Union Activity and the Decline in American Trade Union Membership

Union Activity and the Decline in
American Trade Union Membership
C. TIMOTHY KOELLER
Stevens Institute of Technology, Hoboken, NJ 07030
This paper extends recent research on the determinants of the decline in union membership in the United States. Usirig biennial state-level data for a set of years
between 1958 and 1982, my model tests "union organizing," "structural," "management opposition," and "public policy" hypotheses concerning union membership
and suggests improved specifications of each of these hypotheses. The paper also
examines the relative importance of each factor in explaining the decline in unionization. The results support each of the hypotheses and confirm previous findings
that changes in the structure of the labor force are most important in explaining
union membership decline.
I.
Introduction
Analysis of the decline in American trade union membership continues to be an
important research topic for labor economists and industrial relations scholars. In a
recent contribution to this Joumai, Moore and Newman (1988) examine interstate
differences in union membership during the period 1950-1980. Their regression analysis incorporates hypotheses drawn from three prominent "explanations" of membership decline: "stmctural," "management opposition," and "public policy" factors. To
examine the relative explanatory power of each of these factors, Moore and Newman
decompose the independent effect of each factor into an effect resulting from
changes in its estimated regression coefficient and an effect resulting from changes
in the value of the regressor itself. They conclude that the "stmctural" composition
of the labor force, captured by regressors proxying age, occupational, and gender
factors, and "public policy" factors, reflected in the effect of a state right-to-work
law dummy variable, best explain the decline in union membership during the postWorld War II period. The contribution of the proportion of a state's labor force that is
female is singled out as the most influential effect.
Moore and Newman's (henceforth M-N's) efforts are an important contribution
to understanding the processes which influence American union membership. But a
number of issues remain unresolved or unexplored. Eirst, as Earber (1985) suggests
(and M-N recognize), a regressor capturing the effect of the proportion of a state's
labor force that is female is likely to be correlated with regressors measuring the
occupational and industrial composition of the labor force. In this paper the indepenJOURNAL OF LABOR RESEARCH
Volume XV, Number 1 Winter t994
20
JOURNAL OF LABOR RESEARCH
dent effect of the "percent female" regressor is examined further in the presence of
additional factors capturing these related effects.
Second, a fourth explanation of the decline in union membership is cited by
Ereeman (1985) and by Dickens and Leonard (1985), which M-N refer to as the
"union organizing" hypothesis. Recent declines in union organizing success in
NLRB elections that contribute to the decline in union membership can be linked to
reduced organizing activity by unions. M-N do not explore this hypothesis, citing
data limitations and uncertainty about the appropriate specification. I explicitly test
this hypothesis.
Third, M-N's "public policy" hypothesis is expanded to incorporate a proxy for
political attitudes towards unions used by Elliott (1979) and Koeller (1985, 1992)
which permits examination of the independent effect of the presence of right-towork legislation per se on union membership, having controlled for attitudes
towards unionism.
These and other factors are included in a reduced-form model to explain variation in the proportion of nonagricultural employment that is unionized among the
contiguous states for each of the pooled data samples 1958-1972, 1974-1982, and
1958-1982. As a point of reference, an amended version of M-N's model is first estimated for these data sets. In addition, following M-N, Blinder's (1973) decomposition approach is adopted to identify the factors contributing most substantially to the
decline in the extent of union membership during the period 1958-1982.
II. Model Specification
EoUowing M-N and others (e.g., Lumsden and Petersen, 1975; Koeller, 1985; Hirsch,
1980), the models of the extent of union membership estimated herein are reducedform equations. It is assumed that differences in the equilibrium rate of union membership across states depend on factors which influence the demand for or the supply
of union services in those states.
A number of the regressors included in M-N's model are included in this study.
Because they are fairly "standard" regressors in models of union organizing activity,
I do not repeat the theoretical arguments underlying these variables, but when new
variables are introduced, an explanation is provided. Table 1 contains the variables
used in this study and lists their sources. In addition, there is likely to be a significant
problem of multicollinearity in cross-sectional models. I note instances where regressors included in M-N's model are excluded to minimize this problem.
"Structural" Hypothesis. This hypothesis suggests that a relative decline in
unionization in a state may result from changes in the structure of the state's labor
force. Two arguments have been advanced to explain why the percentage of a state's
labor force that is female (WLF) negatively affects the extent of union membership.
Eirst, as M-N and Hirsch have noted, women as "secondary" workers are expected to
have a lower demand for union services than are men. Second, Earber (1985) found
21
C. TIMOTHY KOELLER
Table 1
Variable Names, Definitions and Data Sources
Dependent
Variable
Percent of nonagricultural employees who are union members
(Statistical Abstracts of the U.S.).
