Union Coverage,Membership and Performance-Related Pay: Are Piece Rates Different?

Abstract

This paper revisits a literature on the links between unionisation and performance-related pay (PRP), which offers a disparate set of results. Part of the reason for this is the usual inability to distinguish between different types of PRP and the lack of rich panel data containing such measures. Analysis of panel data containing six separate PRP schemes reveals that union members are less likely to receive PRP in general and less likely to receive bonus payments, stock options, or profit sharing than non-members. Furthermore, profit sharing is negatively related to both union membership and coverage. However, union members or those covered by a union contract are more likely to be paid piece rates, providing further evidence that piece-rate pay differs from other forms of PRP.

Keywords

performance pay unions piece rates

1. Introduction

This paper revisits an inconclusive body of evidence on the relationship between union status and performance-related pay (PRP). Its major contributions are that it analyses a rich set of panel data containing several measures of PRP and unionisation. There is a large body of research concerning whether unions oppose performance-related pay (PRP). However, relatively few studies employ panel data containing covariates known to be associated with both PRP and unionisation. Nor do many studies contain several measures of PRP. In what follows, I estimate the relationship between various forms of PRP and two measures of unionisation using panel data containing a set of covariates to test whether there are differences in the occurrence of various PRP schemes among union members and those covered by a union contract. Although there is previous evidence on the prevalence of PRP and unionisation using the National Longitudinal Survey of Youth 1979 (NLSY79), none combine all periods of PRP data available to consider explicitly the impact of union status on various PRP schemes. I provide new evidence that there are differences in which PRP schemes are more or less prevalent among unionised and non-unionised workers within a cohort of US workers from the NLSY79. One reason the US labour market is unique is due to the low and declining rates of unionisation. Consequently, it may be of interest to explore how the prevalence of various PRP schemes differs between union members, union covered and non-unionised workers in the presence of lower levels of collectivisation.

Confirming much of the existing literature, union membership is negatively related to most forms of PRP. However, there are no significant differences in the incidence of PRP among those who are only covered by a collective bargaining contract and those who have no union coverage after controlling for other available factors. Analysing each PRP scheme separately reveals that union membership and coverage have a positive influence on the receipt of piece-rate pay, especially among the male subsample. This positive relationship persists despite efforts to control for occupation and individual effects.

Additionally, I find that union members are less likely to receive bonus payments, stock options, or profit sharing. Furthermore, those who are covered by a union contract are less likely to receive profit sharing. Therefore, corroborating prior research by Brown, (1990) and MacLeod and Parent (1999), piece-rates schemes must provide different economic incentives from other forms of PRP. Consequently, there is evidence that unions may not necessarily oppose all PRP schemes. These findings are robust and persist notwithstanding several different estimation techniques including fixed-effects, which control for fixed individual unobservables that might otherwise bias links between union status and PRP.

This paper proceeds as follows. Section 2 sets the background for the relationship between PRP and union status and reviews prior literature. Section 3 describes the data and discusses determinants of PRP and unionisation as well as sample construction. Section 4 describes the estimation techniques and empirical analysis, while section 5 offers discussion and conclusions.

2. PRP and unionisation: setting the context

Setting the background

There are several theories as to why unions and unionised workers are thought to oppose the implementation of PRP. Freeman (1982) theorised that unions would oppose PRP because it weakens union solidarity by increasing earning disparities among those covered by a collective bargaining agreement. He also noted that unions typically prefer standard wage rate systems as protection from supervisory discretion in evaluations. Implicit in this theory is the fact that it would be harder for unions to organise or service a group of heterogeneous workers, suggesting that diversity within the work group may undermine the credibility of policies based on the median voter. Secondly, workers and unions often oppose the use of PRP schemes when worker pay is subject to external factors beyond the control of individual workers’ efforts, such as recessions or weather events. Additionally, unions may oppose PRP because it reduces implicit profit sharing between labour and capital. The implicit profit sharing function occurs because the union is able to capture more of the rents for labour, more than labour could achieve in the absence of a union. If a larger fraction of unionised worker's pay is variable and not subject to union influence, then this likely weakens the union and its ability to capture rents. Furthermore, many forms of PRP are intended to incentivise workers to be more productive, which may leave some workers, especially lower productivity workers, worse off. Lastly, as pointed out by Kruse and Blasi (1995), rigid work rules surrounding union agreements often make the implementation of PRP schemes problematic. Prior research does find that non-unionised workplaces are more likely to use individualised PRP schemes such as piece rates and commissions (Brown, 1990; Long and Shields, 2005; Lemieux et al., 2009). Hence, unions may oppose the implementation of a PRP scheme.

On the other hand, unions may not necessarily be opposed to all forms of PRP. Warhurst and Heery (1994), and Heery (1997a; 1997b) theorise that if a new Human Resource Management (HRM) scheme is objective and modified to union demands, unions may provide an avenue for more effective implementation of the new HRM practice. Warhurst and Heery (1994) find that in response to individualised PRP schemes, most unions try to seek regulation of the schemes through collective bargaining and report considerable success in doing so. Heneman, von Hippel, Eskew, and Greenberger (1997) point out that most unions oppose PRP centred on subjective performance evaluations since unions typically cannot influence subjective evaluations. Consequently, unions may be especially opposed to subjective PRP schemes in which they have no discretion. However, unions may not necessarily oppose objective schemes if they are able to mitigate potential adverse impacts on their constituents. Furthermore, unions may also be able to prevent the well-known ‘ratchet effect’ by enforcing and monitoring the terms of the scheme.¹ Therefore, the relationship between PRP and union coverage is theoretically ambiguous.

In what follows, I apply panel data estimations on an aggregated measure of PRP and individual specific PRP schemes as the dependent variables with union status and a rich set of controls as the independent variables. First, I explore the relationship between the various measures of PRP to other determinants of PRP, including union status, using the following equation:

\begin{array}{l} P R P_{i t} & = & a_{i t} + b 1_{i t} Union Status \\ + & b 2_{i t} Job Characteristics \\ + & b 3_{i t} Individual Characteristics + e_{i t} \end{array}

(1)

where PRP represents the form of performance-related pay and union status represents whether the individual is a union member or covered by a union contract. The vector of job characteristics includes industry, occupation, average hours of work per week, four separate firm size indicators, three different shift indicators, tenure in years and tenure squared, experience in years and experience squared, and the number of jobs held. The vector of individual characteristics includes, gender, race (black, Hispanic, non-black non-Hispanic), age, marital status (never married, married-spouse present, other), education as measured by highest grade completed, and the individual's ASVAB score as a proxy for ability. Summary statistics for all of these variables are presented in table 1.

