 0MG series.

This newer series has some desirable charac-teristics relative to the export data provided di-rectly by manufacturing establishments in theEME, although the older series is generallyviewed as the more reliable of the two series.”One attractive feature of the 0MG is that thedata are available with a lag of months ratherthan years. In addition to manufactured ex-ports, shipments data on nonmanufactured mer-chandise exports are provided. The initial for-eign destination of these goods is provided aswell. Consequently, information about state-to-country export flows is available for the firsttime. Like the EME, however, this series doesnot approximate the extent of value added in astate resulting from manufactured exports.

A COMPARISON OF THETWO SERIES

While there are several reasons why the twoexport series might differ, it is possible that theactual differences are small enough to allow thedata to be used interchangeably. No comparisonof the two series has been possible previouslybecause 1987 was the first year for which the

data in the 0MG were available and the 1987EME was just released in April 1991. First, wecompare each state’s level of exports as in-dicated in the EME and the 0MG. Next, we in-vestigate whether a state’s rank differs betweenthe two measures. To complete the analysis, aparticular facet of the linear relationship be-tween the two export series is examined. Seeappendix A for details on the three methodsused to compare the two 1987 series, as well asthe two 1986 export series discussed later. Ifthe two series are closely related, then the 0MGdata, which are available after a considerablyshorter lag, could be used in place of the EMEdata.

Comparing Levels: 1987 ExportSeries

Table 1 shows the value of 1987 manufac-tured exports according to the two series forthe 51 “states” (50 states and the District of Co-lumbia) and the total of the states. One reasonthe two series differ is that the data in the 0MGinclude transportation costs and wholesale mar-gins, while the data in the EME are the value ofexports at the producing plant. This accountsfor the bulk of the $22.3 billion excess of the0MG state total over the EME state total intable 1.12

If these items were the only source of dif-ference, the export value in the 0MG for eachstate would be higher than the value in theEME. Also, the difference would be greater forthose states farthest from major ports or a for-eign border, reflecting the higher transportationcosts. Table 1 shows that exports according tothe 0MG are higher than the level according tothe EME in just 20 of the 51 states. This is insharp contrast to the expectation that the 0MGmeasure should be higher based on differencesin its coverage and on the difference in the statetotals. This discrepancy occurs primarily becauseof the OMG’s focus on where merchandise beganits export journey. Since this location is often

°Smith(1989, 1990) notes that identifying the productionlocations of exported goods is especially difficult foragricultural and mined commodities. Small shipments ofthese commodities are often combined at storage facilitiesprior to reaching their port of embarkation. Shippers tendto report either the state of consolidation or the port as thestate of origin.

‘°Detailson the methods to generate these estimates canbe found in Lerch 0990).

115ee Farrell and Radspieler (1990) and Little (1990).12Appendix B of the Exports from Manufacturing

Establishments: 1987 shows the difference between thevalue of exports at the port of export and the estimatedplant value to be $28 billion.

69

Table IManufactured Exports by State for 1$

‘lit ‘tip~ä~ DS~cenees RankStat, EM$ OMP Leyels~ e EME GMCAtab tot 135$ $ 8968 18 1 3% 25 26

50 5264 0 88 30na 086 27727 666 9 7 1

Akan 15530 5361 70 530 4

Calfomia 22996 4487 ,4$2,t 324do I818~ 82 4 tOeD it 30 29

Conne toUt 4 ~ 09ft9 I 544.2 34 4 19Ode a 5157 840 3 50 35ThstrotfGolmb 00 8 1 .4 1 3, 50’

F~tonØa 48030 66020 47968 999 13 8

Georot Si 338 taD I 20 aHew ItS 1536 ~t8 4 47 9Idaho 7650 4620 8 ~0 39 4

me 6t7 84t6 6 <5 7 8I 001 4102 498 180 1fo 2 1$ 7983 1 4 28

6582 1448 410 82 <29 31~(eueky 20$4 1939.9 975 3 26to 3408 5,865 2, 72. 2Mane 719 636.1 26 -18 37<Marytsrld 9279 1081 488 —4 8

Masse usetts ft 79 30 9 ii 0 g

Mr 1241 0< 17910 5 4 4Mn spta .1333 850 803 I I 16

rs~ pp~ 172 1 9 -.444 2Mtsor 440 ~0 7 449 go

09 0 3 50 49t4ebraàa 7607 as

73 398 86 1095 48 44Mew mph 1.1604 835,0 4 80 34 39

9025 6346 2305 594 7 10NewMexr 4 2040 149 7 toNot erR tO 0 I 7908 490 4Non a a 56706 ~,808 13<6 1t4ettOko 26 4 2 45

tO’ 13041 099 0.500 31 7Oktahome 3 29 3641 26 3o ~oo 21211 2294 173 8 & 23Pnnsy4vrna 61 8,7346 95 146 8Rhode HId 691 40 8 41 40 43ottrtrfa 254 3<159.6 ,E~ .33 2 4