WLF
Percentage of a state's labor force that is female (Censuses of Population).
NWLF
Percentage of a state's labor force that is nonwhite (Censuses of Popuiation).
SLF
Percentage of a state's labor force that is aged 55 or older (Censuses of Population).
YLF
Percentage of a state's labor force between 16 and 24 years old (Censuses of Population).
BCW
Percentage of a state's labor force employed in blue-collar occupations (craftsmen, foremen,
and kindred workers, operatives, and nonfarm labor) (Censuses of Population).
EGPS
Percentage of a state's nonagricultural employees working in unionizable industries:
Mining, Constmction, Manufacturing and Transportation (Statistical Abstracts of the U.S.).
SOUTH
Regional dummy variable (SOUTH= 1 if a state is a Southem state).
RTW
Right-to-work law dummy variable (RTW=l if law is present in a state).
EARN
Hourly eamings adjusted for geographic differences in the CPI during the year
(Statistical Abstracts of the U.S.).
Number of unfair labor practice cases filed against employers (classification CA) and
received by the NLRB during the previous year (Annuai Reports of the NLRB).
ULP
FS
Percentage of establishments employing more than 500 employees during the year
(County Business Pattems).
UNULP
Number of unfair labor practice cases filed against unions (classifications CB, CC, CD, and
CP) and received by the NLRB during the previous year (Annual Reports of the NLRB).
UAV
Fraction of the time each senator or representative voted in accordance with or was paired in
favor of the AFL-CIO's CO.P.E. position on union-related votes during the year
(Congressional Quarterly).
PCTPT
Percentage of a state's labor force employed part-time (Censuses of Population).
PU
Percent of representation elections in which union status was chosen during the previous
year (Annual Reports of the NLRB).
UN
Unemployment rate during the year (Statistical Abstracts of the U.S.).
that, once occupational factors are taken into account, women appear to be less able,
not less willing, to secure union jobs than are men.
To extend this analysis, regressors are included to capture occupational factors:
the percentage of a state's employment engaged in blue-collar occupations (BCW)
and the percentage of employment that is part-time (PCTPT). The first regressor was
also included by M-N; they expected and found a positive effect, though it was not
always significant. The second, new regressor accounts for individuals working
22
JOURNAL OF LABOR RESEARCH
fewer than 35 hours per week in a state. As is true for WLE, part-time employees, as
"secondary" workers, should have a lower demand for union services than full-time
employees. A negative effect is thus expected for PCTPT.
Eollowing M-N and others, additional regressors measure the percentage of a
state's labor force between 16 and 24 years old (YLF), the percentage of the labor
force aged 55 and older (SLF), and the percentage of the labor force that is nonwhite (NWLF). Based on M-N's findings, it is expected that the first of these regressors will negatively affect, and the latter two will positively affect, the extent of
union membership.
M-N also include regressors measuring regional differences, firm size, and the
industrial composition of employment. Similar regressors are included in my model.
A dummy variable (SOUTH) represents the extent of unionization in Southern
states; its regression coefficient should be negative. The percentage of estabhshments
employing more than five hundred workers (FS) is expected to have a positive effect
on the model's dependent variable if there are economies of scale in union organizing activities.' The percentage of a state's employment working in "unionizable" (or
goods producing) industries — Mining, Construction, Manufacturing, and Transportation — is included, following M-N, because these industries have traditionally
been heavily unionized.^ It is expected that this regressor (EGPS) will have a positive
effect on the extent of union membership.
Einally, unlike M-N, a regressor designed to capture the effect on the demand
for union services of living in "urban" states is not included. Preliminary results
including such a regressor indicated significant collinearity with other regressors.^
"Management Opposition" Hypothesis. Ereeman (1986) suggests that management opposition to unionization may result in few union organizing successes and in
lower union membership. In this study, the number of unfair labor practice cases
filed against employers (ULP) is used to capture employer opposition to unionization. This variable is consistent with recent empirical models of union organizing
activity (e.g.. Hunt and White, 1983, 1985; Koeller, 1992; Ereeman, 1986; Ereeman
and Kleiner, 1990). The empirical findings are mixed: Hunt and White (1985), M-N,
and Ereeman and Kleiner found no effect; Koeller found a positive effect. These conflicting results suggest two possible explanations: union organizing activity may be
enhanced by workers reacting strongly to employer unfair labor practices or
employer unfair labor practices may have proven to be a successful management
strategy. To control for the first of these two effects, a regressor (UNULP) is included
measuring the number of unfair labor practice charges brought against unions
(NLRB cases classified as CB, CC, CD, and CP).'* This regressor should have a positive effect on union membership if it captures the aggressive efforts of unions to
secure organizing success and representation.