Table 1.

Prevalence of performance-related pay

	Incentive pay	Non-production bonuses	Performance-related pay
			Broad definition	Narrow definition
Civilian economy	4	34	37	19
Private sector	5	35	39	21
State and local government	0	25	25	2
Private sector only:
Major industry
Mining	0	46	46	28
Construction	2	27	28	20
Manufacturing	3	47	49	26
Wholesale and retail trade	10	32	39	28
Transportation and utilities	10	35	40	24
Information	6	54	57	36
Financial activities	12	60	67	32
Professional and business services	4	38	40	20
Education and health services	1	31	31	12
Leisure and hospitality	1	17	18	9
Other services	11	24	33	25
Establishment size
1 to 49 workers	6	29	33	23
50 to 99 workers	6	31	34	19
100 to 499 workers	5	37	40	20
500 to 999 workers	3	47	48	19
1000 workers or more	2	52	52	20
Major occupation group
Management, business and financial	5	51	55	28
Professional and related	1	45	45	19
Sales	19	29	42	35
Office and administrative support	3	41	43	23
Service	2	21	22	10
Construction and extraction	1	24	25	17
Installation, maintenance and repair	8	32	38	26
Production	3	41	43	23
Transportation and material moving	9	31	36	21
Work level
1-4	4	26	28	16
5-8	8	40	45	27
9-12	5	52	55	28
13-15	1	61	61	33
Missing work level	3	46	49	25
Position on basis of hourly wage
Lowest 10 per cent	1	13	14	9
Lowest 25 per cent	2	18	20	10
Second 25 per cent	4	36	39	24
Third 25 per cent	7	41	46	24
Top 25 per cent	7	49	54	28
Top 10 per cent	7	54	58	30
Top 5 per cent	9	53	58	31
Top 1 per cent	17	54	62	39
Full-time	6	39	43	24
Part-time	2	14	16	8
Union	1	31	32	15
Non-union	5	36	39	22

Notes: Incidence rates are per cent of hours worked, Employer Costs for Employee Compensation data, first quarter 2013. ‘Civilian’ includes private industry and State and local government, but excludes the Federal Government. Excluded from the private industry scope are the self-employed, farm and private household workers, as well as those who set their own pay. Wage position is based on entire civilian sample.

B. Prior literature

Although the prior literature is far from unanimous, the majority detect a negative relationship between aggregated measures of PRP and union coverage.² Many researchers have also noted a trend of simultaneously declining rates of unionisation (Farber and Western, 2000) and increasing use of PRP (Lemieux, MacLeod, and Parent, 2009). This is especially true within the US labour market where unionisation rates are historically low. For example, Verma (2005) observes increasing use of bonus pay and declining rates of unionisation between 1976 and 1991. Assuming unions oppose PRP, one would expect declining unionisation to be associated with increased usage of PRP. The NLSY also reveals rising rates of PRP usage within the periods PRP data are available. Hence, due to declining unionisation, employers may be encouraged to implement PRP schemes without fear of union interference.

One explanation for the inconclusive literature is that researchers have employed incongruent definitions of PRP. Some past measures include variable pay, gain sharing and piece rates, to name just a few. Often, researchers may only have data that included one or two measures of PRP. An advantage of using NLSY79 data is that the data contain several separate measures of PRP. Obviously, PRP schemes take many forms. Therefore, it is necessary to review the prior literature on the different PRP schemes. Although Bronars and Moore (1995) observed a positive relationship between unionisation and piece rates, others observed a negative relationship (Brown and Phillips, 1986; Ng and Maki, 1994; Heywood, Seibert and Wei, 1997; MacLeod and Parent, 1999; and Long and Shields, 2005). Notably, the majority observe no statistically significant difference between piece rates and unionisation (Brown, 1990; Drago and Heywood, 1995; Heywood and Wei, 1997; MacLeod and Parent, 1999 – using NLSY data; Heywood, Hübler and Jirjahn, 1998; and Geddes and Heywood, 2003). Consequently, prior research is less conclusive concerning the union's response to piece-rate plans.

Prior literature reveals a similar opaque relationship between other types of PRP and unionisation. Although few papers explicitly consider the relation between union coverage and the use of commissions, the vast majority of those observe a negative relationship between the two.³ Similarly, the vast majority observe that bonuses and profit sharing are negatively related to union coverage.⁴ However, recent research by Sesil, Kroumova, Blasi and Kruse (2002) finds that union firms that employ gain sharing plans fare better than those that do not. Additionally, research on group incentive plans reveals that unions may not be as opposed to these types of plans as originally believed (Kruse, 1993). Kroumova and Sesil (2006) observe that it is difficult to predict how union status and group-based compensation would be related because some employers may adopt these types of PRP plans as a deterrent to unionisation. There is very little prior research concerning the relationship between tips and union coverage. Gregg and Machin (1988) observe a positive relation between stock options and union coverage while both Jones and Pliskin (1997) and Long and Shields (2005) find the two to be negatively related. Of course, many of these papers had different definitions of PRP and union coverage as well as methodology and data, so making direct comparisons can be problematic. The next section introduces the data, sample selection and variable construction.

3. Sample choice and data description

The NLSY79 follows a nationally representative cohort of the population born between 1957 and 1965. These individuals were interviewed annually between 1979 and 1994 and semi-annually from 1996. Unfortunately, key PRP data is only available for the 1988–90 and 1996–2000 waves. As a consequence, analysis is necessarily restricted to those six years. Following Bronars and Moore (1995), I eliminate those who report working for the government or in the agricultural sector as well as the self-employed.⁵ Additionally, I remove observations with missing information and drop the unemployed, resulting in an unbalanced panel of 28,166 person-wave observations across 8,551 individuals.

