06/06/2003 12:41 PM

What effect do unions have on wages now and would ‘What Do Unions Do?’ be surprised?

David G. Blanchflower

Bruce V. Rauner 1978 Professor of Economics Dartmouth College and NBER

Alex Bryson Policy Studies Institute and

Centre for Economic Performance, LSE

Friday, June 06, 2003 11:09:45 AM Contact addresses David Blanchflower Alex Bryson Department of Economics Policy Studies Institute, Dartmouth College 100 Park Village East Hanover, NH 03755, USA London NW1 3SR, UK Email: blanchflower@dartmouth.edu a.bryson@psi.org.uk Internet: www.dartmouth.edu/~blnchflr http://www.psi.org.uk We thank Bernt Bratsberg, Bernard Corry, Barry Hirsch, Andrew Oswald, Richard Freeman and Jim Ragan for their help.

1

06/06/2003 12:41 PM

“Everyone ‘knows’ that unions raise wages. The questions are how much, under what conditions, and with what effects on the overall performance of the economy”. What Do Unions Do?, Freeman and Medoff, 1984, p.43

In 1984 Richard Freeman and James Medoff (FM) published their pathbreaking book What Do Unions Do?. The book has had an enormous impact. According to Orley Ashenfelter, one of the commentators in a review symposium on the book published in January 1985 in the Industrial and Labor Relations Review, 38(2), pages 245-258, the response of the popular press to the book ‘has only been short of breathtaking’ (p.245)1. It received rave reviews at the time it was written and unlike most books has withstood the test of time: it is certainly the most famous book in labor economics and industrial relations. One of the other reviewers in the symposium, Dan Mitchell called it ‘a landmark in social science research’ and so it has proved (p.253). We went to the Social Science Citations Index and typed in ‘What do unions do’ (henceforth WDUD) and found it had been cited by other academics more than one thousand times2. In what follows we will show that the vast majority of their commentary written in the early 1980s is still highly applicable today despite the fact that private sector unionization has been in precipitous decline since they put pen to paper or maybe it was even fingers to keyboard! An old adage is that a classic book is one that everyone talks about but nobody reads. This book is not one of those. It is a classic in the true sense because it continues to be a book that anyone – scholar or layman - interested in labor unions needs to read! Central to the thesis propounded by FM is that there are two faces to unions – the undesirable monopoly face – which enables unions to raise wages above the competitive level which results in a loss of economic efficiency. This inefficiency arises because employers adjust to the higher union wage by hiring too few workers in the union sector. The second more desirable face, is the collective voice face which enables unions to channel worker discontent into improved workplace conditions and productivity. This chapter concentrates on the monopoly face of unions and its impact on relative wages. We examine the various claims made by FM about the impact of unions on wages and update them with new data. We examine in some detail the role of the public sector, which was largely ignored by FM. This was a perfectly understandable omission at the time but is less appropriate today given the importance of public sector unionism in the US3. Richard Freeman has devoted a lot of his

1 The symposium included an introduction by the editor John Burton along with reviews by Orley Ashenfelter, Barry Hirsch, David Lipsky, Dan Mitchell and Mel Reder plus a reply by Richard Freeman and Jim Medoff. 2 It has been cited an astonishing 1024 times over its near twenty year life (is it really that long?). In 2002 alone it had 34 cites. It is clear that the book continues to be relevant. 3 In their reply to the various reviewer comments in the symposium issue of the ILRR (1985) they gave a solid defense for these omissions!

“(W)e have only the defense of finite life. In writing a book on such a vast topic as trade unionism, one must give short shrift to some topics or end up like Sisyphus, never completing the task. We admit to sins of omission on the dynamics of wage negotiations, internal union behavior, detailed treatment of the most recent developments in collective bargaining, strikes and, of course, the public sector – a sufficiently difficult terrain to merit quite a different analytical treatment” (Freeman and Medoff, 1985, p261).

2

06/06/2003 12:41 PM

subsequent writings to an examination of unionism in the public sector including Freeman (1986, 1988) and four chapters written with co-authors in Freeman and Ichniowski (1988). In a paper with Ichniowski and Lauer union wage gaps for police were estimated (Freeman, Ichniowski and Lauer, 1989). Freeman has also compared the effects of unions on wages in the US with that in other countries in a joint paper with one of the authors (Blanchflower and Freeman, 1992). Freeman has also written widely on what he argues is the large impact declining unionism has had on rising wage inequality that has occurred in the United States – see for example Freeman, (1993, 1995, 1999); Freeman and Katz (1995); Freeman and Needels (1993) and Freeman and Revenga (1998)4. In Section One we report FM’s main findings. In Section Two we discuss the main labor market changes that have occurred since WDUD was written. Section Three reports our estimates of wage gaps disaggregated by various characteristics used by FM. We also examine wage gaps that FM did not examine including those in the public sector as well as for immigrants and the self-employed. Section Four examines time series changes in the union wage gap. Section Five models the determinants of changes in the union wage premium at the level of the industry and state. In the penultimate section we attempt to explain our findings. In Section Seven we outline our main findings and discuss whether FM would have been surprised about these findings when they wrote WDUD. 1. Summary of FM’s findings on union wage effects. FM reported that early work used aggregate data on different industries, occupations and areas as data on the wages of unionized and non-unionized individuals and establishments was not available. Much of this work was summarized in Lewis (1963). The reason that such aggregated data were used was that ‘data on the wages of unionized versus non-unionized individuals or establishments was neither available nor, given the state of technology, readily amenable to statistical analysis’ (1984, p.44). These studies found a union wage effect on average of 10-15%. The more recent studies FM examined, including a number of their own, used micro data at the level of the establishment but more usually at that of the individual. In Table 3.1 FM showed that the union differential in the 1970s was 20-30% using cross-section data (the 7 numbers in the table averaged out at 25.3%). Such estimates may still suffer from bias because differences due to the skills and abilities of workers are wrongly attributed to unions. FM also considered ‘before and after’ comparisons and argued that, although they represent a way to eliminate ability bias they also suffer from measurement error problems derived from mismeasurement of the union status measure (Hirsch, 2003). FM reported 12 estimates using panel data in their Table 3.2 for the 1970s: these are sizable but smaller than the cross-section estimates they examined, averaging out at 15.7%5. FM used data from the May 1979 Current Population Survey to obtain a series of disaggregated estimates using a sample of non-agricultural private sector blue-collar workers aged 20-65. They

4 For a discussion of the issues involved see Blanchflower (2000a). 5 For further discussions on these issues see Lewis (1986), Freeman (1984) and Blanchflower and Bryson (2003).

3

06/06/2003 12:41 PM

reported that unions raise wages most for the young, the least tenured, whites, men, the least educated, blue collar workers and in the largely unorganized South and West6. Further FM found, using data from the 1975-1979 that out of 62 industries there was considerable variation in the size of the differential7. FM argued that the amount of monopoly power held by unions is related to the wage sensitivity of the demand for organized labor. The smaller response of employment to wages the greater they argued is the ability of unions to raise wages without significant loss in employment. Areas where employment is less responsive to wage changes such as air transport they argued should be where one would expect to find sizable gains. FM went on to argue that the differential likely depends on the extent to which the union is able to organize a big percentage of workers – the higher the percentage the higher the differential (p51). FM found that for blue collar workers in manufacturing a 10% increase in organizing generates 1.5% increase in union wages. In contrast, they argued that the wages of nonunion workers do not appear to be influenced by the percentage workers organized. In terms of the characteristics of firms and plants FM obtained the following results. a) Union differentials depend on the extent to which the firm bargains for an entire sector rather than for individual plants within a sector. b) Wage differentials FM found tended to fall with size of firm/plant/workplace. c) There was no clear empirical evidence on the relationship between product market power and differentials primarily as it is so difficult to measure power. In terms of macro changes in differentials FM found that the 1970s were a period of increases in the union wage premium. FM conjectured that a possible explanation was the sluggish labor market conditions prevailing at the time. Wages of union workers, they argued, tend to be less sensitive to business cycle ups and downs – particularly due to three year contracts. This implies the union wage premium moves counter-cyclically – high in slumps when the unemployment rate is high and low in booms when the unemployment is low. However, FM found that inflation and unemployment explained less than 50% of the rising union differentials in the 70s. Nor they argued did the rising wage differentials of the 1970s represent an historical increase in union power. The early 80s according to FM were a period of ‘givebacks’ where unions agreed to wage cuts. (These have continued in some companies, such as American Airlines and employee-

6 In contrast Lewis (1986), who did not restrict his analysis to non-agricultural private sector blue-collar workers aged 20-65 as FM did, found no differences by either gender or color although he confirmed FM’s other disaggregated results. Lewis reported a number of additional disaggregated results that were not examined by FM. Lewis found the wage gap was greater for married workers; U-shaped in age/experience; U-shaped in tenure/seniority minimizing at 22-24 years of seniority; was higher the higher is the unemployment rate. He found mixed results regarding any relationship with the industry concentration ratio. 7 FM estimates of the variation in the wage effect by industry used 1973-1975 May CPS data. Sample sizes in many cases were likely very small as is made clear from their footnote 11 which says that they limited their sample of industries to ones containing at least five union and non-union members. The rule resulted in only four industries being dropped.

4

06/06/2003 12:41 PM

owned United Airlines). Union wage gains, according to FM were NOT a major cause of inflation. FM ended chapter three by estimating the social cost of monopoly power of unions. Loss of output due to unions they found to be ‘quite modest’: accounting for between 0.2% and 0.4% of GNP or between $5 billion and $10 billion. FM drew six conclusions on the union wage effect. a) The common sense view that there is a union wage effect is correct. b) The magnitude of the differential varies among people, markets and time periods. c) Variation in the union wage gap across people is best understood by union standard rate policies arising from voice. d) Variation in the union wage gap across markets is best understood by union monopoly power and employer product market power. e) Wage premia of 1970s were substantial but it returned to more ‘normal’ levels in the 1980s. f) Social loss due to unions is small. 2. Changes in the Labor Market Since WDUD 1.1. Table 1 shows that union density rates in the US have fallen rapidly from 24% in 1977 to 13% in 20028. The decline was most dramatic in the private sector where in 2002 less than 1 in 10 workers are union members. Density remains higher in manufacturing than in services. 1.2. Table 2 suggests that union membership has roughly the same disaggregated pattern in 2001 as it did in 1977 – union density is higher a) among men than women, b) for older versus younger workers, c) in regions outside the South, d) in transportation and communication and construction. Exceptions are: a) by race where in 1977 rates were higher among non-whites but there is little difference by 2001. b) by schooling – in 1977 membership rates for those with<high school were nearly double those with >high school. In 2001 they were approximately the same. So the highly qualified have increased their share of union employment. 1.3. The number of private sector union members declined over the period 1983-2002 whereas the number of public sector union members actually increased (Table 3). Due to the growth in total employment in the public sector, however, the proportion of public sector workers who were union members was exactly the same in 2001 and 1983 (37%) – see Table 39. By 2002,

8 FM use 1977 union density rates in table 2.1. 9 These estimates are obtained from the ORG files which only contain numbers from the public sector since 1983.

5

06/06/2003 12:41 PM

46% of all union members in the US were to be found in the public sector compared with 32.5% in 1983. 1.4. Unemployment and inflation were both lower in the 1990s than they were in the 1970s or 1980s. Averages for the past six decades and numbers for the most recent three years are given below. Inflation is measured by the annual change in the CPI. Inflation Unemployment 1940s 5.6% 5.2% 1950s 2.1% 4.5% 1960s 2.4% 4.8% 1970s 7.1% 6.2% 1980s 5.6% 7.3% 1990s 3.0% 5.8% 2000 3.4% 4.0% 2001 2.8% 4.8% 2002 1.6% 5.8% Source: http://www.bls.gov 3. Union wage gaps since WDUD 3.1. What has happened to the union wage differential between 1979 and 2001? Table 4 presents union wage gaps obtained from estimating a series of equations for each of the major sub-groups examined by Freeman and Medoff who used the 1979 May CPS file on a sample of non-agricultural private sector blue-collar workers aged 20-65. Their sample was very small, being limited to around 6,000 observations. Rather than use the estimates reported by Freeman and Medoff to ensure large sample sizes we decided to pool together 6 successive May CPS files from 1974-1979 and then compare those to wage gaps estimated for the years 1996-2001 using data from the MORG files of the CPS. Columns 1 and 2 estimate wage gaps for the private sector for 1996-2001 and 1974-1979 respectively. Columns 3 and 4 present equivalent estimates for the sample used by Freeman and Medoff of non-agricultural private sector blue-collar workers aged 20-65. Hirsch and Schumacher (2002) have recently shown that there is what they call a ‘match bias’ in union wage gap estimates due to earnings imputations10. They show that this bias arises because currently 30 per cent of workers in the Current Population Survey have earnings imputed using a “cell hot deck” method. This means that wage gap estimates are biased downward when the 10 We do not deal here with a further problem identified by Card (1996) of misclassification of self-reported union status in the CPS, first identified by Mellow and Sider (1983). Card concludes that about 2.7% are false positives and 2.7% are false negatives. Given that there are more non-union workers than union workers, this means the union density rate is biased upwards. See Farber (2001) for a discussion and a procedure to adjust the union density rate for error. In 1998, the observed private sector rate of 9.7% translates to an adjusted rate of 7.4% (the figures for 1973 were 25.9% and 24.5% respectively).

6

06/06/2003 12:41 PM

attribute being studied (e.g. union status) is not a criterion used in the imputation. By construction, the individuals with imputed earnings have a union wage gap of zero; hence omitting them raises the size of the union wage gap. They show that standard union wage gap estimates such as reported in Blanchflower (1999) are understated by about 3 to 5 percentage points as a result of including individuals who have had their earnings imputed. Unfortunately, it is not a simple matter to exclude those individuals with imputed earnings in a consistent way over time11. Here we follow the procedure suggested by Hirsch and Schumacher (2002) and that we used in Blanchflower and Bryson (2003). All allocated earners are identified and excluded for the years 1996-2001 in the MORG files. Because the May CPS sample files available to us do not include allocated earnings in 1973-81, the series are adjusted upward by the average bias (of .033) found by Hirsch and Schumacher using these May CPS data for 1979-81. For the period 1974-1979 total sample size was approximately 184,000 compared with 547,000 for the later period. In each year from 1996-2001 there are approximately 130,000 observations for the private sector in the MORG; in the May files, sample sizes are approximately 31,000. Comparing FM’s sample and the wider private sector sample for the 1970s (columns 4 and 2 respectively in Table 4), FM’s sample generates a larger wage gap for all, with the exception of the least educated12. The difference between the two samples is large, with the FM sample generating a premium for the whole private sector which is a third larger (28% as opposed to 21%) than the wider sample. However, patterns in the wage gaps across workers are similar. a) By sex, there is little difference in the size of the gap. b) By age, the union effect is U-shaped and largest among the youngest, who are the lowest

paid. c) By tenure, the pattern is a similar U-shape.