A further problem with a regressor representing employer unfair labor practices
is the possibility that it is jointly endogenous with union membership.^ To mitigate
C. TIMOTHY KOELLER
23
this problem, the regressor ULP is lagged one year relative to the year for which the
extent of union membership is measured.
"Public Policy" Hypothesis. Changes in public policy can alter the demand and
supply sides of the "market" for union services. M-N included three regressors to
control for "public policy" effects. Two of these involving "govemment substitution"
(Neumann and Rissman, 1984) are social welfare expenditures and unemployment
compensation payments within states; M-N expected negative effects for these
regressors, but because they found the first of these variables to be generally
insignificant it is excluded from further consideration.
Conceming the second regressor, M-N did not indicate its precise nature, but
described it as a "measure of the expected unemployment compensation benefits in
the state." Eurthermore, two opposite effects can be expected for such a regressor.
The first is the negative one predicted by M-N, but a positive effect is also possible
if this regressor accounts for higher unemployment benefits in heavily unionized
states. Thus, as Stepina and Eiorito (1986) have suggested, there may be a substantial
problem in measuring the "govemment substitution" effect. My models were estimated with and without a regressor for a state's unemployment compensation benefits paid in the previous year in order to capture the net result of these opposite
effects. Because no statistically significant effect was found and inclusion did not
add significantly to the models' explanatory powers, no models containing this
regressor are reported.
M-N's third "public policy" regressor is a dummy variable indicating a state
right-to-work law. As Earber (1984) has reported, the effect of right-to-work laws on
the extent of union membership may reflect a statutory impediment to unionism as
well as worker preferences against union representation. Eollowing Elliott and
Koeller (1985, 1992), an additional regressor is included in this study to capture
political attitudes towards unionism across states; UAV is the state average of the
fraction of instances that each U.S. senator or representative favored the AP'L-CIO's
Committee on Political Education position on selected votes during each year. This
proxy for a pro-union political environment should be positively related to the extent
of union membership, and this effect should be observed even after taking into
account the right-to-work dummy variable (RTW).
"Union Organizing" Hypothesis. A decline in recent union organizing activity
may have reduced the extent of new union membership and the extent of union representation. In this study union organizing activity is measured by two regressors that
are expected to have positive effects on union membership. Eirst, the number of
unfair labor practice charges lodged recently against unions (UNULP) was mentioned earlier in the discussion of the "management opposition" hypothesis. This
regressor may capture recent union "militancy" during organizing activities. Second,
the percentage of recent NLRB elections in which union representation was chosen
(PU) is included as a regressor. Ereeman (1985) has cited the lack of research findings relating NLRB election results to union membership density.^
24
JOURNAL OF LABOR RESEARCH
Two econometric issues arise conceming these regressors. Eirst, it is possible
that UNULP and PU are positively correlated; i.e., aggressive union behavior during
organizing may be collinear with union organizing success. Second, there is the possibility of simultaneity bias of the estimated effect for PU\ union success in organizing manifest in higher membership density.^ This potential problem is dealt with by
imposing a one-year lag on the regressor PU.
State-Level Economic Conditions. Two additional regressors account for interstate differences in economic conditions: real hourly eamings (EARN) and the unemployment rate (UN). Previous researchers (e.g., M-N, Hunt and White, Hirsch)
include measures of hourly eamings or per capita income on the hypothesis that the
income elasticity of demand for union services is positive. In addition, Koeller (1985,
1992) and Lawler and Hundley (1983) suggest that states with relatively high eamings are those that are the most heavily unionized, and those where union organizing
efforts are most successful. Koeller (1985) and Hirsch also suggest that eamings and
unionization are determined simultaneously; following Hirsch, however, it is assumed
that unions are not able to significantly increase aggregate eamings for a state, and
causation is expected to run from the eamings regressor to union membership. The
real hourly eamings regressor is derived by defiating each state's nominal average
hourly eamings by a regional CPI.
Eollowing Hunt and White and Koeller (1992), an unemployment rate regressor
captures employment declines in heavily unionized (e.g., Rust-Belt) sectors of the
economy. Also, as suggested by Ashenfelter and Pencavel, a high unemployment
rate poses a threat to workers' employment security, a threat workers may attempt to
minimize through union membership. In either case, this regressor should have a
positive effect on the extent of union membership.
