ACTIVIDAD DE APRENDIZAJE Elabora una síntesis y al final coméntala frente al grupo.
3.8 LEY DEL MERCADO DE VALORES
Our model of unionised employment is based on the assumption that workers are able to observe differences in the working conditions and compensation packages offered by unionised and non-unionised firms. That is not to say, however, that unions, particularly through the award system, do not influence the working conditions or remuneration of non-unionised workers. Rather, it assumes that a union presence has an additional impact which is manifested as differences in working conditions or rates of compensation, or both. In particular, our model of unionised employment allows (but does not require) the wage structure to differ between the two sectors.
The union wage differential is conventionally estimated by including a union membership dummy in a wage equation, or by allowing full interactions and estimating separate wage equations for union members and non-members. Throughout the thesis we have argued, at some length, that the wage received by an individual in a unionised plant will be independent of his or her union membership status. However, whether this argument is correct is ultimately an empirical question. Accordingly, in this section, we seek to answer three questions: (i) Is there a wage differential associated with unionised employment? (ii) If so, is the unionised employment differential greater than the conventional union membership differential? and (iii) Do members in unionised establishments receive higher wages than non-members?
Union Coverage and Membership
Modelling Wages and Union Coverage
Recall that we assume that the individual i has a wages function of the form
where Wj is the log hourly wage, and Z, is a vector of human capital and other
wage-determining characteristics. The superscript j indicates the union status
of the worker.^^
If we follow the conventional approach by categorising workers
according to their membership status, and allow only the intercept term to vary
between union members and non-members, the wage equation to be estimated
is
^NU TT/SiU iNU. ^
= 60 - So ) 6 1 + 6 3 .
where Uj is a union membership dummy, Z,"^ contains the same elements as Zj
but excludes the constant, and the 61 is a row vector of parameters excluding
the intercept parameter. The union membership differential, d, is measured by
^^Categorising workers by union membership status: j = U for union members and j = NU for non-union members. Categorising workers by both union membership and unionised employment status: j = U U E for union members, j = N U E for non-union members in unionised establishments, and j = NE for workers in non-unionised employ- ment.
Chapter 4
the coefficient of the union membership dummy (ie.
It is often assumed that there the error term in (4.17), €3; (eji =
and the union membership dummy, conditional on Z,, are correlated. If this is so, the OLS estimate of d (given by the coefficient of the
union membership dummy) is biased. However, consistent estimates may be
obtained by using instrumental variables estimation (Duncan and Leigh, 1985),
or by using the Heckman's (1979) control function approach. For the control
function approach the model to be estimated is
= (4-18)
where Ri = ((/)i(U,-4'i))/(<i'i(l-^i)) is the generalised residual from a reduced form
probit model of union membership (Gourieroux et. al. 1987; Vella and
Verbeek, 1993), and a^ is the covariance between the error term in the
membership probit equation, 6,i, and the error term in the wage equation, 63,.
However, Poirier (1980) has shown that the standard control function
approach is incorrect if the selection mechanism consists of more than one
decision, such as when workers who have secured unionised employment
subsequently choose whether to join the union. Under these circumstances, the
^•^The percentage union-non-union wage differential, D , is given by D = e''-1.
Union Coverage and Membership
full model consists of a wage equation and two selection equations.
v: UE-
w;
; _
3iUnion Membership Equation Unionised Employment Equation
Wage Equation
(4.19)
Note that the wage equation has been specified so as to allow all of the
parameters to vary for union members, non-members in unionised
employment, and non-unionised workers. Assuming the errors in the model
are trivariate normally distributed with zero mean and covariance matrix^^
1 ^12 ^13
s = 1 ^23
^31 ^32 O 3 3
(4.20)
then, following Tunali (1986) and Main and Reilly (1992), the double selection
extension of the conventional control function approach can be derived as
follows.
First, separate wage equations are specified for the three different
25
W e have assumed the usual probit normalisation for the t w o selection equations (ie.
. 2 2
169
Chapter 4
groups of workers. The regression equations to be estimated are^^
E(yv^^\U=\,UE=\)
= Z 6 ^ ^ + £ ( e 3 | f / ' > 0 , f / £ : ' > 0 ) U n i o n M e m b e r sI ( / = 0 , C / £ = 1 ) =
Z b ^ ^ ^
+ £ ( € 3 1W < . 0 , UE'>0)
N o n - U n i o n M e m b e r s ( 4 . 2 1 )Eiw"^^
IUE=0) = \ UE'^0)
N o n - U n i o n i s e d W o r k e r sNow, the conditional means of e^, for the three different groups of workers can
be written as
Eie^\U=l,UE=l) = OjjA-j, + o^^X^^
Eie,\U=0,UE=l) =
0,3X2, + (4.22)E(e,\UE-0) = 0,3X0
where Z—TTT A , - - (4.23) ^21 A o V o N '^•22- P 12) P 12)^^The subscript i is omitted in order to avoid notational clutter.
Union Coverage and Membership
and
Finally, the conditional mean of the error term for workers in non-unionised workplaces incorporates a standard Mill's ratio term as there is only a single selection mechanism for these workers.
X
. i M l W
(4.25)Consistent estimation of the wage equations now proceeds by using a modified form of Heckman's (1979) two-step procedure. First, the bivariate probit model of unionised employment and union membership defined by equations (4.1) and (4.10) is estimated. This provides the estimates of jS,, jSj and (7,2 which are necessary to calculate the selection terms in each of the wage equations. The selection terms are included in the wage equations which are then estimated using OLS.
Estimation of the Union Impact on Wages
Using ordinary least squares, we estimate the union wage differential by including a single union status dummy variable in a standard hourly wage