11 The number of wage observations followed by the percentage imputed in parentheses (hourly + non-hourly paid) in the NBER MORG are given below. Note in 1995 allocation information is only available on one-third of the wage observations, hence the small sample. 1979 171,745 (16.5%) 1986 179,147 (10.7%) 1993 174,595 (4.6%) 2000 161,126 (29.8%) 1980 199,469 (15.8%) 1987 180,434 (13.5%) 1994 170,865 (0%) 2001 171,533 (30.9%) 1981 186,923 (15.2%) 1988 173,118 (14.4%) 1995 55,967 (23.3%) 2002 184,137 (30.4%) 1982 175,797 (13.7%) 1989 176,411 (3.7%) 1996 152,190 (22.2%) 1983 173,932 (13.8%) 1990 185,030 (3.9%) 1997 154,955 (22.2%) 1984 177,248 (14.7%) 1991 179,560 (4.4%) 1998 156,990 (23.6%) 1985 180,232 (14.3%) 1992 176,848 (4.2%) 1999 159,362 (27.6%) 12 Although there are some differences in the levels of the wage differences reported in columns 3 and 4 of Table 4 the majority of these results are consistent with the findings reported by FM in their Figure 3.1. Major exceptions are FM’s finding that wage gaps were higher for men than women and for non-whites compared with whites. We suspect such differences arise because of the small sample size in the May 1979 CPS of 6,018 used by FM.

7

06/06/2003 12:41 PM

d) By education, unions raise wages most for the least educated, with the most highly educated having the lowest premium.

e) By race, unions raise wages by a similar amount for whites and non-whites. f) By occupation, although not reported in column 4, Freeman and Medoff (1984: 49-50) report

larger gains for blue-collar than for white-collar workers. The manual/non-manual gap in column 2 bears this out.

g) By region, unions had the largest effects in the relatively unorganized South and West, with

more modest effects in the relatively well organized Northeast. h) By industry, Construction and Services have the highest premia. What has happened since the 1970s? a) By sex, the wage gap has declined for women, but remained roughly stable for men so that,

by the late 1990s, the union wage gap was higher for men than for women. The rate of decline in women’s union premium is underestimated in FM’s restricted sample, but is still apparent.

b) By age, the U-shaped relationship apparent in the 1970s has disappeared because there has

been a precipitous decline in the premium for the youngest workers, while older workers’ wage gap has remained roughly constant. In the full private sector sample, young workers still benefit most from union wage bargaining, though this is not apparent in FM’s restricted sample.

c) By tenure, as in the case of age, the U-shaped relationship between tenure and the union

premium apparent in the 1970s has disappeared, because low and high tenured workers have seen their wage gap fall substantially while middle-tenure workers have experienced a stable union wage gap. Now, it seems the premium declines with tenure.

d) By education, the lowest educated continue to benefit most from union wage bargaining, but

not to the same degree as in the 1970s. Although the trend is not so apparent in FM’s sample, the wage gap has fallen most for high-school drop-outs

e) By race, a three percentage point gap has opened up between the union premium

commanded by non-whites and the lower premium for whites f) By occupation, the union premium has collapsed for non-manual workers. Despite some

decline in the premium for manuals, their wage gap was 17 percentage points larger than that for non-manuals by the late 1990s (compared with only 5 percentage points in the 1970s)

g) By region, the wage gap remains largest in the West and the South though, in FM’s sample,

there is no difference in the premium in the South and Central regions. The wage gap remains smallest in the Northeast.

8

06/06/2003 12:41 PM

h) By industry, the wage gap remains largest in Construction and smallest in Manufacturing.

The decline in the differential was particularly marked in Services. We return to industry differentials later

Two points stand out from these analyses. First, no group of workers in the broader private sector sample has experienced a substantial increase in their union premium. Indeed, the only group recording any increase at all is those aged 45-54 who see their premium rise from 13% to 14%. Clearly, unions have found it harder to maintain a wage gap since FM wrote. Second, with the exception of the manual/non-manual gap, those with the highest premiums in the 1970s saw the biggest falls, so there has been some convergence in the size of wage gaps. This finding is apparent whether we compare trends using FM’s sample (columns 3 and 4) or the broader private sector sample (columns 1 and 2). This trend may be due to an increasingly competitive US economy, where workers commanding wages well above the market rate are subject to intense competition from non-union workers. Nevertheless, with the exception of the most highly educated and non-manual workers, the wage premium remains around 10% or more. 3.2. Public sector FM said little or nothing on the role of unions in the public sector, although, as noted above, Freeman has subsequently written voluminously on the issue. Given that the remaining bastion of unionism in the US is now in the public sector, it is likely if FM were writing today they would have devoted a considerable amount of space in a twenty-first century edition of WDUD to the public sector. So we did some of it for them: more evidence on how the role of unions in the public sector since WDUD was written is presented by Morley Gunderson in another chapter in this volume. It is apparent from Table 3 above that the size of the public sector grew (from 15.6 million to 19.1 million or 22.4%) over the period 1983-2001 but as a proportion of total employment it fell from 18.0% to 16.1%. Union membership in the public sector grew even more rapidly (from 5.7 million to 7.1 million or by 24.6%). As we noted earlier, by 2001 public sector unions accounted for 44% of all union members compared with 32.5% in 1983. Table 5 is comparable to Table 4 above for the private sector in that it presents disaggregated union wage gap estimates. Because sample sizes in the public sector are small using the May CPS files we once again decided to use data from the ORG files of the CPS for the years 1983-1988 for comparison purposes with the 1996-2001 data. It was not possible to use data for the years 1979-1982 as no union data are available. A further advantage of using the 1983-1988 data is that information is available on those individuals whose earnings were allocated who were then excluded from the analysis. The main findings are. a). The private sector union wage gap has fallen over the two periods (21.5% to 17.0%) whereas

a slight increase was observed in the public sector (13.3% to 14.5% respectively).

9

06/06/2003 12:41 PM

b). The majority of the worker groups in Table 5 experienced increases in the size of their union wage premium over the two periods. Wage gaps declined markedly for those under 25 and with less than a high school education.

c). There was little change in public sector union wage gaps for men or women. In marked

contrast to the private sector where men had higher differentials than women, wage gaps in both periods in the public sector were higher for women than for men.

d). Unions benefit workers most in local government and least in the federal government

although the differential for federal workers increased over time. e). Just as was found for the private sector, the wage benefits of union membership are greatest

for manual workers, the young and the least educated. f). There are only small differences in union wage gaps for non-whites compared to whites in

both the public and the private sectors. g). In contrast to the private sector where wage differentials were greatest in the South and West,

in the public sector exactly the opposite is found. Higher differentials are found in the public sector in New England and the Central region in both time periods whereas the reverse was the case in the private sector.

h). Wage gaps increased over time for teachers, lawyers, firefighters and police. 3.3 Immigrants FM also said nothing about the extent to which labor unions in the US are able to sign up immigrants as members and then by how much they are able to raise their wages. Data exists in the CPS files from the mid-1990s to examine this issue: such data were not available in the CPS before then. Table 6 illustrates. The first part of the table includes information on the extent to which union wage gaps vary by length of time the immigrant had been in the US, holding characteristics constant as well as wage gaps for the US born. There is very little variation by time the immigrant had been in the US which is in direct contrast to the results reported in the second part of the table based on source country. Differentials for us Europeans (11.6% for Western Europe and 12.7% for Eastern Europe) are well below those of the native born (16.8%)! Estimates are also in low double digits for Asians and Africans (13.3% and 11% respectively). A very small sample size appears to explain the insignificant Australasian result. A union wage gap of similar size was found in the Rest of South America (12.2%) which contrasts sharply with that for Mexicans of more than 28%. 3.4 Self-employed. According to the 2001 Statistical Abstract of the United States (Table 580), in 2000 8,674,000 non-agricultural workers in the United States were self-employed. This is a considerably lower

10

06/06/2003 12:41 PM

proportion than in most other Western countries13. Despite the fact that the numbers of self-employed were higher in 2000 than in 1980 (7,000,000), they were a slightly lower percentage of the non-agricultural workforce than they were in 1980 (6.6% in 2000 compared with 7.3% in 1980) which is in marked contrast to countries such as the UK which has seen a dramatic increase in the non-agricultural self-employment rate over that period (see Blanchflower, 2000). We know very little about the unionism rates of the self-employed or the extent to which unions are able to raise their wages. FM don’t even reference them in their index. Why is this? The most obvious reason is probably because the self-employed are not asked for their union status regularly in the CPS. As we show below there is some data in a few of the recent February Supplements. One source of data is the General Social Survey14. Unfortunately, sample sizes are small. Pooling surveys together we find the following union membership rate among the self-employed Employees Self-employed 1972-1978 25.8% 6.2% 1980-1989 18.0% 4.5% 1990-2000 16.3% 4.4% When separate regressions were run with the dependent variable the log of annual earnings along with a full set of controls (gender, age, age squared, two race dummies, 8 region dummies, 4 marital status dummies, log of hours and years of education) the results for, 1975-2000 (excluding 1977 and 1982) were as follows Wage gap N Employees only 33% 9,653 Self-employed 3% 1,244 The estimate for the self-employed was not significantly different from zero (t=0.22). 4. Time series changes in the union wage gap. FM reported that the 1970s was a period of rising differentials for unions, although they did not separately estimate year by year results themselves. Table 7a, which is taken from Blanchflower and Bryson (2003) and is reproduced by permission15, reports adjusted estimates of the wage gap using separate log hourly earnings equations for each of the years from 1973 to 1981 using the National Bureau of Economic Research’s (NBER) May Earnings Supplements to the Current

13 Some of these cross-country differences appear to arise because of the way in which the self-employed are defined. Owners of larger businesses – the incorporated self-employed – in the US are usually treated as employees but in most of the rest of the OECD they are treated as self-employed. For a discussion see Blanchflower (2000b). 14 The GSS is not taken in every year. Surveys were not taken in 1979, 1981, 1992, 1995, 1997, 1999 . 15 We have added data for 2002 as the 2002 ORG has recently become available.

11

06/06/2003 12:41 PM

Population Survey (CPS)16 and for the years since then using data from the NBER’s Matched Outgoing Rotation Group (MORG) files of the CPS17. The MORG data for the years 1983-1995 were previously used in Blanchflower (1999)18. For both the May and the MORG files a broadly similar, but not identical, list of control variables is used, including a union status dummy, age and its square, a gender dummy, education, race and hours controls plus state and industry dummies19. The first column of Table 7a reports time-consistent estimates of union wage gaps for the total sample while the second and third columns report them for the private sector. To solve the match bias problem discussed above, as we did in Tables 4 and 5 we followed the procedure suggested by Hirsch and Schumacher (2002) such that all allocated earners were identified and excluded for the years 1983-88 and 1996-2001 from the MORG files. For 1989-95, allocation flags are either unreliable (in 1989-93) or not available (1994 through August 1995). For 1989-93, the gaps are adjusted upward by the average imputation bias during 1983-88. For 1994-95, the gap is adjusted upward by the bias during 1996-98. Because the May CPS sample files available to us do not include allocated earnings in 1973-81, the series are adjusted upward by the average bias (of .033) found by Hirsch and Schumacher using these May CPS data for 1979-81. In each year there are approximately 160,000 observations for the US economy and 130,000 for the private sector in the MORG; in the May files, sample sizes are approximately 38,000 and 31,000 respectively until 1980 and 1981 when sample sizes fall to approximately 16,000 and 13,000, respectively, as from that date on only respondents in months four and eight in the outgoing rotation groups report a wage. Results obtained by Hirsch and Schumacher (2002) with a somewhat different set of controls are reported in the final column of the table. For a discussion of the reason for these differences see Blanchflower and Bryson (2003). The time series properties of all three of the series are essentially the same. The wage gap averages 18 per cent over the period, and is similar in size in the private sector as it is in the economy as a whole. (The private sector differentials in the second column are smaller than those obtained by Hirsch and Schumacher (2002) due to differences in model specification and, in particular, Hirsch and Schumacher’s inclusion of occupational dummies (see Blanchflower and Bryson (2003) for details). The table confirms FM’s comment (1984: 53) that ‘the late 1970s appear to have been a period of substantial increase in the union wage premium’.

16 The May extracts of the CPS extracts in Stata format from 1969-1987 are available from the NBER at http://www.nber.org/data/cps_may.html. 17 Hirsch, Macpherson, and Schumacher (2002) have compared union wage gap estimates obtained from the BLS quarterly Employment Cost Index (ECI) constructed from establishment surveys and from the annual Employer Costs for Employee Compensation (ECEC) with those obtained using the CPS. They find that union/non-union wage trends in the three series ‘are consistent neither with each other nor with the CPS’, and ultimately conclude that ‘we find ourselves relying most heavily on results drawn from the CPS’ (Hirsch, Macpherson, and Schumacher, 2002, p.23). 18 There was no CPS survey with wages and union status in 1982. 19 Following Mincer, it is more usual to include a term in potential experience rather than a direct measure of age. We use education, however, for reasons of comparability as the CPS Outgoing Rotation Group files from 1993 report qualifications rather than years of schooling.

12

06/06/2003 12:41 PM

What is notable is the drop in the size of the union differential after 1978, a jump in 1983/4 (we have no data for 1982) and a slight decline thereafter until 2000 when the series picks up again as the economy started to turn down. There appears to have been a decline in the size of the differential since 1995, as the US economy entered a boom period. The differential turned up again in 2002 as the economy slowed and unemployment rose. Table 7b presents estimates of both the unadjusted and adjusted union wage gaps for the private sector. In the first two columns the samples used exclude individuals with imputed earnings. In column 1 of the Table we report the results of estimating a series of wage equations by year that only include a union dummy as a control. These numbers are consequently different from those reported by Hirsch and Macpherson (2002) in their Table 2a who report raw unadjusted wage differences between the union and non-union sectors but do not exclude individuals with imputed earnings20. It is clear that the unadjusted gap has declined more rapidly since 1983 than has been the case for the adjusted estimates. In 1983 the unadjusted estimate was 128% higher than the adjusted estimate: in 2001 the difference had fallen to 97.5% higher. To establish what is driving this effect, Hirsch, Macpherson and Schumacher (2002) decompose the unadjusted wage gap into its three components – employment shifts, changes in worker characteristics, and in the residual union wage premium. Using CPS data for the private sector only, they find almost half (46%) of the decline in the union-nonunion log wage gap over the period 1986-2001 is accounted for by a decline in the regression-adjusted wage gap. Sixteen per cent of the decline is accounted for by changes in worker characteristics and payoffs to those characteristics, chief among these being the increase in the union relative to nonunion percentage of female workers. The remaining 38% of the decline in the unadjusted wage gap Hirsch, Macpherson and Schumacher found was due to sectoral shifts and payoffs to the sectors workers are located in. The sectoral changes that stand out are the substantial decline in union relative to nonunion employment in durable manufacturing, and the decline in relative pay (that is, the industry coefficient) in Transportation, Communications and Utilities, a sector with a large share of total union employment. The results reported in Table 7a appear to be broadly comparable to the estimates obtained by H. Gregg Lewis (1986) in his Table 9.7, which summarized the findings of 165 studies for the period 1967-79. Lewis concluded that during this period the US mean wage gap was approximately 15 per cent. His results are reported below21: Year # studies mean estimate Year # studies mean estimate 1967 20 14% 1973 24 15% 1968 4 15% 1974 7 15% 1969 20 13% 1975 11 17% 1970 8 13% 1976 7 16%

20 Similarly in the 2003 edition of Hirsch and Macpherson’s Union Membership and Earnings Data Book, recently received. 21 There is a dissonance between the estimates Lewis offers by way of summary in his introductory chapter and those given in his Table 9.7 which are produced here (Lewis, 1986, p. 9).