III. Regression Results
Estimates from an amended version of M-N's model are provided as a benchmark for
the estimates of my expanded model. Ordinary least-squares was used for each of
three sample periods with pooled biennial data: i958-1972, 1974-1982, and 19581982. The amended M-N model is:
^, ajNWLF-, + a^SLF^ + a^YLF^, -I- a^BCW-,
^ - , -F ajSOUTH-f + a^RTW^, + a^EARN^ + a^^ULP^,
+ «ll^5,-, + «,r
The expanded model developed in this study is:
UM., = Amended M-N Model H- b^2UNULP.^, -\- b^^lJAV^,
+ b.^PCTPT,, + b,,PU,, + b.^UN,, + «,,,
where u is the error term, and data are pooled across states (0 and time (0.
The data refiect values of the models' variables in 1958, 1960, 1968, 1970,
1972, 1974, 1976, 1978, 1980, and 1982. Data on the characteristics of each state's
C. TIMOTHY KOELLER
25
labor force (WLE NWLE SLF, YLF, BCW, and PCTPT) are only available for the
census years 1960, 1970, and 1980, so that census-year data are repeated for off-census
years for these regressors on the assumption that these characteristics of the labor
force did not change significantly within a two- to four-year period before and after
each census year; e.g., 1970 values for these regressors were also reasonably valid as
late as 1974.^
The three sample periods were chosen for a number of reasons. Eirst, the sample
period 1958-1982 is reasonably comparable with M-N's period 1950-1980, and the
sample period 1958-1972 permits comparisons with the results of studies by Hunt
and White and Koeller (1985, 1992) using data for the early to mid-1970s. Second,
Neumann and Rissman have reported that union membership as a percentage of
nonagricultural employment was higher in 1960 than in earlier or later years studied
by M-N (especially 1950). Data starting around 1960 and continuing to the 1980s
thus captures the period of decline of American union membership density. Third,
the two samples 1958-1972 and 1974-1982 provide a basis for examining changes in
the effects of the determinants of union membership over the period 1958-1982. The
substantial degrees of freedom afforded by these two pooled samples should also
increase the likelihood of estimating statistically significant regression coefficients
and lessen M-N's difficulty with multicollinearity in implementing Blinder's decomposition method. Einally, pooling data sets for 1958-1972, 1974-1982, and 1958-1982
permits a covariance analysis of the effect of time on union membership; dummy
variables are introduced to represent the biennial years included in each of the pooled
data sets.
Cross-sectional regression results for the amended M-N model and the
expanded model are reported in Table 2. The dependent variable used for each of
the estimated equations is union membership as a percentage of a state's nonagricultural employment.
Amended M-N Model. As reported in Table 2 (and especially equation (3)), the
amended M-N model yields estimated results which are similar to those reported by
M-N for the period 1950-1980. Exceptions to this result are the effects of the regressors WLF, SLF, BCW, EGPS, and EARN. In addition, adjusted R^ statistics for equations (l)-(3) are comparable to those reported by M-N (about .70).
The estimated effect of the regressor WLF is negative but statistically insignificant in equations (1) through (3) of Table 2. This result was anticipated by M-N,
who speculated that increased stability of employment and decreasing employment
discrimination may account for a weakened effect of the regressor WLF. In addition,
WLF exhibits substantial collinearity with other regressors, especially YLF, ULP,
and SLF.^
The effect of the regressor SLF is mixed: positive (if insignificant) for 1958-1972
and negative for 1974-1982 and 1958-1982. The pre- and post-1973 comparison suggests, ceteris paribus, a decline in membership among older workers. In contrast,
M-N found a positive effect for this regressor. They speculated that variations in the
JOURNAL OF LABOR RESEARCH
26
y ^
Table 2
Regression Results: Determinants of the Extent of Union Membership
Amended Moore-Newman Model
Equation
Data Set
Adjusted R^
F
mean of dep var.