Sotrth0 3 60 14 79 44Teen a 3662 23093 1 9 3537 as 140$3 6 8516 $3UtIr 589 157 674 3 4 3Vernon 5986 704 157 7 41Yrgrra 6566 708 10942 .289 6W (wtqtn 108417 1 793 9522 88 $WeVr re 1126 9200 06 1 5 34W’ on~te 41085 35004 608.1 - 48 7

Wyornhg 395 2 2 1877 4752 51 46

Total $19 $710 $25863 5 92

urces. EME U S Departm ci o Corome ce Ru ea o the ens 5M $ ore Manufa to 9Es 3bl,s men 1957 GPO 198) aMP- Ma hu a Irtstrtute for ctal nO E oaomResearch tjnrv ty f Massa husett U S xpert by State of Cog it ot Movemendata ape (1990)

Mflhons of do lar20MG We mmu LInE value

((0MG EME)IEME)100.

JULY/AUGUST 1991

70

identified as a port, the 0MG estimates of ex-ports are more concentrated in states that con•tam, or are near, major ports. Also, in somestates where transportation costs might be ex-pected to be relatively high, such as Nebraskaand Kansas, the export value in the 0MG islower than the value in the LIME, again con-trary to the expectations based on transporta-tion costs alone.

The importance of ports in the 0MG data isfurther illustrated in figure 1, which plots thelevel of exports in each series. If each state’s ex-ports were identical in both series, all pointswould fall on the line labeled line of equality.”The points below the line of equality indicatethat states’ exports in the 0MG often are lowerthan reported in the EME. In seven of the stateslabeled in figure 1— California (CA), Florida (FL),Louisiana (LA), Michigan (MI), New Jersey (NJ),New York (NY) and Texas (TX)—the value usingthe 0MC is much higher than the value usingthe LIME. In these states, total exports using theformer measure exceed exports using the lattermeasure by almost $37 billion. This pattern isconsistent with the fact that data in the LIME in-dicate the value of exports shipped from a state’smanufacturers, while the data in the 0MG aremore likely to indicate the value of exports ship-ped from the state of consolidation or the port.Therefore, using the value in the 0MG as ameasure of a state’s export activity can bemisleading.

As table 1 shows, the percentage differencesare also considerable for many states. For exam-ple, the value of exports from Wyoming mea-sured in the 0MG is nearly six times higher thanthat reported in the EME, while the 0MG esti-mate for South Dakota is 79.3 percent lower.On average, the absolute value of the differencefor a state is 44 percent; excluding Wyomingreduces the average difference to 35.3 percent.

The overall correspondence between the dol-lar levels of the two series is highlighted by cal-culating the simple correlation between them.This measure ranges from negative one to posi-tive one and equals one when the two measuresare perfectly correlated. In the present case, thecorrelation of .95 is high. Thus, when a state’s0MG export value is higher than the average0MG export value using all states, the state’sLIME export value also tends to be higher thanthe average LIME export value.

Figure 1 shows this general correspondenceby plotting the states’ 1987 exports as indicated

by the two series. Most observations clusteraround the line of equality. Still, the substantialdifference between the two series for severalstates indicates that the two measures are notidentical.

Comparing Ranks: 1987 ExportSeries

Another useful comparison between the twoseries involves ranking the states to see if stateswith larger (smaller) export values using onemeasure also have larger (smaller) export valuesusing the other measure. Ranks are often usedas a summary measure of a state’s relative ex-port performance. If each state had the sameor, at least, a similar rank among the 51 statesusing either series, then the more current datain the 0MG could be used to rank the states ina more recent year.

In view of the high simple correlation betweenthe two series, it is no surprise that table I indi-cates a general similarity between a state’s ex-port ranks using the two measures. Californiaand Texas, for instance, rank first and secondin both series. This general impression is cor-roborated by the calculation of a Spearman rankcorrelation that allows for pairwise comparisonsof the alternative proxies. This coefficientranges from negative one to positive one; itequals one when measures yield identical rank-ings and minus one when the rankings are iden-tically inversely related. The correlation be-tween the two series’ export ranks is .96, whichis very close to one.