Within the NLSY79, respondents were asked if any of their earnings were based on piece rates, commissions, bonuses, stock options, tips, or other forms of performance-related pay at their current or most recent job. Unfortunately, individuals were not asked what amount of their pay was associated with PRP, so it is impossible to explore the relationship between union coverage and the amount of PRP based compensation. Consequently, the variables for each category of PRP will necessarily be dichotomous. I exclude ‘other PRP’ as a separate category since it is impossible to ascertain what ‘other PRP’ entails.⁶ Also, respondents were asked if they were offered any fringe benefits in the form of profit sharing at their current or most recent job. Although profit sharing is considered to be PRP according to the U.S. Bureau of Labor Statistics (BLS), it is analysed separately from PRP because the NLSY considered it as a fringe benefit distinct from PRP. Also, since the PRP and profit sharing questions were asked in separate areas of the survey, differing attrition rates result in differing numbers of observations when estimations involve profit sharing in comparison to PRP. I also create an aggregate PRP indicator which takes a value equal to 1 if the respondent reported any pay in the form of piece rates, commissions, bonuses, stock options or tips. Within the subsample, 25 per cent of those working report receiving some kind of PRP as defined above, while 33 per cent report receiving profit sharing.

There are also three union related variables available over the period of examination. For each year under investigation, individuals were asked whether they were “a union member or a member of some association similar to a union”. For the 1988–90 waves, individuals were asked whether their “wages or salary were set by a collective bargaining agreement between a union or employee association”. For the 1996–2000 waves, individuals were asked whether “they were covered by a union or employee contract.”⁷ From this information, I construct two separate indicators of unionisation. Union covered equals one if individuals reported that (i) their wages or salary in their job is set by collective bargaining agreement between their employer and a union or employee association or (ii) individuals reported they were covered by a collective bargaining agreement. Union member equals one if the respondent reports that they are a member of a union or employee association, zero otherwise. Unsurprisingly, all union members were either covered by a union contract or had their wages set by collective bargaining. Among those covered by a union contract, 25 per cent were not union members. Within the selected sample, 15 per cent reported that they had some union coverage while only 11 per cent reported being actual union members. Prior research notes that this distinction is meaningful (Jones, 1982). Consequently, analysis will explore the impact of both union coverage and membership on the proclivity of receiving various forms of PRP.

Summary statistics for the variables reveal that, as expected, usage of PRP appears more prevalent within the non-union sector and is much lower among union members than among those who are only covered by a union contract. Absent controls, all PRP schemes except piece rates appear to be negatively correlated with unionisation. Confirming Freeman (1982), wage levels are higher among unionised workers (both members and affiliated) than among those with no union coverage. Also, those receiving PRP receive higher average wages than those not receiving PRP, confirming similar findings presented by Booth and Frank (1999). Comparisons of means show expected gender and racial differences in union coverage and PRP. Specifically, there were more men in unions receiving PRP than women and women disproportionately receiving piece rates.⁸ Comparisons of means also reveal higher union coverage among non-whites but lower incidences of PRP. However, comparisons of means can be deceiving because there might be other confounding factors involved. Therefore, the next section will discuss estimation techniques and empirical results.

4. Empirical analysis

Given the dichotomous nature of the variables of interest and the availability of panel data, focus will be placed on estimating simple linear probability models and a pooled probit as a benchmark. One obvious advantage of panel data is the ability to control for person-specific factors. However, there are typically two basic panel data models to choose from, the random-effects model and the fixed-effect model. A Hausman test reveals that the random-effects estimates are inconsistent and biased. Therefore, fixed-effects estimation may be more desirable.⁹ Furthermore, the fixed-effects estimator is more robust concerning problems of selection bias than the random-effects estimator (Kennedy, 2003) because the fixed-effects procedure is better at controlling for selection characteristics. Consequently, I estimate panel fixed-effects because this model eliminates bias arising from fixed individual unobservables that might otherwise bias the estimates. This is also done to explore possible OLS bias. I also estimate a bivariate probit model. This is done as a comparison to the pooled probit estimates and to determine whether there are unobservables that jointly influence both union coverage and PRP. Next, I test whether the main results differ by gender or within certain occupations known to be associated with PRP. This is performed in order to determine whether the results are being influenced by factors such as occupational crowding or gender difference, which in turn might be influencing the observed relation between PRP and unionisation. However, if there is non-random selection with respect to occupational choice, then restricting the sample to those occupations could bias downward the unionisation coefficients.

There are well known selection effects associated with PRP. For example, workers who excel under a piece-rate scheme may be choosing piece-rate occupations, irrespective of whether or not the job is unionised. Therefore, restricting the sample to these occupations could downward bias the union status coefficients.¹⁰

The results for the baseline linear specifications are presented in table 2 while the results for the pooled probit estimations appear in columns 1 and 2 of tables 4 and 5. As displayed in tables 4 and 5, most of the independent variables take the sign and significance observed in prior research.

Table 2.

Baseline OLS

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
	PRP	Piece rates	Commissions	Bonuses	Stock options	Tips	Profit sharing
Linear probability model for union covered
Union covered	−0.039	0.005	−0.009	−0.02	−0.01	−0.011	−0.048
	(5.27)***	(1.83)**	(2.47)**	(3.31)***	(3.68)***	(3.68)***	(5.97)***
Obs.	28,166	28,166	28,166	28,166	28,166	28,166	26,716
R2	0.09	0.04	0.16	0.07	0.04	0.17	0.1
Linear probability model for union members
Union member	−0.078	0.001	−0.017	−0.049	−0.015	−0.014	−0.078
	(9.20)***	−0.42	(3.91)***	(6.99)***	(4.82)***	(4.38)***	(8.59)***
Obs.	28,143	28,143	28,143	28,143	28,143	28,143	26,694
R2	0.09	0.04	0.16	0.08	0.04	0.17	0.11

Notes: Absolute value of t-statistics in parentheses.

* significant at 10%;

** significant at 5%;

*** significant at 1%.

Each estimation includes a full set of controls, including industry and occupation at the one digit level, average hours of work per week, four separate firm size indicators, three different shift indicators, tenure in years and tenure squared, experience in years and experience squared, and the number of jobs held, gender, race (black, Hispanic, non-black non-Hispanic), age, marital status (never married, married-spouse present, other), education as measured by highest grade completed, and the individual's ASVAB score as a proxy for ability.

