WAYNE VROMAN Urban Institute JOHN M. ABOWD Cornell University Disaggregated Wage Developments UNEMPLOYMENT has declinedsubstantiallyin the United States in recent years, from 9.6 percent of the civilianlaborforce in 1983to 5.7 percent in the firstquarterof 1988,while wage and price inflationhave remained low. Annual increases in the private nonfarmaverage hourly earnings index, for example, rangedfrom2.0 percentto 3.8 percentandaveraged 3.0 percentbetween 1983and 1987.Duringthe firstthreemonthsof 1988 averagehourlyearningsincreasedat an annualrate of 2.5 percent. This reportexamines nominalwage growth duringthe 1980s. It first documentsthe slowdown in wage inflationand then examines possible sources of that slowdown. Recent Wage Inflation Table 1 shows data on average rates of unemploymentand of wage andpriceinflationsince 1965.Wageinflationratesforthe privatenonfarm sector are shownfor three series: the hourlyearningsindex, the employmentcost index, andeffective wage changesfor majorunionagreements (agreementsaffecting 1,000 or more workers). Price inflationis shown for the consumerprice index andthe GNP deflator.In each series, wage and price inflationaveragedless than 4 percent a year during1983-87. Wagegrowthduringthat period was at least 5 percentagepoints below its average during1980-81. Not since the early 1960shas wage growth The authors acknowledge financial support from the National Science Foundation and the Ford Foundation. Ron Ehrenberg provided helpful comments on an earlier draft. 313 314 Brookings Papers on Economic Activity, 1:1988 Table 1. Average Wage Inflation, Price Inflation, and Unemployment Rates, 1965-87 Percent per year Price inflation Wage inflation Employment cost indexb Major union agr eementsc Consumer price indexd Implicit GNP deflatore Unemployment rate Men aged Civilian 24-54 Period Hourly earnings indexa 1965-69 1970-74 1975-79 5.4 7.2 7.7 n.a. n.a. 7.6U n.a. 8.2 8.4 3.8 6.7 8.2 4.2 6.8 7.6 3.8 5.4 7.0 2.0 3.0 4.4 1980 1981 1982 1983 1984 9.3 8.1 6.1 3.8 3.2 9.0 8.8 6.3 5.0 4.1 9.9 9.5 6.8 4.0 3.7 12.5 8.9 3.8 3.8 3.9 9.9 8.7 5.2 3.6 3.4 7.1 7.6 9.7 9.6 7.5 5.2 5.5 8.0 8.2 5.9 1985 1986 1987 3.2 2.0 2.6 4.1 3.1 3.3 3.3 2.3 3.1 3.8 1.1 4.4 3.1 2.2 3.3 7.2 7.0 6.2 5.6 5.6 5.0 Source: Hourly earnings index, consumer prices, and unemployment rate are from U.S. Department of Labor, Bureau of Labor Statistics, Eniploynietit anid Earnings, various issues. Employment cost index and major union agreements are from Bureau of Labor Statistics, Current Wage Developments, vol. 40 (March 1988). GNP deflator is from National Income and Product Accounts. n.a. Not available. a. December to December percent change for private nonagricultural workers. b. December to December percent change of wages and salaries of private industry workers. c. Average annual wage adjustment effective in the year for all industries. Major agreements are those covering 1,000 or more workers. d. December to December percent change. e. Fourth quarter to fourth quarter percent change. f. 1976-79. been so slow. The trend is particularlystrikingbecause of the drop in unemployment,from9.6 percentto 6.2 percent, between 1983and 1987. Aggregate Wage Growth To estimate the size of the unexpected component in the recent slowdown in wage growth, we fit a series of augmentedPhillipscurves, in which aggregatewage growth is modeled as a function of the unemploymentrateandthe expected rateof priceinflation.Expectedinflation is measured with a three-year distributed lag on the actual rate of inflation.The regressionsexplainannualizedquarterlypercentagechanges in the average hourly earnings index (HEI) for the private nonfarm economy. We use changes in quarterlyaveragesof unadjustedmonthly Wayne Vroomanand John M. Abowvd 315 HEI data to have availablea full rangeof comparabledata for the nine industrialdivisions thatunderliethe aggregateindex. Because we do not use seasonally adjusted data, we include additive quarterlydummy variablesin each regressionto controlfor seasonality. Two specificationissues are important:the unemploymentvariable and the variable to test for effects of internationaltrade on wages. Because the demographicmix of unemploymenthas changed since the 1960s and 1970s, the choice of the cyclical control variable strongly influencesthe performanceof the regressions.Associated with the high overallunemploymentof the 1980shas been a largecomponentof longterm unemploymentand of unusuallyhigh adult male unemployment. In every yearbetween 1981and 1987the numberof workersunemployed 27 weeks or longer exceeded 1 million, and the share of those workers in total unemploymentwas at least 14 percent. Between 1948and 1979 long-termunemploymentsharesof 14percentor higheroccurredinjust seven other years: 1958, 1959, 1961, 1962, 1975, 1976, and 1977. The exceptionally high adult male unemploymentrate may be due to the increased pace of U.S. industrialrestructuringand increased foreign trade competitionas well as to the persistentlyhigh overall unemployment of the decade. In 1987, when the overall unemploymentrate was 6.2 percent, the unemploymentratefor men aged 25-54 was 5.0 percent, or0.81of the overallrate.In 1976,anearlieryearwhentheunemployment ratefor menaged25-54 was 5.0 percent,the overallratewas 7.7 percent. Figure 1 illustratesthis relative change in adult male unemploymentof the 1980s.The relativeunemploymentratefor men aged 25-54 averaged 0.79 between 1980and 1987,but 0.59 in the 1970s. The change in the structureof relative unemploymentrates makes the choice of which unemploymentrate series to use in the regressions particularlyimportant.' We use both the overallunemploymentrateand the rate for men aged 25 to 54. Equationsbased on the unemployment ratefor adultmen predictlower wage growthfor the 1980s. Given the long-run trend toward increased openness of the U.S. 1. The importance of changes in the demographic structure of unemployment and its implications for labor market inflation were first emphasized by George L. Perry, "Changing Labor Markets and Inflation, " BPEA, 3:1970, pp. 411-41. For a recent analysis of demographic changes in the unemployment of the 1980s, see Lawrence H. Summers, "Why Is the Unemployment Rate So Very High near Full Employment?" BPEA, 2:1986, pp. 339-83. 316 Brookings Papers on Economic Activity, 1:1988 Figure 1. Ratio of Unemployment Rates for Men Aged 25-54 to the Civilian Unemployment Rate, 1953:1-1987:4 Ratio 0.9 - 0.8 0.7 0.6- 0.5- 0.4 I - 1955 1960 1965 I I 1970 I I I 1975 I 1980 I I 1 F 1985 Source: Eniplovy,etit and Earnings,variousissues. economy and the persistence of large U.S. merchandisetrade deficits since 1984, we test for the effect of foreign trade on domestic wage growth, using both price variables and trade flow variables.2 Price variablesinclude changes in importprices and changes in nominaland real exchangerates.3Tradeflow variablesincludeimportsas a share of GDP or of sales, as well as export sharesand net export shares. We find that importprice inflationexhibits consistently small but significantly 2. John M. Abowd codirected a recent large-scale project at the National Bureau of Economic Research that assessed the effects on the labor market of international trade and immigration. In a later section of this paper we use international trade data and equation specifications drawn partly from that project in a microeconomic analysis of union wage settlements in manufacturing. See John Abowd and Richard Freeman, "The Internationalization of the U.S. Labor Market" (National Bureau of Economic Research, March 1987). 3. Rudiger Dornbusch and Stanley Fischer, "The Open Economy: Implications for Monetary Policy and Fiscal Policy," Working Paper 1422 (NBER, August 1984); Robert J. Gordon, "U.S. Inflation, Labor's Share and the Natural Rate of Unemployment," paper presented at International Seminar on Recent Developments in Wage Determination, Mannheim, Germany (October 5-6, 1987). Wayne Vroman and John M. Abowd 317 positiveeffects on the growthin privatenonfarmhourlyearnings.Import price inflationnet of petroleumand petroleumproductshas coefficients rangingfrom 0.08 to 0.14 in aggregateequations. Expected negative effects of changes in the trade-weightedreal or nominal U.S. dollar exchange rate are not significant.Also insignificantare the potential positive effects of the export share and the net export share (as a percentageof real GDP) and an expected negative effect of the import share. As will be reported later, import shares are significantin an analysis of microeconomicdata from manufacturing.We attributethe contrast in results to the greaterrange of variationof import shares in the microeconomicdata. Table 2 shows a set of six regressionresults and equationprojections for 1980-87. Price inflationis measuredwith the consumerprice index, entered as a twelve-quarterdistributedAlmon lag startingwith a onequarterlag. Importprice