1
1958-1972
.66
31.72
23.95
Intercept
-8.50
(-.63)
3
1958-1982
.68
51.99
22.12
4
1958-1972
.74
34.42
23.95
21.01**
(1.73)
12.35a
(1.48)
-58.43*
(-3.94)
-.22
(-1.00)
-.01
(-.06)
-.15
(-1.03)
-.05
(-.23)
-.38*
(-2.41)
-.25**
(-1.87)
NWLF
.03
(.33)
.09"
(1.47)
.07
(1.30)
.12"
(1.43)
.06
(1.04)
.05
(1.15)
SLF
.24
(.66)
-1.26*
(-4.31)
-.51*
(-2.20)
1.99*
(4.56)
-.98*
(-3.78)
.17
(.73)
YLF
-.20
(-.66)
-.65*
(-2.70)
-.51*
(-2.65)
1.27*
(3.10)
-.46**
(-1.88)
-.02
(-.09)
BCW
-.13a
(-1.48)
-.26*
(-3.44)
-.18*
(-3.08)
-.06
(-.67)
-.17*
(-2.73)
-.09**
(-1.78)
EGPS
.26*
(4.69)
.21*
(4.28)
.23*
(6.18)
.28*
(5.59)
.15*
(3.72)
.22*
(6.64)
SOUTH
-1.97
(-1.17)
-4.64*
(-4.05)
-3.84*
(-3.95)
-2.84**
(-1.87)
-2.81*
(-2.96)
-3.14*
(-3.60)
RTW
-5.35*
(-5.91)
-6.07*
(-8.90)
-5.86*
(-10.37)
-.87
(-.85)
-3.04*
(-4.73)
-2.88*
(-4.83)
EARN
14.40*
(12.31)
11.25*
(13.41)
12.35*
(17.44)
11.58*
(10.47)
8.28*
(11.43)
9.92*
(14.84)
WLF
ULP
FS
2
1974-1982
.70
38.72
20.30
Expanded Model
.04x10-2
..09x10-2"
(.27)
(-1.55)
10.03*
(1.97)
14.95*
(3.88)
-.05x10-2
(-.98)
10.15*
(3.30)
.05x10-2
(.40)
5
6
;1974-1982 1958-1982
.81
.75
51.76
58.49
20.30
22.12
17.01**
(1.66)
-6.78
(-.80)
-.05x10-2 -.04x10-2
(-1.05)
(-.84)
5.66
(1.22)
14.35*
(4.55)
7.61*
(2.72)
UNULP
.01*
(4.57)
.01*
(7.29)
.01*
(6.95)
UAV
.08*
(3.63)
.08*
(5.52)
.08*
(5.85)
-.96*
(-3.60)
.25
(1.38)
-.20
(-1.29)
.08*
(2.27)
.08*
(2.83)
.07*
(3.11)
1.37*
(2.27)
.60*
(4.01)
.59*
(3.69)
PCTPT
PU
UN
—
—
Numbers in parentheses are estimated t-statistics.
» (**, ") = Significant at the 5 (10, 15) percent level (two-tail test).
C. TIMOTHY KOELLER
27
estimated effect of the regressor SLF may reflect changing attitudes towards unions
of workers in the "senior" age cohort over the 25-year period 1958-1982.'° If SLF is
reflecting pro-union sentiment among these workers, it will be useful to examine its
effect once a regressor is added to explicitly capture interstate differences in political
attitudes towards unions (UAV, as reported below).
A somewhat Weaker contrast applies to the regressor BCW. As reported in equations (1) through (3) of Table 2, this regressor generally exhibits a negative effect on
union membership, suggesting a decline in union membership in traditionally unionized occupations. M-N found a positive effect for this regressor for the period 19501980, but few statistically significant effects for other sample periods.
The regressor EGPS consistently has a positive effect on union membership, as
expected, and in contrast to M-N's finding of no effect for a similar regressor.
Finally, the regressor EARN exerts a positive effect on the extent of union membership, as expected. This result is in contrast to M-N's finding of no significant
"income" effect; union membership thus appears to be a normal good.
Expanded Model. Equations (4) through (6) of Table 2 indicate the increased
explanatory power of my expanded model as compared with the amended M-N model.
In fact, as will be discussed below, each of the regressors added within the expanded
model has a statistically significant effect on the extent of union membership.
"Structural" Hypothesis. In contrast to the amended M-N model (equations
(l)-(3)), in the expanded model (equations (4)-(6)) the estimated effect of the regressor WLE is generally negative, and especially after 1972. This negative effect is consistent with results reported by M-N and by Hunt and White (1985) and Koeller
(1992). Most noteworthy is that for the 1958-1972 sample period the estimated effect
of WLE is not statistically significant, while the effect of the extent of part-time
employment in a state (PCTPT) is strongly negative. This comparison is consistent
with M-N's speculations conceming the independent effect of WLE in the presence
of other "structural" regressors.