The basic model for the fixed-effects estimation is:

y_{i t}^{*} = a_{i} + β^{'} x_{i t} + ε_{i t}

(2)

where

\begin{matrix} ε_{i t} \sim N (0, 1) & and & y_{i t} = {\begin{array}{l} 1 if y_{i t}^{*} > 0 \\ 0 if y_{i t}^{*} \leq 0 \end{array} \end{matrix}

where i refers to the individual and x_it is a vector of explanatory variables including union coverage or membership with estimated parameters of β and

y_{i t}^{*}

is a latent index that captures the tendency of an individual to receive various forms of PRP. The bivariate probit model is specified as:

y_{1}^{*} = β^{'} x_{1} + ε_{1}

(3)

where

ε_{1} \sim N (0, 1) and ε_{2} \sim N (0, 1)

y_{2}^{*} = β^{'} x_{2} + ε_{2}

and

\begin{matrix} y_{1} = {\begin{array}{l} 1 i f y_{1}^{*} > 0 \\ 0 i f y_{1}^{*} \leq 0 \end{array} & and & y_{2} = {\begin{array}{l} 1 i f y_{2}^{*} > 0 \\ 0 i f y_{2}^{*} \leq 0 \end{array} \end{matrix}

where ρ represents the correlation between ε₁ and ε₂. If ρ = 0, then the model collapses into separate probit models for y₁ and y₂. However, when ρ ≠ 0, then the variables appearing in one part of the model have an indirect influence on the second part of the model. According to Booth and Francesconi (2000), if higher ability workers sort into jobs that have a low probability of being covered by collective bargaining, then one should detect a negative correlation in unobservables (ρ < 0). Alternatively, if lower ability workers sort into jobs covered by collective bargaining where they receive PRP pay, then a positive correlation in unobservables will result (ρ > 0). Consequently, unmeasured factors may be jointly influencing the decision to take jobs that entail both PRP and union coverage. Booth and Francesconi (2000) explain the positive correlation between unobservable components as workers sorting into jobs that entail both PRP and union coverage. They point out that this provides indirect evidence that there are no inherent contradictions between collectivism and individualism which could be interpreted as PRP being positively associated with union coverage. Consequently, I replicate their analysis using NLSY79 data so as to explore this issue in the context of the United States.

Table 2 presents the results of the linear probability analysis for each of the corresponding PRP indicators defined above. As shown in table 2, column 1, workers who are covered by a collective bargaining contract or are union members are less likely to receive PRP. The results for each of the individual PRP schemes are presented in columns 2–7 of table 2. Of these, piece rates are the only PRP scheme that is positively associated with union coverage. The positive association between piece rates and union coverage provides preliminary support for the hypothesis that unions may not be as opposed to piece-rate plans since they may be able to influence the terms of the piece-rates system. Additionally, linear probability estimation reveals that union coverage and membership are negatively related to commissions, bonuses, stock options, tips and profit sharing.

The association between union status and PRP type observed in the baseline models may, however, be attributable to fixed individual unobservables biasing the estimates. One of the advantages of panel data is that it enables one to control for individual heterogeneity, which likely influences both union coverage and pay schemes. Jobs associated with PRP or unions are known to attract higher-ability workers. Union firms may comprise a disproportionate number of higher-ability workers since the higher pay and termination restrictions attract a larger number of applicants. Consequently, unionised firms may carefully screen applicants and ‘cream skim’ those who appear to be higher-productivity workers, potentially resulting in a disproportionate number of higher ability workers within union establishments. Sorting on PRP may occur because higher ability workers earn higher returns to their ability in PRP jobs than they could in salaried or time rate positions. Additionally, PRP is known to elicit more effort than standard time rates. Accordingly, jobs that entail PRP and/or union coverage may be attracting a disproportionate number of higher ability workers. Therefore, it is essential to control for individual heterogeneity.

In order to compensate for individual heterogeneity, a fixed-effects model is applied to the panel as per equation (2) above. Also, in all estimations, rho is significantly different from zero, implying that fixed-effects estimation is capturing a significant amount of individual heterogeneity. Consequently, the fixed-effects estimations are an improvement over those presented earlier. The coefficients for the fixed-effects estimation appear in table 3 for both union coverage and union membership. These estimates employ the same specification as described in equation (1), however, all of the time invariant variables drop out. There are some significant differences between the coefficients of the baseline models and the fixed-effects model. Union coverage is no longer negatively related to PRP; however, union membership remains negative and significantly related to the aggregate measure of PRP. This may be due to the stronger voice union members have over others who may only be covered by a union contract. Accordingly, with more covered members, the union will be stronger and more able to resist the imposition of PRP.

Table 3.

Fixed-effects estimations

	(1)	(2)	(3)	(4)	(5)	(6)	(7)
	PRP	Piece rates	Commissions	Bonuses	Stock options	Tips	Profit sharing
Individual fixed-effects for union covered
Union covered	0.006	0.017	0.003	0.008	−0.003	−0.007	−0.028
	−0.57	(4.33)***	−0.62	−0.87	−0.75	(1.76)*	(2.58)***
Obs.	28,166	28,166	28,166	28,166	28,166	28,166	26,716
Individuals	8,551	8,551	8,551	8,551	8,551	8,551	8,339
ρ	0.44	0.48	0.52	0.36	0.29	0.53	0.477
F-test (u_i=0)	2.05	2.4	2.83	1.63	1.21	2.83	2.64
Individual fixed-effects for union members
Union member	−0.044	0.015	−0.007	−0.034	−0.012	−0.013	−0.058
	(3.35)***	(2.97)***	−1.07	(2.95)***	(2.25)**	(2.74)***	(4.29)***
Obs.	28,143	28,143	28,143	28,143	28,143	28,143	26,694
Individuals	8,551	8,551	8,551	8,551	8,551	8,551	8,339
ρ	0.44	0.48	0.52	0.36	0.29	0.53	0.476
F-test (u_i=0)	2.04	2.38	2.83	1.62	1.21	2.83	2.63

Notes: Absolute value of t-statistics in parentheses.