inflation,measuredwith a fixed-weightindex of nonpetroleumimportpricesfromthe nationalincomeaccounts, enters as an Almon variable with a four-quarterlag. The data period for all regressionsbegins in 1964:2and ends in either 1979:4or 1987:4. The first two equationsin table 2 show the importanceof the choice of unemploymentmeasure. In augmentedPhillipscurves fittedthrough 1979:4,unemploymentandlaggedinflationare both significant,with the inflationrate the more significantof the two. Equation2, which uses the unemploymentrate for men aged 25-54, fits somewhat better over the sampleperiodand makes much moreaccuratewage growthprojections for 1980-87. The average overprediction errors over the eight-year projectionperiodare, respectively, 1.50percentand0.49 percent.When import price inflationis added, in equation 3, the average eight-year projectionerror drops to 0.02 percent. The addition of import prices causes the equationto make lower wage growthprojectionsfrom 1981 through1985.As the averageprojectionerrorsin table 2 show, the use of the unemploymentratefor men 25-54 causes abouttwo-thirdsof the difference between the projections made by equations 1 and 3 in the 1980s.4 4. A similardecompositionbetween the averageerrorsfrom equations 1 and 3 was obtainedwhen importprice inflationwas addedto equation 1. In a regressionthat used importprice inflationand the unemploymentrate for persons aged 16 and older, the averageoverpredictionerrorsfor 1980-87and 1984-87were, respectively,0.88 percent and 1.11percent.Thusunderbothpossibledecompositions,the use of the unemployment rate for men aged 25-54 made a largercontributionthan importprice inflationto the superiorperformanceof the projectionsfromequation3 comparedwithequation1. Brookings Papers on Economic Activity, 1:1988 318 Table 2. Regressions Explaining the Growth in the Hourly Earnings Index, 1964:2-1979:4 and 1964:2-1987:4a Eqluation Independent variable Constant (1) 4.95 (11.2) .. . 4.93 (10.8) ... Laggedimportpricec 0.81 (7.1) ... -0.81 (4.2) 0.81 (9.1) ... - 0.89 (4.0) 0.74 (8.6) 0.11 (3.0) Predictionerror (actualminus predicted) 1980 1981 1982 1983 1984 0.70 2.17 1.65 1.02 1.87 - 0.06 - 1.59 - 0.28 0.50 - 0.91 Civilianunemployment rate Unemploymentrate, men aged 25-54 Lagged CPIb 5.65 (8.6) - 0.59 (3.3) ... 1964.2-1979:4 (2) (3) 1964.2-1987:4 (4) (5) (6) 6.87 (12.1) - 0.89 (7.6) ... 5.26 (15.0) 4.91 (14.5) . ... 0.87 (13.7) ... - 0.85 (9.9) 0.79 (16.9) ... - 0.83 (10.4) 0.75 (16.2) 0.10 (3.6) 0.50 - 0.69 0.50 0.99 - 0.29 - 0.22 - 1.56 - 0.28 0.51 - 0.90 0.26 - 1.26 0.12 0.85 - 0.70 0.38 -0.84 0.24 0.76 -0.44 1985 1986 1987 - 1.33 - 1.71 - 1.56 -0.41 -0.66 -0.48 0.60 - 0.68 - 0.80 - 0.42 - 0.82 -0.92 - 0.22 - 0.48 - 0.32 0.39 -0.73 -0.79 1980-87 1984-87 -1.50 -1.61 -0.49 - 0.62 0.02 - 0.29 -0.58 -0.76 - 0.22 - 0.43 -0.13 -0.39 0.68 0.97 1.77 0.73 0.89 2.02 Summarystatistic R2 Standarderror Durbin-Watson 0.65 1.02 1.65 0.75 1.14 1.33 0.80 1.01 1.68 0.83 0.94 1.85 Source: Authors' calculations based on data from the Bureau of Labor Statistics and the National Income and Product Accounts. a. The dependent variable measures the annualized quarterly percent change in the average hourly earnings index. Data are not seasonally adjusted, and each regression includes quarterly seasonal dummies. Numbers in parentheses are t-statistics. b. Lagged one quarter. The variable enters as an Almon variable with a 12-quarter distributed lag. c. Fixed-weight index of nonpetroleum import prices. The variable enters as an Almon variable with a four-quarter lag. Equations 4-6 apply the specificationsof equations 1-3 to the full period 1964:2through1987:4.Equations5 and 6 are similarto equations 2 and 3 in both their coefficients and average projection errors. In equation 4 the average projection errors of the 1980s are reduced, comparedwith those of equation 1, because these later observations help to determine the estimated coefficients. Although the average Waynie Vroman and John M. Abowd 319 residualsin the 1980s are generally smallerin equations 4-6, the rank orderingacrossthe threeequationsis generallypreserved-that is, those from equation4 are the largest. All equations overpredictin 1986 and 1987.Importprice inflationof these years adds to the overpredictions. Wage Adjustments in Broad Industrial Divisions As a check on ouraggregatewage changeequations,we next conduct disaggregatedanalysis on the nine broadindustrialdivisions that make up the hourlyearningsindex. As table 3 shows, the divisions have wide variationin the demographicmix of employment,the degreeof openness to internationaltrade, and the extent of unionization. In general,the industrialdivisionsin whichmenaged25-54 constitute a large share of total employmentare also comparativelyopen to trade and have above-averageunionization.Miningand durablemanufacturing, for example, rankhigh on all three dimensions, while retail trade, finance, and services rank low on all three. The other four industry divisions are more mixed, with the two biggest exceptions being wholesale trade,whichhas moreadultmaleemployeesthanmightbe expected, and construction,which is less open to internationaltradethanmightbe expected. Because demographic mix, openness to internationaltrade, and unionizationhave common patternsacross broad industrialdivisions, an analysis of wage dynamics at this level of disaggregationis less revealingthan might be expected. In a longer, unpublishedversion of this paper, we report wage growth regression results for the nine industriesusing the same specificationas in the aggregateanalysis of table 2. While interestingin several respects, the results for the nonpetroleum import price inflation variable were not consistent across industrialdivisions. Specifically, wage growth in wholesale and retail trade showed more sensitivity to internationaltradethan was plausible giventhese industries'tenuousconnectionto the internationaleconomy. To help isolate the separate effects of demographics,international trade, and unionwage developments,we conduct two additionalanalyses. First, we examineunionwage adjustmentsfromthe manufacturing sector, using detailed internationaltrade data by industry to explore bothtradeanddemographiceffects on unionwage adjustments.Second, we examinemajorcollective bargainingagreementsin the 1980s. Brookings Papers on Economic Activity, 1:1988 320 Table 3. Demographic, International Trade, and Unionization Characteristics of Major Industrial Divisions Expor-t employinent share gained, Import employment shar-e lost, Industry Employment share of men aged 25-54, 1980 1984a 1984b Union r-epresentation share, 1987 Mining Construction Durable manufacturing Nondurable manufacturing Transportation and utilities 0.606 0.604 0.505 0.400 0.546 0.245 0.016 0.154 0.093 0.101 0.460 0.022 0.254 0.251 0.076 0.195 0.222 0.263 0.224 0.362 Wholesale trade Retail trade Finance Services 0.481 0.243 0.286 0.253 0.060c 0.060c 0.026 0.025 0.014c 0.014c 0.027 0.023 0.091 0.074 0.032 0.076 All industries 0.372 0.065 0.078 0.146 Sources: Employment shares for men aged 25-54 taken from U.S. Bureau of the Census, Censuis of Popiulation 1980, Detailed Populationi Characteristics, U.S. Surnmnary,Sec. A: United States (Government Printing Office, 1980). Export employment shares taken from Kan Young, Ann Lawson, and Jennifer Duncan, "Trade Ripples Across U.S. Industries" (U.S. Department of Commerce, January 1986), tables 3A, 3B, and appendix table 4B. Union representation data from Employmenit anid Earnings, vol. 35 (January 1988). a. Share of industry's total employment gained from export activity. b. Share of industry's total employment lost due to import activity. c. Average shares for wholesale trade and retail trade combined. Major ManufacturingSettlements In this section we disaggregatefurtherand examine data from major manufacturingcollective bargainingagreementsbetween January1959 andDecember 1984.Ourgoal is to quantifythe determinantsof nominal wages at the bargainingunit level. Using the estimatedwage settlement equation, we show that settlements from 1980 through 1984 were predicted well by the same determinantsthat explained wages in the precedingtwo decades. We also examinethe extent to which wage settlementsare influenced by changing import and export competition in the industry. Because manufacturingindustriesdifferin theirsensitivityto importcompetition, we simulatedour estimated equationfor each majorindustrygroup to study the effects of a permanentincrease in the rate of importson wage settlements. The dependentvariableis the annualizedrate of wage growth over the life of the collective bargainingagreement.This rate includes wage changeseffective on the dayof settlement,scheduleddeferredincreases, Wayne Vromanand John M. Abowd 321 and realizedcontingentcost-of-livingadjustments(COLAs). The wage settlementsare datedfromthe beginningof the new contractso that, for example, the wage growth