Public interest and concern has been stirred by recent articles thatpresage a decline of middle income earners. Those who support this viewcontend that such earners are declining as a proportion of the U.S. workforce because more of the new jobs are at the top and bottom of theearning structure. They warn that this trend could lead to politicaland social unrest stemming from a two-tiered society, fewer advancement opportunities for those on the lower range of the earnings ladder, andeven economic disaster as the great purchasing power engine of themiddle class loses steam.
Discussions of the declining proportion of middle income earnerscan focus on changes in the distribution of earnings of individuals orchanges in the distribution of earnings of families. Changes in thedistribution of earnings of individuals may be caused by changes in theoccupational structure of the economy that reflect changes in industrialstructure and technology. In addition, changes in the distribution ofearnings within each occupation and changes in relative earnings amongoccupations can affect the distribution of earnings of individuals.Changes in the distribution of earnings of families are affected notonly by these same factors but also by changes in family structure. Forexample, increasing numbers of dual earning families can lead to anincrease in the proportion of families with high earnings and increasingnumbers of single person families can lead to an increase in theproportion of families with low income. This article focuses primarily on how changes in occupationalstructure affect the distribution of earnings of individuals. It alsoconsiders the contribution of changes to the distribution of earnings ofindividuals caused by changes in the distribution of earnings byoccuption over the 1973-82 period. Essential points in discussion Proponents of the declining middle thesis suggest that a variety offactors are causing a decline in the proportion of our work force in themiddle income levels.
These factors can be categorized as affectingeither the occupational structure of employment or relative wages amongoccupations. The more significant of these concern the occupationalstructure of employment: (1) the decline of employment in the so-called smokestack industries that have a large number of production workerswho, according to most proponents, exemplify workers in the middle ofthe earnings spectrum; (2) the rapid growth of high tech industries thatsome argue have a bipolar occupational structure; (3) the large numberof job openings and large numerical growth in low paying occupationsindicated by the BLS industry and occupational projections; and (4) theshifting industrial structure of the United States from goods-producingindustries that, according to the arguments, have a large proportion ofmiddle income workers to service-producing industries that areconsidered to have many high and low income earners with relatively fewin the middle. The economic structure of the United States, however, is verycomplex and many factors, in addition to those cited above, affect theearnings distribution of American workers. Not all of these factorswill cause bipolarization of earnings. Some will decrease the number oflow income workers and increase middle income workers and work againstbipolarization. Actual changes in the earnings distribution of Americanworkers are determined by the combined effect of many factors.
The past Data from the Current Population Survey (CPS) on usual weeklyearnings and on employment of full-time wage and salary workers bydetailed occupation for 1973 and 1982 were used to examine the merits ofthe declining middle income earner thesis. The first analysisidentifies the effect of changes in occupational structure on thedistribution of employment of full-time workers in three income groups:low, middle, and high. The second analysis illustrates the combinedeffect of changes in occupational structure and changes in relativeearnings among occupations on the earnings distribution of full-timeworkers over the 1973-82 period. A third analysis is identical to thefirst, but includes part-Time as well as full-time workers. The 1982 CPS provided data on usual weekly earnings of full-timewage and salary workers for 416 detailed occupations.
To test theeffect of changes in occupational structure on the distribution ofworkers into low, middle, and high earnings groups between 1973 and1982, I (1) arrayed the 416 occupations in 1982 by earnings and arrangedthem into thirds (bottom, middle, or top), with each third containingthe same number of occupations; (2) summed the number of workers in theoccupations in each third and calculated a percent distribution of theemployment; and (3) arrayed employment in 1973 for each occupation inthe same order as in 1982, and calculated the 1973 percent distributionfor each third. Consequently, an occupation was in the same third in1973 as it was in 1982. If the middle income earners are declining, the proportion of totalemployment in the middle third would show a decline between 1973 and1982, and the bottom and top thirds, an increase. The followingtabulation shows the distribution of employment in 1973 and 1982 byusual median weekly earnings in 1982: The top third increased the bottom decreased, and the middledecreased modestly. From this analysis, we can conclude that changes inoccupational structure alone from 1973 to 1982, whether caused bytechnological change, the shift from goods- to service-producingindustries, or other factors, do not support the notion ofbipolarization. As indicated, changes in wages levels also effect the earningsdistribution of workers.