* significant at 10%;

** significant at 5%;

*** significant at 1%.

Interestingly, both union members and those covered by a union contract are significantly more likely to receive piece rates. However, many of the statistically significant marginal effects are rather small, implying little practical significance. For example, although union membership is negatively significant, the coefficient is −0.04. This implies that if all other x_it are fixed, then for each 1 unit change in union membership, PRP changes 0.04 units. Also, union coverage and membership remain negatively related to profit sharing. Another difference that emerges is that union coverage no longer has any significant impact on commissions, bonuses or stock options. Union membership, on the other hand, remains negatively related to bonuses, stock options, tips and profit sharing. Consequently, after controlling for individual fixed effects, union membership has a larger impact on whether one receives PRP than simply being covered by a collective bargaining contract or having your wages set by a union. This confirms a strong negative relationship between union membership and PRP observed in prior estimates. This tends to validate previous research by Jones (1982), who observes significant differences between the characteristics of union members and non-members covered by a union contract.

The bivariate probit estimates reveal a negatively significant value for ρ for both the aggregated measure of PRP as well as for most of the individual PRP schemes except for piece rates.¹¹ This implies that variables appearing in one part of the model have an indirect influence on the other part of the model. Within the NLSY79, the unobserved components of the likelihood of having union coverage are negatively correlated with the unobserved components of the likelihood of having a job that entails PRP, except for piece rates. This contradicts evidence from Great Britain analysed by Booth and Francesconi (2000). One interpretation is that higher ability workers are sorting into PRP jobs that have a low probability of being unionised. Alternatively, higher ability workers sort into piece-rate jobs that have a high probability of being unionised. This suggests that unions may be less opposed to the introduction of a piece-rate scheme than other schemes since they may have more discretion over the conditions of the piece-rate scheme. Alternatively, unions may be more opposed to other schemes besides piece rates because they have little discretion over the terms of the PRP plan.

Tables 4 and 5 show a comparison between the pooled probits and the bivariate probit for those covered by a union contract and union members respectively. Many of the coefficients take the expected sign and remain significantly different from zero in both specifications. Tenure and experience have a positive but small effect on union coverage and membership, while black or Hispanic status has a positive effect on unionisation. Also, females are less likely to be either union members or covered by a union contract. Older workers are more likely to be unionised but less likely to receive PRP. However, many of the coefficients are small. Smaller firms are more likely to use PRP but are less likely to be unionised. Additionally, shift workers and off-shift workers are less likely to receive PRP in comparison to those on an irregular shift. As expected, there is a great deal of variation in the incidence of PRP and unionisation by industry and occupation. Furthermore, the average hours of work per week and the number of jobs held have a small but statistically significant impact on PRP and unionisation.

Table 4.

Bivariate probit estimations for union covered

	Pooled probit		Bivariate probit
	(1)	(2)	(3)	(4)
	PRP	Union covered	PRP	Union covered
Bivariate predicted probability	−	−	0.243	0.092
Actual probability	−	−	0.263	0.112
Union(a) covered	−0.038(4.93)***	−	−−	−−
PRP(a)	−	−0.023	−	−
		(5.07)***	−	−
Tenure (years)	0.003(1.40)	0.011(8.05)***	0.007(1.17)	0.055(8.02)***
Tenure squared	0.000(1.35)	0.000(3.59)***	0.000(1.27)	−0.001(3.55)***
Experience (years)	0.014(2.06)**	−0.004(0.74)	0.043(2.07)**	−0.02(0.80)
Experience squared	−0.001(1.71)*	0.001(1.18)	−0.003(1.73)*	0.003(1.23)
Non-white(a)	0.001	0.032	−0.002	0.16
	(0.12)	(7.00)***	(0.08)	(7.00)***
Female(a)	−0.011	−0.033	−0.031	−0.165
	(1.70)*	(6.81)***	(1.52)	(6.76)***
Age	−0.002	0.003	−0.007	0.015
	(1.68)*	(3.25)***	(1.80)*	(3.30)***
ASVAB score	0.001(5.24)***	0.000(4.18)***	0.002(5.34)***	−0.002(4.34)***
Education (grade)	0.012(7.29)***	0.005(4.29)***	0.036(7.15)***	0.024(4.08)***
Small firm(a)	0.036	−0.096	0.127	−0.46
	(3.14)***	(11.93)***	(3.49)***	(12.04)***
Medium firm(a)	0.013	−0.008	0.044	−0.043
	(1.10)	(1.07)	(1.15)	(1.10)
Largest firm(a) firm	0.005(0.37)	0.010(1.15)	0.014(0.35)	0.048(1.37)
Shift work(a)	−0.054	−0.013	−0.164	−0.055
	(3.83)***		(3.79)***	(0.98)
Off shift(a)	−0.036	0.043	−0.125	0.206
	(2.54)**	(3.45)***	(2.69)***	(3.55)***
Professional(a)	−0.028	−0.027	−0.086	−0.139
	(0.36)	(0.48)	(0.35)	(0.48)
Managerial(a)	0.080	−0.055	0.245	−0.327
	(0.97)	(1.08)	(0.99)	(1.12)
Sales(a)	0.239	−0.020	0.657	−0.138
	(2.62)***	(0.36)	(2.62)***	(0.47)
Clerical(a)	−0.074	0.011	−0.251	0.063
	(1.00)	(0.19)	(1.01)	(0.22)
Craft(a)	−0.024	0.076	−0.088	0.332
	(0.31)	(1.14)	(0.35)	(1.14)
Operative(a)	0.021	0.133	0.049	0.533
	(0.27)	(1.84)*	(0.20)	(1.83)*
Transport(a) operative	0.011(0.14)	0.096(1.35)	0.021(0.08)	0.392(1.34)
Labourer(a)	−0.065	0.123	−0.236	0.494
	(0.88)	(1.67)*	−0.94	(1.69)*
Service(a)	0.052	0.048	0.152	0.212
	(0.64)	(0.75)	(0.61)	(0.73)
Private(a) house	−0.201(3.11)***	0.026(0.33)	−0.990(3.13)***	0.147(0.40)
Construction(a)	0.000	0.036	−0.004	0.167
	(0.00)	(1.78)*	(0.04)	(1.77)*
Mfg(a)	0.043	−0.009	0.133	−0.051
	(1.60)	(0.52)	(1.61)	(0.58)
Transportation(a)	0.080	0.100	0.223	0.404
	(2.72)***	(4.45)***	(2.56)**	(4.36)***
Wholesaled(a)	0.093	−0.023	0.283	−0.131
	(3.33)***	(1.33)	(3.38)***	(1.44)
Finance(a)	0.077	−0.078	0.238	−0.520
	(2.63)***	(4.98)***	(2.73)***	(5.07)***
Business(a)	0.085	−0.038	0.258	−0.221
	(2.93)***	(2.21)**	(2.99)***	(2.30)**
Personal(a) services	0.172(5.17)***	−0.058(3.28)***	0.490(5.27)***	−0.373(3.47)***
Entertainment(a)	0.037	−0.001	0.113	−0.008
	(1.04)	(0.03)	(1.05)	(0.06)
Professional(a) services	−0.089(3.53)***	−0.013(0.70)	−0.300(3.51)***	−0.056(0.60)
Public(a) administrator	−0.206(3.89)***	0.131(2.93)***	−1.055(3.95)***	0.526(3.06)***
Hours of work per week	0.003(11.37)***	0.000(1.96)*	0.010(11.44)***	−0.003(2.30)**
Number of jobs ever held	0.003(4.59)***	0.001(2.11)**	0.009(4.52)***	0.005(1.98)**
Married(a) Spouse present	0.008(1.45)	0.001(0.25)	0.025(1.42)	0.004(0.20)
Young(a) child (<6 yrs)	0.008(1.15)	−0.019(3.48)***	0.029(1.26)	−0.092(3.52)***
Year = 1988(a)	−0.051	0.092	−0.179	0.410
	(2.08)**	(4.19)***	(2.22)**	(4.26)***
Year = 1989(a)	−0.062	0.082	−0.213	0.372
	(2.78)***	(4.15)***	(2.91)***	(4.24)***
Year = 1990(a)	−0.069	0.066	−0.238	0.307
	(3.45)***	(3.74)***	(3.57)***	(3.85)***
Year = 1996(a)	−0.044	0.036	−0.147	0.174
	(2.76)***	(2.69)***	(2.84)***	(2.78)***
Year = 1998(a)	−0.035	0.02	−0.117	0.099
	(2.74)***	(1.90)*	(2.79)***	(1.98)**
Constant	−	−	−1.529	−1.907
			(4.86)***	(5.15)***
Log likelihood	−15,001.60	−10,395.97	−25,412.15
rho			−0.064(4.65)***
Likelihood-ratio test of rho=0:	χ²(1)		[21.63]
Observations	28,166	28,166	28,166