dated 1984covers 1984-86 for a three-year collective bargain. We considerthree distincttypes of explanatoryvariablesin the wage settlement equations-aggregate labor market conditions, aggregate price inflationexpectationsand realizations,and industry-levelproduct marketconditions. We discuss each below. The labormarketcontrols are the adultmale unemploymentrate the monthbefore settlement, the change in that rate since the last contract settlement(annualized),and the most recent wage settlement in either the automobileor steel industry(annualrate).The adultmaleunemploymentrateis most naturallyviewed as anindicatorof the generaltightness of the labor market. One expects the current wage settlement to be inversely related to the unemploymentrate. The most recent nominal wage settlement in the automobile or steel industry is a well-known leadingindicatorof the new wage settlementsin general.If it is providing informationthataugmentsthe informationintheprevailingpriceinflation expectations, it shouldbe positively relatedto wage settlements. We use two differentmeasuresof the expected inflationrate. The first is based on the predictedannualrateof changein the CPIas determined by the most recent annual rate of change and two annual lags. The expected inflationrate is calculatedfromthe regression: (1) ln CPI(t + 12) - In CPI(t) = 0.913 + 0.855 [ln CPI(t) - ln CPI(t - 12)] (4.847) (19.516) - 0.375 [ln CPI(t - 12) - ln CPI(t - 24)] (7.329) + 0.307 [ln CPI(t - 24) - ln CPI(t - 36)] (7.994) Period:January1950-July1986;K2 = 0.58; Durbin-Watson= 0.085. Since we require a forecast of the annual inflation rate based upon inflationduringthe previous contract, this specificationis the natural extensionof the distributedlag specificationswe fit to the quarterlydata for the aggregateequations. The CPI-basedexpected inflationrate is 322 Brookings Papers on Economic Activity, 1:1988 calculatedas the predictedannualinflationrate from this regressionin the month before the contract was settled. The CPI-based inflation surpriseis calculatedas the realized inflationrate duringthe life of the contract (annualrate) minus the CPI-basedexpected inflationrate the monthbefore the contractwas settled. Our second measureof expected inflation,the Livingstonexpected inflationrate, is the consensus expected annualinflationrate from the Livingstonsurveyjust priorto the contractsettlement.5The Livingston inflationsurpriseis calculatedas the realized inflationduringthe life of the contract (annualrate) minus the Livingstonexpected inflationrate the month before the contract was settled. We also considered an expected inflationrate and inflation surprise based on the Survey of Consumer Finances, but the results did not differ substantiallyfrom those obtainedwith the other two measuresand are not reportedhere. The inflationsurpriseover the life of the currentcontractis relevant only for contracts that contain a contingentCOLA clause. Because of the effects of triggers(minimumchangesin the CPIrequiredto produce an adjustment),such clauses typically increase realized wages more when the inflationsurpriseis positive. For this reason we allowed the inflationsurpriseto enter separatelyfor positive and negativesurprises. Only the positive inflationsurprises over the life of the contract had consistently large effects. The inflation surprise affects wage growth more in contractsthat do not cap the COLA formulathan in contracts that cap the cost-of-livingadjustmentat a specified amountregardless of the level of inflation.For this reason we estimate the effects of the inflationsurpriseseparatelyfor cappedanduncappedCOLAcontracts, and we suppressthe currentinflationsurprisefor contractsthat do not includea COLA. The effects of laggedinflationsurprisesarecalled "catch-up"effects. Wagecatch-upoccurs when the previouswage settlementwas based on an inflationexpectationthat turnedout to be substantiallyin error.This erroris measuredby the laggedinflationsurprise(the differencebetween the realized inflationrate over the life of the previous contract and the inflationrate expectation the month before that contract was settled). 5. The Livingstonsurvey, conductedby J. A. Livingstoneach June and December, surveys6d economistsfrombanking,labor,government,anduniversitiesregardingtheir predictionsfor variousleadingeconomicindicators.Surveyresultsare maintainedby the PhiladelphiaFederalReserveBank, Departmentof Research. WayneVromanand John M. Abowd 323 The effects of catch-upare symmetric,so no distinctionis requiredfor positive and negative surprises. However, catch-up should be a less importantphenomenonfor contractswith uncappedCOLAclauses. We estimate separate catch-up terms for the cases in which the previous contractcontainedno COLA,anuncappedCOLA,anda cappedCOLA. Finally, we consider two types of productmarketcontrols. The first is the annualrate of change in industryprices as measuredby the value of productshipmentsdeflatorfrom the AnnualSurvey of Manufactures (ASM). The second productmarketmeasureis a decompositionof the ASM value of shipments into the weighted sum of the annual rate of changein apparentdomestic consumption(D), definedas shipments(S) plus imports(M)minusexports(X);the annualrateof changein exports; and the absolute change in IPR, the importpenetrationratio (100 times importsdividedby apparentdomesticconsumption).Thedecomposition of productshipmentsis given by: dlnS - ( X)dlnD + -dlnX- -dIPR, andall changeshave been multipliedby 100.This decompositionallows us to measurethe effects of increasedimportand export activity in the productmarketholdingconstanttotal apparentdomesticconsumption.6 In particular,if the absolutevalue of the effect of the importpenetration ratio on wage settlements exceeds the absolute value of the effect of apparentdomestic consumption,then importcompetitionhas a greater effect on wage settlementsthanjust the effect impliedby the reduction in domesticoutputthataccompaniesan increasein the IPR withindustry prices and domestic consumptionheld constant. To complete our dynamicspecification,we also include the nominal wage settlementfromthepreviouscontractandthe annualrateof change of bargainingunit employmentover the life of the previous contractas controlvariablesin our estimatedequation.7 Table4 comparesalternativespecificationsof the dynamicmodel for 6. This decompositionis discussed and used in RichardB. Freemanand Lawrence Katz,"WagesandEmploymentinanOpenEconomy,"preparedfortheNBERconference on Immigration,Trade,andLaborMarkets(September1987). 7. John M. Abowd and Thomas Lemieux, "The Effects of InternationalTrade on Collective BargainingAgreements:Evidence from Canada," preparedfor the NBER Conferenceon Immigration,Trade,and LaborMarkets(revisedJanuary1988). " _ ~ Cu .1~i- '*- ~~~~ro* Cu ~~~ 0 - N r 00~~ N < 0 r C). -4 r -t t 00 N, . . o (7\ C? b i- f r- rr N 6- 06 N 10t Nt C) . " . -. CC 0 r. Cl 0 0? C00 ao 0 000 C It 00 . . ) DC rl~6 \6O b - ' 0 6r~6W ON r6 - N r N C) 00N 4.o otb< ^^~~~ 0 O^N N4 ClX e~~~ rj~~~~~~~~~~~~~~~~~~rv XS Cl N 0 0 o 0 0 N N 66r 66 >6r ; < ol CO N 0~~~~~~~~~~~0 ClCl 01 ;- -~00 00 - -m Cu -ON ON * . - O- 6 W~~~ ~ Zu 6i N 00 o~~- C-q 0 0 0 c 0 Cli < 000 V0f e7 6 rf 6&6ri C00 -ON ',C- Not C 00 0- C r \ o N ObN oorb<o N 00 i - i-~ . 0 o o -Cl-ON-NO 00 -N "C -00 N o r^ 0rl No -N r ri ON r~ -~ rf 0 N 0000 C - Cl 000 0" \-C Cl X r) r- 00 00 -- C) 0 C) C) ON) - 0 N oNNo ON 00 o o ON ?i 00 O C4 0 ON - oo U.tNoo_obb - 0 _ - N N ooo ON N 0 "C . < 0~~~~~~~~~~~0 X~~~~N 0 ON4 o C~~~~~~V) em~C Sv) X ~~~~O o b s o l o 0 E C'N t~~~~ .i ~~~~Cu 4 < a E S E 0n N C' o t o - o- t1 C'soO ~~~~~~~~~~ ~ ~ ~ ~ ~ ~ ~~ @;, 0 A c e O s = C's0 ~~~~~~~ ,Eo , *- -< ~~~' i. . *:: - cin.> e 9~>In 0 in u ~~~~~~~~~I. in E'?- C' C'sQ o in. I: O l C XN O? N00 - O .t t -X t C - m e^ 646 C) 6 -ON NON "C iC)- ci NNr C)O "C QN "CN CONC " b - e = - 'C 0 ?- ON t C) 00 rC ON,I .m >. 5F. Cl rl Cl4 oo 0N o C r-N "CN oo~ CC, Cl 6i Q ONN C,\"C 0 's -p rC's C's en en m in r- 00 m 00 "t IC - C's C's C's C's IU C's C's V) C'sIU C's IU ct z C's 00 C's > C) 00 M 00 W) a,\ 00 C) 't IC m 6. IU 00 C) C's C's 00 C) co 17) rl 00 m 00 c-, a, C's tn 00 00 IC - C's C's C's m m a,\ m 00 00 IC m M v) "t "t "C 00 m C's N00 00 CN C\ 'It IC lu C's m C's C's I-Q d) .6 421 C's C's C's lu U C's U-a C's 'A Ca > C's >C's= C's C's C"s C's lu r- C's 4u. . . C's . 20 Ca, C's 0 6. 0. C's = C's E C. >, C's d) 0. cl, 0 -C's O,c = - t A d) 0 > U C's Q C's C's0 C's 0 4.- C's C's = C's : v. C's C's 4. 