To illustrate the combined effect of changesin relative wages and in occupattional structure on the earningsdistribution of workers over the 1973-82 period, I (1) rankedoccupations in the 1973 CPS into thirds based on 1973 earnings; (2)summed employment in each of the thirds and calculated a percentdistribution of employment; and (3) compared the resulting distributionwith the 1982 distribution of employment in each of the three earningsgroups. The following tabulation shows the distribution of employmentby usual median weekly earnings in 1973 and 1982: The data show that the proportion of total employment increased inthe top and middle thirds and decreased in the bottom third. Thiscalculation does not show a trend toward bipolarization, but insteadindicates a shift of workers from the low to the middle and highearnings levels, with the middle having the largest increase. Thus,according to this tabulation, changes in occupational structure, whencombined with changes in relative wages and other factors, moved workersup the earnings distribution over the 1973-82 period. However, bipolarization can occur without significant shifts ofemployment to the top and bottom thirds of the earnings distribution ifthe earnings of those at the top were to increase significantly fasterthan those at the bottom. For example, if the earnings distribution ofthe bottom third remained at the 1973 level in 1982, but the top thirdincreased, it could be said that bipolarization occurred even thoughthere were no significant shifts in employment.
However, the data do notindicate that this occurred. As shown in the following tabulation, theaverage of the median earnings for the detailed occupations weighted byemployment increased in each third by about the same amount from 1973 to1982, although the increase was slightly larger in the top third andslightly lower in the bottom third than in the middle: Part-time workers. Including part-time workers in an analysis ofhow changes in occupational structure have affected the earningsdistribution of workers is very complex. Part-time workers may workfrom 1 to 34 hours per week and, therefore, weekly earnings are probablyaffected more by the number of hours worked than by wage rates. Inaddition, most part-time workers (about two-thirds in 1982) are onpart-time schedules by choice. Some are students who work only a fewhours a week for spending money, some are older workers drawingretirement income who work part-time at least in part to providediversity, and some are members of a household having a wage earner witha high income. Thus, the earnings of many part-time workers have littlesignificance to issues related to concerns about the declining middle,such as lack of advancement opportunities and social and politicalunrest. Some part-time workers, however, are on part-time schedules foreconomic reasons such as slack work rather than by choice.
The earningsof these workers would be higher if they were able to work full time,and their employment and earnings problems are therefore relevant to thedeclining middle issue. Over the 1973-82 period, the proportion ofworkers on part-time schedules for economic reasons increasedsignificantly, from 3.1 percent to 6.5 percent of total employment. Alarge part of this increase resulted from the recessionary conditionsprevalent in 1982, but not in 1973. Still, some structural changes inthe economy may also have occurred between 1973 and 1982 which affectednot only the distribution of occupational employment of part-timeworkers but also the level of part-time employment. In turn, thesechanges could have affected the proportion of workers in the middleincome group. Because of the complexities of dealing with part-time workers in ananalysis of the decline of middle income earners, only the effect ofpart-time workers on changes in occupational distribution from 1973 to1982 is considered in this article.
Issues concerning such factors aschanges in hours worked and in the proportions of those who workedpart-time voluntarily or for economic reasons are not considered. Therefore, part-time workers were combined with full-time workers.Total employment (combined part- and full-time employment) for 1973 and1982 was distributed into the top, middle, and bottom thirds of theoccupational earnings structure, based on median usual weekly earningsin 1982. Part-time workers were placed in the same third of theoccupational distribution by earnings as full-time workers in the sameoccupation. Also, they were given an employment weight equal to afull-time worker.
Part-time workers are heavily concentrated in occupations in thebottom third of the earnings structure. Therefore, the inclusion ofpart-time workers resulted in a larger proportion of workers in thebottom third than when only full-time workers were included. Thefollowing tabulation shows the distribution of total employment in 1973and 1982 by usual weekly earnings in 1982 (part-time workers weredistributed according to the 1982 usual weekly earnings of full-timewage and salary workers in the same occupation): The data show that changes in the distribution of total employmentamong the top, middle, and bottom thirds of the earnings distributionbetween 1973 and 1982 were very similar to the changes that were shownwhen only full-time workers were considered. The top third increased,the bottom third declined, and the middle third declined very slightly(but not as much as the bottom third). These results also do not support the notion of bipolarization.Most importantly, none of the three analyses shows an increase in thebottom third, which is an important part of the bipolarizationhypothesis. In fact, they all show a decline in the share of employmentin the lowest group. Data limitations.