Although this high correlation indicates aclose overall correspondence between the rank-ings according to the two series, policymakersor researchers who rely on the more currentranking available from the 0MG as an indicatorof the relative scope of export activity in aspecific state can easily be misled. The ranking

of each state in the 0MG is not identical to themore reliable ranking in the LIME. Florida andLouisiana, for example, rank considerablyhigher according to the 0MG, due to the majorports in those states from which a large volumeof merchandise is shipped. Missouri, on theother hand, ranks only 20th using the morecurrent 0MG measure, but is 11th according tothe LIME.

I~nC0At P~cPt~S PMJV OF ST fllt]S

1 71

IIIIIIIIIIIIIIIII

Figure 1A Graphical Comparison of TwoState Export Series:1987

C)50‘a4)

4~I

•00EE0C)

0

C4)E0S

0C0)I-

0

40

30

20

10

00

Exports from Manufacturing Establishments (EME)

A Closer Look at the LinearAssociation Between the 1987Export Series

A more rigorous criterion to assess the inter-changeability of the two measures reveals asubstantial difference between the two series.This criterion, termed difference preservation,requires that the two export series differ by nomore than some constant across states. If this

25

criterion is met, one export series could bereliably used as an index for the other.

If the 0MG data preserved the difference inthe EME data, the association between the twoseries could be illustrated by a line indicatingequality of a state’s exports, give or take someconstant. In figure 1, such a line would be par-allel to the line of equality. This is not the case,however. The dashed line, based on the actuallinear association between the two series, isclearly not parallel to the line of equality. Con-

5 10 15 20

JULY/AUGUST 1991

72

sequently, one measure is not interchangeablefor the other. This means that researchers andother users of state export data in statisticalstudies should not use one measure as a proxyfor the other because the results can vary de-pending upon which measure is used. In prac-tice, this finding applies to the use of the moretimely 0MG-based measure as a proxy for theEME-based measure.

A NEW STATE EXPORT MEASURE

BASED ON VALUE ADDED

Existing state export series indicate the valueof export shipments rather than export valueadded. As such, they reflect both the value add-ed in a state’s factories as well as the value add-ed embodied in intermediate goods which mayhave been produced in other states. For exam-ple, an airplane assembled and exported fromthe state of Washington may have componentsmanufactured in California and Texas. Conse-quently, these series fail to identify the trueamount of state economic activity used to pro-duce manufactured exports.

To address this problem, we estimate a mea-sure of each state’s value added associated withmanufactured exports. In conjunction with theEME, the Census Bureau provides data for eachstate regarding the number of manufacturingworkers producing manufactured exports ineach industry as well as the number of non-manufacturing workers in jobs related to theproduction of manufactured exports. In fact, ap-proximately the same number of nonmanufac-turing jobs as manufacturing jobs are related tomanufactured exports. This reflects the factthat manufacturing requires the productive ef-forts of workers (such as lawyers, accountantsand transportation and communication workers)from various nonmanufacturing industries.

Unlike the value of export shipments, thelevel of export-related employment is directlyrelated to the value added of exports in a state;such employees directly generate the value add-ed. This employment information is used to esti-

‘~Thisis consistent with the Washington State input-outputmodel for 1982 (Bourque, 1987). For example, inWashington’s largest export sector, aerospace, inputs fromother states equal 56,2 percent of the sector’s shipments.

mate state export value added. The estimatesare based on the assumption that the productiv-ity (output) of each export-related employee is nodifferent than the average worker’s productivityin that industry and state. Consequently, exportvalue added in a state is equal to the sum overall industries of the number of export-relatedemployees in a state multiplied by theirproductivity.

One data series necessaxy for such estimates,gross state product—the market value of thegoods and services produced within a state dur-ing a year—is not currently available for 1987.Since this precludes calculating export valueadded for 1987, our measure of exports isestimated for 1986. Appendix B provides adetailed discussion of the methodology used inestimating state export value added.

Comparing Export Value Added Vs.

EME

Figure 2 and table 2 compare the value of thenewly constructed series of manufactured ex-ports, export value added, with the value ofstate export shipments from the LIME. Sum-ming over all states, the export value added total($196,656.2 million) exceeds the total in the LIME($159,374.5 million) by $37,281.7 million. Three-fourths of this difference is due to transporta-tion costs and trade margins that are includedin our calculations, hut are not in the EMLItotal.