Notes: z and P<|z| correspond to the test of the underlying coefficient being 0. Absolute value of z statistics in parentheses.

* significant at 10%;

** significant at 5%;

*** significant at I%.

(a) dF/dx is for discrete change of dummy variable from 0 to I.

Table 5.

Bivariate probit estimations for union members

	Pooled probit		Bivariate probit
	(1)	(2)	(3)	(4)
	PRP	Union member	PRP	Union member
Bivariate predicted probability	−	−	0.243	0.092
Actual probability	−	−	0.263	0.112
Union(a) member	−0.078(9.00)***		−	−
PRP(a)	−	−0.033(9.22)***	−−	−−
Constant	−	−	−I.53(4.85)***	−2.444(5.42)***
Log likelihoodrho	−14,959.53	−8,259.84	−23,264.89−0.139(8.91)***
Likelihood-ratio test of rho=0:	χ²(1)		[80.09]
Observations	28,143	28,143	28,143

Notes: z and P>|z| correspond to the test of the underlying coefficient being 0. Absolute value of z statistics in parentheses.

* significant at 10%;

** significant at 5%;

*** significant at 1%.

(a) dF/dx is for discrete change of dummy variable from 0 to I. Each estimation includes a full set of controls, including industry and occupation at the one digit level, average hours of work per week, four separate firm size indicators, three different shift indicators, tenure in years and tenure squared, experience in years and experience squared, and the number of jobs held, gender, race (black, Hispanic, non-black non-Hispanic), age, marital status (never married, married-spouse present, other), education as measured by highest grade completed, and the individual's ASVAB score as a proxy for ability.

As a robustness check, all the specifications were performed for both the male and female subsamples. Doing so revealed some differences.¹² Within the male subsample, fixed-effects estimation confirms a positive association between union coverage and piece rates and a negative association between coverage and profit sharing. Among the female subsample, there is a positive relationship between the aggregated measure of PRP and union coverage as well as between union coverage and piece rates. However, there is a negative relationship between union coverage and tips. Moving from coverage to membership, being a union member is negatively associated with PRP overall and bonuses and profit sharing in particular among the male subsample. Also, the piece-rate coefficient remains positive but becomes insignificant. Among the female subsample, female union members are much more likely to receive piece rates but less likely to receive tips or profit sharing. Consequently, this evidence supports previous literature that observes females being more likely to be in unionised piece-rate positions in comparison to males.

In many occupations, there is virtually no use of PRP, raising the possibility that the role being attributed to union coverage may be a function of unmeasured occupational differences, and hence not associated with a union's ability to influence PRP schemes. Furthermore, some occupations have extremely low unionisation rates. In order to account more fully for these occupational differences, estimates are limited to occupational categories in which the various types of PRP are most prevalent. Within the sample, piece-rate jobs tend to be found in Craft, Operative and Labourer occupations, whilst commission jobs tend to be limited mostly to sales occupations. Also, bonuses and stock options mostly occur within Professional and Managerial occupations while profit-sharing jobs are predominantly found within Professional, Managerial, Clerical, Craft and Operative occupations. Fixed-effects estimates within these occupations mimic many of the prior results, and hence are omitted for brevity. Union membership remains positively related to piece rates and negatively related to stock options and profit sharing. However, bonuses are no longer significantly different from zero among both members and those covered by a union contract.