0 Cr C'S C's >1 C's Ob a) C's in. 0 u C's C's C's C's C's lu lu 00 C's C', in. u C's u C's 00 00 m 00 C's 00 CN Z 0 = IC) = C's C C4 a:IZ 0. C's t4-. Ob t 0. C's ; C's " i. C's C's e : 44 326 Brookings Papers on Economic Activity, 1:1988 wage settlementsestimatedfor 1959-84. Columns1-4 reportregression coefficientsand theirassociated t-statistics, with columns 1-3 using the CPI-basedmeasure of expected inflationand column 4 using the Livingstonexpected measure. The first question of interest is the extent to which the period from 1980 through 1984 represents a break from the structure of wage determinationin the previous two decades. Column 1 reports the regression estimates of the dynamic wage settlement equation from a specificationthat includes year effects (dummies)for each year from 1980 to 1984. These year effects show that the equation overpredicts wage settlements in four of the five years, althoughonly the 1980and 1981overpredictionsare statisticallysignificant.The relevant standard for interpretingthese year effects as forecast errorsis the standarderror of the equation, which is 2.028; hence, none of the errorsin the 1980s exceeds one-halfof the standarderrorof the equation. Althoughoverprediction 4i 1980 is large, it is about the same as overpredictionsin 1960, 1969, 1970,and 1971.The errorsin 1981-84are aboutthe same as those of the previousdecades. The equationwiththe Livingstonmeasure of expected inflationreaches a similarconclusion. Figure2 breaksthe predictedwage settlements into five key components. Not shownis a sixth component,averageannualeffect of changes in industryconditionsincludingimports,discussed below, that contributes an average of 88 basis points a year and explains about 50 basis points of the decline in the 1980s. The figure shows that from 1980 through1984nominalwage settlementsdeclinedabout600basis points. Of thattotal the decline in expected inflationcontributedabout200basis points; the decline in key settlements in the automobile and steel industriescontributedanother 200 basis points; positive inflationsurprises contributednothing;catch-up(laggedinflationsurprises)contributed about 100 basis points; and the level and change in adult male unemploymentcontributedabout 50 basis points. The next questionof interestis the extent to which the decomposition of the aggregatefactors affectingthe wage settlementsis sensitive to the choice of expected inflationrate measureor the use of the key auto and steel settlement variable. Table 4 reports results in column 4 for the Livingstonexpected inflationrate measure. The equationbased on the Livingston expected inflationrate fits the data slightly better than the equationbased on the CPI expected inflationrate (column2). The only Wayne Vroman and John M. Abowd 327 Figure 2. Components of Predicted Wage Settlement, 1959-84 Percent 12- 10 Predicted wage settlement 8 - 6- 42 -> 0- A -2 - I 1960 1965 1970 1975 1980 Key to components + Key auto and steel settlements. O Expected inflation rate. * Positive inflation surprise in current contract. x Total inflation surprise in previous contract. V Adult male unemployment rate and its change since previous contract. Source: Authors' estimates based on equation 2 (using CPI-based expected inflation) of table 4. estimated effects that are materiallydifferent are the lagged inflation surpriseterms. Withthe CPI-basedmeasure,the estimatedcatch-upfor contractswith no COLA is smallerthanthe catch-upfor capped COLA contracts.Withthe Livingstonmeasure,thereis no significantdifference between the no-COLAand the capped-COLAcatch-upestimatedelasticities. Columns2 and 3 of table 4 provide comparableestimates with and withoutthe key settlementsvariable.The expected inflationrateelasticity in column3 aboutequals the sum of the key settlementand expected inflationelasticities in column2. The positive inflationsurpriseelasticities in column 4 are greater than those in column 2, and the catch-up effects are also slightlylarger.The effects of the level and change in the adultmaleunemploymentratearealso largerin column3 thanin column 1 r 00 O~~~~~~~~~" em 4k = I 1 CD o tr r4 I I I I Y I O I I Iot a, r- r r- 00 I II4 I a\ a\ 00I O) I,\ I (1 rt >o It It oo CD No 0XY0 - L. t ID I- It It I0 r, U t Er; c~~~~~~~~~~~~~~~ em~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~' >~~~~~~~~~~1 tr ?0 ?- < N rno rai "t (1 O > U t k X O mO?O X > oO X -N : X a w m X tL.o Cq mo e t > o t - O 0~~~~~~~~~~~00 tr (1 tr o 0 o - a 00 r- 0o a\o (1 C, a\ 4 or N(1 ( ?C' to C' C' o r-o (14 t) 0 o~rn It (1 cr4l 4 rf I Z ~~~~~~~~~~~~~~~~~ C' .0 C's VA< a t CA > > = O E t ?~~~~~~~~~~~~~-, o~~~~~~~~~~~~~ o o m m 00- c o N c) ) C) _ oot o0 oo oo ) oD W It rf q r-- C-i r--4 0 06 E Lz m ^ me t e m o t W) re) rf 06 tQs E b C, V) rn r- 3 00 C ) cq 00 cq a' 00 r- C1 r - > rf a'\ C) =3 m o o Y ;~~~~~~~~~~~~~, emO~~~~~~~" 00 V) r- r-00 00 C) 00 x = S C->-? < E = S= C7 = cq : Edi3 (~~~~~~~~~~~ =~~~~~~~~~~~~~~ 0 X (t D D e ?~~~~~~~~ ;0 oc 2 o t c oo m r > oo ooo ooo oo rN 3 O n oo = t X Vk s r? E3 06 6 C; 6 Z Ce w~~~~~~~~~~~~~~r z~~~~~~~~~~~~~cq oo-m c r q 4 00 em ~~~~~~~~~~~~~~~~~ Gtn ~~~~~~~~~~~~~~~~~~~~~~~~ O- r~6 6 Y 6 Q~~~~~~~~~' C, 4 O- r4, tr ,, Oi C, ' 0 (1 rc N >Nt 00 0 oo oc . >n =) CZ L", .=,, oo tr r Nc? D o a Dc <,,~~~~~~~~~~~~~O ~ ~ ~ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ C'soe0 > 3 ^ _ 3 t _ X2 o t Qo 5 > n o S U M ; =~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ o Lz f Q @ 4 < o S t_~~~~~~~~~~~~~~) 4 C) ee eeeee to- S~~~~~~~~~~~~~~~~~~~~~ *a t t_ < &.E i ~~~~ Q Pw ; OH C' s 9= C's,3 '-" m CQ eX E _r 'D r ~~~~~~~~~~~~~~~ Wayne Vroman and John M. Abowd 329 2. However, the equationin column 3 fits the data substantiallyworse than that in column2. (The F-statisticfor the significanceof the change in R2 is 234.46, which is the squareof the t-statisticon key settlements in column2.) The final question of interest is the extent to which changes in the productmarkets,particularlychanges relatedto importactivity, affect wage settlements. As noted above, the cumulative contributionof all product marketfactors to the decline in wage settlements duringthe early 1980swas about50 basis points. This estimatemasksconsiderable heterogeneityamong industrygroups. As is clear from the estimated effect of changesin importpenetrationon nominalwage settlementsfor all of the models in table 4, the coefficient is six times greaterthan the coefficient on the change in apparentdomestic consumption. In other words, the effect of importpenetrationon nominalwage settlementsis approximatelysix times greater than one would predict from the lost domestic outputalone. Table 5, which shows the effects on wage settlements of increasing imports,reportsthe resultsof simulatingthe equationin table4, column 3, separatelyfor 20 two-digit StandardIndustrialClassification(SIC) code industries.Each settlementis firstsimulatedto approximatesteady state, using the 1984 values of expected inflation, unemployment, changes in the nominalwage over the previous contract, and change in employment over the previous contract. The 1984 positive inflation surprises,laggedinflationsurprises,andchangein adultmaleunemployment are enteredas initialconditions, then zeroed to obtain the steady state. Finally, the 10-year industry average annual growth rates for prices, shipments,exports, and importsare used to generatethe simulation values for the change in industry prices, apparent domestic consumption,exports, and the importpenetrationratio. We simulated the transportationequipmentindustryfirstandused laggedvalues of the predictedsettlements in this industryas the key settlementvariablein the steady-statesimulation. Once the industryis in steady state (approximately10 years into the simulation),the rate of growthof importsis permanentlyincreased by one standarddeviationfor each industry.The table reportsthe resulting differencesin predictedwage settlements in that industryfor the first, sixth, and eleventh years following this change. The simulationholds constantthe growthrate of industryprices, shipments,and exports, so 330 Brookings Papers on Economic Activity, 1:1988 thereis no changein the rateof growthof real domesticproduction.The entirechangeis in the rate of growthof apparentdomestic consumption and in the import penetration ratio. The simulated wage settlement effects, then, may be interpretedas the net effects of increasedforeign productmarketshareholdingconstantthe rateof growthof realdomestic shipments. Wage increases in all industriesare slowed by the increasedgrowth of imports.The leatherproducts,primarymetals, andapparelindustries have the largestpredictedeffects in the firstyear of the simulation.The largest long-term effects occurred in the tobacco products, lumber products,and primarymetals industries. The results using microeconomicdata contrast with the aggregate results mentioned earlier. With microeconomic data the effect of expected inflationis only 0.54 when the key settlementsvariableis omitted, rangingup to 0.61 if one thinks that expected inflationis measuredby addingits own coefficient and the coefficient on key settlements when they are entered together. By comparison, the elasticities from the macroeconomicresults in table 2 range from 0.74 to 0.87. The larger expected inflationcoefficientsin table 2 probablymean that these price terms are