The estimated effect of PCTPT varies between the two data sets 1958-1972 and
1974-1982. Negative and statistically significant in the earlier period (despite high
positive correlations with WLE, SLF, and YLF), this effect tums weakly positive in
the later period, possibly the result of high negative correlations between PCTPT and
NWLF and SOUTH.^^
The estimated effect of the regressor SLF is positive and significant for the
period 1958-1972 in equation (4), in contrast to no significant effect in equation (1);
its effect remains negative for the period 1974-1982. Thus, having accounted for
interstate differences in attitudes towards unions (the regressor UAV), the change in
the effect of SLF over the period 1958-1982 is strengthened: change occurred over
this time period in the "senior" age cohort's demand for union membership.
"Management Opposition" Hypothesis. The regressor ULP generally exhibits
no significant relationship to the extent of union membership. This result is comparable to that reported by M-N.'^ However, the regressor UNULP is positively related
28
JOURNAL OF LABOR RESEARCH
to the dependent variable. If UNULP refiects workers' reactions to employer practices, these results suggest a possible indirect effect of employer unfair practices and,
as noted by Hunt and White and Koeller (1992), they may indicate that employer
unfair labor practices are counterproductive as a union avoidance strategy.'^
"Public Policy" Hypothesis. The estimated effects of the regressors CMVand
RTW are as expected. UAV is positively related to union membership in equations
(4)-(6), as found by Elliott and Koeller. The RTW regressor exerts a negative effect
after 1972, lending support to M-N's proposition that state right-to-work laws may
be increasing the cost of union organizing because of free-rider problems or may
be weakening unions' bargaining positions and thus lowering demand for union
representation.
"Union Organizing" Hypothesis. The estimated effect of the regressor UNULP
on union membership is generally positive, as expected; i.e., union membership is
high in states where unions have recently opposed management's initiatives. The
effect of recent union success in representation elections (PU) is also to sustain relatively high membership, as indicated by the positive effect reported for this regressor
in equations (4)-(6).
Einally, the regressor UN has a positive effect on union membership, as
expected. Union membership may thus be perceived as a way for workers to minimize a possible threat to employment security, or it may reflect employment in
declining sectors of the economy. In addition, the regression coefficients for the time
dummy variables introduced into the models are not reported in Table 2 in order to
conserve space. However, these coefficients are negative in each equation, indicating
a decline in the extent of union membership, and these negative effects grew larger
as time progressed.
IV. Decomposition Results
Eollowing M-N, Blinder's method is used to evaluate the relative importance of the
four hypotheses advanced to explain the decline in union membership over the
period 1958-1982. The decomposition can be expressed as:
I/3/-XA - J^pE^E ^ j^pi^^xt - Xf) + IX-HPl^ -Pf),
where B,. is the regression coefficient associated with regressor i, X,- is the mean value
of this regressor, and each of the values are estimated for an earlier (E) and later (L)
time period.''* The amount attributable to the regressors (the first summation on the
right-hand side of the equation) is the difference in union density resulting from
changes in the mean values of the regressors, with the unionization model estimated
for the later sample period. The amount attributable to the coefficients (the second
summation on the right-hand side) is the difference in union density resulting from
how the regressors would have affected union density in the later sample period as
compared with how the regressors actually affected the dependent variable in the earlier sample period.
C. TIMOTHY KOELLER
29
As reported in Table 3, the regressors and regression coefficients of my
expanded model predict a substantial decline in the extent of union membership,
which decline more than offsets the predicted increase in the intercept value.'^ This
analysis predicts a 23.6 percent net decrease in union membership, about twice the
decrease predicted by M-N.
Especially noteworthy is the substantial negative effect attributable to the
regressor SLF. This effect dominates all others, including that attributable to the
regressor WLF, and stands in sharp contrast to M-N's results for the WLF regressor.
The different relative effect reported for WLF in Table 3 can be explained by the
presence of other regressors in the model, notably PCTPT as explained earlier in the
discussion of the results reported in Table 2. Eurthermore, the result attributable to
the change of the coefficient for the regressor SLF is consistent with a decline in
demand for union membership among older workers, as suggested earlier.
Table 3
Decomposition of the Decline in
Union Density, 1958-1972 to 1974-1982
Total
Amount
Attributable
Amount
Attributable
to Regressors
Amount
Attributable
to
Coefficients
WLF
NWLF
SLF
YLF
BCW
EGPS
SOUTH
RTW
EARN
ULP
FS
UNULP
UAV
PCTPT
PU
UN
-46.6%
-1.0
-188.1
-160.0
-27.9
-28.7
-0.6
-6.1
-3.0
-1.5
11.4
0.2
-3.6
92.1
9.0
-4.2
-11.7%
0.5
11.6
-14.1
2.3
-1.8
0.0
0.0
10.9
-0.3
0.4
2.4
-0.1
-0.5
-5.3
4.9
-34.9%
-1.5
-199.7
-145.9
-30.2
-26.9
-0.6
-6.1
-13.9
-1.2
11.0
-2.2
-3.5
92.6
14.3
-9.1
Subtotal
Shift Coefficient
Total
-358.6%
335.0%
-23.6%
-0.8%
-357.8%
Independent
Variable
30
JOURNAL OF LABOR RESEARCH
Other noteworthy factors contributing to the decline in union membership
include an increased negative effect of the regressor YLF and a decreased negative
effect of the regressor P C r / T between the sample periods 1958-1972 and 1974-1982.