13

06/06/2003 12:41 PM

1971 20 14% 1977 10 19% 1972 7 14% 1978 7 17% 1979 3 13% The left panel contains estimates for the six years prior to our starting point in Table 7. It does appear that the unweighted average for this first period, 1967-72, of 14 per cent is slightly below the 16% for the second interval, 1973-79. The estimates for the later period are in turn somewhat smaller than those we obtained in Table 7a – which averaged 20 per cent – but appear to have the same time-series pattern; for example, 1979 has the lowest value in both sources. In part, the low number Lewis obtained for 1979 is explained by the fact that the 1979 May CPS file included allocated earners and hence the estimates were not adjusted for the downward bias caused by the imputation of the earnings data22. Figure 1 plots the point estimates of the union wage premium for the US, taken from the first column of Table 7a, against unemployment for 1973-2000. The premium moves counter-cyclically. This observation is supported by regressing lagged unemployment, time, and union density on the yearly premium as reported below. Equation 1 includes a time trend and equation (2) replaces the time trend with the lagged wage premium. Equation 3 takes the one year change in the premium as the dependent variable. We also tried using the unemployment rate in period t but the lagged rate worked best. The unemployment effect is unaffected by the replacement of the time trend with the lagged premium. Premiumt = 21.72 + .66 Unemploymentt-1 -.22 Time - .24 Densityt-1 (1) (2.33) (2.98) (1.31) (0.77) (N=29, R2=.59) Premiumt = 5.148 + .64 Unemploymentt-1 + .36 Premiumt-1 + .11 Density t-1 (2) (2.18) (3.21) (2.34) (1.61) (N=28, R2=.65) ∆Premium = 5.148 + .64 Unemploymentt-1 - .64 Premiumt-1 + .11 Density t-1 (3) (2.18) (3.21) (4.13) (1.61) (N=28, R2=.48) Note: t-statistics are in parentheses The regression supports the notion that the union wage premium is counter-cyclical, moving positively with changes in the lagged unemployment rate. Thus, as unemployment falls, as it has done since 1995, the union wage premium falls; and as unemployment rises the wage premium rises.23 A time trend, although negatively signed, is not statistically significant, suggesting the 22 Lewis (1986) had 35 studies using the CPS, 1970-1979; 16 studies using the 1967 Survey of Economic Opportunity; 25 studies using the Panel Study of Income Dynamics, 1967-78; 15 studies using Michigan Survey Research Center survey data other than the PSID, including the 1972-73 Quality of Employment Survey; 22 studies using the National Longitudinal Surveys of 1969-72; and 8 studies exploiting other sources. 23 Bratsberg and Ragan (2002) also find a counter-cyclical union wage effect in their industry-level analysis similar in size to that found by Lewis (1986: 154).

14

06/06/2003 12:41 PM

decline is not secular. The change in the premium is negatively correlated with the starting level of the premium. There are three factors which influence the degree of counter-cyclical movement in the wage gap. The first, cited by FM (1984: 52-53) as the reason for the widening wage gap during the Depression of the 1920s and 1930s, is the greater capacity for unionized workers to ‘fight employer efforts to reduce wages’ when market conditions are unfavorable. Conversely, when demand for labor is strong, employees are less reliant on unions to bargain for better wages because market rates rise anyway. The second factor is the fact that union contracts are more long-term than non-union ones and, as such, less responsive to the economic cycle. This means that union wages respond to economic conditions with a lag. When inflation is higher than expected, this can result in a greater contraction in the premium because non-union wages are more responsive to higher inflation. However, the third factor, which should reduce the cyclical sensitivity of the union wage premium, is the cost-of-living-adjustment (COLA) clauses in union contracts which increase union wages in response to changes in the consumer price level. According to FM (1984: 54) the percentage of union workers covered by these agreements rose dramatically in the 1970s, from 25% at the beginning of the decade to 60% at the end of the decade. However, FM’s estimates for manufacturing suggest that COLA provisions ‘contributed only a modest amount to the rising union advantage’ in the 1970s. Bratsberg and Ragan (2002) revisit this issue and find the increased sensitivity of the premium to the cycle is due in part to reduced COLA coverage from the late 1980s, but we find no such evidence. Commenting on the growth of the union wage premium during the 1970s, FM (1984: 54) suggested that “at least in several major sectors the union/nonunion differential reached levels inconsistent with the survival of many union jobs”. They were right. In the 1970s and early 1980s, the wage gap in the private sector rose while union density fell, as predicted in the standard textbook model of the way that employment responds to wages where the union has monopoly power over labor supply. In the classic monopoly model, demand for labor is given, so a rise in the union premium will result in a decline in union membership since the premium hits employment. The fact that unions pushed for, and got, an increasing wage premium over this period, implies that they were willing to sustain membership losses to maintain real wages, or that unions were simply unaware of the consequences of their actions. From the mid-1990s, the continued decline in union density was accompanied by a falling union wage premium. This occurred because demand for union labor fell as a result of two pressures. The first was increasing competitiveness throughout the US economy: increasing price competition in markets generally meant employers were less able to pass the costs of the premium onto the consumer, so that pressures for wages to conform to the market rate grew. Secondly, unionized companies faced greater nonunion competition. Declining union density, by increasing employers’ opportunities to substitute non-union products for union products, fuelled this process. So too did rising import penetration: if imports are non-union goods, then regardless of union density in the US, they increase the opportunity for non-union substitution.

15

06/06/2003 12:41 PM

These same pressures also increased the employment price of any union wage gap (what economists call the elasticity of demand for unionized labor). 5. Industry and state-level wage premia So far, we have focused primarily on union wage effects at the level of the individual and the whole economy. However, the literature on the origins of the union wage premium focuses largely on firms and industries. This is because the conventional assumption is that unions can procure a wage premium by capturing quasi-rents from the employer (see Blanchflower, Oswald and Sanfey, 1996 for more on this). If this is so, there must be rents available to the firm arising from their position in the market place, and unions must have the ability to capture some of those rents through their ability to monopolize the supply of labor to the firm. Individual-level data can tell us little about these processes. Instead, the literature has concentrated on industry-level wage gaps. In this section we model the change in the union wage premium at two different units of observation - the industry and state. 5.1. Industries. As we noted above FM reported estimates by the extent of industry unionism. They (1984: 50) comment on substantial variation in the union wage effect by industry, with gaps ranging between 5% and 35% in the CPS data for 1973-75. FM’s results are reported in the first column below. We used our data to estimate separate results by two digit industry for 1983-1988 and 1996-2001. We chose these years as it was possible to define industries identically using the 1980 industry classification. Using these data we also found considerable variation by the size of the wage gap by industry. Full results are reported in Table 8 which are summarized in four bands below. Estimates by industry FM 1973-5 1983-88 1996-‘01 <5% 13 11 10 5-15% 17 15 19 15-35% 24 12 12 >=35% 8 6 3 # industries 62 44 44 It does appear that there is less variation in the wage gap by industry in the later period than in the earlier period with only 3 industries, construction (41%), transport (36%) and repair services (37%) having a differential of over 35%, compared with 6 in the earlier period which includes the same three - construction (52%); transport (44%); repair services (37%) – plus agricultural services (41%); other agriculture (56%) and entertainment (47%). Where is the union wage premium rising, and where is it falling? Analysis of trends in the premium by industry reveal a very different picture to the aggregate level analysis, or indeed analyses by demographic characteristics, which show uniform decline. Instead, in four of the ten (39%) industries we analyzed, the union wage gap rose. The details are presented in Table 8. It shows the regression-adjusted wage gaps in [44] industries during the 1980s (1983-1988) and then in the late 1990s (1996-2001). The wage gap rose in 17 industries and declined in 27. The

16

06/06/2003 12:41 PM

gap rose by more than ten percentage points in autos (+12%) and leather (+19%). It declined by more than twenty percentage points in other agriculture (-33%) retail trade (-20%) and private households (-29%). The decline in the wage gap for the whole economy, presented earlier, is due to the fact that the industries experiencing a decline in their wage gap make up a higher percentage of all employees than those experiencing a widening gap. The results are similar to those presented by Bratsberg and Ragan (2002). They find that, over the period 1971-1999, the regression-adjusted wage gap closed in 16 industries and increased in 16 others Their analysis is not directly comparable to ours, but where industry-level changes are presented in both studies, they tend to trend in the same direction. Only in one industry (Transport Equipment) do Bratsberg and Ragan report a significant increase in the wage gap where we find a decline in the wage gap. 24 We decided to explore further the determinants of changes in the union wage premium by industry over time. (In the section that follows we perform an identical exercise where the unit of observation is the state). We then ran 855 separate first-stage regressions, one for each of our 45 industries in each year from 1983-2001 with the dependent variable the log hourly wage along with controls for union membership, age, age squared male 4 race dummies, the log of hours and 50 state dummies. The sample was restricted to the private sector and excluded all individuals with allocated earnings. Three sectors with very small sample sizes (toys, tobacco and forestry and fisheries) were deleted. We extracted the coefficient on the union variable, giving us 19 years * 42 industries or 798 observations in all. As we pointed out in the notes to Table 7, in 1989-95 allocation flags are either unreliable (in 1989-93) or not available (1994 through August 1995). For 1989-93 for each state the wage gaps were adjusted upward by the average imputation bias during 1983-88 (.031). For 1994-95 each of the wage gaps were adjusted upward by the bias during 1996-98 (.046). The coefficient on the union variable was then turned into a wage gap taking anti-logs, deducting 1 and multiplying by 100 to turn it into a percentage. We used the ORG files to estimate the proportion of workers in the industry who were union members both in the private sector and overall and mapped that onto the file. Unemployment rates at the level of the economy are used as industry specific rates are not meaningful: workers move a great deal between industries and considerably more than they do between states. Appendix Table 1 provides information on the classification of industries used and the average number of observations each year. Regression results are reported in table 9. The number of observations is 756 as we lose 42 observations in generating the lag on the wage premium and the union density variables. Equation (1) of Table 9 treats the percentage change in the union premium as a function of union density in the first period and the size of the wage premium in the first period. The change in the wage premium is negatively associated with the size of the wage premium in the first period: the

24 Bratsberg and Ragan (2002) also use CPS data. But their analysis differs in several ways. First, they assess trends over the period 1971-1999 whereas we present trends over the period 1983-2001. Second, we adjust for wage imputation as recommended by Hirsch and Schumacher (2002) whereas Bratsberg and Ragan do not. Third, specifications producing the regression-adjusted estimates differ somewhat. Fourth, the samples differ. In particular, Bratsberg and Ragan exclude government workers and they present results for some different industries. 19 of our industries do not appear in Bratsberg and Ragan’s analysis, while 4 of their industries do not appear in our analysis. Fifth, their wage premium relates to weekly wages whereas all of our estimates are derived from (log) hourly wages.

17

06/06/2003 12:41 PM

overall union density term, which includes both public and private sector workers, is not significantly different from zero. The result is the same in column 2 when private sector density is used. Adding industry fixed effects makes union density significantly negative in column 3 and again in column 4 when private sector density is used. Columns 5 and 6 report the result of adding the (aggregate) unemployment rate and a time trend instead of the year dummies. As we found above, the unemployment rate enters positively, although with a huge coefficient. The lagged dependent variable and the union density variables are unaffected. Column 7 uses the same list of controls as in column 6 but uses a GLS to estimate an equation with industry specific AR(1) process in the error term and where each observation in the GLS regressions is weighted by the industry observation count of the first step following Bratsberg and Ragan (2002). Column 8 drops the unemployment rate term and the time trend and adds back year dummies. The results using GLS are very similar to those obtained using OLS in the first six columns except there is a big rise in the time trend coefficient, which becomes significantly negative between the OLS specification in (6) and GLS in (7). Contrary to the aggregate time series evidence presented in equation 1 above there is some evidence here of a secular decline in the private sector premium over time. This decline is more apparent below when we use state level data which includes disaggregated data on the unemployment rate. Finally, column 9 estimates the same GLS equation but this time using Bratsberg and Ragan’s data for the years 1973-199925. In all specifications there is clear evidence of regression to the mean, with changes in the wage premium negatively associated with the size of the wage premium in the first period. Thus, those industries experiencing a rising (falling) union wage premium tended to have a low (high) premium at the outset. This confirms similar findings presented by Bratsberg and Ragan (2002). In an exercise similar to the one reported here, which modeled the union premium, using weekly wages, at the level of the industry Bratsberg and Ragan (2002) reported that the premium was influenced by a number of other variables. In particular they found that COLA clauses reduced the cyclicality of the union premium and that increases in import penetration were ‘strongly associated with rising union premiums. They also found some evidence that industry deregulation had mixed effects. Their main equations (their Table 2) did not include a lagged dependent variable or the union density measure that they included in their Table 1. Table 10 reports results using their data for the years 1973-1999 using their method and computer programs that they kindly provided to us: the dependent variable is the level not the change in the wage premium. Column 1 of the Table reports the results they reported in column 2 of their table 2. Column 2 reports our attempt to replicate their findings. We are unable to do so exactly. There are several similarities – we find import penetration both in durables and non-durables, COLA clauses, deregulations in communications and the unemployment rate all have positive and significant effects. We also found as they did that deregulation in Finance lowered the premium. In contrast to Bratsberg and Ragan, however, the inflation rate and the two interaction terms with the unemployment rate were insignificant. In column 3 when we add the lagged density, a time trend and the lagged premium, which are all significant, only the

25 We thank Bernt Bratsberg and Jim Ragan for providing us with their data and helping us with our many queries.

18

06/06/2003 12:41 PM

unemployment rate and import penetration variables are significant26. COLA clauses and deregulation have no effect on the wage premium. Several points are worth making about Bratsberg and Ragan's data file. a) Many industry year cells use very small numbers of observations. Average cell size for the years 1973-1981 was 369 while for the years 1983-1999 it was 4829. Indeed, in the earlier period which was based on data drawn from the March 1971 and May 1973-1981 some of the cell sizes were very small – Tobacco averaged 10 observations (4 in 1976); Petroleum averaged 34 observations while leather averaged 51 observations. Out of 32 industries in the first period, 15 averaged less than 150 observations. In the second period which used data as we do from the ORG files of the CPS, Tobacco averaged 90; Petroleum 281 and Leather 263. Small cell sizes are likely to produce unstable estimates year to year in the size of the union premium. Bratsberg and Ragan have re-estimated their analyses excluding the small industries and find their findings are robust to their exclusion which of course is unsurprising. The problem comes when industries with large cell sizes are omitted. b) Industries defined by Bratsberg and Ragan are very different in size. Some industries are very broadly defined – for example industry 32 Services covers SIC codes 721-900 while Tobacco, for example covers one SIC code #130. Cell sizes in one or two cases are very large. In the first period Services averaged 3407 observations and 56,378 in the second. Food averaged 2196 in the first period and 28,012 in the second. No other industry had cell sizes in the first period of over 700 observations or 12000 in the second period. Bratsberg and Ragan impose weights based on the number of observations obtained at the first stage. Inevitably then imposing the weights will result in the estimates being driven by what happened in industries such as Services and Food. Again, responding to our concerns, Bratsberg and Ragan re-ran their estimates excluding Services and Food and confirmed their original results. It is unclear what happens to the results with other sample exclusions. c) No data exists for 1982 or 1972 hence the lagged premium and lagged density variables Bratsberg and Ragan use for 1983 and 1973 respectively have to use data from 1981 or 1971 or some form of interpolation needs to be used. We therefore decided to re-estimate the equation reported in column 3 of Table 10 first with the weights removed, second with the sample restricted to the years after 1983 – we drop 1983 because of the problem of constructing lagged variables as noted above. We report estimates for the later period using the frequency weights in column 5 and without them in column 6. A number of findings emerge from the unweighted model for 1973-99 in column 4. i) The coefficient on the lagged dependent variable is much smaller without the weights than with them. This is unsurprising given that there is much less likely to be variation in the union