5. Discussion and conclusions

Theory predicts that unions oppose PRP. However, the empirical literature does not seem to convey a clear message as to which PRP schemes unions find most objectionable. This paper presents new evidence on this issue by estimating the relationship between various forms of PRP and two measures of unionisation using a rich set of longitudinal controls. Within the cohort analysed, union members are less likely to receive PRP in general and less likely to be paid bonuses, stock options and profit sharing in particular. Moreover, those covered by a collective bargaining contract are less likely to receive profit sharing. In addition, both union membership and coverage have a positive influence on the likelihood of receiving piece rates. However, the results on union membership are more robust. Hence, piece rates are somehow different from the other forms of PRP. As argued by Heery (1997a; 1997b), unions may not be as opposed to a PRP scheme if they have the ability to influence the scheme to ensure pay compression among their constituency. If unions are able to influence piece rates to the benefit of the union and the workers it represents, unions may not be opposed to their implementation, hence the positive relation between piece rates and unionisation. Also, there is a strong but small negative relationship between union coverage and profit sharing. However, there is no reason a priori why unions would be opposed to group based PRP plans such as profit sharing and stock options. As pointed out by Balkin (1989), these plans avoid competition between employees and hence do not erode union solidarity. Furthermore, Ng and Maki (1994) observe that unions are more opposed to individual based plans than group based plans. However, Kruse (1993) observes that the presence of a union increases the likelihood of firms adopting cash based profit sharing but not deferred profit sharing. Unfortunately, there is no information on whether the profit sharing plans are contemporaneous or deferred within the NLSY79. These results are robust to several different estimation techniques including fixed-effects and tend to support the hypothesis that collectivism and individualisation of the employment contract are not necessarily at odds for all forms of PRP. Consequently, future researchers should be cognisant as to which types of PRP are included or excluded in their measures of performance pay, since they may have differing impacts on variables of interest such as union coverage or membership. The empirical findings presented fall in line with previous literature that often observes similar patterns. Consequently, the view that unions oppose PRP may require modification.

Future research should focus on explicitly modelling the union's ability to influence the implementation and operation of the PRP scheme. Often, PRP is introduced by management to sharpen incentives and improve productivity. In light of the common perception that unions will oppose any introduction of PRP, management may feel that eliminating the union is the only available strategy. However, if unions are allowed some voice in setting the performance norms and rewards, then there may be no inherent contradiction between collectivism and individualism. If this is the case, both employers and employees should be able to split fairly the resulting productivity gains from the implementation of PRP.

However, the presented results are far from conclusive. Ideally, future research would involve developing a model that explicitly defines the mechanisms linking PRP and union status. This model would illustrate whether or not PRP exists in a job, including unionisation as a determinant and combined with another model of how the union affects the implementation of PRP. Future efforts will focus on development of a more extensive theoretical model which may shed light on an improved empirical specification. Another caveat is there is no indication of how much of the worker's pay is dependent on PRP, only whether the worker receives any PRP. Consequently, the estimates may be biased, arising from combining those who receive miniscule amounts of PRP with those whose pay may be entirely determined by PRP.

Another interesting avenue for future research would involve exploration of differences between covered members and covered non-members with respect to PRP. One would anticipate that the causal effects on the use of PRP are mostly attributable to coverage as opposed to membership. However, given the change in the coverage question between the two periods mentioned above, the two periods would likely have to be analysed separately to ensure comparability.

Footnotes

1

The ‘ratchet effect’ refers to a situation whereby employers lower the piece rate in response to observed increases in output, believing that the job was not as difficult as they originally thought, resulting in workers having to work harder in order to make the same amount of earnings.

2

The papers that observed a negative relationship between PRP and union coverage are, in reverse chronological order: Pendleton, Whitfield and Bryson (2009); Hanley and Nguyen (2005); Long and Shields (2005); Brown and Heywood (2005); Black, Lynch, and Krivelyona (2004); Barkume (2004); Garen (1999); MacLeod and Parent (1999); Gunnigle, Turner and D'Art (1998); Heywood, Siebert and Wei (1997); Brown and Phillips (1986); and Freeman (1982). Three others observed a positive relationship: Booth and Francesconi (2000); Heery (1997a,b); Booth and Frank (1997), while two observed no relation: Machin and Wood (2005); Gittleman, Horrigan and Joyce (1998).

3

The papers that observed a negative relationship between commissions and union coverage are, in reverse chronological order: Long and Shields (2005); Geddes and Heywood (2003); Bronars and Moore (1995); and Ng and Maki (1994), while two others showed no relationship between commissions and union coverage: MacLeod and Parent (1998); .

4

Out of the seven papers to consider explicitly bonus receipt, six observed a negative relationship between bonuses and union coverage: Long and Shields (2005); Barkume (2004); Parent (2002); MacLeod and Parent (1999); Bronars and Moore (1995); Bell and Neumark (1993), while one revealed no significant relationship: .

5

Results are not sensitive to these sample restrictions and the main analytical findings are very similar to those presented herein.

6

All regressions for the overall PRP indicator were run inclusive of those receiving other types of PRP with no significant variation in results. Estimation results including other PRP in the definition of PRP are available from the author by request.

7

There were no periods where these two indicators overlap, so it is impossible to give any indication of their relationship. However, there were greater incidences of individuals reporting that their wages or salaries were set by a collective bargaining agreement between a union or employee association than reported being covered by a collective bargaining agreement. This likely arises since the NLSY did not collect information as to whether the union members were covered by a union contract during the 1996–2000 waves. Hence, the incidence of union coverage between the two periods is different. To address this issue, I recoded the union coverage variable to I if the individual is a union member in the 1996–2000 waves. Doing so, the coverage – membership difference varies slightly between the two periods.

8

These comparisons of means are available from the author by request.

9

Hausman test resulted in an F-statistic equal to 84.56271 with a p-value of 0.000326 implying one would reject the null hypothesis that the random-effects estimator is consistent.

10

I would like to thank an anonymous referee for alerting me to this possibility.

11

These estimates can be obtained from the author by request.

12

These estimates can be obtained from the author by request.

References

Balkin

D.B.

(1989), ‘Union influences on pay policy: a survey’, Journal of Labour Research, 27(3), pp. 1–27.

Barkume

A.J.

(2004), ‘Using incentive pay and providing pay supplements in U.S. job markets’, Industrial Relations, 43(3), pp. 618–33.

Bell

L.A.

Neumark

(1993), ‘Lump-sum payments and profit-sharing plans in the union sector of the United States economy’, The Economic Journal, 103(418), pp. 602–19.

Black

S.E.

Lynch

L.M.

Krivelyona

(2004), ‘How workers fare when employers innovate’, Industrial Relations, 43(1), pp. 44–66.

Booth

A.L.

Francesconi

(2000), ‘Collectivism versus individualism: performance-related pay and union coverage for non-standard workers in Britain’, ILR Working Papers No. 061.

Booth

A.L.

Frank

(1999), ‘Earnings, productivity, and performance-related pay’, Journal of Labor Economics, 17(3), pp. 447–63.

Bronars

S.G.