The data used in the three analyses have somelimitations that should be recognized. These limitations result fromsampling and response errors in the CPS as well as from differences indata definitions. The data for 1973 include workers who reported theywere self-employed but who had not incorporated their business. Theseindividuals are not included in the 1982 data.
However, the number ofthese workers in relatively small and should not significantly affectthe data. Also, the 1973 data reflect only one month, May, whereas the1982 data are annual averages. The future Data on changes in occupational structure and occupational wagelevels for the 1973-82 period do not support the declining middle incomeearners thesis. But what about the future? The basic tenets of thethesis could perhaps be more applicable to the future than to the recentperiod of back-to-back recessions. It is very difficult to forecast the future in terms ofoccupational structure and associated earnings by occupation, but someinsights can be gained by looking at the BLS 1982-95 occupationalprojections. The projections are based on the occupational classification systemused in the Occupational Employment Statistics (OES) survey, rather thanon the classification system used in the CPS.
Because earnings data arenot collected in the OES survey, a similar analysis could not beconducted for detailed occupations as was done for the 1973-82 period.However, CPS and OES data are similar enough to permit analysis ofdevelopments for the standard major occupational groups of workers. Thedata indicate the following: * Workers who typically have a high level of earnings–professionaland technical workers and managers–are projected to increase as aproportion of total employment. * Craftworkers, who also have higher than average earnings, butwith slightly more workers in the middle third than in the top third,also are projected to increase as a proportion of all workers over the1982-95 period. (See table 1.) * Among those occupational groups with low earnings, laborers andfarmworkers are projected to decline as a proportion of the totalemployment, and service workers and clerical workers are expected toincrease their shares. However, if the four occupational groups withlower than average earnings (operatives, laborers, service workers, andfarmworkers) are combined, they are projected to decline as a proportionof total employment.
The projected data are generally consistent with the findings forthe 1973-82 period. Namely, they show an increasing proportion ofemployment in higher than average earnings occupations and a decliningproportion in occupations with lower than average earnings, rather thana trend toward bipolarization. Specific issues As noted, the declining middle income earners thesis is based on anumber of widely discussed developments, including the decline ofsmokestack industries, the rapid growth of high tech industries, thelarge number of openings in low paying occupations, and the shift fromgoods- to service-producing industries. However, the extent to whicheach of these factors has contributed or can be expected to contributeto the decline of middle income earners is open to debate.
Thefollowing discusses these four factors in terms of their significance tothis phenomenon. Decline of smokestack industries. Proponents of the decliningmiddle income thesis argue that the long-term employment decline of someof the major so-called smokestack industries–automobile manufacturing,blast furnaces and basic steel products, and iron and steelfoundries–is a major cause of bipolarization. These industries dodemonstrate declining trends in employment. Employment peaked in themid-1960’s in the blast furnaces and basic steel products industry,and in the mid-1970’s in iron and steel foundries. Automobile manufacturing employment peaked in 1978 at about 1 million workers, andmost industry analysts do not expect employment to rebound to that levelin the foreseeable future. (Employment trends in these and otherindustries are shown in table 2.
) These smokestack industries pay relatively high wages. Averagehourly earnings of production workers in each of the three industriesare well above the average for production or nonsupervisory workers inall private nonagricultural establishments. (See table 2.) Theseindustries also have a higher than average proportion of productionworkers. Thus, if it is assumed that production workers in theseindustries exemplify middle income earners, and that those displaced from these industries end up on low wage jobs or become unemployed, thedecline of employment in these three industries would tend to causeincome polarization. However, the effect of the employment decline in smokestackindustries on the overall economy is not significant. Since 1973 (thehigh point of combined employment in automobile manufacturing, blastfurnaces and basic steel products, and iron and steel foundries), therehas been a notable decline in the number of workers in these industries.