In addition, the variables reflecting union organizing efforts (UNULP and PU)
appear to have contributed little to the decline in union membership, although for PU
there is a positive total effect resulting from an increased infiuence of PU on the
dependent variable. That is, between the two sample periods decreases in union organizing success contributed to the decline in union membership, a finding consistent
with results reported by Dickens and Leonard for the period 1950-1980.
As with practically all of the independent variables (except real hourly eamings,
EARN, and employer unfair labor practices, ULP), most of the total amount of
decline of membership attributable to each regressor can be accounted for by
changes in values of the regression coefficients. That is, even if the mean values of
the regressors had remained constant between the sample periods 1958-1972 and
1974-1982, changes in the coefficients of these regressors would have resulted in a
decline in union membership density. Thus, stmctural changes refiecting the process
of unionization, rather than changes in the characteristics of workers and of the
industrial relations environment, appear to be the principal source of the decline in
union membership.'^
V. Conclusions
Much of the decline in union membership during the period 1958-1982 is explained
by changes in effects of "stmctural" variables, especially the proportion of women,
of older workers, and of blue-collar workers in a state's labor force, which is consistent with M-N's findings. However, this study does not support M-N's conclusion
that the most important factor in explaining this decline is the increasing proportion
of women in the labor force. This difference may refiect the sample periods chosen
for analysis. It also refiects the addition of a regressor capturing part-time work status,
as well as substantial changes in the effects of the extent of blue-collar and older
workers on union representation.
Conceming the "union organizing" hypothesis, this study suggests some positive
effects of union militancy and union organizing success on the extent of union membership. However, decomposition analysis reveals that these effects have not contributed substantially to the decline in union membership over this period; i.e., these
effects are modest in comparison with those attributed to the "stmctural" variables.
There appears to be little basis for attributing a substantial portion of the decline
in union membership to aggressive employer unfair labor practices, to changes in
states' political attitudes towards unions, or to the presence of state right-to-work
laws. Of course, reduced-form models such as mine, M-N's, and others inevitably
obscure important interactions among these and other factors underlying the movement away from union representation. It is unlikely, therefore, that repeated model-
C. TIMOTHY KOELLER
31
ing of this type, especially using aggregated data, will increase our ability to sort out
the effects of these factors. This study also suggests Stepina and Eiorito's conclusion
that recent research on workers' entry into and exit from unions (e.g., studies examining the results of individual certification and decertification elections) will continue to be a key source of information on the processes contributing to changes in
union representation.
NOTES
*I am indebted to John W. Ballantine, Frederick W. Cleveland, and an anonymous referee for their comments on an earlier version of this paper.
' M - N did not precisely define their firm size regressor; the regressor FS captures large-firm effects on the
extent of union membership.
2Leonard (1992) also recently found that a substantial portion of the decline in union representation may
be due to relatively slow employment growth in unionized plants.
^The extent of a state's population living in urban areas was found to be highly collinear with the variables
FS and UAV within the 1958-1982 data set.
"•AS noted below for the measure of employer unfair labor practices, the regressor UNULP is lagged one
year relative to the year for which the extent of union membership is measured.
^See, e.g., Roomkin and Harris (1984), Koeller (1992), and Elliott and Huffman (1984).
^See Freeman (1985, p. 47).
''M-N and Neumann and Rissman also recognized this possible source of bias.
'Census of Population data for these variables for 1960 are repeated for 1958; data for 1970 are repeated
for 1968, 1972, and 1974; data for 1980 are repeated for 1976, 1978, and 1982.
'Zero-order correlation coefficients between WLF and YLF, ULP, and SLF are .72, .35, and -.37, respectively, within the 1958-1982 data set.
'"it also reflects collinearity between SLF and WLF (correlation coefficient = .21), NWLF (-.49), and
SOUTH (-.43) in the sample period 1958-1972.
"The probability value for the estimated coefficient for PCTPT in equation (5) is 17 percent.