The differences between the two measures atthe level of individual states, however, reflectmuch more than transportation costs and trademargins. Rather, they reflect the fundamentaldistinction between value added and value ofshipments accounts. The state of Washington isespecially noteworthy as the level of exportvalue added is approximately one-half the levelof exports in the EMLI. This suggests that manu-facturing export shipments from Washingtoncontain a large percentage of intermediate in-puts produced elsewhere.” Using the exportshipments value as a measure of this state’s ex-port activity is clearly misleading.

rr,n,rnA,fl~OCt,%,CoA,,,V fir itt fl~ to

I 73

Figure 2A Graphical Comparison of TwoState Export Series:1986

a)-U‘5

a)c0>t00.xUi

30

20

10

00

Exports from Manufacturing Establishments (EME)

Using export shipments values also can causeinaccurate inferences in terms of understating astate’s export activity. For example, the exportvalue added in 12 states exceeds the values inthe EME by more than 50 percent. Wyoming,with an export value added that is more thannine times its EME-based export value, is by farthe most extreme example. The primary reasonis that firms in Wyoming process large quanti-ties of oil and coal that are shipped to otherstates for use in manufactured exports.

While Wyoming is a small exporter regardlessof the measure used, the large percentage dif-ferences are not restricted to relatively small cx-

20

porters. California, the nation’s leading exporterby both measures, is estimated to export 53.5percent more on the basis of value added thanit does on the basis of shipment data. Thus,California firms are supplying large amounts ofgoods and services ultimately exported in theform of manufactured exports from other states.

These differences for many states betweenexport value added and the LIME-based measureof state exports raise the issue of the generalassociation across all states between the mea-sures. As was done above, the ranking of states’export value added was compared with theranking of exports reported in the LIME to

IIIIIIIIIIIIIIIII

5 10 15

S JULY/AUGUST 1991

Table 2Manufactured Exports by State for 1986

1986 Exports1 Ditterences Rank

Value ValueState EME Added - Levels2 Percentage3 EME Added

Alabra 51 5829 92427’ S742 2 44’¾ 29 22Aiasca 7129 5459 167 1 --234 38 43Anzcna ‘ 755 8 2 283 7 527 9 30 1 27 24A”.ansss 1.0654 I 2992 2338 21 9 33 33ca:.forrna -7216.: 264214 92050 535 1GoIo-adc 1 4777 20870 6093 4’ 2 30 28Gonnect-cut 3.9964 4.9680 971 6 243 13 13Delaware 429 5 794 5 365 0 85 -3 42 39D,slr-ctoI GnLmbia 91 0 1943 ‘033 1135 50 49Floriaa 33726 49662 15940 473 16 i4Georgia 28267 36850 8583 304 20 19

-lawau 2143 ‘608 635 —250 4D 51Idalo 5026 5481 454 90 10 42Illinois 7.2092 10 ‘073 2898 c 402 7 6Indiana 4.787 t. 53528 6654 1’ 5 11 11lcaa 19324 18930 —394 -20 24 29Karsas 18350 6903 .1447 —79 26 31

Kentucky i~398 21912 2514 30 23 26Louisia’a 30203 2.3749 6454 —21 4 8 23Vamp 8006 ‘996 - -0 36 38Vary:ara 1 7405 2 :627 4222 243 28 27Massa&r.jsptts 55138 7’398 16260 295 9 9Michgan 10.8780 102739 6041 —56 3 5Minnesota 3.6q1 9 43277 6358 172 14 ‘6

Miss;ssipp 1 337 i ‘ 401 7 645 48 31 32M ssoun 42679 38963 - 371 6 87 2 17

Mcntana 101 2 2328 ‘3’ 6 130 I 49 49Neo’asica 753 3 808 0 54 7 7 3 37 37Nevada 167 1 2435 764 457 ‘8 471~iewHampsnw 8926 1 ‘933 3004 J37 35 34Newiesey 35481 72485 37000 1043 15 6New -Vexco ‘777 3354 1577 88? 47 44NowYcrk 94124 156605 62450 664 6 2NorInGaro-na 53506 59’67 6559 125 0 10No0n Dakota 2147 289 c 7-13 346 44 45

10.6530 115F17 9087 85 4 4Oic.ahorna 1 0646 1 8196 7350 678 32 3COre-go, 6627 22645 401 8 21 6 25 25Fer.rsylvania 60266 93732 33470 555 8 7P.r-ode Isand 281 9 748 266 8 55 4 41 40South Cac,li.na 23950 2451 8 538 72 22 21South 0a~o1a 2127 2508 351 79 46 46Tenr’essee 29104 3212’ 3017 134 9 20texas ‘09815 131951 22140 202 2 3Utah 6685 9440 2755 412 39 36Ve’monl 3840 651 3 2673 69 C 43 41Virgin,a 2.7040 3701 8 9978 369 21 ‘6Wasiingtoi 98628 51761 —‘6870 475 5 12V/psI Virgina 9832 10491 659 67 .i4 35Wisconsin 33’BS 41636 8501 257 17 16Wyoming 19: - 732 1541 8069 5 50

loin. $1593745 $1966662 5372817 234 -- -

Sources EME U S Depar’ment ct C~rr-er~eB. esu o’ the cens_s Exports ‘torn Manufac;unng