But, if the decline had not taken place, total employment in 1983 wouldhave been only .5 percent higher. Even if all of these workers were inthe middle third of the earnings structure, the overall distribution ofworkers by earnings would not be significantly different than it was in1983 because they would be such a small part of the total. We can conclude that the decline of smokestack industries is afactor that could cause bipolarization. However, we cannot concludethat international competition and technological change, factors thatare largely responsible for the declining employment in the smokestackindustries, cause bipolarization without looking at other industrieswhich also face the same problems and which also have experiencedemployment declines over the past decade–textile, apparel, and leatherproducts manufacturing. (See table 2.) Because these latter industriespay relatively low wages, the decline in the number of workers in thebottom of the earnings scale that resulted from their employmentdeclines (600,000 from 1973 to 1983) more than offset the decline in thehigher paying smokestack industries.
Growth of high tech industries. An additional argument advanced byproponents of the declining middle income earners thesis indicates thatthe rapid growth of high tech industries contributes to bipolarizationbecause these industries are characterized by large proportions of highand low paid workers and few in the middle. If this argument has merit,these industries would have relatively high proportions of highly paidprofessional and managerial workers, and of low paid clerical andservice workers; production workers would have to be relatively low paidunless there were very few of them in these industries. In previous studies, the BLS has shown that high tech employment,under each of three groups of high technology industries, is growingfaster than total employment. However, the analysis also showed thathigh tech industries comprise a relatively small proportion of totalemployment and total employment growth.
BLS defines the three groups ofhigh tech industries as: group I–industries with a proportion oftechnology-oriented workers (engineers, life and physical scientists,mathematical specialists, engineering and science technicians, andcomputer specialists) at least 1.5 times the average for all industries;group II–industries with a ratio of R&D expenditures to net sales at least twice the average for all industries; and groupIII–manufacturing industries with a proportion of technology-orientedworkers equal to or greater than the average for all manufacturingindustries, and a ratio of R&D expenditures to sales close to orabove the average for all industries (two non-manufacturing industrieswhich provide technical support also are included). The followingtabulation shows the percent of total employment in each of the threegroups of high tech industries in 1972, 1982, and 1995, and the percentchange for 1972-82 and 1982-92: In 1982, under the broadest definition (group I), high techindustries only accounted for 13.4 percent of total employment, up from131. percent in 1972. Under a more narrow definition (group III), hightech comprised only 6.
2percent of total employment. An even narrower definition (group II),shows high tech employment accounting for only 2.8 percent of the total.Group III is probably the definition that would be used by proponents ofthe declining middle income earners thesis because the broadestdefinition includes, among other industries, automobile manufacturing. In about half of the high tech industries included in the group IIIdefinition, professional and managerial workers combined accounted for ahigher proportion of total employment than in the economy as a whole,and very few were significantly below the average. Nearly all of thehigh tech industries have a higher proportion of highly paid workersthan manufacturing as a whole.
However, the proportion of workers inthese industries, are low paid. But nearly all of the productionworkers in these industries have average hourly earnings above averagefor production workers in all manufacturing and production ornonsupervisory workers in all private nonagricultural establishments.(See talbe 3.) All these factors combined would tend to work againstpolarization when the entire economy is considered. Therefore, data onearnings and on employment growth provide little evidence that high techindustry growth is contributing to bipolarization. growth are Job openings in low paying occupations.
Another point made by someproponents of the declining middle income earners thesis is that amajority of the occupations having the largest number of job openingsand large projected employment growth are on the low end of the earningsspectrum. (See table 4.) This point is often made using the latest BLSprojections of occupational growth, 1982-95.
In these projections, manyof the occupations that are expected to have the largest numericalemployment growth over the 1982-95 period do have low earnings.However, these factors do not necessarily imply that low paying jobswill increase their share of employment and cause the proportion ofworkers earning low wages to rise. The BLS data on job openings indicate that most openings are causedby the need to replace workers rather than by growth in the number ofjobs. This is especially true in low paying occupations that employlarge numbers of young people and women, who may periodically leave thelabor force to attend school or to care for their families. In lowpaying jobs there also is significant movement between occupations.However, despite the large number of openings in these occupations,there is no indication that the number of workers having low earnings inincreasing because the rate of increase in employment in these jobs isgenerally not faster than that for the total economy.
Similarly, in analyzing the composition of employment by occupationimplied by projected growth, the growth rate must be considered inpreference to numerical change. A very large occupation with a growthrate close to that for all occupations will show large numerical growthbut will not increase as a proportion of total employment. For example,building custodians are projected to have the largest numerical growthbetween 1982-95, but with only an average projected rate of growth, thisoccupation is not expected to increase as a proportion of totalemployment.