'^In contrast with M-N, the ULP regressor (as well as UNULP) is not expressed per thousand employees
in a state. The absolute amounts were used to construct these regressors to avoid possible spurious correlation with the dependent variable.
''There may also remain some joint endogeneity between this regressor and the extent of union membership, i.e., charges of employer unfair labor practices may be filed more frequently in states having substantial union membership.
'''Like M-N, in conducting the decompositions the model was reestimated using the natural logarithm of
the extent of union membership as the dependent variable.
" i t is, of course, impossible for the measure of unionization to decline by more than 100 percent. However, the orders of magnitude of the predicted effects attributed to each of the model's regressors provide a
basis for comparing their relative explanatory powers.
'*M-N's decomposition analysis reveals a similar finding, except that most of the effect on union membership attributed to their WLF regressor is accounted for by changes in its mean over the period 1950-1980.
32
JOURNAL OF LABOR RESEARCH
REEERENCES
Ashenfelter, Oriey, and John Pencavel. "American Trade Union Growth: 1900-1960." Quarterly Joumai
of Economics 42 (August 1969): 435-48.
Blinder, Alan. "Wage Discrimination: Reduced Form and Structural Estimates." Journal of Human
Resources 8 (Fall 1973): 436-55.
Dickens, William T, and Jonathan S. Leonard. "Accounting for the Decline in Union Membership, 19501980." Industrial and Labor Relations Review 38 (April 1985): 323-34.
Elliott, Ralph. "Do Right-to-Work Laws Have an Impact on Union Organizing Activities?" Journal of
Social and Politicai Studies 4 (Spring 1979): 81-93.
. and James R. Huffman. "The Impact of Right-to-Work Laws on Employer Unfair Labor
Practice Charges." Joumat of Labor Research 5 (Spring 1984): 165-76.
Farber, Henry. "Right-to-Work Laws and the Extent of Unionization." Joumai of Labor Economics 2 (July
1984): 319-53.
. "The Extent of Unionization in the United States." In Challenges and Choices Facing
American Labor, ed. Thomas Kochan. Cambridge: MIT Press, 1985.
Freeman, Richard. "Why Are Unions Faring Poorly in NLRB Representation Elections?" In Challenges
and Choices Facing American Labor, ed. Thomas Kochan. Cambridge: MIT Press, 1985.
.. "The Effect of the Union Wage Differential on Management Opposition and Union Organizing Success." American Economic Review 76 (May 1986): 92-6.
and M. Kleiner. "Employer Behavior in the Face of Union Organizing Drives." Industriai
and Labor Reiations Review 43 (April 1990): 351-65.
Hirsch, Barry. "The Determinants of Unionization: An Analysis of Interarea Differences." Industriai and
Labor Relations Review 33 (January 1980): 147-61.
Hunt, Janet C , and Rudolph A. White. "The Effects of Right-to-Work Legislation on Union Outcomes:
Additional Evidence." Joumai of Labor Research 4 (Winter 1983): 47-64.
. "The Effects of Management Practices on Union Election Returns." Journal of Labor
Research 6 (Fall 1985): 389-404.
Koeller, C. Timothy. "Wages, Trade Union Activity, and the Political Environment of Unionism: A Simultaneous Equation Model." Joumai of Labor Research 6 (Spring 1985): 147-66.
. "Employer Unfair Labor Practices and Union Organizing Activity: A Simultaneous Equation Model." Joumai of Labor Research 13 (Spring 1992): 173-87.
Lawler, James J., and Greg Hundley. "Determinants of Certification and Decertification Activity." Industrial Relations 22 (Fall 1983): 335-48.
Leonard, Jonathan S. "Unions and Employment Growth." Industrial Relations 31 (Winter 1992): 80-94.
Lumsden, Keith, and Craig Petersen. "The Effect of Right-to-Work Laws on Unionization in the United
States." Joumai of Political Economy 83 (December 1975): 1237-48.
Moore, William J., and Robert J. Newman. "A Cross-Section Analysis of the Postwar Decline in American Trade Union Membership." Joumai of Labor Research 9 (Spring 1988): 111-26.
Neumann, George R., and Ellen R. Rissman. "Where Have All the Union Members Gone?" Joumai of
Labor Economics 2 (April 1984): 175-92.
Roomkin, Myron J., and Dawn A. Harris. "Interindustry Pattems in Unfair Labor Practice Cases." Joumai
of Labor Research 5 (Spring 1984): 113-26.
Stepina, Lee P., and Jack Fiorito. "Toward a Comprehensive Theory of Union Growth and Decline."
Industrial Relations 25 (Fall 1986): 248-63.