26 The import penetration variables are calculated as the ratio of imports to industry shipments. Bratsberg and Ragan (2002) in their footnote 19 report that they tabulated shipments through 1994 from Feenstra (1996) and thereafter from the U.S. Bureau of the Census, U.S. Merchandise Trade, series FT900 (December) and Manufactures’ Shipments, Inventories and Orders (http://www.census.gov)

19

06/06/2003 12:41 PM

wage gap estimates in industries with large sample sizes that have higher weights in the former case. It isn’t obvious that weights should be used if we regard each industry as a separate observation. In cross-country comparisons which, say compared outcomes for Switzerland, the UK and the US it wouldn’t make a lot of sense to weight by population and thereby make the observation from the US 4.67 times more important than that of the UK and 39.3 times more important than Switzerland27. ii) Lagged union density and the unemployment rate are no longer significant. iii) Deregulation in railroads is now significantly positive whereas the coefficient on the variable for Trucking is significantly negative. iv) The import penetration variables remain significant and positive. When the sample is restricted to the period since 1984 with the weights imposed in column 5 the unemployment variable is positive and significant again. The last deregulation occurred in 1984 and hence we do not report the deregulation variables as they are primarily picking up the industry fixed effects28. The lag of the dependent variable is large in column 5 but declines dramatically when the weights are removed. Otherwise, the results in column 6 are very similar to those in column 5 with the exception that the import penetration variable for non-durables is significant in column 6 whereas it was insignificant in column 5. Increases in import penetration are strongly associated with rising union premiums. We find no evidence that COLA clauses influence the cyclicality of the union premium. The evidence on the impact of union density is more mixed. The dispersion of union premiums across industries has narrowed over time as high premiums have tended to fall and low premiums to rise. Note column 5 suggests a decline in the size of the premium; in column 6 the time trend approaches statistical significance. 5.2. States In the US, unions are geographically concentrated by town, county, district and state. Often towns next to each other differ – one is a union town, the other is non-union. Waddoups (2000) used this interesting juxtaposition of union and non-union zones to estimate the impact of unions on wages in Nevada’s hotel-casino industry.29 Although they share many features, and are subject to broadly similar business cycles, most of the 51 states in the US are comparable in size and economic significance to many countries. They also differ markedly in their industrial structures and unionization rates. Assuming union density proxies union bargaining power, this implies different premiums across states. However, as noted earlier, FM found the union premium at regional level was inversely correlated with 27 According to the 2002 Human Development Report Table 5 (http://hdr.undp.org/) the population of the US in 2000 was 283.2 million compared with 60.6 million in the UK and 7.2 million in Switzerland. 28 We are grateful to Bernt Bratsberg for this point. 29 He finds wages of highly unionized occupations in Las Vegas’s hotel and gaming industry are significantly higher than wages of identical occupations in less unionized Reno.

20

06/06/2003 12:41 PM

union density, with the premium highest in the relatively unorganized South and West. To explore this issue further, and to assess changes over time, we estimated separate wage gaps for two time periods at the level of the 50 states plus Washington DC. Results at the level of the state are reported in Table 11, and are summarized in the table below. The data used are from the Outgoing Rotation Group files of the CPS. It was not possible to identify each state separately in the May CPS so FM did not report such results. Hence, we compared results from a merged sample of the 1983-1988 with those obtained from our 1996-2001 files. Estimates by state 1983-8 1996-01 <5% 0 0 5-15% 6 21 15-35% 43 30 >=35% 2 0 # states (+DC) 51 51 The first thing we notice is that the variation in the union wage premium is much less by state than it is by industry. Only two states in the earlier period had gaps of at least 35% - North Dakota (35%) and Nebraska (37%) and none in the later period. There has been a downward shift in the premium generally, as indicated by the movement from the 15-35% category to the 5-15% category. The mean state union wage gap was 23.4% between 1983 and 1988, falling by 6.2 percentage points to 17.2% in 1996-2001. The premium fell in all but five states, with South Dakota recording the biggest decline (16.8 percentage points). In four of the five states where the premium rose, it only increased by a percentage point or two (Vermont, Massachusetts, Wyoming, Hawaii). The premium only rose markedly in Maine, where it increased 9 percentage points (from 7% to 16.1%). Since the early 1980s, union density fell by an average of 5.7 percentage points, with Pennsylvania (-10.6%) and West Virginia experiencing the biggest decline (-11 percentage points). The premium appears to have declined more in smaller states than it has in bigger states. It is apparent from Table 11 that the five biggest states of California, Texas, Florida, New York and Illinois had small changes in their wage gaps (-1.4%; -6.7%; -10.1%; –0.6% and –4.5% respectively). For example the five smallest states measured by employment tended to have big declines in the differentials New Mexico (-14.10%); Alabama (-14.20%); Nebraska (-15.00%); Arkansas (-15.20%); South Dakota (-16.80%). We then ran 969 separate first-stage regressions, one for each state in each year from 1983-2001 with the dependent variable the log hourly wage along with controls for union membership, age, age squared, male, 4 race dummies, the log of hours and 44 industry dummies. The sample was restricted to the private sector and excluded all individuals with allocated earnings. We extracted the coefficient on the union variable, giving us 19 years * 51 states (including D.C.) or 969 observations in all. As we did for the industry level analysis described above for 1989-93 for each state the wage gaps were adjusted upward by the average imputation bias during 1983-88 (.031). For 1994-95 each of the wage gaps was adjusted upward by the bias during 1996-98 (.046). We then mapped to that file the unemployment rate in the state year cell30 plus the

30 Source: http://data.bls.gov/labjava/outside.jsp?survey=la

21

06/06/2003 12:41 PM

proportion of workers in the state who were union members31. Data on union density rates are available for all workers and then separately for the private sector, public sector, private construction and private manufacturing. Once again we ran a series of second-stage regressions where the dependent variable is the one year change in the premium (obtained by taking anti-logs of the union coefficient and deducting one) on a series of RHS variables including the lagged premium and lagged unemployment and union density rates32. Results are reported in Table 12 – the number of observations is 918 as we lose 51 observations in generating the lag on the wage premium and the union density variables. Each column of the table is a separate regression33. Both unweighted and weighted results are presented where the weights are total employment in the state by year. The main determinants of the change in the premium are the lagged premium which enters negatively and significantly everywhere, confirming the time series results reported earlier. The higher the starting level of the premium, the more likely it will decline, and vice versa. The lagged unemployment rate enters positively and significantly once the state fixed effects are included. This confirms FM’s finding that the premium moves counter-cyclically – high when the economy is in recession and the unemployment rate is high and low in booms when unemployment is low. Lagged union density in the state – whether for the private sector or the rate for the public and private sectors combined – enters negatively and significantly with or without state fixed effects. Farber (2003) argues that there remain potential unobserved variables which simultaneously determine density and wages, but which are time-varying, and thus not picked up in fixed effects, which might explain such a (perhaps) counter-intuitive finding. An alternative possibility is that state-level density is not a meaningful measure for bargaining power. Indeed, one may assert that, where so many other workers – including comparable workers – lack the bargaining power offered by a union – it is no surprise that the few with bargaining power will gain significantly – and perhaps more so than unionized workers in a generally more unionized environment. Perhaps surprisingly the overall density variable in columns 8 and 9 works better than private sector density. Public sector density, private construction or private manufacturing densities were always insignificant whether entered alone or together (results not reported). We also found no role for the change in density variables defined either using overall density or using private sector density .

31 The source of the data is the Union Membership and Coverage Database which is an Internet data resource providing private and public sector union membership, coverage, and density estimates compiled from the Current Population Survey (CPS) using BLS methods. Economy-wide estimates are provided beginning in 1973; estimates by state, detailed industry, and detailed occupation begin in 1983; and estimates by metropolitan area begin in 1986. The Database, constructed by Barry Hirsch (Trinity University) and David Macpherson (Florida State University), is updated annually. The Database can be accessed at http://www.unionstats.com. 32 We experimented with both the level of the unemployment rate and the log and the latter always worked best. 33 We estimated the impact of RTW laws on change in the union wage premium by adding a RTW dummy to the model presented above. It is not significant and makes no difference to the results: the lagged premium remains negative and significant. In part this likely arises because few states change their Right-to-Work laws over this period. Only Texas who introduced the law in 1993 and Idaho which became an RTW state in 1985 and Oklahoma in 2001 changed status.

22

06/06/2003 12:41 PM

The finding that lagged unemployment has a positive impact on the wage premium confirms the weaker findings at the industry and aggregate levels which were based on unemployment rates at the economy wide level. Here we have variation in unemployment across both state and year. The union wage premium is counter-cyclical – high in slumps and low in booms. Union wages are more rigid than non-union wages. High union density in period t-1? in the state causes the wage premium to decline. Low union density in period t-1 in the state causes the wage premium to rise. As we found when we used industries as the unit of observation, the dispersion of union premiums has narrowed over time as high premiums have tended to fall and low premiums to rise. There is now further evidence in Table 12 that the private sector premium has trended downwards. There was relatively weak evidence for this at the industry level in Table 9 but none at the aggregate level in section 4 above. The addition of the weights in columns 6, 7 and 8 does reduce the size and significance of the time trend. 6. Explanation for the findings Why are there such stark differences in the industry-level wage premium and trends in the premium measured other ways? Freeman and Medoff (1984: 50) comment on substantial variation in the union wage effect by industry, with gaps ranging between 5% and 35% in the CPS data for 1973-75. This variance continued to characterize the US economy by the late 1990s. FM (1984: 50-52) identified three economic forces underlying the vastly different effects of unions on wages in different industries: union monopoly power; employer market power; and the level at which bargaining occurs. We consider each in turn. In addition, we consider the roles played by employer opposition to unions and union effects on non-union wages. 6.1. Unions’ monopoly power FM (1984: 50) considered unions’ monopoly over the supply of labor to be the ‘principal source’ of industry variance in the wage premium. This power relates to the wage sensitivity of the demand for organized labor, that is, the change in employment induced by a given change in wages. Where employment is relatively insensitive to a rise in industry wages, unions will be less constrained in pushing for higher wages. Thus, other things equal, one would expect the largest union wage gap in these industries. Employment is least likely to be responsive to union-induced wage rises where there are few non-union producers competing with unionized employers. This will be the case where the majority of producers are unionized, a situation often proxied in empirical research by high industry-level union density. 6.2. Increasing competition Some have argued that the convergence in premia is a natural consequence of increased competitiveness across the US economy (Hirsch and Schumacher, 2001: 501). As FM note (1984: 54) some employers can afford to pay a union wage premium due to their privileged or protected product market positions. This status allows them to make ‘monopoly profits’ which they can share with labor in the form of a union wage premium. If increased competition

23

06/06/2003 12:41 PM

reduces the rents employers can share with labor by lowering entry barriers and introducing price competition, this may well explain why the wage premium has declined in some industries. The problem with the standard competition story in explaining convergence in industry wage gaps is that, although it may offer a good explanation as to why large premiums are less sustainable than in the past, it does not explain the increase in premiums in industries with previously lower premiums. However, recent studies identify ways in which competition can bring about increases, as well as reductions, in the premium.

It is not easy to measure changes in competition in the economy, but the deregulation of highly regulated markets offers a ‘natural experiment’ allowing us to establish how the union wage premium responds to heightened competition. It is usually assumed that deregulation results in the removal of entry barriers and raises price competition, thus lowering rents that unions target. Case studies indicate this does occur in some industries, such as trucking (Hirsch and Macpherson, 1998). But in other industries such as railroads (Hendricks, 1994), the opposite happens. This is partly due to the structure of these markets. In trucking, entry barriers are low, allowing for new entrants to come in offering low wages, something that is less likely in railroads where entry barriers are high. Rising premiums after deregulation occur because in some cases regulation can create no rents for unions, or even dissipate them. The classic example is the regulation of services to the public where franchises must maintain unprofitable lines or routes as part of public service. Hirsch and Macpherson (1998, 2000) reach similar conclusions contrasting trucking and airlines. In the latter, regulation had protected wages of non-union workers and, with deregulation, the power of unions meant that union wages fell less rapidly than the higher non-union wages. These results from case studies are confirmed by Bratsberg and Ragan’s (2002) CPS analysis. They find deregulation effects differ by industry, raising the premium in telecommunications and lowering it in finance.34 We also find mixed results, as presented earlier. Thus, deregulation can explain divergent trends in the wage gap for a sub-set of industries, with the direction of the impact depending on factors like who received the rents in the regulated environment and the effect of deregulation on union power.

Other measures of increased competition are more difficult to obtain and results are less robust. However, import penetration is a good proxy for competition in the traded goods sector. Although the impact of imports on the US wage distribution is often overstated (Blanchflower, 2000a), it may be expected to play a role where imports permit substitution of union products for non-union products. If imports reduce demand for domestic output and, in turn, demand for labor, this should reduce union and non-union wages (assuming the supply of non-union labor is not perfectly elastic). Whether the premium rises or falls with increased import penetration depends on the relative responsiveness of union and non-union wages to demand shifts resulting from foreign competition. Table 8 shows that many of the industries experiencing a rise in the union premium between 1983 and 2001 would have been subject to intensifying international trade (Machinery, Electrical Equipment, Paper, Rubber and Plastics, Leather) but this is equally true for those experiencing declining premiums (such as Textiles, Apparel and Furniture). Horn (1998) found that increases in import competition led to increases in union density and decreases in the wage premium within manufacturing industries. This occurred because union membership fell slower that 34 Their other results, though not statistically significant, are consistent in direction with the case study literature, indicating positive trends for railroads and airlines and a negative effect for trucking.

24

06/06/2003 12:41 PM

overall employment when faced with import competition. Horn also found that imports from OECD countries decreased union density, while imports from non-OECD countries tended to raise union density within an industry. In Table 10, we find import penetration in durable and non-durable goods sectors increases the premium, suggesting union wages are more resilient than non-union wages to foreign competition. Bratsberg and Ragan (2002) present panel analyses also showing increased imports raise the union premium. However, their repeat cross-section analyses (excluding industry fixed effects) indicate a negative relationship between imports and the premium. The implication is that import penetration is correlated with unmeasured industry characteristics that depress the premium inducing a negative bias that is removed once industry characteristics are controlled for. They conclude that import penetration has reduced demand for union and non-union labor, with union wages holding up better than non-union wages, but at the expense of reduced union employment. There are theoretical and empirical reasons as to why this might occur. For instance, since union wages tend to be less responsive to market conditions generally, it may be that union wages are sluggish in responding to increased competition from imports. Alternatively, industries characterized by ‘end-game’ bargaining may witness perverse union responses to shifts in product demand as the union tries to extract maximum rents in declining industries (Lawrence and Lawrence, 1985). Another possibility is that increased import penetration reduces the share of union employment in labor intensive firms, increasing it in capital intensive firms. Greater capital intensity reduces elasticity of demand for union labor, allowing rent maximising unions to raise the premium (Staiger, 1988). 6.3. Changes in bargaining structures

An industry’s bargaining structures may affect unions’ ability to negotiate a wage premium. Since the early 1980s when Freeman and Medoff were writing, industry-level wage bargaining has disappeared in the private sector, with plant- or organization-level bargaining all-pervasive. Plant-level bargaining reduces unions’ monopoly power because, as noted by FM (1984: 51), workers and managers in a plant will worry that wage increases in their plant may shift product demand, and hence jobs, to other plants. One might expect the biggest declines in the wage premium in industries where strong industry-level bargaining arrangements have disappeared. In the public sector, which is still characterized by industry-level bargaining, the premium has remained roughly constant. 6.4. Employer attitudes towards unions

Another explanation for convergence in industry wage gaps is the impact of employer opposition to trade unions. It seems probable that, above a certain point, employer opposition to unionization will harden as the wage premium rises (Blanchflower and Freeman, 1992: 70-71).35 It is therefore conceivable that, where unions have established a sizeable premium, they will face concerted efforts by employers in the next period to keep the premium down – a factor which will not come into play with those starting from a low base. Although we have no direct evidence of employer attitudes towards unions by industry over time, it is reasonable to assume

35 Blanchflower and Freeman (1992) suggest union density and the premium will rise together initially, driven by increasing desire for membership as the premium rises. However, once employer incentives to oppose unions rise at a faster rate than worker incentives to organize, so union density begins to decline.