incorporatingcatch-up as well as expectational effects on nominalwageadjustments.Theeffects of unemploymentarealso smaller in the microeconomicdata where both level and rate-of-changeeffects are identified.Together, their coefficients imply that it is lagged unemploymentthataffectscurrentwage settlements.Separateproductmarket and labor marketeffects on wage settlementsare successfully isolated in the microeconomicdata. In both the microeconomicand macroeconomic data, the effects of internationaltradeappearto operatethrough imports but not exports and to moderate wage inflationin the 1980s. Aggregatedatashow thatwages are affectedby importprices. Disaggregated data show wages affectedby the importpenetrationratio. Union Wage Settlements Duringthe 1980swage growth has slowed more in union bargaining situations than elsewhere in the private sector. Possible explanations include changes in unemploymentdemographics,internationaltrade, andothercompetitivefactors. In tradedgoods industriesin which union workers are paid more per hour than nonunion workers, increased Wayne Vroman and John M. Abowd 331 Table 6. Average Annual Rates of Wage Growth, Selected Periods, 1964-87a Percent per year Hourly earningsindex Period High-union industriesc 1964-69 1970-75 1976-82 1983-87 5.1 7.6 8.1 2.5 Low-union industriesd 5.7 6.7 7.7 3.0 Employmentcost indexb Union workers n.a. n.a. 8.6 2.8 Major Nonunion workers union agreements n.a. n.a. 7.4 4.0 n.a. 8.3 8.5 3.3 Source: Based on data from Currett Wage Developmetnts, various issues. n.a. Not available. a. December to December rates of growth. b. Indexes for wages and salaries. c. Weighted average of hourly earnings indexes for mining, construction, manufacturing, and transportation and utilities based on each industry's percentage of hours in 1967. d. Weighted average of hourly earnings indexes for wholesale, retail, finance, and services. importcompetitioncould have adverseeffects on both employmentand wages for union workers. This section examines overall union wages and employment with attention to developments in the 1980s. Before discussingthe 1980s,however, we brieflyreview earlierunion-nonunion wage trends. Because comprehensivetime series data distinguishingunion from nonunion wage growth became available only with the advent of the employment cost index in 1975, inferences for earlier periods must compare hourly earnings growth in industries with high unionization rates with that in industrieswith low unionizationrates. In an examination of the growth in average hourly earningsamong private nonfarm three-digitindustriesbetween 1953and 1976, Dan Mitchellfound that hourly earningsin highly unionized industriesgrew more rapidlythan those in low-unionizationindustriesbetween 1953and 1958and again between 1971 and 1976.8 During 1971-76 the wage growth differential favored high-unionindustriesby about 1 percentagepoint a year. By contrast, between 1964and 1968,hourlyearningsgrew more rapidlyin industrieswith low unionizationrates. Table6, whichshows averageannualratesof wagegrowthfor selected periods for three wage series, confirmsMitchell's findings.The hourly earningsindex data show that wages grew more rapidlyin low-union industriesin the late 1960sand againduring1983-87,while the converse held during 1970-75 and 1976-82. Over the entire 1964-87 period, 8. See DanielJ. B. Mitchell, Unions, Wages, and Inflation(Brookings,1980),chap. 2, pp. 23-61. CI 11 r- I- X ,0 Ib ir) bNt \I IbbXt cn r- CI VIt oc \I \I 6 06 'i 06 \I r- \I \I \IC r7 6 Z N 04~~~~~~~~~~~~\ \. ZX. ~~~~~~~~~~~~~~~~~~~~\C0 C-A tl \,C a- oow r a) oc oc C-Aa)a 00 X~~~~~~~~C 00 CIA \, C t r- , LW C) CIA )\, -0 O 00 r) a) - -i) o obbbo. ????<<<<<<t^Y , LW : Z t~~~~~~~~~~~~~~N ~ \, ~~~~ 66\6, ~ t 060 \, (01 (0, - 00 1- r(01 ol ol ol ol , It ~~40 ol It t n 1- r- r- r- 1 O ol ol ol , )1-I ,01 00 ?t 00 00 00 C: ,01 ,01 ,01 ,01 ,01 (0 V0 ?0 E N CN ?o Wayne Vroman and John M. Abowd 333 however, averageannualwage growthwas aboutthe same (5.7 percent) in low- and high-unionindustries.The explicit union-nonunioncomparison using employment cost index data shows a 1 percentage point differentialin favor of union workers during 1976-82 (8.4 percent, as against7.4 percent), but then a 1 percentagepoint lower rate of union wage growth during 1983-87 (3.3 percent, as against 4.3 percent). The wage deceleration since 1983 has been 2 percentage points larger for union workers than for nonunion workers (5.1 points, as against 3.1 points). Table 6 also shows the similarityof union wage increases in the employmentcost index and the effective wage changes in majorunion agreements(those affecting 1,000 or more workers). The simple correlation of the two annualwage change series from 1976to 1987is 0.990. Workerscovered by majoragreementsconstitute more than half of all union workers in the private nonfarmeconomy. Althoughthe ECI has wage changedatafor all unionworkers,the majorunionagreements data have some importantadvantagesfor present purposes. They go back to 1968for all privateindustriesand includedetail not availablein the ECI data:effective wage changes brokendown by source (currentyear negotiationsor first-yearincreases, deferredincreases from prior settlements,andcost-of-livingincreases),the numberof workersreceiving each type of wage change, and the coverage of COLAclauses. Table 7 summarizesinformationon employmentand effective wage changes in majoragreementssince 1968. One importantfeature of the datais the declinein the numberof workerscoveredby these agreements, as shown in column 1-from 10.8millionworkersin 1968,to 9.7 million in 1977, to only 6.3 million in 1987. Employment covered by major agreementsfell from 19.3percentto 7.4 percentof privatenonfarmwage and salary employment during these 20 years. The decline in the employmentshareparallelsthe generaldecline in privatesector unionization. For 1973-81 and 1983-87, the years when comparisons with CurrentPopulationSurvey (CPS)-based unionizationestimates can be made, the two series move almost identically. The major settlements employment shares of column 2 averaged 62.8 percent of the CPS unionizationrate estimates, and theirtime series correlationwas 0.973. About 85 percent of workers covered by majoragreementswork in the highly unionized mining, construction, manufacturing,and transportationandutilitiesindustries,industrieswhose shareof privatewage and salaryemploymentshrankfrom50.0 percentin 1968to 35.6 percent 334 Brookings Papers on Economic Activity, 1:1988 in 1987.The changingmix of employmentby industrythus also affects the majoragreementsemploymentshare in column 2. Column3, however, shows thatthe shareof employmentcovered by majoragreements in the highly unionized industries declined from 34.3 percent to 17.0 percentbetween 1968and 1987.Because majoragreementsemployment has declined relativeto total employmentin all privateindustries,most of the decline in the shareof overallemploymentshown in column2 has been due to developments within industriesand not to changes in the industrymix of employment.The employmentpercentagein column 2 for 1987 would have been 9.5 percent, rather than 7.4 percent, if the highlyunionizedindustrieshad retainedtheir 1968shareof total private nonfarmemployment. The first three columns of table 7 suggest that the decline in major agreementsemploymentwas morerapidin the late 1970sand 1980sthan in earlier years. The explanationfor the decrease in the employment base for majoragreements,and for privatesector union employmentin general,includesregionalemploymentrelocationtowardless unionized states in the Sun Belt; changes in the industrialrelations practices of employers, particularlylargeemployers, towardunions;reducedunion organizingactivity and lower success rate in representationelections; changesin workerattitudestowardunions;deregulationin the transportationandcommunicationindustries;andincreasedimportcompetition.9 The increase in the union wage premium implied by the union and nonunionwage growthtrendsduring1970-82probablygave employers increasedfinancialincentives to use nonunionworkers. Table 7 shows that even after the economy emergedfrom the 1981-82 recession, the employmentbase for the majoragreementscontinuedto decline, from 7.9 millionworkersin 1983to 6.3 millionin 1987. To explain the decline in employmentcovered by majoragreements we fittime seriesandcross-sectionregressions.Bothregressionanalyses include controls for the effects of internationaltrade. Time series variationin the major agreementsemployment percentage within the highly unionized industries(P), column 3 of table 7, is explained with 9. For analysesof the decline in privatesector unionization,see WilliamT. Dickens and JonathanS. Leonard, "Accountingfor the Decline in Union Membership,19501980,"Industrial andLaborRelations Review, vol. 38 (April1985),pp. 323-34;andHenry S. Farber,"TheDeclineof Unionizationin the UnitedStates:WhatCanBe