Among the 20 occupations that are projected to grow fastest overthe 1982-95 period, most are in the top third earnings category and mostof the remainder are in the middle third. (See table 5.) However,looking only at the fastest growing occupations can be misleading. Acomprehensive analysis should include the entire occupational spectrum(which was done in an earlier section of this article). It is necessaryto use data for all occupations because, individually, the fastestgrowing occupations are numerically small and have little effect onchanging the overall distribution of workers by earnings level. Shift from goods- to service-producing industries.
Data on thechanging distribution of industry employment clearly show a shift fromgoods-producing to service-producing industries. To support theconclusion that this trend leads to bipolarization of earnings, the datawould have to show that the distribution of low and high earningsoccupations is concentrated to a greater extent in service-producingindustries than in goods-producing industries. An analysis of this nature was conducted by Thomas Stanback, Jr.
and Thierry J. Noyelle for 10 major occupational groups in 18 industrycategories. This analysis showed a tendency towards bipolarization thathas been used by many of the other proponents of the declining middleincome earners thesis as a basis for their conclusion. Stanback and Noyelle applied 1975 earnings data for majoroccupational groups to data on employment by major occupational group byindustry for 1975 and 1960. Using constant earnings data, they analyzed how changes in the occupational distribution alone would affect thedistribution of employment by earnings.
Their analysis was, therefore,similar to that presented in this article for the economy as a whole.However, the Stanback and Noyelle analysis was done at a majoroccupational group level, rather than by detailed occupation. Theiranalysis showed that employment of middle income earners declinedbetween 1960 and 1975, and that employment at both the top and bottom ofthe earnings scale increased. Their study also showed that growth ofservice-producing industries was largely responsible for this trend.Their analysis does lend considerable support to views that the middleis declining. However, there are some concerns about the validity of theanalysis.
Data on the occupational employment distribution ofindustries used by Stanback and Noyelle for 1960 were from theindustry-occupation matrix developed by BLS based on the occupationalclassification used in the 1960 census. Earnings data, however, weretaken from the Survey of Income and Education collected as a supplementto the CPS in 1975, which used the 1970 census classification. Althoughsimilar to the 1960 census classification, some occupations shifted fromone major group to another and could have affected the analysis. In addition, employment data in the industry-occupation matricesinclude part-time workers. Given that part-time workers are generallyfound in low paying occupations and that part-time workers increasedsignificantly as a proportion of the work force between 1960 and 1975,these data would tend to show an increase in low paid workers. Also,1975 was a recession year and thus had a larger proprotion of workers onpart-time schedules for economic reasons than 1960.
Finally, becausethe calculations were done by major occupational group, the analysiswould not have captured the changing structure among detailedoccupations within each major group. Thus, it is possible that somestructural changes are masked by the broad data used. Interestingly, a study by Peter Henle and Paul Ryscavage thatmeasured the tend toward inequality in earnings for a similar periodproduced results similar to Stanback and Noyelle. This study, based ondata from the CPS over the 1958–77 period, used a Gini index to measurethe equality of earnings distribution for a number of factors, includingoccupations. In general, the study showed greater inequality over time, but withconsiderable slowing of the long-term trend for the 1973-77 period. Forsome major occupational groups, however, there is a trend toward greaterequality over time or an uncertain trend. For those showing greaterinequality over time, there was less change later in the period.
The Stanback and Noyelle and Henle and Ryscavage studies both showcomparable results for a period beginning about the early 1960’s tothe mid-1970’s which suggest some bipolarization of earnings.However, my analysis of occupational trends for the 1973-82 period showsthat the tendency toward bipolarization, if it did exist, seems to havebeen reversed since the mid-1970’s. IS THE MIDDLE DECLINING? Some trends in the industrial andoccupational structure of employment could cause a degree of earningsbipolarization. However, a multitude of factors have an effect on theoccupational structure of our economy and on the earnings of workers inspecific occupations. Although not all can be quantified, an analysesof available data indicates that the combined effect of all factorsapparently has not caused bipolarization over the 1973-82 period.
Also,given BLS projections of employment by occupation, bipolarization is notlikely to occur between 1982 and 1995.