25

06/06/2003 12:41 PM

that employer support – or at least acquiescence – will be highest in the sector where union density has remained consistently high throughout the period, namely the public sector. Here the premium, although smaller than for the private sector as a whole, has remained roughly constant such that the gap between the union premium in the two broad sectors has closed from 10.2 log points in 1977 to 2.7 log points in 2001 (Table 2). Consistent with a declining premium would then be declining employer and worker opposition to unions and this is what we find.

6.5. Employee perceptions of trade unions As well as unions being able to extract a wage premium because they wield bargaining power emanating from their ability to monopolize the supply of labor to employers, unions may also affect wage outcomes through power in the political process which helps them influence the economic and legislative framework in which bargaining occurs. These two facets of union power are nicely captured in an alternative measure based on citizens’ perceptions of union power. Table 13 includes responses to the question “Do you think that trade/labor unions in this country have too much power or too little power?” Options available were a) far too much power b) too much power c) about right d) too little power e) far too little power. Similar questions are also asked in relation to ‘business and industry’ and the federal government. The table reports the percentage saying ‘far too much power’ in five countries – Australia, West Germany, Great Britain, the USA and Italy. Data are obtained from three sweeps of the International Social Survey Program (ISSP).36 It is clear from the top panel that union power in the US is perceived to have fallen precipitously since the mid-1980s, in common with 3 of the other 4 countries shown in the table. Furthermore, this has occurred at a time when the perceived power of business and industry has been stable (second panel), and the power of federal government has been on the increase (third panel). So, since the time FM were writing, the relative strength of the three main players in the industrial arena – unions, employers and government – has moved to the detriment of unions. All the evidence is that employees view the improvement of pay and conditions at work as unions’ core business. For example, in 1994 - just before the union wage premium began to fall - Freeman and Rogers (1999: 79) found union members and non-union employees identified ‘better pay/working conditions’ as “the most important thing a union does for its members”. Forty-eight per cent of union members gave this response, 17 percentage points ahead of the second most popular choice ‘more respect/fair treatment on job’. The 17 percentage point gap is also apparent for non-union employees (39% citing ‘better pay’, 22% citing ‘more respect’). What is more, 72% of union members expressed satisfaction with the say they and their fellow members had in union decisions about bargaining about wages and benefits (1999: 78). Presumably the decline in the premium, even if cyclical, by signaling a fall in the gross benefits of unionization, should have been accompanied by declining support for unionization? Well, no. Figure 2 shows support for unions has risen since the early 1990s among the general public. Figure 3 is even more compelling: it shows that, since the 1980s support for unions has been

36 For details see http://www.issp.org.

26

06/06/2003 12:41 PM

growing among non-members and, in 2002, the proportion wishing to vote for a union exceeded the proportion wishing to vote against for the first time. It is clear that, despite recent difficulties in procuring a wage premium for members, support for unions is growing, suggesting that workers look for more than simply what is in their wage packet. Perhaps this is heartening for unions wishing to fill the representation gap. According to the AFL/CIO more non-union workers in non-supervisory roles would vote for a union today than was true 15 years ago37. When Americans who don't have a union are asked how they would vote if a union election were held in their workplace tomorrow, 43 percent say they would definitely or probably vote for a union. This percentage has increased by nearly half over the past 15 years. In 1984 only 30% of such workers said they would vote for a union. 6.6. The impact of unions on non-union wages. So far we have focused on the impact of union density on the wage gap through its link to union bargaining power. However, union density can also influence wage setting in the non-union sector in three ways. First, non-union firms may raise their wages above market rates to keep unions out (the ‘threat’ effect). Where union density is falling (rising) the ‘threat’ of unionization falls (rises), so that the effect should lower (increase) non-union wages, thus increasing (lowering) the union wage premium. Second, if the union premium leads to job loss in the union sector, displaced workers seeking work in the non-union sector will lower the wages offered in the non-union sector (the ‘spillover’ effect). Where union density is falling (rising), the spillover effect diminishes (increases), thus increasing (reducing) non-union wages which, in turn, reduces (increases) the wage gap. Third, as FM (1981) note, falling (rising) density can reduce (increase) non-union wages if this density fall (rise) reduces (increases) the relative cost of union labor by shifting demand for labor in favor of (against) union producers. The impact of a change in density on the union wage premium depends on the relative magnitude of the likely positive impact on union wages and the positive or negative impact on nonunion wages. There is only patchy evidence regarding the relative importance of these three factors on the wage gap by industry. Hirsch and Schumacher (2001: 501) suggest that there are substantial threat effects at industry level, whereas spillover effects are weak. Freeman and Medoff (1981) suggest spillover effects will exceed the impact of demand shifts where labor is immobile, but the opposite will be the case if labor is geographically mobile. One might infer which wage determination mechanisms are most important by industry from the relationship between density change and trends in the premium. Where industries face a declining premium and declining density, this implies that it is an industry where spillover effects dominate threat effects. The premium falls here because the pressure to raise non-union wages resulting from the reduction in spillover effects dominates the pressure to lower non-union wages arising from the reduction in the threat effect. Of course, if the density effect on union bargaining power predominates, the decline in this power and its impact in lowering union wages, may be all that is required to explain the narrowing gap between union and non-union wages. Conversely, where industries face a rising premium and declining density, this implies that it is an industry where threat effects dominate spillover effects. The premium rises here 37 Downloadable from the AFL/CIO at http://www.aflcio.org/mediacenter/resources/1999labordaypoll.cfm.

27

06/06/2003 12:41 PM

because the pressure to lower non-union wages resulting from a declining threat effect dominates the pressure to higher non-union wages arising from the reduction in the spillover effect. Although we have no direct evidence in our data on the relative importance of threat and spillover effects, the fact that we observe a rising premium in the presence of falling density in nearly half our industries suggests that threat effects in these industries could be very important. Farber (2003) argues industry-level density is a poor proxy for the threat effect because what matters is the marginal probability of unionization, something union density may not capture. Using three alternatives (individuals’ predicted probabilities of unionization, changes in states introducing RTW legislation, and changes in deregulated industries) he finds mixed evidence of the existence and trends in threat effects. Most interesting are the results relating to changes in the threat effect associated with deregulation in air, trucking and telephones. Running wage regressions with fixed effects for state, year and industry, he compares non-union (union) earnings pre- and post-deregulation in the affected industries with non-union (union) earnings elsewhere. He finds evidence for threat effects in telephones and airlines, where non-union wages fell while union wages did not (and actually rose). In trucking, non-union wages fell – consistent with a threat effect, but union wages also fell.38 6.7. The impact of compositional change. There are two ways in which compositional change in the workforce could affect the size of the aggregate union wage premium. First, compositional change within the union sector could result in a rise or fall in the aggregate premium depending on which unionized sectors survive and prosper. Obviously, if unions with smaller premiums prosper relative to those with bigger premiums the aggregate premium will fall, and vice versa. The second way that workforce compositional change plays a part is through alterations in the nature of unionized employment relative to non-unionized employment. Such changes may play a role in explaining trends in the unadjusted wage gap. The unadjusted wage gap has been higher than the regression-adjusted gap throughout the period indicating that union members possess attributes, or have jobs or are employed in workplaces that raise their wages, regardless of their union membership status. However, the unadjusted wage gap has fallen relative to the regression-adjusted wage gap over time, indicating that the advantages that union members possess have diminished over time (Table 7b). 7. What have we learned and would FM have been surprised about these results when they wrote WDUD? 7.1. The private sector wage premium is lower today than it was in the 1970s. This would not have surprised FM. Indeed, they predicted that the premiums of the 1970s were unsustainable due to impact on union density (FM 1984: 54). Perhaps it is surprising that the premium remains as high as it does?

38 The results for airlines and communication are in line with earlier work. Earlier case studies indicated a fall in the premium in trucking but Farber’s finding of a negative non-significant effect is in line with Bratsberg and Ragan (2002).

28

06/06/2003 12:41 PM

7.2. The union wage premium is counter-cyclical. The decline in the premium in good times is what seems to explain the decline in the premium since the mid-1990s. Far from being a surprise to FM, they identified the counter-cyclical nature of the premium. We show the premium is counter-cyclical at the economy, state and industry levels. FM said COLA’s could dampen counter-cyclical movements in the premium, but thought their significance had been overplayed. This is confirmed: we find little or no COLA effect, though this finding is contested by others. 7.3. There is some evidence of a secular decline in the union wage premium. There is some evidence, primarily at the level of the state but also using industry level data, of a downward trend in the private sector union wage premium accompanying the marked decline in union presence in the private sector. Interestingly the wage gap appears to have declined most in the smallest states (e.g. New Mexico, Alabama, Nebraska, Arkansas, South Dakota) and declined least in the bigger states (e.g. California, Texas, New York; Illinois). It would not have been a surprise to FM that there had been some reduction in the ability of unions over time to raise wages as the proportion of the workforce they bargain for has declined. 7.4. There remains big variation in the premium across workers. Patterns in the premium across worker types resemble those found by FM. The FM sub-sample generally overstated the size of the premium in the population as a whole but we suspect this would not have surprised FM. A decline in the premium over time seems to have occurred for all types of worker in the private sector, but there is regression to the mean, the biggest losers being among the most vulnerable and certainly the lowest paid workers (the young, women, high school dropouts). But perhaps the real surprise is just how large the premium still is for some of these workers (26% for high school dropouts, 19% for under 25s). One puzzle is the remarkable rise in the share of employment taken by the highly qualified, yet they continue to receive the lowest union premium. Why? Do they look for something else from their unions, e.g. professional indemnity insurance or voice, or are their unions less effective? In contrast to the private sector, the public sector has experienced a small increase in the premium, an increase apparent for most public sector employees. It is a sector where industry-level bargaining remains the norm, maintaining union bargaining power. So, perhaps FM would not have been surprised by this result. We look at two groups of workers that FM did not consider: immigrants and the self-employed. The premium varies little by year of entry to the US, but does depend on the country of birth, with Mexicans benefiting from the highest premium. Whether this reflects the human capital, occupational mix, or costs of immigration faced by different groups is a matter for further

29

06/06/2003 12:41 PM

research, but we suspect the results would not have surprised FM. The absence of a union premium among the self-employed is also unlikely to have surprised them, since the self-employed are more likely to join unions for non-wage related benefits such as indemnity insurance. 7.5. There is big variation in industry-level union wage premia. FM also found wide variation in industry-level premiums, and might have expected this to persist because unions’ ability to push for a premium, and employers’ ability to pay, is determined by industry-specific factors (such as union organization and the availability of non-union labor, regulatory regimes, bargaining, product market rents). However, we find convergence in industry-level premiums since FM wrote, that is, a falling premium where it was once large, and a rising premium where it was once small. (Overall decline at the economy level is due to the fact that the former constitute a larger share of employment than the latter). FM may well have been surprised by this regression to the mean because in 4/10 industries we examine there has actually been a rise in the premium. This contrasts starkly with analyses at the level of worker types and state, which show almost universal declines in the premium. What accounts for it is industry-specific changes that ‘unravel’ the starting position. For example, we show as do Bratsberg and Ragan (2002), that the premium can both rise and fall with industry deregulation. Where unions were originally strong in the regulated period, deregulation can increase worker competition for the rents of employers, placing downward pressure on the union wage premium. New market entrants can also hit employer rents and thus their ability to pay a premium. Conversely, deregulation can sometimes raise the union wage premium. It can do so by opening up opportunities for unions to extract rents that were previously ‘regulated away’, or else can subject all workers in a sector to downward wage adjustment, an adjustment which organized workers are best placed to combat. Other possible factors are: a) Industry-specific changes in bargaining structures, the expectation being that, because as FM note, union monopoly power is highest with industry-level bargaining, industries facing the biggest premium falls will include those shifting away from industry bargaining to organization or plant level bargaining. b) An inverse relationship between employer preparedness to concede a high union premium in the next period and the premium in the current period, as suggested by Blanchflower and Freeman (1992). c) Industry-specific differences in the importance of union wages for non-union wage setting: a rising premium alongside falling density in nearly half our industries suggests threat effects are more important than FM originally suggested. 7.6. State level union wage premia vary less than industry level premia.

30

06/06/2003 12:41 PM

FM did not explicitly compare variations in the premia at the state and industry levels. Richard Freeman expressed the view to us that it made sense to him that there would be more variation at the industry level as they are more closely approximated to markets. No surprise here. 7.7. Higher union density is associated with a lower union wage premium. Although density was not significant in estimating the economy-level wage premium and the evidence is a little mixed at the industry-level, we found union density is negatively associated with the size of the union wage premium and growth in the wage premium at both the state and industry level. Despite the evidence they present on the inverse relationship between density and the premium at regional level, this result might have surprised FM who argued higher union density increases union bargaining power and, because density has a greater impact on union wages than it does on non-union wages (through threat effects, spillover effects, and shifts in labor demand), it is the bargaining power effect which should predominate. So why does it turn out differently? The key issue is: does union density determine the size of the wage premium, or does the size of the wage premium determine density? It is likely that they interact with one another in the following ways. First, a union’s bargaining power is likely to rise with the proportion of employees it counts among its members. Consider the issue at the level of the employer, since this is usually the level at which bargaining occurs. Where few have joined the union the employer can choose to wholly ignore the union and set wages unilaterally. Even if the union has won a vote by gaining support from at least half the bargaining unit, the employer is more likely to have regard to the union’s position where it commands support from the vast majority of employees, rather than a bare majority. This is because, to the extent that the union controls the supply of labor to the firm, failure to agree terms may disrupt production, sales and profits. (It may be that the union only covers a sub-set of workers employed by the employer, for example, because it is a craft union. If so, what will matter is the proportion of craft workers organized by the union, rather than employer-level union density). Of course, if there is plentiful suitable non-union labor outside the firm, the employer may have no difficulties in circumventing the union, irrespective of union density within the firm. This points to the importance of union density in the labor market which the employer operates in. In practice, this is often proxied by union density in the employer’s industry. Where industry-level union density is high it is hard for the employer to find non-union labor. This will strengthen the union’s bargaining hand, but it will also work to the advantage of the employer. This is because, even if the employer concedes higher wages to the union, it is less likely that the employer will lose market share through undercutting by non-union producers. In other words, the price elasticity of demand for labor falls where industry density is high. In such circumstances, a high wage premium is compatible with high union density. The situation is likely to change where the threat of competition from non-union producers starts to rise. This may occur with a decline in industry-level union density, or with an increase in foreign non-union competition, two trends in the US economy. In these circumstances, demand for unionized labor will become more price-sensitive so that, if unions continue to hold out for a sizeable wage premium, it is likely that the union’s share of labor will fall as it is replaced by