Learnedfrom RecentExperience?"WorkingPaper2267(NBER, May 1987). WayneVromanand John M. Abowd 335 regressionsthat also include controls for the business cycle (URM, the unemploymient ratefor menaged25-54), the regionalmixof employment in these industries(EMIX, employmentin the Northeast and Midwest as a percentageof nationalemployment),anda controlforthe accelerated downward trend that started in 1976 (T76). The internationaltrade variable in the time series analysis is real imports measured as a percentage of real GDP(RMS). The regression shown below, with t-statistics in parentheses, explains 99.1 percent of variation in the employmentpercentageduring1968-87. P = 6.41 + 0.564 URM + 0.598 EMIX - 0.540 776 - 0.991 RMS (0.7) (3.8) R2 = (4.7) (6.2) (4.9) 0.991;standarderror = 0.51; Durbin-Watson= 1.45. All four explanatoryvariablesare statistically significant.In recessions, majoragreementsemploymentdeclines more slowly than overall employmentin the highlyunionizedindustries,perhapsbecauseworkers are still covered by the agreementseven if they are not all employed. The downwardtrend in the majorsettlements employmentpercentage is explained by regionalemploymentmix, the trend accelerationafter 1976,andgrowthin the real importshare. Because the real importshare grew from 7.9 percent of real GDP in 1968to 14.8 percent in 1987, the equation implies that growth in internationaltrade accounted for 41 percent of the total decline in the majoragreementsemploymentpercentage. The cross-section analysis examines the change in the majoragreements employmentsharebetween 1978and 1987at the level of two-digit SIC industries. For 36 private nonfarm industries in which major agreementscovered at least 5 percent of industryemploymentin 1978, themajoragreementsemploymentsharewas lowerin allbutone industry (SIC 24, lumber)in 1987.To explainthe decline, we fit regressionsthat include the level or the change in import-relatedemployment losses between 1977 and 1984.10Also included are controls for industry employmentgrowthand a dummyvariablefor three industrieswith major 10. Kan Young, Ann Lawson, and JenniferDuncan, "Trade Ripples Across U.S. Industries"(U.S. Departmentof Commerce,January1986). 336 Brookings Papers on Economic Activity, 1:1988 regulatorychanges (trucking,airlines, and communications).Employment growth and deregulationdo not add significantlyto explained variationin this analysis. The simpleregressionsthat use, respectively, the average level and the change in import-relatedemploymentlosses are as follows: DP = -33.86 (88.3) K2 DP 0.30 A VIMP (2.0) = 0.081;standarderror = 18.31, = -36.42 (10. I) K2 - - 0.324 DIMP (1 .7) = 0.048;standarderror = 18.65. In the regressionsthe variablesare measuredas follows: DP = the percentage change between 1978 and 1987 in major agreements employment as a percent of total industry employment MEL77 = employmentlosses due to importsin 1977measuredas a percentof domestic employmentin the industryin 1977 MEL84 = employmentlosses due to importsin 1984measuredas a percentof domestic employmentin the industryin 1984 AVIMP = average employment loss due to imports between 1977 and 1984 = (MEL77 + MEL84)/2 DIMP = difference in employment loss due to imports between 1977 and 1984 = MEL84 - MEL77. The cross-sectionregressionsallowfor changesin importpenetration by industryin ways not permittedin a more aggregateanalysis. Both import variables have negative coefficients as expected, but note the size and significance of the intercepts. The regressions suggest that import penetration accounted for 10-20 percent of the employment losses between 1978and 1987experiencedby workerscovered by major agreements. Taken together the time series and cross-section results implythat internationaltradehas contributedto the employmentlosses in majoragreements,but that other factors combinedhave had quantitatively largereffects. The steady decline in membershipand in the size of bargainingunits has undoubtedlyinfluencedunion objectives in collective bargaining. Wayne Vrornan and John M. Abowd 337 Greaterjobsecurityhas become such an importantobjectivein the 1980s thatunionshave been willingto tradeaway wage increasesto get it. The 1980shave seen the spreadof so-calledconcession bargainswhose wage provisions include freezes and reductions, an increased prevalence of lump sum payments, and two-tier pay arrangements.Coincidentwith these changesin contractualwageprovisionshave been sharpreductions in work stoppages. Table7 displays selected indicatorsof increasedunion willingnessto forgo wage increases in recent years. All three componentsof effective wage increases(columns5, 6, and7) declined sharplybetween 1981and 1983and remainedlow through1987.From 1983through1987the total annualeffective wage change was below 4 percent, and first-yearand COLA increases remainedbelow 1 percent a year. Columns 8 and 9 show thatunionwillingnessto accept wage freezes andwage reductions increased sharplyduring1983-87, contributingto the smallerobserved first-yearwage increases. In any one year in this five-yearperiod about one-fifthof workers received no wage increase. In the decade 1973to 1982,thatproportionreachedas highas 6 percentonly twice. Union bargainershave also been willing to give up cost-of-living escalators in recent years. Following nearly a decade (1976 to 1984) when COLAcoverage averagedabout60 percent, coverage declinedto 50 percent in 1985and then to roughly40 percent in 1986and 1987. Of workers with COLA coverage, the proportion who actually receive COLAs, the ratioof column 11to column 10, droppednoticeablyduring 1983-87,averaging0.76 between 1976and 1982but only 0.57 from 1983 through1987.The decline in the importanceof COLAs as a component of unionwage changes has been due to threefactors identifiablein table 7: the low inflationrate of 1983-87 (column 12), the decline in COLA coverage(column10),andthe decline in the proportionof those covered who actuallyreceive COLAs (column 11as a fractionof column 10). The full extent of the union wage deceleration and the component parts of the decelerationare clearly illustratedby the data in columns 4-7 of table 7. First-yearadjustmentsand COLA increases have both been especially smallduring1983-87. The explanation for the deceleration of union wage gains, both absolutelyand relativeto nonuniongains, could be internationaltrade. At least four considerations, however, suggest that other factors are importantas well. First, the accelerateddecline in the employmentbase for the majoragreements, and for union employmentmore generally, 338 Brookings Papers on Economic Activity, 1:1988 predates by at least seven years the emergence of the large U.S. merchandisetrade deficits in 1984. Second, domestic economic developments are capable of explaining why recent union wage increases have been unusuallysmall. In both the time series and the cross-section analysis of majorsettlementsemploymentshares, we findthat international trade is importantbut does not account for the majorityof the employmentlosses. The growthof nonunionemployment,deregulation in the transportationandcommunicationindustries,the developmentof dualunion-nonunionindustrialrelationsstrategiesamonglargeemployers, and worseningworkerattitudestowardunions have all contributed 11 The emergenceof a substantialmerchandise to hardtimes for unions. trade deficit has contributedto the decline in unionized employment, but so have all these other developments.Third,our analysis of microeconomic data from manufacturingdoes not suggest that international tradeis the maincausefor smallerunionsettlementsin the 1980s.Fourth, an analysis by Mitchellconcludes that internationaltradeand deregulation have not been the root cause for the large increase in concession bargainingsince 1982.12He finds that in industriesnot directly affected by trade or deregulation,union wage concessions have been at least as prevalentand as largeas elsewhere. The preceding suggests continuing moderation in union wage demands. The short-termprospect for union employment is for further reductions,both generallyand in the majorbargainingsituations. With a decliningemploymentbase, union negotiatorswill continue to focus on job securityand downplaywage demands. In summary,the slowdown in wage inflationof the 1980sis due not only to reducedinflationaryexpectations, but to the increase in relative unemploymentamongadultmen, the decline in unionwage settlements, and the competitive effects of importedgoods and services. Because these developmentsare likely to persist, they pointto continuedmodest wage growthfor the rest of 1988and into 1989. 11. The changingenvironmentof industrialrelations is documentedin Thomas A. Kochan, Harry C. Katz, and Robert B. McKersie, The Transformationof American IndustrialRelations (Basic Books, 1986). One analysis of changes in workerattitudes towardunionsis found in Farber,"The Decline in Unionization."His analysissuggests that nonunionworkersbecame less favorablyinclinedtowardunions between 1977and 1984. 