Esfab4shmoots 1986 (GPO 1999) Exam Vu-uc Adood A.2’-ars cal-zulu: cns-M,llions at mioliars

‘Expo’t Vaije Aoded min~.sEFv1E vaiue

~ifExpot VaLe Adoeo- FMSE-VEI11)O

1 75

determine whether states with larger (smaller)export values using one measure in 1986 alsohad larger (smaller) export values using theother measure in the same year. The twomeasures yield a general similarity between astate’s export ranks. The Spearman rank cor-relation is .98, which is virtually one.

The ranks of a number of states, however,differ substantially across the two measures.Nine states have ranks that differ by five placesor more. The largest changes involve Washing-ton, which drops from fifth place using exportshipments to twelfth place using export valueadded, and New Jersey and Alabama, both ofwhich moved up seven places (New Jersey from15 to 8 and Alabama from 29 to 22) when usingvalue added.

The overall correspondence between these twomeasures is also indicated by the simple correla-tion between the two measures. The simple cor-relation is .95. This close correspondence is evi-dent when the two series for each state areplotted as in figure 2. Many observations clusteraround the line of equality reflecting the linearassociation between the two measures.

This strong association, however, does notmean that the two measures are interchange-able. In terms of figure 2, the actual linear asso-ciation (identified by the dashed line) varies sig-nificantly in a statistical sense from the line ofequality.” Consequently, one measure is not areliable proxy for the other when used instatistical studies.

CONCLUSION

Despite some improvement in available infor-mation on state export activity in recent years,the two existing state export series are deficientin several ways. Their most important limitationis that they both measure the value of shipmentsand not the extent of a state’s economic activity(value added) related to manufactured exports.Nonetheless, as general indicators of export ac-tivity across all states, the two measures providesimilar information Despite this overall similari-ty, the two series can lead to substantially dif-ferent conclusions when used for some states.Furthermore, the OMC series, which is availableon a more timely basis, is not a satisfactory

‘4The dashed line indicates that a state’s value added tendsto be higher than its shipments. In particular, each dollarincrease in state export shipments in 1986 was associatedwith a $1.36 rise in export value added, on average.

proxy for the more accurate EME data on ex-port shipments according to the criterion usedin this article.

The estimate of a state’s export value addedgenerated in this article is inherently superiorto the existing measures available for assessingstate export performance. This new measurecan produce different conclusions than ship-ments-based data when used for some states orwhen used in statistical studies. Consequently,users should reconsider their use of the existingexport series when they desire an accurate mea-sure of a states value added related to manufac-tured exports.

The evidence presented here on export valueadded and its deviation from the EMLI-based ex-port shipments measures is for one year only,however. The behavior of this discrepancy overtime is unknown. This reinforces the importanceof developing historical data on state exportvalue added for analyses involving state exportactivity.

IIIIIIIIIIIIIIIII

REFERENCES

Bourque, Philip J. The Washington State Input-Output Studyfor 1982, Graduate School of Business Administration,University of Washington (March 1987).

Coughlin, Cletus C., and PhUllip A. Cartwright. “An Examina-tion of State Foreign Exports and Manufacturing Employ-ment,” Economic Development Quarterly (No. 3, 1987), pp.257-67.

Farrell, Michael C., and Anthony Radspieler. ‘Origin ofMerchandise Exports Data,” unpublished manuscript,Foreign Trade Division, Census Bureau, US. Department ofCommerce (March 20, 1990).

Jackson, John D., and James A. Dunlevy. ‘The Interchange-ability of Alternative Measures of Permanent Income inRegression Analysis,” Applied Economics (October 1982),pp. 455-68.

_______ - “The Orthogonal Least Squares Slope Estimator:Interval Estimation and Hypothesis Testing Under theAssumption of Bivariate Normality,” Proceedings of theBusiness and Economic Statistics Section of the AmericanStatistical Association (1981), pp. 294-97.

Lerch, Stephen C. “State of Origin Export Data: An Interpre-tive View,” unpublished manuscript, Office of FinancialManagement, State of Washington (1990).

Little, Jane Sneddon. ‘New England’s Links to the WorldEconomy,” Federal Reserve Bank of Boston New EnglandEconomic Review (November/December 1990), pp. 33-50.

Malinvaud, Edmond. Statistical Methods of Econometrics(North-Holland, 1980).

JULY/AUGUST 1991

76

Ott, Mack. “Have U.S. Exports Been Larger ThanReported?” this Review (September/October 1988),pp. 3-23.