31

06/06/2003 12:41 PM

cheaper non-union labor. Of course, there will continue to be product markets that remain largely immune to price competition, so that employers may continue to share their rents with union labor. There will be other circumstances where unions chose an ‘end game’ bargaining strategy, whereupon they will continue to hold out for a premium in the knowledge that the industry is in decline in any event. 7.8. Union workers remain better able than non-union workers to resist employer efforts to reduce wages when market conditions are unfavorable. This was cited as one reason for the counter-cyclical premium by FM (1984: 52-53). We find further evidence in our industry-level analysis where import penetration in the durables and non-durables sectors has raised the union wage premium, implying union wages are more resilient to foreign competition than non-union wages. 7.9. There has been a decline in the unadjusted wage gap relative to the regression-adjusted wage gap. Union members do have wage-enhancing advantages over non-members, but these have diminished in recent years, implying changes in the selection of employees into membership. Unlikely FM would have predicted this. 7.10. The social costs of unions remains unclear. FM found the union premium effect on output was modest, and inflation effects were negligible. Here we take an alternative approach in considering the social costs of the union premium by addressing the fundamental question of where the premium comes from. It still remains unclear what the source of the premium is. FM assume that it necessarily originates in the monopoly rents of employers in privileged market positions. What are the alternatives. These are: a) unions increase the size of the pie because unionized workers are more productive than like

workers in a non-unionized environment, b) unions operate in firms with excess profits arising from a privileged market position (this

may arise either because unions only seek to organize where there are excess profits available, or because these sorts of employers have something particular to gain in contracting with unions),

c) the premium is simply a tax on normal profits. It really matters which of these three options it is. If a) is true then there’s no implication for workplace survival or union density, indeed, if this were the case, one might expect a growth in unionization as firms recognize the advantages in unionizing. If it’s b) then there is limited damage, but with c) there are really problems for firms, with potential knock-on effects for investment, jobs and prices. Of course, different employers may be in different positions at any

32

06/06/2003 12:41 PM

point in time and the weight attached to the three options may differ over time with the business cycle, structural change in the economy and so on. What evidence is there on the above? i) FM (1984: 54) speculate that, at least in some areas, the wage gap has ‘reached levels inconsistent with the survival of many union jobs’. Blanchflower and Freeman (1992) take this issue further and argue that the levels of the union wage differential were so high this gave incentives to employers to remove the union – the benefits of removing the unions appeared to outweigh the costs. Well, what has happened to union jobs? There is a growing body of evidence that employment growth rates are lower in the unionized sector, suggesting b) and or c). The evidence is for the U.S. (Leonard, 1992), Canada (Long, 1993), Australia (Wooden and Hawke, 2000) and Britain (Blanchflower, Millward and Oswald, 1991; Bryson, 2001). There is little evidence that union bargaining has resulted directly in workplace closure, and Freeman and Kleiner (1999) and DiNardo and Lee (2001) show no link between unionization and closure. Indeed, FM record instances in which union workers accepted cuts in normal wages (‘givebacks’), sometimes to keep employers in business. Can employers still obtain wage concessions from unions by citing survival threats? United Airlines is an interesting case study: it got into difficulties and became an employee-owned company; unionized employees withheld demands for traditional pay increases for 5 years in return for share options but, when they resumed bargaining, the company went under because no remedial action was taken in the interim to treat the underlying problems in the business. Recently the pilots agreed to substantial pay cuts. Cheaper labor can give a competitive advantage but this is not always enough for firms to survive and, where there is a union in place, it will want to revert to bargaining on wages at some point. ii) Unionization is more common where employers operate in monopolistic or oligopolistic product markets, suggesting that unions do try to extract surplus rents. iii) The premium falls when one accounts for workplace heterogeneity. The implication is that some of the premium usually attached to membership is, in fact, due to unionized workplaces paying higher wages than non-unionized workplaces. If this is so, why? Perhaps unions target organization efforts on workplaces that have rents to share, as noted above, they would be foolish not to. Or employers use unions as agents to deliver lower quit rates to recoup investment in human capital (which makes them more productive/profitable). Abowd, Kramarz and Margolis (1999) say higher paying firms are more profitable and/or more productive than other firms. If this is so for unionized firms, they are operating at a higher level of performance than other firms. It makes sense that unions are located in higher performing workplaces since, despite higher wages and a negative union effect on performance, they continue to survive. A central thesis of WDUD was that unions were more productive so this would not be a surprise to FM. iii) The evidence that unions have a substantial negative impact on employment growth suggests that the social cost of unions may be larger than FM calculated. If the premium reduces the competitiveness of unionized firms, they’ll lose employees and, as a consequence, union organizing will get tougher for unions. This is exactly what has happened.

33

06/06/2003 12:41 PM

On the other hand, if unions do not command a premium, they lose their best selling point for prospective customers. It’s Catch 22. And then finally we asked Richard Freeman whether he was surprised by any of these findings. He kindly read the paper and told us, a) he was most surprised by the fact that the public sector wage effects are so large and so similar to those in the private sector, b) that we know little about the social costs of unions in the twenty first century. Consequently he was unsure about the magnitude of the social costs, but would like to see empirical work on the issue although he said would be ‘stunned’ if there were large effects, but as any good empiricist he would let the data speak, c) he was not surprised by any of our other findings. So there you have it!

34

06/06/2003 12:41 PM

Table 1. Union membership rates 1973-2002 1973 1977 1983 1987 1992 1997 2001 2002 All 24 24 20 17 16 14 14 13 Private sector 24 22 17 13 11 10 9 9 Manufacturing 39 36 28 23 20 16 15 14 Non-manufacturing 17 16 13 10 9 8 8 7 Public sector 23 33 37 36 37 37 37 38 Federal 19 16 17 17 18 19 Postal 74 74 72 72 68 70 State 28 28 29 30 31 31 Local 42 48 42 43 43 43 Source: Hirsch and Macpherson 2002.

35

06/06/2003 12:41 PM

Table 2. Disaggregated union membership rates, 1977 & 2001 1977 2001 All 24 14 Private sector 22 9 Public sector 33 37 Private sector employees Men 27 12 Women 11 6 Whites 20 9 Non-white 27 10 Ages 16-24 12 4 Ages 25-44 23 9 Ages 45-54 27 13 Ages >=55 22 10 < high school 23 7 High school 25 12 > high school 13 8 North East 24 12 Central 25 12 South 13 5 West 22 9 Agriculture, forestry & Fisheries 3 2 Mining 47 12 Construction 36 19

36

06/06/2003 12:41 PM

Manufacturing 34 15 Transportation, communication, And other public utilities 48 24 Wholesale & retail trade 10 4 FIRE 4 3 Services 7 6 Source: 1977 “What Do Unions Do’. 2001 own calculations from the ORG file of the CPS.

37

06/06/2003 12:41 PM

Table 3. Union Membership and Employment by Sector (‘000’s). Union membership Total employment a) 1983 Private sector 11,933 71,224 Public sector 5,737 15,633

Federal Government 469 2,417 Post Office 523 705 State Government 1,071 3,800 Local Government 3,673 8,711

% union members in 32.5% the public sector b) 2002 Private sector 8,652 100,581 Public sector 7,327 19,379

Federal Government 443 2,382 Post Office 615 874 State Government 1,756 5,654 Local Government 4,514 10,488

% union members in 45.9% the public sector Source: Based on data from Table 1c of Union Membership and Earnings Data Book: Compilations from the Current Population Survey by B.T. Hirsch and D.T. Macpherson, 2003.

38

06/06/2003 12:41 PM

Table 4. Private sector union/non-union log hourly wage differentials, 1974-1979 & 1996-2001 Private sector Freeman/Medoff’s sample 1996-2001 1974-1979 1996-2001 1974-1979 Men 17% 19% 28% 27% Women 13% 22% 24% 27% Ages 16-24 19% 32% 23% 35% Ages 25-44 16% 17% 28% 26% Ages 45-54 14% 13% 27% 22% Ages >=55 16% 19% 28% 29% Northeast 11% 14% 22% 21% Central 15% 20% 27% 27% South 19% 24% 26% 29% West 22% 23% 34% 31% < High school 26% 33% 29% 31% High school 21% 19% 28% 25% College 1-3 years 15% 17% 28% 28% College >=4 years 3% 4% 14% 17% Whites 16% 21% 27% 28% Non-white 19% 22% 30% 28% Tenure 0-3 years 20% 20% n/a 28% Tenure 4-10 15% 16% n/a 19% Tenure 11-15 11% 10% n/a 12% Tenure 16+ 8% 17% n/a 28% Manual 21% 30% n/a n/a Non-manual 4% 15% n/a n/a Manufacturing 10% 16% 19% 19% Construction 39% 49% 45% 55% Services (excl construc.) 16% 34% 29% 43% Private sector 17% 21% 28% 28% Notes: 1996-2001 data files exclude individuals with imputed hourly earnings. Controls for 1996-2001 are 50 state dummies, 46 industry dummies, gender, 15 highest qualification dummies, private non-profit dummy, age, age squared, log of weekly hours, 4 race dummies, 4 marital status, year dummies + union membership dummy (n=546,823). Estimates for 1974-1979 are adjusted upwards by the average bias found during 1979-81 of .033. Controls for 1974-1979 are 9 census division dummies, 46 industry dummies, years of education, age, age squared, log of weekly hours, 4 race dummies, 4 marital status dummies, 5 year dummies + union membership dummy (n=183,881). Tenure estimates for 1974-1979 obtained from the May 1979 CPS and for 1996-2002 files and February 1996 and 1998 Displaced Worker and Employee Tenure Supplements and January 2002 and February 2000: Displaced Workers, Employee Tenure, and Occupational Mobility Supplements. Freeman/Medoff’s sample consists of non-agricultural private sector blue-collar workers aged 20-65 (n=64,034 for 1974-1979 and 142,024 for 1996-2001).

39

06/06/2003 12:41 PM

Table 5. Union wage differentials in the public sector 1983-1988 1996-2001 Wage gap Sample Wage gap Sample Size Size Private 22% (754,056) 17% (567,627) Public 13% (165,276) 15% (110,833)

Federal 2% (33,633) 8% (20,938) State 9% (42,942) 10% (34,919) Local 16% (88,642) 20% (60,981)

Male 8% (77,528) 10% (48,298) Female 17% (87,748) 16% (62,534) Age <25 28% (15,603) 23% (7,771) Age 25-44 13% (93,676) 15% (53,798) Age 45-54 8% (32,127) 11% (33,830) Age >=55 13% (23,870) 14% (15,433) New England 17% (33,540) 17% (20,148) Central 16% (38,863) 16% (25,930) South 10% (51,785) 12% (33,522) West 10% (41,088) 13% (31,232) <High school 26% (13,217) 18% (29,775) High school 15% (48,037) 13% (21,536) College 1-3 13% (35,097) 11% (9,672) College >= 4 years 8% (68,925) 11% (50,029) Whites 13% (131,676) 14% (85,893) Non-whites 15% (33,600) 16% (24,939) Manual 18% (17,874) 18% (9,679) Non-manual 13% (147,402) 14% (101,150) Registered nurses (95) 5% (2,945) 6% (1,854) Teachers (156-8) 15% (25,147) 21% (19,484) Social workers (174) 12% (2,870) 12% (2,716) Lawyers (178) 5% (1,014) 17% (1,184) Firefighters (416-7) 15% (1,866) 19% (1,227) Police, sheriffs, bailiffs& correction 16% (6,068) 18% (5,503) officers (418-424) Sample excludes individuals with allocated earnings. Controls as in Table 4. Source: Outgoing Rotation Group files of the Current Population Survey. Numbers in parentheses towards the bottom of the first column are 1980 Occupational Classification codes

40

06/06/2003 12:41 PM

Table 6. Private Sector – Union Wage Gaps for Immigrants, 1996-2001 US born 16.8% (496,643) 1) Year of entry to US Pre 1975 13.9% (10,215) Post 1974 15.7% (60,508) 1975-1983 10.7% (8,171) 1984-1989 11.5% (16,830) 1990-1989 14.9% (8,870) 1990-1993 14.0% (14,533) 1994 13.3% (23,249) 2) Country of birth Europe 11.6% (7,838) East Europe 12.7% (3,110) Asia 13.3% (17,927) South America 19.7% (38,803) Mexico 28.5% (20,473) Rest of S. America 12.2% (17,610) Africa 11.0% (1,427) Australasia 0% (525) Canada 15.4% (2,084) Notes: Controls as in Table 4. Year of entry dummies also include 7 country of origin dummies.

41

06/06/2003 12:41 PM

Table 7a. Union Wage Gap Estimates for the United States, 1973-2001 (%) (excludes workers with imputed earnings)

All sectors Private sector Private sector Year Blanchflower/Bryson Blanchflower/Bryson Hirsch/Schumacher 1973 17.9% 16.5% 17.5% 1974 18.4% 17.6% 17.5% 1975 19.0% 18.2% 19.2% 1976 19.4% 18.4% 20.4% 1977 23.0% 22.3% 23.9% 1978 22.8% 22.6% 22.8% 1979 16.6% 15.3% 19.7% 1980 17.7% 17.0% 21.3% 1981 16.1% 16.3% 20.4% 1983 19.5% 21.2% 25.5% 1984 20.4% 22.4% 26.2% 1985 19.2% 21.0% 26.0% 1986 18.8% 20.1% 23.9% 1987 18.5% 20.0% 24.0% 1988 18.4% 19.1% 22.6% 1989 17.8% 19.2% 24.5% 1990 17.1% 17.6% 22.5% 1991 16.1% 16.6% 22.0% 1992 17.9% 19.2% 22.5% 1993 18.5% 19.6% 23.5% 1994 18.5% 18.2% 25.2% 1995 17.4% 18.0% 24.5% 1996 17.4% 18.4% 23.5% 1997 17.4% 17.7% 23.2% 1998 15.8% 16.1% 22.4% 1999 16.0% 16.9% 22.0% 2000 13.4% 14.3% 20.4% 2001 14.1% 15.1% 20.0% 2002 16.5% 18.6% 1973-2001 average 18.0% 18.4% 22.4% Notes: Wage gap estimates calculated taking anti-logs and deducting 1. Columns 1 and 2 are taken from Table 3 of Blanchflower and Bryson (2003). Data for 1983-2001 are taken from the Matched Outgoing Rotation Group files of the Current Population Survey (MORGs). Controls comprise usual hours, age, age2, four race dummies, 15 highest qualifications dummies, male, union, 46 industry dummies, four organizational status dummies, and 50 state dummies. Sample is non-agricultural workers working at the time of interview, aged at least 16 years. For 1989-95 allocation flags are either unreliable (in 1989-93) or not available (1994 through August 1995).

42

06/06/2003 12:41 PM

For 1989-93, the gaps are adjusted upward by the average imputation bias during 1983-88 (.031 in column 3). For 1994-95 the gap is adjusted upward by the bias during 1996-98 (.046 in column 3). Column 3 is taken from column 5 of Table 4 of Hirsch and Schumacher (2002). Data for 1973-81 are from the May CPS Earnings Supplements. The Hirsch and Schumacher (2002) wage gap reported in column 3 is the coefficient on a dummy variable for union membership in a regression where the log of hourly earnings is the dependent variable. The control variables included are years of schooling, experience and its square (allowed to vary by gender), and dummy variables for gender, race and ethnicity (3), marital status (2), part-time status, region (8), large metropolitan area, industry (8), and occupation (12). Because the sample does not include allocated earnings in 1973-78, the “not corrected” series are adjusted upwards by the average bias found during 1979-81 of .033 in column 3. All three columns include only workers reporting earnings. All allocated earners are identified and excluded for the years 1973-88 and 1996-2001. For 1989-95 allocation flags are either unreliable (in 1989-93) or not available (1994 through August 1995). For 1989-93, the gaps are adjusted upward by the average imputation bias during 1983-88 (.031 in column 3). For 1994-95 the gap is adjusted upward by the bias during 1996-98 (.046 in column 3). Sample includes employed private sector nonagricultural wage and salary workers aged 16 years and above with positive weekly earnings and non-missing data for control variables (few observations are lost) Source: Blanchflower and Bryson (2003).

