As in earlier downturns, the impacts of recession during 1981-82 werenot evenly distributed among the many demographic groups in the laborforce. For example, the rise in the unemployment rate was greatest, inrelative terms, for men. The decrease in labor force participation wasmost pronounced among teenagers, while the labor force participationrate for women actually increased during this period of general economicdecline. To what extent were these and other differential impacts of therecession the result of differences in the behavior of the labor forceparticipants? To what extent were they instead the result of differinglabor market opportunities? These, of course, are very difficultquestions to answer, particularly when dealing with aggregate data.
Toillustrate, a decrease in labor force participation can be the result oftwo factors–an increase in the rate at which individuals leave thelabor force, or a decrease in the rate at which workers enter the laborforce. Because these and other types of labor force transitions canhave different behavioral interpretations (that is, they may have”different kinds of sources”), it is important to identifywhich transitions generate demographic differences in labor forceparticipation and unemployment experience. To address these issues, Iexamine, by age, sex, and race, the monthly flows into and out of thelabor force and between employment and unemployment from January 1981 toJanuary 1984, well into the current recovery period. Distribution of economic impacts Race, sex, and age differences in the levels of unemployment andlabor force participation rates can be seen in table 1. The entries areaverages over the period December 1980 to December 1983 of data fromCurrent Population Survey “Gross Change Tabulations,” whichgive monthly estimates of the numbers of people employed, unemployed,and out of the labor force during the preeceding month.
The entries inthe table therefore are not based on or equivalent to the unemploymentand participation rates published by the Bureau of Labor Statistics. Inspection of the table indicates, however, that the well-known race, sex, and age differences found in the published estimates are alsofound here. Blacks and members of other races, on average, have higherunemployment rates than whites, and lower levels of labor forceparticipation, regardless of sex or age.
Women have slightly lowerunemployment rates than men (a relatively recent phenomenon), and lowerlabor force participation rates, regardless of age or race.Unemployment rates are seen to decrease with age for all sex/racegroups, while labor force participation rates increase and then decreasewith age, peaking in the 25- to 59-year-old “prime-age”category. Although the point estimates from the gross change data maydiffer from the published BLS estimates, the age, race, and sexrelationships seem to be the same. The focus of this study is not on differences in the levels ofunemployment and participation, however, but rather on differences intheir behavior over the most recent business cycle. The National Bureauof Economic Research has identified the peak of that cycle as July 1981and the trough as November 1982. The corresponding changes in theunemployment rates during the period for each demographic group arepresented in table 2, along with changes since the recovery began, forthe November 1982 to December 1983 period.
During the downturn, theunemployment rate increased more on average for men than for women, morefor whites than for blacks and others, and more for older (over 59)workers than for teenagers (age 16 to 19), youth (20 to 24), orprime-age workers (24 to 59). The greatest increases were felt amongolder women, who experienced growth in their unemployment rate of morethan 158 percent. The sex difference was reversed for nonwhite teens,youth, and older workers, with nonwhite women experiencing greaterrelative unemployment increases than nonwhite men. The racialdifference was reversed for teenagers. Of course, the lags in the impacts of an economic downturn can varyacross demographic groups, so that the “official” definitionof the timing of the downturn may not be the appropriate timeframe forthis type of analysis. For example, the unemployment rate for black mendid not peak until July 1983. To account for this, I computed thepercentage change in the unemployment rate for each demographic groupbetween the month the group’s rate was at its minimum and the monthit reached its maximum.
These estimates are presented in the followingtabulation, for the “all ages” and “teens”subgroups: Most of the qualitative conclusions noted above do not change. Therelative increases in the unemployment rate were still worse for menthan for women and worse for whites than for members of other races(except among teens). One differences is that, by this measure, teenssuffered greater than average unemployment rate increases, while onemight conclude the opposite using the measure in table 2. Referring again to table 2, we see that the pattern in the recoveryperiod differs somewhat from that of the recession. For instance, theeffect of the recovery was relatively stronger for women than for men,while the opposite was true of the recession.
The racial differenceremained the same: the effect of the recovery was felt more, on average,by whites than by nonwhites. The sex difference is primarily due to thefact that the unemployment rate continued to rise for nonwhite men wellinto the recovery period. Again, these observations are consistent withthose based on the published unemployment rates. Many explanations have been offered for these differences. Forexample, the effect of the downturn has been said to have been greaterfor men than for women because the economic decline affected primarilythe goods-producing, as opposed to the service-producing, sector.Construction and auto-related industries, including steel manufacturing,were especially hard hit. In contrast, some service industries actuallyincreased employment (although at a decreasing rate) throughout most ofthe recession. Along the same lines, blue-collar workers suffered worseemployment losses than white-collar workers.