Smith, Tim R. “Regional Export Growth: Lessons from theState-Level Foreign Trade Data,” Regional Science Perspec-tives (No. 1, 1990), pp. 21-38.

“Regional Exports of Manufactured Products,”Federal Reserve Bank of Kansas City Economic Review(January 1989), pp. 21-31.

U.S. Department of Commerce, Bureau of the Census,Exports from Manutacturing Establishments: 1985 and 7986(GPO, January 1989).

Appendix A

Exports from Manufacturing Establishments: 1987(GPO, February 1991).

________ - “State Offices That Provide Export Assistance,”Business America (No. 2, 1991) pp. 24-25.

U.S. Office of Management and Budget. Standard IndustrialClassification Manual 1987

Webster, Elaine, Edward J. Mathis, and Charles E. Zech.“The Case for State-Level Export Promotion Assistance: AComparison of Foreign and Domestic Export EmploymentMultipliers,” Economic Development Quarterly (August1990), pp. 203-10.

Interchangeability of Alternative MeasuresThe existence of alternative export measures

raises the issue of the extent to which themeasures are interchangeable. In other words,do these measures provide virtually identical in-formation about state export performance? Dif-ferent criteria exist for assessing this issue.

Three criteria are used here: 1) rank correla-tion criterion; 2) simple correlation criterion;and 3) orthogonal regression criterion. The rankcorrelation criterion focuses on the degree towhich measures have identical rankings for cor-r’esponding observations. As a first step, states(including the District of Columbia) are rankedfrom 1 (the state with the largest value of ex-ports) to 51 (the state with the smallest value ofexports) for each export measure. To make pair-wise comparisons of the rank-order, a Speat-manrank correlation coefficient, H, is calculated asfollows:

(1) H, = 1 —[6!N(N2— 1)1 (0—

where i is a subscript denoting specific states(i= 1, 2,...,51); 0~is the rank of the ith stateusing one measure; A1 is the rank of the ithstate using the alternative measure; and N is thesample size.

If the rank-orders of the two measures areidentical, then R,= 1. For example, using onemeasure and assuming that three states—A, B

and C—had export values of 300, 200 and 100,the states would be ranked 1, 2 and 3. If, usingthe other measure, states A, B and C had exportvalues of 5, 4 and 1, then an ordering of thestates based on the two export measures wouldbe identical and the Spearman rank correlationcoefficient is one. A rank-order of zero meansthat the rank-orders of the two measures arenot related. A rank-order of minus one meansthat the rank-orders at-c the reverse of eachother. Thus, the closer R, is to one, the moresimilar the rank-orders and the more inter-changeable the measures for ranking puiposes.

As indicated in the text, the Spearman rankcorrelation coefficient was approximately onefor- the comparison of the 1987 export measuiesin LIME and OMC (.96) and the 1986 measuresin LIME and our export value added (.98). Conse-quently, the measures provide highly correlatedlankings. These results suggest that, as a sum-mary indicator of states’ relative export perfor-mance, the measures provide roughly identicalinformation. Satisfying this criterion, however,does not preclude large differences in a specificstate’s r’ank across the measures. For- example,recall that the state of Washington droppedfrom fifth place using the shipments-based datain the LIME for 1986 to 12th place using our ex-port value added measure.

A stronger condition than rank correlation in-volves the simple correlation of the levels of the

rrnrnn,nec-otrn,onA,,v no ot I fli HO

I 77

alternative measures. The simple correlationcoefficient provides information concerning theextent of a linear relationship between the alter-native measures. The simple correlation coeffi-cient is calculated as follows:

(2) r = I(x-~)(y--z)/~(I(x—~’Z(y-~)’),

where ~ and ~ are the sample means of thealternative export series, x and y.

For any given state, if the value of exportsusing one measure exceeds the mean of thismeasure based on all states by a certain amountand the value of exports using the other measurealso exceeds (or falls below) its series mean by aset amount, then a perfect linear correlationexists. The value of the correlation coefficientwill equal one (or minus one in the event of anegative relationship). For example, as in thenumerical example above, assume states A, Band C had export values of 300, 200 and 100using one measure. Using the other measure,assume states A, B and C had export values of450, 250 and 50. Thus, A’s exports exceed thefirst series mean of 200 by 100 using onemeasure, and exceeds the second series mean of250 by 200 using the other measure. For a per-fect linear correlation, C’s exports (which are100 less than the first series mean) must be 200less than the second series mean (that is, equal50). Since this is the case, the correlation equalsone. A correlation coefficient of zero meansthat no linear relationship between themeasures exists.