43

06/06/2003 12:41 PM

Table 7b. The Ratio between Unadjusted and Adjusted Union Wage Gap Estimates for the United States, 1983-2001 (%) - Excludes Workers with Imputed earnings. Unadjusted Adjusted Unadjusted/adjusted 1983 48.3% 21.2% 2.28 1984 48.3% 22.4% 2.16 1985 47.0% 21.0% 2.24 1986 44.8% 20.1% 2.23 1987 45.2% 20.0% 2.26 1988 44.6% 19.1% 2.34 1989 38.0% 19.2% 1.98 1990 34.3% 17.6% 1.95 1991 32.8% 16.6% 1.98 1992 32.5% 19.2% 1.69 1993 34.0% 19.6% 1.74 1995 34.6% 18.0% 1.92 1996 35.8% 18.4% 1.95 1997 36.1% 17.7% 2.04 1998 33.2% 16.1% 2.07 1999 32.5% 16.9% 1.92 2000 29.4% 14.3% 2.06 2001 29.8% 15.1% 1.98 2002 35.6% 18.6% 1.91 Notes: column 1 obtained from a series of private sector log hourly wage equations that only contained a union membership dummy and a constant. Reported here is the antilog of the coefficient minus 1. Column 2 is from Table 7a. Column 3 is column 1/column 2. Sample includes employed private sector nonagricultural wage and salary workers aged 16 years and above with positive weekly earnings and non-missing data for control variables (few observations are lost) Source: ORG files of the CPS 1983-2001.

44

06/06/2003 12:41 PM

Table 8. Wage gap estimates by private sector industry 1983-1988 1996-2001 density wage gap N density wage gap N ∆density ∆wage gap Agricultural services 1.8% 41.2% 1,852 2.2% 32.6% 4,115 0.4% -8.6% Other agriculture 2.5% 55.9% 12,451 1.8% 23.0% 4,828 -0.7% -32.9% Mining 18.3% 15.8% 9,311 12.8% 9.4% 4,142 -5.5% -6.4% Construction 22.7% 51.6% 44,026 18.8% 40.9% 31,878 -3.9% -10.7% Lumber 15.1% 15.7% 6,662 9.9% 14.0% 4,251 -5.2% -1.7% Furniture 16.8% 11.3% 5,449 7.8% 2.7% 3,288 -9.0% -8.6% Stone, clay & glass 30.6% 14.5% 5,143 21.6% 12.6% 3,149 -9.0% -1.9% Primary metals 48.1% 5.8% 7,021 36.0% 8.0% 3,852 -12.1% 2.2% Fabricated metals 28.1% 12.5% 11,117 16.5% 13.7% 7,033 -11.6% 1.2% Machinery excluding electrical 17.5% 3.9% 22,343 11.0% 8.2% 13,061 -6.5% 4.3% Electrical equipment 19.6% 6.4% 19,162 10.6% 10.1% 10,387 -9.0% 3.7% Autos 56.6% 10.0% 10,365 38.8% 21.7% 6,749 -17.8% 11.7% Aircraft 34.6% 0.4% 4,929 27.8% 5.7% 2,347 -6.8% 5.3% Other transport equipment 23.9% 4.8% 5,517 16.3% 9.2% 2,933 -7.6% 4.4% Photographic 10.0% -0.9% 6,104 5.1% -3.1% 4,069 -4.9% -2.2% Toys 16.3% 2.2% 1,094 6.4% 1.5% 828 -9.9% -0.7% Miscellaneous manufacturing 13.8% 7.6% 3,344 8.4% 22.0% 2,416 -5.4% 14.4% Food 31.1% 16.9% 16,396 22.5% 13.0% 9,627 -8.6% -3.9% Tobacco 33.0% 26.0% 553 22.4% 17.4% 248 -10.6% -8.6% Textiles 10.0% 4.0% 7,076 5.6% -1.4% 2,681 -4.4% -5.4% Apparel 20.8% 5.1% 10,666 8.1% 2.7% 4,144 -12.7% -2.4% Paper 44.6% 7.6% 6,544 29.9% 9.7% 3,479 -14.7% 2.1% Printing 13.4% 24.4% 14,983 8.2% 18.3% 9,331 -5.2% -6.1% Chemicals 17.2% 0.8% 10,953 11.0% 4.7% 6,908 -6.2% 3.9% Petroleum 32.7% 2.3% 1,678 21.9% -0.1% 940 -10.8% -2.4% Rubber & plastics 23.3% 11.4% 6,732 15.2% 9.2% 4,716 -8.1% -2.2% Leather & nes 20.7% 1.4% 1,953 17.5% 20.4% 633 -3.2% 19.0% Transport 35.8% 44.2% 29,626 25.9% 36.2% 22,091 -9.9% -8.0% Communications 39.4% 6.0% 14,075 22.3% 10.2% 9,884 -17.1% 4.2%

45

06/06/2003 12:41 PM

Utilities & sanitary 35.1% 5.4% 9,731 29.6% 15.0% 6,085 -5.5% 9.6% Wholesale trade 8.1% 15.7% 35,107 5.7% 9.3% 24,987 -2.4% -6.4% Retail trade 7.3% 34.2% 155,875 5.1% 13.9% 110,741 -2.2% -20.3% Banking 1.6% 1.4% 28,151 2.0% -3.2% 20,459 0.4% -4.6% Insurance & real estate 3.6% 4.8% 28,075 2.9% 6.5% 20,034 -0.7% 1.7% Private households 0.3% 29.2% 10,567 1.0% 0.2% 3,723 0.7% -29.0% Business services 5.1% 15.3% 31,916 3.0% 6.9% 30,867 -2.1% -8.4% Repair services 5.8% 36.6% 8,259 3.3% 37.0% 6,504 -2.5% 0.4% Personal services excl households 8.4% 11.1% 22,431 6.8% 12.1% 15,731 -1.6% 1.0% Entertainment 11.7% 46.8% 8,863 6.8% 29.3% 9,949 -4.9% -17.5% Hospitals 11.8% 7.4% 32,573 8.1% 10.2% 25,514 -3.7% 2.8% Health excl hosp 5.1% 5.4% 28,681 4.6% 3.0% 31,345 -0.5% -2.4% Educational 9.7% 21.2% 16,421 12.8% 18.6% 15,522 3.1% -2.6% Social services 3.0% 24.0% 10,338 2.9% 16.6% 13,478 -0.1% -7.4% Other professional 3.5% 29.3% 28,175 2.4% 15.0% 27,600 -1.1% -14.3% Notes: see table 4. Union estimates are obtained from the ORG data files (weighted).

46

06/06/2003 12:41 PM

Table 9. Industry level analysis of the change in the private sector union wage premium (1) (2) (3) (4) (5) (6) (7) (8) (9) Premiumt-1 -.3077 -.3112 -.7532 -.7540 -.7548 -.7555 -.7353 -.7292 -.3649 (11.94) (26.64) (20.23) (20.25) (20.46) (20.49) (20.75) (20.60) (13.97) Log unemployment ratet-1 5.4235 5.3328 4.5115 (1.95) (1.92) (2.42) Time -.1935 -.1918 -.3372 (1.57) (1.58) (4.01) Union densityt-1 -.0075 -.3664 -.3581 -.1011 (0.26) (2.43) (2.41) (2.82) Private sector density t-1 -.0305 -.3700 -.3586 -.3349 -.3892 (0.97) (2.56) (2.52) (2.66) (3.16) Year dummies Yes Yes Yes Yes No No No Yes Yes State dummies No No Yes Yes Yes Yes Yes Yes Yes Method OLS OLS OLS OLS OLS OLS GLS GLS GLS R2/Wald Chi2 .1887 .1897 .3918 .3923 .3752 .3757 439.2 473.1 417.4 N 756 756 756 756 756 756 756 756 832 GLS regression estimated with industry specific AR(1) process in the error term and each observation in the GLS regressions is weighted by the industry observation count of the first step following Bratsberg and Ragan (2002). Column 9 uses Bratsberg and Ragan’s data for the years 1973-1999. Our analysis excludes the following industries with small numbers of observations – toys, tobacco and forestry and fisheries.

47

06/06/2003 12:41 PM

Table 10. Industry level analysis of the level of the union wage premium in the private sector using Bratsberg & Ragan’s data. (1) (2) (3) (4) (5) (6) 1973-99 1973-99 1973-99 1973-99 1984-99 1984-99 weighted weighted weighted unweighted weighted unweighted Premiumt-1 .5937* .2395* .5689* .1773* (.0275) (.0356) (.0337) (.0448) Densityt-1 -.0951* -.0568 -.0850 -.1068 (.0395) (.0472) (.0621) (.0710) Unemployment rate .0187* .0131* .0081* .0056 .0078* .0062 (.0017) (.0017) (.0015) (.0031) (.0019) (.0034) Unemployment rate*Inflation .0026* .0012 .0005 .0005 .0014 -.0002 (.0009) (.0009) (.0009) (.0015) (.0012) (.0200) Unemployment rate*COLA -.0092* -.0047 -.0036 .0039 -.0014 .0003 (.0038) (.0036) (.0026) (.0055) (.0030) (.0045) COLA .0763* .0767* .0259 -.0146 -.0043 -.0051 (.0313) (.0303) (.0218) (.0399) (.0254) (.0367) Inflation -.0182* -.0077 -.0010 -.0009 -.0083 -.0017 (.0065) (.0069) (.0065) (.011) (.0081) (.0141) Deregulation Rail .0329 .0400 .0206 .0666* n/a n/a (.0905) (.0844) (.0600) (.0285) Deregulation Trucking -.0716 -.0617 -.0249 -.0601* n/a n/a (.0560) (.0570) (.0413) (.0269) Deregulation Air .0554 .0684 .0079 .0180 n/a n/a (.1262) (.1217) (.0847) (.0311) Deregulation Communications .0752* .0609* .0363 .0586 n/a n/a (.0316) (.0244) (.0196) (.0313) Deregulation Finance -.0614* -.0599* .0205 -.0500 n/a n/a (.0191) (.0188) (.0162) (.0263) Import penetration .2048* .2201* .1308* .1182* .1115* .1408* Durables (.0427) (.0414) (.0351) (.0426) (.0444) (.0441) Import penetration .1655* .1459* .0833* .0907* .0517 .2014* Non-durables (.0513) (.0525) (.0316) (.0215) (.0471) (.0331) Time -.0015 -.0001 -.0017* -.0013 (.0003) (.0005) (.0004) (.0007) Wald Chi2 2325.01 2781.32 1227.3 1234.5 8701.1 2054.1.7 N 832 832 832 832 512 512 GLS regression estimated with industry specific AR(1) process in error term. Each observation in the GLS regressions is weighted by the industry observation count of the first step following Berntsberg and Ragan (2002). Standard errors in parentheses.

48

06/06/2003 12:41 PM

Table 11. Wage gaps and union density by state 1983 - 1988 1996 - 2001 State Density N Wage gap Density N Wage gap Change in Change in (1983) (2000) density wage gap Alabama* 16.9% 8,908 27.3% 9.6% 7,463 13.1% -7.3% -14.2% Alaska 24.9% 9,407 22.9% 21.9% 5,280 15.5% -3.0% -7.4% Arizona* 11.4% 8,291 32.3% 6.4% 8,858 21.7% -5.0% -10.7% Arkansas* 11.0% 9,037 31.3% 5.8% 7,317 16.1% -5.2% -15.2% California 21.9% 59,161 23.5% 16.0% 44,410 22.1% -5.9% -1.4% Colorado 13.6% 10,384 26.4% 9.0% 9,923 13.5% -4.6% -12.8% Connecticut 22.7% 9,671 12.2% 16.3% 6,249 8.3% -6.4% -3.9% Delaware 20.1% 8,451 22.9% 13.3% 5,759 14.1% -6.8% -8.8% Dist of Columbia 19.5% 5,758 17.8% 14.7% 4,211 11.1% -4.8% -6.8% Florida* 10.2% 30,682 26.9% 6.8% 24,244 16.8% -3.4% -10.1% Georgia* 11.9% 11,547 21.9% 6.3% 8,904 20.7% -5.6% -1.2% Hawaii 29.2% 7,634 18.8% 24.8% 5,202 20.8% -4.4% 2.0% Idaho* 12.5% 9,042 31.0% 7.6% 8,143 26.6% -4.9% -4.4% Illinois 24.2% 30,071 18.5% 18.6% 23,510 14.0% -5.6% -4.5% Indiana 24.9% 13,139 22.0% 15.6% 8,554 17.6% -9.3% -4.4% Iowa* 17.2% 9,874 24.9% 13.6% 8,474 20.0% -3.6% -4.9% Kansas* 13.7% 9,456 25.4% 9.0% 8,254 18.6% -4.7% -6.7% Kentucky 17.9% 8,884 22.9% 12.0% 6,545 18.9% -5.9% -4.0% Louisiana* 13.8% 7,867 24.9% 7.1% 5,922 20.9% -6.7% -3.9% Maine 21.0% 8,730 7.0% 14.0% 6,662 16.1% -7.0% 9.0% Maryland 18.5% 11,169 22.6% 14.6% 6,322 10.8% -3.9% -11.8% Massachusetts 23.7% 28,695 12.0% 14.3% 11,895 13.5% -9.4% 1.6% Michigan 30.4% 29,789 16.9% 20.8% 20,047 12.7% -9.6% -4.1% Minnesota 23.2% 12,061 26.5% 18.2% 10,194 16.5% -5.0% -10.0% Mississippi* 9.9% 8,941 25.0% 6.0% 5,909 14.3% -3.9% -10.6% Missouri 20.8% 11,983 31.3% 13.2% 7,386 21.9% -7.6% -9.4% Montana 18.3% 8,605 31.1% 13.9% 7,444 21.7% -4.4% -9.5% Nebraska* 13.6% 9,523 37.3% 8.4% 9,067 22.3% -5.2% -15.0%

49

06/06/2003 12:41 PM

Nevada* 22.4% 8,446 25.9% 17.1% 8,555 18.5% -5.3% -7.3% New Hampshire 11.5% 8,228 21.4% 10.4% 6,901 19.6% -1.1% -1.8% New Jersey 26.9% 29,018 9.0% 20.8% 15,060 7.9% -6.1% -1.1% New Mexico 11.8% 7,345 34.6% 8.1% 6,909 20.4% -3.7% -14.1% New York 32.5% 44,351 11.7% 25.5% 27,394 11.2% -7.0% -0.6% North Carolina* 7.6% 27,599 27.8% 3.6% 13,458 25.7% -4.0% -2.0% North Dakota* 13.2% 8,915 35.0% 6.5% 7,817 28.0% -6.7% -7.0% Ohio 25.1% 33,118 17.7% 17.3% 21,484 11.6% -7.8% -6.1% Oklahoma* 11.5% 9,339 23.5% 6.8% 7,614 23.4% -4.7% -0.1% Oregon 22.3% 8,243 20.3% 16.1% 7,371 17.5% -6.2% -2.9% Pennsylvania 27.5% 32,228 15.6% 16.9% 23,276 10.8% -10.6% -4.8% Rhode Island 21.5% 8,325 15.8% 18.2% 6,117 11.3% -3.3% -4.5% South Carolina* 5.9% 9,960 23.9% 4.0% 6,195 13.3% -1.9% -10.5% South Dakota* 11.5% 9,981 28.9% 5.5% 8,351 12.1% -6.0% -16.8% Tennessee* 15.1% 9,922 25.9% 8.9% 6,883 14.2% -6.2% -11.6% Texas* 9.7% 37,316 25.7% 5.8% 27,668 19.0% -3.9% -6.7% Utah* 15.2% 9,308 29.3% 7.3% 8,741 18.6% -7.9% -10.7% Vermont 12.6% 7,968 7.0% 10.3% 6,330 8.4% -2.3% 1.4% Virginia* 11.7% 12,777 28.5% 5.6% 8,587 24.0% -6.1% -4.5% Washington 27.1% 8,685 22.6% 18.2% 7,947 14.3% -8.9% -8.3% West Virginia 25.3% 7,257 27.1% 14.3% 6,633 21.3% -11.0% -5.8% Wisconsin 23.8% 12,288 20.2% 17.6% 10,188 12.9% -6.2% -7.3% Wyoming* 13.9% 6,259 32.0% 8.3% 7,336 33.1% -5.6% 1.1% Source: Outgoing Rotation Group files of the Current Population Survey. Union density obtained from Statistical Abstract of the United States 2001, Table 639 Notes: * implies right-to-work state.