Because men and women aredistributed differently among industries and occupations, with men inthe more cyclically sensitive ones, men would be expected to sufferrelatively greater increases in their unemployment rates. The fact thatthe industries and occupations that incurred the greatest losses indemand are also those with traditionally higher than average layoff rates could have further aggravated their employment declines. The contribution this makes to the sex difference in the employmentdeclines is unclear, however. We know that men have higher layoff ratesthan women, but that is probably primarily because of the sex differencein the occupational distribution. Any sex differences in the cyclic sensitivity of layoff rates are also probably due to the industrial oroccupational distributions. To fully understand the role of layoffrates in explaining the sex differences in the cyclic behavior ofunemployment rates, we need to know whether the responsiveness of thelayofff rate is less for women than for men in the same industry andoccupation.
Evidence presented by Norman Bowers suggests that in thethree previous recessions the responsiveness of the layoff rate wasactually greater for women than for men, both on average and by industryand occupation. Findings by Francine Blau and Lawrence Kahn, however,seem to show that there is little, if any, sex difference in thecyclical component of layoffs after controlling for industry,occupation, and other worker characteristics. Differences in cyclical variations in layoff rates also fail toexplain the racial difference in changes in the unemployment rate.
Nonwhites suffered relatively smaller unemployment rate increases thanwhites during the last recession, yet their layoff rates havehistorically been more cyclically responsive, even after controlling forworker and job characteristics. Instead of layoff rate disparities, theracial difference in the unemployment response is probably due, at leastin part, to the fact that members of racial minorities never fullyrecovered from the 1980 recession. Their unemployment rates werealready high when the most recent downturn began, so that the increasesit brought about were relatively small.
One other factor that could be important in explaining thedifferential unemployment rate impacts both by race and by sex is thepropensity, as unemployment rates increase (or, put differently, asemployment opportunities decline), for labor participation rates todecrease. If women and nonwhites tend to drop out of the labor force ata greater rate than white males in response to a given change inemployment opportunities, then their unemployment rates will not rise byas much as those for white males. The “economic impact” formen and women could therefore be the same–women could suffer as much asmen–but it would not be reflected in the unemployment rate. It is forthis reason that many analysts argue that unemployment rates are notappropriate measures of the welfare of a demographic group, and preferto study the “employment to population ratio” instead.
Iprefer to examine the problem directly and look at the behavior of boththe unemployment and labor force participation rates. In particular, weneed to examine the relationships between the two. Estimates of the percentage changes in (seasonally adjusted) laborforce participation rates for the July 1981-November 1982 period and theNovember 1982-December 1983 period are presented in table 2.
As withthe cyclic behavior of the unemployment rate, differences existaccording to age, race, and sex. Note that the participation ratedecreased for men during the economic decline, while it increased forwomen. The rate rose for whites, but the increase was small relative tothe increase for blacks and others.
Referring to the previousdiscussion, we find these results suggest that the unemployment ratemeasure actually overstates the burden of the recession for women andmembers of racial minorities relative to white men, rather thanunderstating it as had been hypothesized above. Certainly, these changes may be due to recent trends more than tothe business cycle. To correctly interpret changes in the unemploymentrate, we need to look at its relationship with participation rates netof trend. I do this by examining the coefficient on the unemploymentrate variable in the following equation: (1) log (LFPR).sub.t =[BETA].sub.
0 + [BETA].sub.1.TIME.sub.1 + [BETA].sub.
2.URATE.sub.t.-1 +*(seasonal dummies) + u.sub.
1 where LFPR is a given group’s laborforce participation rate in period t, and URATE.sub.t.-1 is theunemployment rate (for that group, for the entire population, or forsome reference group, such as prime-age men), lagged one period. Lagging the unemployment rate is one way to eliminate the problems created bythe fact that sampling errors in URATE and LFPR may be highly correlated at any point in time. Estimates of BETA.sub.
1 and BETA.sub.2 arepresented in table 3, by age, race, and sex. (The estimates are derived using the Cochrane-Orcutt technique, assuming first-order serialcorrelation. The unemployment rate variable is here defined as theaverage unemployment rate for the population as a whole.
) The results indicate that the relationship between the unemploymentrate and the labor force participation rate (as measured by thecoefficient on URATE) did not differ much by race, except for maleteenagers. For nonwhite male teens, a 1-percent increase in theunemployment rate (that is, from 10.0 to 10.01) is associated with a.3483-percent decrease in their labor force participation rate. Thatresponse is almost four times the response exhibited by whites. For thepopulation as a whole, however, the magnitudes of the responses varylittle by race. Some differences do exist by sex, with males exhibitinga strong tendency to decrease their participation as unemployment raterise.