12. See DanielJ. B. Mitchell,"WageTrendsandWageConcessions:Implicationsfor MediumTermEconomicExpansion,"WorkingPaper 114(Institutefor IndustrialRelations, Universityof Californiaat Los Angeles, July 1986). Comments and Discussion Gary Burtless: WayneVromanandJohnAbowd have writtena rather discursivepaperon three aspects of recent wage inflation:the changing demographicprofileof unemploymentandits effect on wage pressurein labormarkets, the impact of internationaltrade on wage bargains,and the declining role of unions in aggregate wage determination.In my comments, I shall focus on the regressionresults discussed in the first two-thirdsof the paper. The authorsbeginby analyzingaggregatewage inflation,as measured by the hourlyearningsindex. The econometricspecificationand results of this analysisare shown in table 2. The Phillipscurves estimatedin the table are based on an extremely parsimoniousspecification.The equations include only the level of the unemploymentrate and the lagged value of price inflation.The authorsargue-and I certainlyagree-that the unemploymentrate of men between the ages of 25 and 54 is a more reliableindicatorof labor markettightness than the overall unemployment rate. The results in table 2 suggest that a 1 percentagepoint rise in male unemploymentis associated with a dropin nominalwage inflation of 0.8-0.9 percent. The specification of the effects of price inflation will raise a few eyebrows. In the simplest equation, Vroman and Abowd include the distributedlagged rate of change in the consumerprice index over the previous three years. The coefficientson this variableimplythat within three years of a jump in prices, hourly earningsrise enough to offset about 80 percent of the price hike. When the authorswant to juice up thisparsimoniousspecification,they addthe four-quarterdistributedlag of import price inflation (excluding petroleum). Under one interpretationof this procedure,the authorsare tryingto capture the impact of greater or lesser price competition in product markets on the pace of domestic wage increases. If importprices fall 339 340 Brookings Papers on Economic Activity, 1:1988 relative to domestic prices, wages will be held down in the domestic sectors that compete with importedgoods. The estimated coefficient, which significantlyimprovesthe fit of the equation,especially in recent years, appearsto confirmthis hypothesis. Onemightthinkthatthis kindof product-marketpressurewouldgrow as the share of importedgoods in the U.S. market rises. In the mid1960s, U.S. importswere less than 5 percent of GNP; today, they are around11-12 percent. So one would anticipatethat the effect of import price movements on U.S. wages would be greater today than in the 1960s. Perhapsthis is the reason that the regressionsin table 2 consistently overpredictaggregatewage inflationin recent years, even when they include the termmeasuringimportprice inflation.Note, however, that the errorin predictionis consistently smallerin the equationsthat includeimportprice changes thanin those that exclude this term. Although the effect of trade on the overall U.S. economy may be growing, the effect on average wage bargainsmay be declining. My impressionis that goods-producingindustriesthat are heavily involved in trade account for a decliningfractionof U.S. employment.So while the effect of tradeon these sectors may be rising, the importanceof the wage settlementsstruckin these sectors is probablyshrinking. It would be reasonableto expect thatimportprice movementsshould have the greatestinfluencein sectorsthatdirectlycompetewithimported goods-namely, the goods-producingindustries, excluding construction. But such was not the patterndiscoveredby the authorsin sectoral regressionsthat are mentioned, thoughnot presented, in this paper. In those sectoral regressions the authors found that wage increases in mining and durable and nondurablemanufacturingare significantly affected by importprices. That is, wages in those sectors are restrained if overall import prices decline or grow slowly. But the authors also found that wages in transportationand in retailand wholesale tradeare just as stronglyinfluencedby importprice movements as wages in the goods industries. Why should this be the case when workers in these service industriesdo not directlycompete with importedservices? Partof the problemis that relevantimportprice pressuresare poorly measured by the regressor included in table 2 and in the sectoral regressionsmentionedinthe text. Inprinciple,the authorsshouldinclude a variablethatcapturesimportpricesof goods andservices thatcompete directlywith those of the affected industry.Overall importprice movements mightdo a badjob. Wayne Vroman and John M. Abowd 341 Vromanand Abowd addressthis issue in the most novel partof their paper-the section dealing with collective bargainingsettlements in manufacturing.Here their data set includes detailed informationabout tradepressurewithinspecificindustrialproductlines. Beforedeveloping this point, though, I should brieflydescribe the data set, which, as just mentioned,pertainsto union settlementsin manufacturing. It is unfortunatethatmanufacturingemploymentis a decliningportion of all U.S. employment.And it is still more unfortunatethat collective bargainingcovers a steadily dwindlingshare of workers, even in manufacturing.These things are not necessarily unfortunatein themselves, eitherfor the economy or for workers-although there are still a few of us who thinkso. They are unfortunatefromthe point of view of analysts who take a great interest in collective bargainingarrangementswithin manufacturing.However intriguingthese arrangementsare, their relevance to generalwage patternsis receding. But I have come to praise this data set, not to bury it. The authors have collected in one place a truly astonishingfile of informationabout collective bargainingagreementsin manufacturing.The file covers a 26year span from 1959through 1984, and includes informationon more than2,700 individualagreements.The datahave been organizedin such a way that the analyst can derive the implications of an individual contractforwagechangesin each yearcoveredby the contract,including the effect, if any, of the COLA clause. Whatis more, the authorshave tabulateddetailed informationabout the specific manufacturingindustries representedin the file, includingsuch data as the level of industry shipments,imports,exports, domesticconsumption,andpricechanges. Consequently,the regressions reportedin table 4 are estimated with a huge cross-sectionaltime series data set, packed with detailedinformation that is generallybelieved relevantto wage settlements. Veteranmacroeconometriciansmightbe astonishedat the numberof observationsused to estimatethe equations. As a microeconometrician myself, I am less easily impressed. Articles in leading microeconomic journalsoften containas manyor moreobservations.Themanyavailable observations here present some problems of interpretationto macroeconomistsaccustomedto dealingwithcompactdatasets. Givenenough observations,one could enter "Numberof Light-yearsto the Dog Star" as a right-hand-sidevariable and get a "significant"-'coefficient. A coefficient that is statistically significantis not necessarily practically significant,in the sense that it has a meaningfullylarge effect on the 342 Brookings Papers on Economic Activity, 1:1988 dependentvariable. On the other hand, a very low t-statistic is informative. If a variable is reasonably well identified in the data-and I would expect all of Vromanand Abowd's are-an insignificantcoefficient does not mean "undecided," as it might in smaller data sets. It means "This variableis unimportant."Even at the outer boundsof the 95 percentconfidenceinterval,the variablehas no meaningfuleffect on the dependentvariable. The specificationused by Vromanand Abowd in table 4 includes the prime-agemale unemploymentrate, anticipatedinflationat the time of the contractsettlement(measuredtwo differentways), the "inflationary surprise"over the life of the contract(relevantmainlyfor wage bargains containinga COLA clause), and several measures of demand for the industry'sproductline. Thislast set of variablesis aclever decomposition of demand changes into product price changes, changes in domestic shipments,changes in exported shipments,and changes in the share of imports in domestic consumption. The last variable-changes in the importpenetrationratio-is clearly included to capture the effects of importpressureson wage settlements. What should we conclude from the results in table 4? First, the male unemploymentratehas a remarkablysmall, althoughstatisticallysignificant, effect on wage settlements. If I read the results correctly, the effect is only a smallfractionof thatimpliedin the aggregateregressions reportedin