50

06/06/2003 12:41 PM

Table 12. State level analysis of the change in the private sector union wage premium, 1984-2001 (1) (2) (3) (4) (5) (6) (7) (8) (9) Premiumt-1 -.438 -.452 -.460 -.799 -.801 -.468 -.795 -.803 -.795 (16.15) (16.40) (16.77) (23.49) (23.91) (17.06) (23.62) (23.97) (23.66) Log unemployment ratet-1 2.607 .701 1.453 2.738 3.878 2.811 4.986 4.116 5.317 (4.22) (0.97) (2.15) (2.48) (4.24) (4.45) (6.51) (4.45) (6.76) Private sector density t-1 -.160 -.193 -.205 -.311 -.316 -.254 -.291 (3.69) (4.36) (4.63) (1.87) (2.00) (7.24) (2.07) Total density t-1 -.428 -.392 (2.49) (2.45) Time -.187 -.224 -.106 -.095 -.231 -.094 (4.04) (3.22) (2.73) (1.51) (3.59) (1.66) Constant 6.385 13.494 11.507 15.437 13.150 8.609 9.487 16.643 12.387 Year dummies No Yes No Yes No No No No No State dummies No No No Yes Yes No Yes Yes Yes Weighted by state employment No No No No No Yes Yes No Yes R2 .2244 .2651 .2380 .4256 .4058 .2465 .4034 .4073 .4046 N 918 918 918 918 918 918 918 918 918 Union density obtained from http://www.unionstats.com/ – see Hirsch, B.T. and D.A. Macpherson (2003).

51

06/06/2003 12:41 PM

Table 13. Do you think that trade <USA:labor> unions in this country have too much power or too little power? (% saying ‘far too much power’). 1985 1990 1996 Australia 49 38 18 West Germany 12 9 8 Great Britain 25 11 6 USA 25 16 16 Italy 30 28 32 How about business and industry? Do they have too much power or too little power? 1985 1990 1996 Australia 10 8 6 West Germany 12 15 12 Great Britain 5 5 7 USA 13 10 13 Italy 32 37 46 And what about the <USA: federal> government, does it have too much power or too little power? 1985 1990 1996 Australia 12 10 5 West Germany 7 9 11 Great Britain 18 16 7 USA 17 13 25 Italy 21 33 46 Source: International Social Survey Programme, 1985, 1990, 1996.

52

06/06/2003 12:41 PM

Figure 1. Movements in the Wage Premium in

the US, 1973-2001

3

8

13

18

23

1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001

Unempl. Blanchflower/Bryson premium

53

06/06/2003 12:41 PM

Figure 2 Attitudes to Labor Unions

Improving Attitudes Toward Unions (general public)

Source: Peter Hart Research, 1993-1999

54

06/06/2003 12:41 PM

Figure 3. Support for Union Representation

Labor Day 2002/Hart Research for AFL-CIO

Surge In Support For Union Representation

50%

42%43%44%39%39%

30%

43%

51%51%52%53%52%

65%

1984 1993 1996 1997 1999 2001 Today

Vote for forming union Vote against forming union

Among workers not currently in a union

Source: Peter Hart Research, 1993-2002, and 1984 Louis Harris poll http://www.aflcio.org/mediacenter/resources/upload/LaborDay2002Poll.ppt

55

06/06/2003 12:41 PM

Appendix Table 1. Industry definitions and average sample size per year (1980 Industry classification) Number SIC code Average # Observations Agriculture Service 1 12-30 745 Other Agriculture 2 10-11 1300 Mining 3 40-50 1135 Construction 4 60 6671 Lumber and wood products, except furniture 5 230-241 955 Furniture and fixtures 6 118 242 812 Stone clay, glass and concrete product 7 250-262 748 Primary metals 8 270-280 993 Fabricated metal and Metal nes 9 & 10 281-300, 169, 301 1546 Machinery, except electrical 11 310-332 3172 Electrical Machinery, equipment, and supplies 12 340-350 2593 Motor vehicles and equipment 13 219 351 1565 Aircrafts and parts 14 227 352 639 Other transportation equipment 15 360-370 765 Professional and photographic equipment 16 371-382 ,935 Toys, amusements, and sporting goods 17 390 176 Miscellaneous and nes 18 391-392 530 Food and kindred products 19 100-122 2348 Tobacco manufactures 20 130 72 Textile mill products 21 132-150 929 Apparel and other finished textile prod. 22 151-152 1343 Paper and allied products 23 160-162 932 Printing, publishing and allied industries 24 171-172 2232 Chemicals and allied products 25 180-192 1694 Petroleum and coal products 26 200-201 233 Rubber and miscellaneous plastics products 27 210-212 1033 Leather and leather products 28 220-222 219 Transportation 29 400-432 4760 Communications 30 440-442 2122 Utilities and Sanitary Services 31 450-472 1493 Wholesale Trade 32 500-571 5467 Retail Trade 33 580-691 22831 Banking and Other Finance 34 700-710 4461 Insurance and Real Estate 35 711-712 4372 Private Household Services 36 769 761 1256 Business Services 37 721-750 5645 Repair Services 38 751-760 1372 Personal Services, Except Private Household 39 762-791 3350 Entertainment and Recreation Services 40 800-810 1706 Hospitals 41 838 831 5385 Health Services, Except Hospitals 42 812-830, 832-840 5588 Educational Services 43 842-860 2851 Social Services 44 861-871 2208 Other professional Services 45 841, 872-893 5215 Forestry and Fisheries 46 031-032 12

56

06/06/2003 12:41 PM

57

References

Abowd, J.M., F. Kramarz and D. Margolis (1999), ‘High Wage Workers and High Wage Firms, Econometrica, vol. 67, no. 2, March, pp. 251-333. Blanchflower, D. G. (1996), ‘The Role and Influence of Trade Unions in the OECD’, Report to the Bureau of International Labor Affairs, Washington, D.C.: U.S. Department of Labor, August, downloadable at www.dartmouth.edu/~blnchflr/Projects.html. Blanchflower, D. G. (1999), ‘Changes Over Time In Union Relative Wage Effects in Great Britain and the United States’, in S. Daniel, P. Arestis, and J. Grahl (eds.), The History and Practice of Economics: Essays in Honour of Bernard Corry and Maurice Peston, Volume. 2, Northampton, Mass.: Edward Elgar, pp. 3-32. Blanchflower, D. G. (2000a), ‘Globalization and the Labor Market’, Report to The Trade Deficit Review Commission, downloadable at http://www.ustdrc.gov. Blanchflower, D.G. (2000b), ‘Self-employment in OECD Countries’, Labour Economics, 7, September, pp. 471-505. Blanchflower, D.G. and A. Bryson (2003), ‘Changes Over Time in Union Relative Wage Effects in the UK and the US Revisited’ in International Handbook of Trade Unions edited by John T. Addison and C. Schnabel. Edward Elgar, Cheltenham, UK. Blanchflower, D. G., and R. B. Freeman (1992), ‘Unionism in the U.S. and Other Advanced OECD Countries’, Industrial Relations, 31 (1), 156-179; reprinted in M. Bognanno and Morris M. Kleiner (eds.), Labor Market Institutions and the Future Role of Unions, Oxford: Blackwell, pp. 56-79. Blanchflower, D. G., Millward, N. and A.J. Oswald (1991), ‘Unionism and Employment Behaviour’, Economic Journal, 101, pp. 815-834. Bratsberg, B., and J. F. Ragan (2002), ‘Changes in the Union Wage Premium by Industry – Data and Analysis’, Industrial and Labor Relations Review, 56 (1) October, pp. 65-83. Bryson, A., (2001), ‘Employee Voice, Workplace Closure and Employment Growth’, Policy Studies Institute Research Discussion Paper No. 6 Card, D. (1996), ‘The Effect of Unions on the Structure of Wages: A Longitudinal Analysis’, Econometrica, 64 (4), pp. 957-979. Farber, H. S. (1990),’ The Decline of Unionization in the United States: What Can Be Learned From Recent Experience’, Journal of Labor Economics, 8(1), Part 2, January, pp. S75-S105. Farber, H. S. (2001), ‘Notes on the Economics of Labor Unions’, Working Paper 452, Princeton University Industrial Relations Section.

06/06/2003 12:41 PM

58

Farber, H. S. (2003), ‘Nonunion Wage Rates and the Threat of Unionization’, Working Paper 472, Princeton University Industrial Relations Section. Farber, H. S. and B. Western (2001), ‘Accounting for the Decline of Unions in the Private Sector, 1973-1998’, Journal of Labor Research, 22 (3), pp. 459-486. Farber, H. S. and B. Western (2002), ‘Ronald Reagan and the Politics of Declining Union Organization’, British Journal of Industrial Relations, vol. 40, no. 3, September, pp. 385-401. Freeman, R. B. (1984), ‘Longitudinal Analyses of the Effects of Trade Unions’, Journal of Labor Economics, 2 (1), pp. 1-26. Freeman, R. B. (1986), ‘Unionism Comes to the Public Sector’ Journal of Economic Literature, 24, March, pp. 41-86. Freeman, R. B. (1988), ‘Contraction and Expansion: The Divergence of Private Sector and Public Sector Unionism in the United States’, Journal of Economic Perspectives, 2 (2), pp.63-88. Freeman, R.B. (1993), ‘How Much Has De-unionization Contributed to the Rise in Male Earnings Inequality?’ in Uneven Tides edited by P.S. Danziger and P. Gottschalk, Russell Sage Foundation, New York. Freeman, R.B. (1995), ‘Are Your Wages Set in Beijing?’, Journal of Economic Perspectives, Summer, pp. 15-32. Freeman, R.B. (1999), ‘The New Inequality in the United States’ in Growing Apart: the Causes and Consequences of Global Wage Inequality edited by A. Fishlow and K. Parker, Council on Foreign Relations Press, New York. Freeman, R. B. and C. Ichniowski (1988), When Public Sector Workers Unionize, NBER and University of Chicago Press. Freeman, R. B., C. Ichniowski and H. Lauer (1989), ‘Collective Bargaining Laws, Threat Effects and the Determination of Police Compensation’, Journal of Labor Economics, 7(2), pp. 191-209. Freeman, R.B. and L.F. Katz (eds) (1995), Differences and changes in wage structures and published by University of Chicago Press and NBER, Chicago. Freeman, R. and Kleiner, M. M. (1999), ‘Do Unions Make Enterprises Insolvent?’, Industrial and Labor Relations Review, Vol. 52, No. 4, pp. 510-527. Freeman, R.B. and J.L. Medoff, (1984), What Do Unions Do?, Basic Books, New York. Freeman, R.B. and J.L. Medoff, (1985), ‘Reply’, Industrial and Labor Relations Review, 38(2), January, pp. 259-263.

06/06/2003 12:41 PM

59

Freeman, R.B. and K. Needels (1993), ‘Skill Differentials in Canada in an Era of Rising Labor Market Inequality’, in Small Differences that Matter edited by D. Card and R. Freeman, University of Chicago Press and NBER. Freeman, R.B. and A. Revenga (1998), ‘How Much Has LDC Trade Affected Western Job Markets?’, in Trade and Jobs in Europe: Much Ado About Nothing? Edited by M. Dewatripont, A. Sapir and K. Sekket, Clarendon Press, Oxford. Hirsch, B. T. (2003), ‘Reconsidering Union Wage Effects: Surveying New Evidence On An Old Topic’, mimeo. Hirsch, B.T. and D. A. Macpherson (2002), Union Membership and Earnings Data Book: Compilations from the Current Population Survey (2002 Edition). Washington: Bureau of National Affairs. Hirsch, B.T. and D.A. Macpherson (2003), ‘Union Membership and Coverage Database from the Current Population Survey: Note,’ Industrial and Labor Relations Review, Vol. 56, No. 2, January, pp. 349-54. Hirsch, B. T., D. A. Macpherson, and E. J. Schumacher (2002), ‘Measuring Union and Non-Union Wage Growth: Puzzles in Search of Solutions’, paper presented at the 23rd Middlebury Economics Conference on the Changing Role of Unions, Middlebury, Vermont, April. Hirsch, B. T., D. A. Macpherson, and W. G. Vroman (2001), ‘Estimates of Union Density by State’, Monthly Labor Review, 124 (7), pp. 51-55. Hirsch, B. T., and E. J. Schumacher (1998), ‘Unions, Wages, and Skills’, Journal of Human Resources, 33 (1), pp. 201-219. Hirsch, B. T. and E. J. Schumacher (2002), ‘Match Bias in Wage Gap Estimates Due To Earnings Imputation’, Mimeograph, Trinity University, available at www.trinity.edu/bhirsch/ or www.ssrn.com. Horn, J.T. (1998), ‘Unions, Firms and Competition: The Effects of Foreign Competition on Unionization and the Impact of Unions on Firm Investment’, Ph.D. thesis, Department of Economics, Harvard University. Lawrence, C. and R. Lawrence (1985) ‘Manufacturing Wage Dispersion: An End Game Interpretation’, Brookings Papers on Economic Activity, Vol. 19, No. 2, pp. 415-33 Leonard, J. S., 1992. ‘Unions and Employment Growth’, Industrial Relations, Vol. 31, No. 1, pp. 80-94. Lewis, H. G., (1963), Unionism And Relative Wages In The United States, University of Chicago Press, Chicago.

06/06/2003 12:41 PM

60

Lewis, H. G., (1986), Union Relative Wage Effects; A Survey, University of Chicago Press, Chicago. Long, R. J., (1993), ‘The Impact of Unionization on Employment Growth of Canadian Companies’, Industrial and Labor Relations Review, Vol. 48, No. 4, pp. 691-703. Mellow, W., and Sider, H., (1983), ‘Accuracy Of Response In Labor Market Surveys: Evidence And Implications’, Journal of Labor Economics, 1, pp. 331-344. Staiger, R.W. (1988), ‘Organized Labor and the Scope of International Specialization’, Journal of Political Economy, vol. 96, no. 5, October, pp. 1022-1047. Waddoups, C.J (2000), ‘Unions and Wages in Nevada's Hotel-Casino Industry’, Journal of Labor Research, 21(2), Spring, pp. 345-361. Wooden, M. and Hawke, A., (2000), ‘Unions and Employment Growth: Panel Data Evidence’, Industrial Relations, Vol. 39, No. 1, pp. 88-107.