This is true for all groups except white teens. The coefficientson TIME indicate that the increases in the participation rates of womenduring the period (recall the results in table 2) were indeed largelythe effect of a trend component rather than a cyclic one. Relatingthese results back to our interpretation of the “burdens” ofthe recession, the fact that declines in aggregate demand seem togenerate relatively larger decreases in participation for men and teens,and especially minority male teens, suggests that the unemployment ratesfor those groups may understate the true relative burden of therecession. Explanations for the differing participation rate responses includethe notion that teens and men exhibit greater than average decreases inparticipation as unemployment rates rise because they suffer greaterthan average decreases in demand for their labor. A decrease in demandcan have two effects: first, assuming some degree of wage rigidity,there is a direct effect on employment, and hence a direct effect onparticipation–if the number of people employed declines then, otherthings equal, the participation rate will decline. Second, there is the”discouraged worker effect,” the decline in participationbecause persons think they will have little success finding a job. Analternative explanation is that demand does not decrease more for teensor men, but rather that, given a change in the demand for their labor,teens and men simply respond more. Results from another study haveshown that resources of differences in participation responses includedifferential costs of search, differential wage rates, and differentiallevels of (not changes in) labor demand, in addition to differential”preferences” for work.
Possible explanations for the relatively small decreases inparticipation exhibited by women may therefore include the following:(1) demand for women’s labor does not decline much as unemploymentrates rise; (2) women have stronger preferences for work and lower costsof search; or (3) women will enter the labor force as unemployment ratesrise to compensate for income lost because of the unemployment of otherfamily members (the “added worker effect”). Evidence of thevalidity of each of these hypothesis is presented later in this study. In sum, using relative changes in the unemployment rate as ameasure of the impact of the recent recession, the evidence indicatesthat the heaviest burdens were placed on male, white, and prime-aged andolder workers. The magnitude of the burdens is open to question,however, if one keeps in mind that changes in labor force participationrates affect measured unemployment rates, and that the participationrate is endogenously determined.
Inspection of the relationship betweenlabor force participation rates and aggregate demand suggests that theunemployment rate variable probably understates the recession’srelative impact on men and on teens. The nature of differential impacts According to the gross change data, 3,293,000 workers becameunemployed during December 1983. Some 1,837,000 entered unemploymentfrom employment, while 1,456,000 entered unemployment from outside thelabor force. During the same month, 3,576,000 workers leftunemployment–1,745,000 into employment and 1,831,000 into thenon-participation state. At this example illustrates, the labor marketis in continual motion. The goal of the following discussion is toexamine the cyclical variations in unemployment and labor forceparticipation noted earlier in the context of such labor market flows. Let us denote the number of workers who make a transition fromstate I to state J (for example, from employment (E) to unemployment(U), or from unemployment to non-participation (N) during month t as IJ.Define the probability of making such a transition, given that one is instate I in month t-1, as lambda.
sub.IJ = IJ.sub.t./I.sub.
t.-1., whereI.sub.t.-1 is the number of people in state I in period t-1. It canthen be shown that unemployment rates and labor force participationrates can be expressed as explicit functions of the six transitionprobabilities lambda.sub.
, lambda.sub.UE., and lambda.sub.UN..
The relationshipsare such that the unemployment rate increases with increases inlambda.sub.NU and lambda.
sub.EU and decreases with increases inlambda.sub.UE and lambda.sub.UN. The effects of changes inlambda.
sub.NE and lambda.sub.EN depend on the relative magnitudes of theother transition probabilities. The participation rate will increasewith increases in lambda.sub.NE and lambda.sub.
NU., and decrease withincreases in lambda.sub.
EN and lambda.sub.UN.. The efects oflambda.
sub.UE and lambda.sub.EU depend on the relative magnitudes oflambda.sub.UN and lambda.
sub.EN. Whatever their size or direction,changes in these transition probabilities are the sources of changes inunemployment and labor force participation rates. We can thereforeanalyze cyclical changes in unemployment and participation rates interms of cyclical variations in transition probabilities. Before proceeding to that analysis, however, it may be useful toexamine age, race, and sex differences in levels of transitionprobabilities. The average over the December 1981-December 1983 periodare presented in table 4 for the population as a whole, and for theteenage and prime-aged groups.