Moore

(1995), ‘Incentive pay, information, and earnings: evidence from the NLSY’, U.S. Department of Labour, Bureau of Labour Statistics, Discussion Paper, Report, NLS 95-23.

Brown

(1990), ‘Firms’ choice of method of pay’, Industrial and Labour Relations Review, 43(3), Special Issue: Do Compensation Policies Matter?, pp. 165-S–82-S.

Brown

(1992), ‘Wage levels and methods of pay’, Rand Journal of Economics, 23(3), pp. 366–75.

10.

Brown

Heywood

J.S.

(2005), ‘Performance appraisal systems: determinants and change’, British Journal of Industrial Relations, 43(4), pp. 659–79.

11.

Brown

Philips

(1986), ‘The decline of piece rates in California canneries: 1890–1960’, Industrial Relations, 25(1), pp. 81–91.

12.

Drago

Heywood

J.S.

(1995), ‘The choice of payment schemes: Australian establishment data’, Industrial Relations, 34(4), pp. 507–31.

13.

Farber

Western

(2000), ‘Round up the usual suspects: the decline of unions in the private sector’, Princeton University, Industrial Relations Sections, Working paper no. 437.

14.

Freeman

R.B.

(1982), ‘Union wage practices and wage dispersion within establishments’, Industrial and Labour Relations Review, 36(1), pp. 3–20.

15.

Garen

(1999), ‘Unions, incentive systems, and job design’, Journal of Labour Research, 20(4), pp. 589–603.

16.

Geddes

Heywood

J.S.

(2003), ‘Gender and piece rates, commissions and bonuses’, Industrial Relations, 42(3), pp. 419–44.

17.

Gittleman

Horrigan

Joyce

(1998), ‘“Flexible” workplace practices: evidence from a nationally representative survey’, Industrial and Labour Relations Review, 52(1), pp. 99–115.

18.

Gregg

P.A.

Machin

(1988), ‘Unions and the incidence of performance linked pay schemes in Britain’, International Journal of Industrial Organization, 6(1), pp. 91–107.

19.

Gunnigle

Turner

D'Art

(1998), ‘Counterpoising collectivism: performance-related pay and industrial relations in greenfield sites’, British Journal of Industrial Relations, 36(4), pp. 565–79.

20.

Hanley

Nguyen

(2005), ‘Right on the money: what do Australian unions think of performance-related pay?’, Employee Relations, 27(2), pp. 141–59.

21.

Heery

(1997a), ‘Performance-related pay and trade union membership’, Employee Relations, 19(5), pp. 430–42.

22.

Heery

(1997b), ‘Performance-related pay and trade union de-recognition’, Employee Relations, 19(3), pp. 208–21.

23.

Heneman

von Hippel

Eskew

Greenberger

(1997), ‘Alternative rewards in unionized environments’, ACA Journal, 6(2), pp. 42–55.

24.

Heywood

J.S.

Hübler

Jirjahn

(1998), ‘Variable payment schemes and industrial relations: evidence from Germany’, KYKLOS, 51(2), pp. 237–57.

25.

Heywood

J.S.

Siebert

W.S.

Wei

(1997), ‘Payments by results systems: British evidence’, British Journal of Industrial Relations, 35(1), pp. 1–22.

26.

Heywood

J.S.

Wei

(1997), ‘Piece rate payment schemes and the employment of women: the case of Hong Kong’, Journal of Comparative Economics, 25(2), pp. 237–55.

27.

Jones

D.C.

Pliskin

(1997), ‘Determinants of the incidence of group incentives: evidence from Canada’, Canadian Journal of Economics, 30(4B), pp. 1027–45.

28.

Jones

E.B.

(1982), ‘Union/nonunion differential: membership or coverage’, Journal of Human Resources, 17(2), pp. 276–85.

29.

Kennedy

(2003), A Guide to Econometrics, 5th Ed, Cambridge, MA, MIT Press.

30.

Kroumova

M.K.

Sesil

J.C.

(2006), ‘Intellectual capital, monitoring, and risk: what predicts the adoption of employee stock options?’, Industrial Relations: A Journal of Economy and Society, 45(4), pp. 734–52.

31.

Kruse

D.L.

(1993), Profit Sharing: Does It Make a Difference?, Kalamazoo MI, W.E. Upjohn Institute.

32.

Kruse

D.L.

Blasi

(1995), ‘Employee ownership, employee attitudes, and firm performance’, NBER Working Paper, (w5277).

33.

Lemieux

MacLeod

W.B.

Parent

(2009), ‘Performance pay and wage inequality’, The Quarterly Journal of Economics, 124(1), pp. 1–49.

34.

Long

R.J.

Shields

J.L.

(2005), ‘Performance pay in Canadian and Australian firms: a comparative study’, International Journal of Human Resource Management, 16(10), pp. 1783–811.

35.

Machin

Wood

(2005), ‘Human resource management as a substitute for trade unions in British workplaces’, Industrial and Labour Relations Review, 58(2), pp. 201–18.

36.

MacLeod

W.B.

Parent

(1999), ‘Job characteristics and the form of compensation’, Research in Labour Economics, 18, (ed. S. Polachek), pp. 177–242.

37.

Maki

(1994), ‘Trade union influence on human resource management practices’, Industrial Relations, 33(1), pp. 121–35.

38.

Parent

(2002), ‘Incentive pay in the United States: its determinants and its effects’, in Brown

Heywood

(eds), Paying for Performance: An International Comparison, Armonk NY, M.E. Sharpe Press, pp. 17–51.

39.

Pendleton

Whitfield

Bryson

(2009), ‘The changing use of contingent pay in the modern British workplace’, Chapter II in Brown

Bryson

Forth

Whitfield

(eds), The Evolution of the Modern Workplace, Cambridge University Press, pp. 256–84.

40.

Sesil

J.C.

Kroumova

M.K.

Blasi

J.R.

Kruse

D.L.

(2002), ‘Broad-based employee stock options in US ‘New Economy's Firms’, British Journal of Industrial Relations, 40(2), pp. 273–94.

41.

Verma

(2005), ‘What do unions do to the workplace? Union effects on management and HRM policies’, Journal of Labour Research, 25(3), pp. 415–49.

42.

Warhurst

Heery

and Kingston Business School, Kingston upon Thames (United Kingdom) (1994), Performance Related Pay and Trade Unions Impact and Response.