As indicated in the text, the linear relationshipbetween both sets of measures is strong. Forthe 1987 measures in LIME and 0MG, the cor-relation coefficient is .95. For the 1986 measuresin LIME and of export value added, the correla-tion coefficient is also .95. Since these coeffi-cients are virtually one, the measures can beviewed as interchangeable using this criterion.

When using a more stringent criterion, how-ever, this is not the case. This criterion for inter-changeability requires that the measures are notonly highly correlated, but that they consistentlydiffer by a constant, possibly zero. Once again,

‘In contrast to simple regression, the fitted line in or-thogonal regression is the one that minimizes the meansquare of the perpendicular (rather than the vertical) devia-tion of the sample points from the fitted line. See Malinvaud(1980), for a thorough discussion of the differences betweenorthogonal and simple regression.

assume states A, B and C had export values of300, 200 and 100 using one measure. Using theother measure, assume state A had exports of350. For “difference preservation,” states B andC’s exports must be 250 and 150. In this case,the two measures differ by 50 for each state.This difference preservation is known as an or-thogonal regression criterion.’

Jackson and Dunlevy (1982) illustrate thiscriterion, in a time-series context, by a simpleexample of estimating a consumption functionwith different permanent income measures.Assume perfect correlation between two incomemeasures y, and y,, so that:

(3) y, = a + by,,

where a is the intercept and b is the slope.Suppose the following consumption function isestimated:

(4) c = d + ey, + E,

where d is the intercept, e is the slope and £ 18

the random disturbance term. The slope is calledthe marginal propensity to consume and is thechange in consumption associated with each $1change in income. If y, is used rather than y,~however, the consumption function becomes:

(5) c = (d+ea) + eby, + £.

The two measures of income will yield the sameestimate of the marginal propensity to consumeonly if b equals one.

Using orthogonal regression, we generatedestimates of b for the alternative export mea-sures discussed in the text. Specifically, weestimated two equations similar to equation 3.In one regression, the 1987 measures of stateexports in 0MG and LIMLI were used as y, andy,, respectively. In the other regression, the1986 measures of state exports based on ourcalculations of export value added and in EMLIwere used as y, and y,, respectively.

The orthogonal regression criterion is notsatisfied by alternative export measures. For the1987 measures in LIME and 0MG, the or-

IIIIIIIIIIIIIIIIII

78

thogonal least squares estimate of b equals 1.37.A t-statistic can be used to test the null hypoth-esis that b equals one.’ The critical t-value for a5 percent significance level with 49 degrees offreedom is 2.01, which is far below the actualt-value of 7.03. Consequently, the null hypothesisis rejected. Similarly, the 1986 measures in LIMEand of export value added produce an ortho-gonal least squares estimate of b (1.38), whichyields a rejection of the null hypothesis that bequals one; the critical t-value of 2.01 at the

5 percent level of significance is far less thanthe actual t-value of 8.35.’

The implication of this analysis is that thelevels of alternative export measures are not in-terchangeable using this criterion and their usewould generate different regression results inotherwise identical estimations. Specifically, thecoefficient estimates for the impact of a changein state exports on some variable of interest,say state economic growth, would differ depen-ding on the measure used.~

‘Because of random variation in the data it is unlikely thatb exactly equals one. Therefore, the t-statistic is used totest whether we can reasonably infer that the estimatedvalue of b equals one. See Jackson and Dunlevy (1981)for additional details on hypothesis tests involving the or-thogonal least squares slope estimator.

‘A related issue involves whether the two measures areconsistently proportional to one another, that is, whetherthey tend to differ by a given percentage. This is investi-gated by testing whether the orthogonal least square slopeestimator between the logarithms of the two measuressignificantly differs from one. Using the logarithms of the1987 measures in EME and OMC, the slope estimateequals 1.01. The associated t-statistic is 0.109, which isless than the critical t-value of 2.01 (5 percent significancelevel). Consequently, the null hypothesis that the slope

equals unity cannot be rejected. These results suggestthat the logarithmic forms of the two 1987 exportmeasures in EME and OMC are interchangeable. On theother hand, the logarithmic forms of the 1986 measures inEME and of export value added yield an orthogonal leastsquares slope estimate of 0.892. Because the associated t-statistic of 2.95 exceeds the 2.01 critical value, the nullhypothesis is rejected, suggesting that the two measuresof 1987 exports are not interchangeable.

4See Coughlin and Cartwright (1987) for an empirical ex-amination of the effect of manufacturing exports onemployment for individual states. This is an example of astudy where the regression results could be altered by us-ing different export series.