Given the race, sex, and age differencesin unemployment and participation rates, the differences in transitionprobabilities are not surprising. Women have lower probabilities ofmaking the transitions from N-to-E and N-to-U, and much higherprobabilities of moving from E-to-N and U-to-N. All of thesedifferences contribute to the lower labor force participation rates forwomen. Members of racial minorities have much lower rates of transitionfrom U-to-E than do whites, and slightly higher transition rates fromE-to-U, which contribute to their higher unemployment rates.
Racialdifferences also exist in the N-to-U and U-to-N transition rates, withnonwhites more likely to enter unemployment on the one hand, more likelyto leave it on the other. These differences tend to cancel one anotherout. A significant racial difference also exists for the N-to-Etransition for teenagers, with nonwhites much less likely to make thetransition. On average, teenagers are much more volatile than otherlabor force groups, with higher than average probabilities for theN-to-E, N-to-U, E-to-N, E-to-U, and U-to-N transitions. The U-to-Etransition rate does not differ much by age. Prime-aged workers differfrom others primarily in their lower E-to-N and U-to-N transitionprobabilities.
The hypothetical relationship between aggregate demand and each ofthe transition probabilities are relatively straight-forward for someflows ad very complex for others, depending on one’s model andassumptions. In a fairly general model, all of the effects of a changein demand are indeterminant. A decline in aggregate demand will tend todecrease lambda.sub.UE and lambda.sub.NE because the number, frequency,and attractiveness of job offers will decline. A decrease in thefrequency of job offers can cause worker’s reservation wages tofall, however, which would tend to increase lambda.
NE. A decline in aggregate demand can increase the flowsfrom E-to-U and E-to-N due to an increase in layoffs nd terminations,but it can decrease the same flows if it lowers workers’ propensityto quit a job. As aggregate demand falls, we might expect lambda.sub.UNto increase and lambda.sub.
NU to decrease as a result of declining joboffers, but this conclusion depends critically on the relativemagnitudes of the levels of changes in job offer rates to people in theU and N states. In addition, lambda.sub.UN may decrease andlambda.
sub.NU may increase when aggregate demand falls, as individualsrespond to the unemployment of other family members. The actualrealtionships between aggregate demand and transition probabilities are,at best, empirical issues. Using the lagged population-average unemployment rate as a measureof aggregate demand, I have explored these relationships by estimatingthe parameters of the following equation for each transition rate andfor the entire population, teens, and the prime-aged group: (2)log(lambda.
sub.IJ.).sub.T = beta.sub.
0 + beta.sub.1.TIME.sub.t +beta.sub.
2.URATE.sub.t-1 + *(seasonal dummies) + u.
sub.t These estimates of beta.sub.1 and beta.sub.2 are presented in table5. Theresults indicate that some transition probabilities were much morecyclically responsive than others and that the responsiveness variedsignificantly across demographic groups. First, the N-to-E transitionrate declined with aggregate demand, for the population as a whole andfor each of the subgroups except nonwhite teenage females.
The declineis especially large for nonwhite males. Nonwhite male teenagersexhibited the strongest response, which would contribute to theirstronger participation rate response. (See table 3.
) Overall, theN-to-E transition rate seems more responsive for racial minorities thanfor whites, and more responsive for men than women. The responsivenessof the N-to-U transition rate differs primarily by race, not only inmagnitude but also in direction. The N-to- transition rate tends toincrease for whites as aggregate demand falls, but decreases for blacksand others (though the effect is often statistically insignificant).The effect of this difference is to decrease labor force participationamong nonwhites and boost it among whites.
The E-to-N transition ratedeclines as aggregate demand falls, for all age, race, and sex groups.The effect is stronger for nonwhites, with little difference by sex.The U-to-N transition rate also decreases with aggregate demand for thepopulation on average, although it increases for female teens. Both ofthese transition rate responses (for E-to-N and U-to-N) are counter tostandard views of the effects of declines in aggregate demand.
Inparticular, they tend to increase rather than decrease labor forceparticipation. The strong negative relationship between the unemploymentrate and participation rates exhibited by many of the demographic groupstherefore is not the result of an increased tendency to drop out of thelabor force. Rather, the relationship is the result of a decrease inthe tendency to enter the labor force, particularly directly intoemployment.