table 2. Second, wage earners obtain a substantiallysmaller share of price increases than is suggested in table 2, where the aggregateequations showed thatwages eventuallyrose by 75-80 percentof the changein the CPI. The results in table 4 suggest that on average workers might get only one-thirdto two-thirdsof the price rise, with workerscovered by COLA clauses getting the most and workers not covered by COLAs gettingthe least. Third,wages appearto be relativelyinsensitiveto demandconditions in productmarkets.Althoughthe coefficients are often significant,as I just noted, small coefficients can be statistically significantin large samples without any practical significancefor the dependentvariable. Even underthe highestestimateof wage rate sensitivity, for example, a 10 percentagepoint rise in industryproduct prices leads to only a 0.2 percenthike in negotiatedwages. The notable exception to this generalizationappearsto be imported Wayne Vroman and John M. Abowd 343 goods competition, althougheven here the simulationresults reported in table 5 suggest that the cumulativeeffect of importsis small. Import penetrationhas a significant,butnot especiallylarge,effect in restraining wages. Fourth,in a nod towardinstitutionaltheories of wage bargaining,the authors include a variable reflectingthe wage gain in the most recent auto or steel settlement. This variableis always found to have a large and reliablyestimatedeffect on new wage bargains.Withoutquibbling with the institutionalknowledgelying behindthis specificationchoice, I thinkthe inclusionof the steel and auto settlementvariableis probably a bad idea. Presumably,the most recent steel and auto wage increase already reflects the influenceof some or all of the factors in the specificationthe currentunemploymentrate, expected inflation,general inflationin productprices, and so on. Hence, part of the effects of these variables is being capturedby the coefficient on the auto and steel settlement, leaving the readerwith no idea of the net equilibriumimpactof, say, a changein unemploymenton new wage settlements. Of course, even with the currentspecificationthese effects could be derived using appropriatesimulations.But the authorsdo not provide these, so the readeris left in the dark. Even thoughI have emphasizedthe differencesbetween the Phillips curves impliedby the manufacturingregressions, on the one hand, and the aggregateequations, on the other, I think it importantto note an interestingsimilarity.Nearly all of the recentresidualsin the regressions have been negative. The aggregateas well as the microeconomicequations tend to overpredictnominalwage gainsin the pastfew years. Many of the errorsare quite large,particularlyin view of the slow pace of real wage movements-and even nominalwage movements-in the 1980s. Finally, we shouldtake note of the absence of any productivityterm in the equationsshown on table 4. While this is customaryin aggregate Phillips curve equations, it may not be as reasonablewith this kind of data set. If there is a sectoral slowdown in productivity growth, and if the equationsreportedin table4 remainvalidpredictorsof wagemovements, the combinedeffect would be higherunemploymentor higherinflation, or both. Unless workers obtain a rising share of value added, either overall inflationor joblessness must rise to slow down the rate of real 344 Brookings Papers on Economic Activity, 1:1988 wage increase. The estimated coefficient on the unemploymentrate is quitelow, so it would presumablytake a large-and permanent-rise in unemploymentto effect the requiredwage slowdown. An increase in general price inflation would accomplish the same result, because nominal wages apparentlyrise only a fraction of a point for each 1 percentagepoint rise in inflation. Perhaps the specification in table 4 is generally correct. Neither workers nor employers pay any attention to productivitytrends when strikingwage bargains.But it is at least possible that some employers take account of worker productivity when making wage offers or acceptinga union's counteroffer.This propositioncould be examinedin the authors' data set by includingsectoral productivitymovements in the specification.Even if the industry-levelproductivitydataare bad, it may be easier to test the idea here thanin a normalmacroeconomicdata set. I do think this data set offers some intriguingpossibilities. The authors should be congratulatedfor assemblingit and showing how it can be exploitedin a macroeconomiccontext. General Discussion Several panelists suggested modificationsin the authors' aggregate wage equations.RobertGordonagreedwith Burtlessthatthe equations are overly parsimonious,and he suggestedincludingseveral lags of the unemploymentrate. He noted thatin his own paperin this volume, rateof-changeeffects are important;withoutlags such effects are precluded and the coefficient on the unemploymentrate may be biased. Gordon also suggestedallowinglaggedwages to compete with laggedprices. He arguedthatthe priceeffects were likely to have variedduringthe period, as the use of indexed wage contracts increased markedlyduring the 1960s and decreased in the 1980s; ideally a variable reflecting this differencein coverage would be used in the equations. EdmundPhelps noted that the price variable may be serving as a proxy for expected wage change. He suggested adding productivity growth to the wage equationsas a variableindicatingboththe capacityof firmsto pay higher wages andworkers'realwage aspirations.Such a variable,he reasoned, mightaccountfor the slowdownof wage growthin the 1980s. Gordon observed that wages are only half the story. The ultimate Wayne Vroman and John M. Abowd 345 object is to understandhow wages get translatedinto prices. The recent theoreticalemphasison price rigidity,for example in the paperby Ball, Mankiw,andRomerinthisvolume,has reopenedthe questionof whether wages are importantto prices in the shortrun-an issue, Gordonnoted, that is ignoredby Vromanand Abowd. He suggested that large swings in labor's shareargueagainsta simplemodel in whichprices are marked up over wages. The recent decline in labor's shareis the counterpartof the recentoverpredictionof the wage equations.Gordonalso noted that these movementsin labor's share mirrormovements in the ratio of the adult male unemploymentrate to the total unemployment rate. He speculatedthat when adult males are doing relativelywell, they bid up wages in general,resultingin an increasein labor'sshare.Vromanlinked the reductionin labor's shareto the decline in unionjobs in manufacturing. Robert Hall noted the existence of simultaneitybias in the microeconomic as well as in the macroeconomicwage equations. In the face of shocks to wages, the dependentvariablewould be expected to feed positively into contemporaneousprice changes, causingan upwardbias in that coefficient. Similarly,feedbackfrom wage change to unemploymentwouldbe expected to bias the coefficienton unemploymenttoward zero. Finally, bothof these effects wouldlead to an underestimateof the standarderrorof the equation.Because it neglects these feedbacks, Hall reasoned, the estimated response of wages to aggregatedemandfrom this type of equation is misleading, and the implicationthat inflation changesdo not originatewith wages cannot be accepted. Abowd acknowledgedthe possibility of bias for the macroeconomic variablesbut arguedthat the variationin many of the variablesin the paper'smicroeconomicequationsis primarilycross-sectional.He added that the estimatedequationswere properlyspecifiedwage models with seriallyuncorrelatederrors.He reportedthatthe resultsare similareven if annual dummy variables are included. For example, the effects of trade penetrationand contract-specificinflation surprises are consistentlyestimated.However, ChristopherSimsobservedthatbiases could exist even for these variables; for example, trade penetration in a particularindustrycould resultfromwage shocks in that industry. Charles Schultze noted that the coefficient on expected inflationin the microeconomicequations should vary dependingon whetherthere is a COLA clause and on the type of COLA clause in the contract. He 346 Brookings Papers on Economic Activity, 1:1988 found it puzzling that the equations implied that contracts without COLAs had respondedvery differentlythan contracts with COLAs to the combinedeffects of pastandexpectedpricechanges.Vromanreplied thatattemptsto estimate separateexpected inflationeffects on different types of contractshad not indicateda significantdifference. Gordonconjecturedthat the price coefficient in the aggregatewage equationswas less thanunity-the value some theorieswouldpredictbecause the consumerprice index substantiallyoverestimatedhousing costs in 1980-81.Vromanrepliedthatalthoughthe coefficientsestimated using other price indexes were sometimes closer to unity, the recent overpredictionsof the wage equationsremained.