Appendix BEstimating Value Added Related to ManufacturedExports by State

In this appendix, we identify the data andmethodology used to calculate the value addedrelated to manufactured exports by state. Webegin by identifying the variables used in thecalculations and the data sources.

Various employment, shipments and gross stateproduct data are essential for our calculations.Manufacturing employment (ME), export-relatedmanufacturing employment (XME) and export-related nonmanufacturing employment (XNME)for each state are published in Exports fromManufacturing Establishments: 1985 and 1986,

U.S. Department of Commerce (1989).’ ME isreported by respondents in the U.S. CensusBureau’s Annual Survey of Manufactures, whileXME and XNME are calculated by the CensusBureau.

Three other series are used. The first is un-published data from the U.S. Department ofCommerce (1991) on nonmanufacturing employ-ment (NME). The second data series, which ispublished in Exports from ManufacturingEstablishments: 2985 and 1986, is total stateemployment. ‘Fhe third data series, which ispublished by the U.S. Department of Commerce(1989), is gross state product (GSP). GSP is the

‘For a detailed explanation of how these data weredeveloped, see this publication’s Introduction and Appen-dix C

1 79

market value of the goods and services pro-duced within a state during a year and is thestate analog of U.S. gross domestic product. GSPdata for individual manufacturing and non-manufacturing industries were used,

Methodology: calculating ExportValue Added

To calculate total value added related tomanufactured exports in state s (XTV,), wesummed estimates of value added within thestate’s manufacturing sector (XMV,) and valueadded in nonmanufacturing sectors related tothe export of manufactured goods (XNMV,).That is,

(1) XTV, = XMV, + XNMV,.

Because identical data were not available foreach manufacturing sector, the components ofXMV, were calculated in one of two ways.’ Forindustries in which export-related data arepublished, XMV, was estimated by applying thefollowing equation:

(2) XM%ç, = (GSP,1/ME,J(XME,).

As defined above, GSP is gross state product,ME is manufacturing employment and XME isexport-related manufacturing employment. Thesubscript i designates the different SIC manufac-turing industry groups. In our calculations, weused the two-digit manufacturing industrygroup so i ranged from SIC 20 to SIC 39. Thismethod implicitly assumes, for each industry,that output per worker in the production of ex-port goods (XMV,1/XME,I) is the same as outputper worker in the production of all goods(GSP,

1IME,). Equation 2 was applied using data

for individual industries rather than for totalmanufacturing, because this assumption is moreplausible for each industry than for manufac-turing as a whole.

For those manufacturing sectors with no pub-lished export-related employment at the two-

‘In some states, export-related manufacturing employmentdata were not published for certain industries either toavoid disclosing data for individual companies or becausethe estimate did not meet publication standards. Summingover all states, unpublished export-related manufacturingemployment accounts for I .8 percent of total 1986 export-related manufacturing employment.

digit industry level, XMV, was estimated usingthe following equation:

(3) XMV,m = (GSP,rn/MEsn,RXMEs,J,

where the m subscript refers to the total ofthose sectors not reported. For example, tocompute a state’s total unreported export-re-lated manufacturing employment (XME,0,), wesimply subtracted the amount reported fromthe total export-related manufacturing employ-ment. Consequently, XMV, is the sum of theestimates for the reported industries (XMV,1)plus the single estimate for the missing in-dustries (XMV,,,,).

To compute a state’s value added in its non-manufacturing sectors related to manufacturedexports (XNMV,), we first estimated the follow-ing measure for each of a state’s four nonnianu-facturing industries [where 3 = 1, trade; j = 2,business services; j = 3, transportation, com-munication and utilities; and j =4, other}:

(4) XNMV,~ = (GSPS~/NME,~)(XNME,~),

where GSP for the “other” sector is calculatedas total state GSP minus manufacturing andminus the three nonmanufacturing industries,(3 = 1.. .3); NMES~is nonmanufacturing employ-ment in industry j; and XNME,J is export-relatednonmanufacturing employment in sector j.’“Other” employment is total state employmentminus employment in manufacturing, trade,business services and transportation, com-munication and utilities.

‘rhe state total for the value added in thesenonmanufacturing sectors, XNMV,, is simply thesum of the value added in the four nonmanu-facturing sectors. The accuracy of this calcula-tion, similar to the calculation for the manufac-turing sectors, rests on the degree to which theproductivity of export-related workers in sector

(XNMVS~/XNME,~)is equal to the productivity ofall workers in that sector (GSP,

1INME,,).

‘Export-related nonmanufacturing employment in the“other” sector accounts for 31.8 percent of the 1986 na-tional total for such employment.

IIIIIIIIIIIIIIIIIS H H V’s’