The E-to-U and U-toE transition rates increase and decrease,respectively, as aggregate demand falls. There is little difference inthe E-to-U response by race or by sex, excpet for teens and perhapsprime-age men. Large race and sex differences do exist for the U-to-Etransition rate, however, which are probably the primary source of thedifferential unemployment rate responses noted earlier. As aggregatedemand fell during the recession, the U-to-E transition rate declinedmore for whites than for racial minorities (except prime-age men), andmore for males than for females except, again, among teens. Thesedifferences may be the result of the disproportionate distribution ofthe sexes and races across occupations and industries. All of these differences in the responsiveness of transitionprobabilities can be related to race, sex, and age differences in thecyclic responsiveness of unemployment and labor force participationrates, and can help identify their sources. The fact that theunemployment rate increased more for men than for women during therecession seems to be the result of the sex differences in theresponsiveness of the U-to-E transition probability.
This may beinterpreted as support for the hypothesis that the demand for labordeclined relatively more for men. The fact that the participation ratedeclined more for men than for women seems to be the result of atendency for the N-to-E transition rate to decline more for men. Thisfact could suggest that the differential participation rate response isa labor demand, rather than a labor supply, phenomenon. The addedworker effect as an explanation for the sex differences in theparticipation response does not get much support here, because theN-to-U transition probability does not respond any more for women thanit does for men, at least among whites. The racial difference in the responsiveness of the unemploymentrate during the recession is primarily the result of racial differencesin the responsiveness of the N-to-U and U-to-E transition probabilities.Both tend to boost unemployment rates more for whites than fornonwhites. The N-to-U difference indicates that the added anddiscouraged worker effects may be important explanations here, withwhites being the added workers and nonwhites the discouraged ones. Thiscould simply be the result of the racial difference in the distributionof single-parent households.
However, it could also be an indicationthat members of racial minorities feel that they are at a considerablelabor market disadvantage because of their race. The relatively largedecline in the N-to-E transition rate for nonwhites may very well meanthat nonwhites do suffer larger decreases in demand for their labor asaggregate demand declines. The major age differences in the responsiveness of unemployment andparticipation rates can also be related to specific transition rates.The unemployment rate of teenagers rose less than average as aggregatedemand fell because the U-to-E transition rate did not decline by asmuch for teens as for other groups, and because the U-to-N transitionrate increased for teens (except nonwhite males) while decreasing forother groups. The first phenomenon could indicate that reservationwages fell more for teens than for other workers, or that the demand forteenage labor decline less than the demand for others, while the secondphenomenon suggests that teens were more likely to become discouragedand quit looking for work. The response of the U-to-N transitionprobability also obviously contributes to age differences in theresponsiveness of the labor force participation rate.
Other factors arethe age differences responses of the N-to-E and N-to-U transition rates,especially for nonwhite males. The large N-to-E response could indicatethat a substantial portion of the participation rate decline for teensis the result of a decrease in the demand for their labor. The result presented here lend support to many of the hypothesesput forth earlier regarding the sources of demographic differences inunemployment and participation rate behavior. The male/femaledifference in unemployment rate behavior is indeed probably due todifferential changes in demand, which may be attributable to theoccupational distribution of the sexes. There is no support, however,for the hypothesis that the participation rate differences arise becausewomen are more likely than men to be “added workers.”Differences between the participation responses of whites and nonwhitesand between those of teens and other workers appear to be due both todifferences in relative responses of the demand for their labor (withthe demand for labor decreasing more for racial minorities and teens),and to difference in “supply.” Suggestions for further research This analysis of gross change data from the Current PopulationSurvey provides insights into the nature of the differential effects ofthe recent recession which cannot be obtained from an analysis ofunemployment or participation rates alone. Many questions remainunanswered, however.
Foremost, of course, is, what exactly causes eachof the differential transition rate responses? If men are discouragedmore than women, hwy? That is a difficult question even with microdata.There are also some questions relating to the methodology, includingthose related to the timing of the effects of the recession and theappropriate lag structures to use for the URATE variable in equations 1and 2. Further, exactly what is the effect on the unemployment rate ofa 1-percent decrease in a given transition rate? Does the effect differby race or sex? One last question we may want to address is, how do theeffects fo the 1981-82 recession differ from those of earlier downturns?Have there been structural changes in the relationships betweenaggregate demand and transition rates which may indicate, for example,that there is less sex or race discrimination in the labor market today,or that there has been a profound and lasting change in women’sattitudes toward work outside the home? Many researchers address theseissues in other contexts, but a comparison of the results presented herewith those from studies of earlier periods could lead to betterunderstanding. Finally, it should be noted that many cyclical changes inemployment status are not between employment, unemployment, andnonparticipation, but rather between full-time and part-time employment.
The data used in this study do not distinguish between full- andpart-time employment. An analysis of gross flow data that make such adistinction could be very fruitful, as could further study of grosschange data broken down by industry of employment.