Who Benefits Most From Employee Involvement: Firms or Workers?¹

Richard B. Freeman* and Morris M. Kleiner*

Employee involvement (EI) programs are the leading edge form of personnel and labor relations in the US. While many managers believe that these programs raise productivity and profits, the statistical evidence that EI improves the performance of firms is equivocal. The coefficients on measures of EI in production functions are usually positive but often insignificant or small (Commission on Labor-Management Relations, 1994, Chapter 2; Cappelli and Neumark, 1999) or contingent on other factors (Ichniowski, Shaw Prennushi, 1997; Black and Lynch, 1997). A detailed case study of EI has further confirmed these small effects that were found in large data sets (Kleiner, Leonard, and Pilarski, 1999). If EI programs do not greatly affect productivity, why does business think so highly of them? In this study, we argue that the main beneficiaries of EI are workers and managers. We estimate the effects of EI on productivity using panel data on firms and the effects of EI on workers using a survey of employees and find that EI barely affects firm productivity but substantially improves worker well-being. We offer two explanations for this result.

Firm-based productivity

The data for our analysis of firm productivity comes from a 1993 mail survey of firms of the Society for Human Resource Management (SHRM) conducted by Cheri Ostroff. The SHRM survey was sent to 3402 firms, of whom 373 responded, giving an 11% response rate similar to other studies that attempt to measure human resource practices by mail survey (Kato and Morishima, 1996). We matched these data with firm-level data on production and financial outcomes from Compustat from 1983 to 1993 and obtained 273 usable observations. We use a difference in difference design to compare the performance of firms year by year as their EI programs change; contrast firms by EI status in the final year of the survey; and examine changes in productivity and EI practices among firms over a decade.

The SHRM survey asked about eight EI practices: self-managed work teams, worker involvement in the design of EI programs, TQM, committees on productivity, worker involvement in work processes, suggestion or complaint systems, information-sharing with employees, and opinion surveys. The survey also asks whether the firms' use of the practice was "very great", covering 80+ % of jobs/workers; "great", 60-79%; "moderate", 40-59%; "some", 20-39%; or "little", 1-19% of workers; and when the practice was implemented: 10 or more years earlier; 5-9 years earlier; 1-4 years earlier, or within a year. Both the number and use of EI practices grew greatly between 1983 and 1993: in 1983 nearly half of the sample had no EI practices and only four companies had all of the practices, whereas in 1993 virtually all had some practice and 94 had all of them. In addition, firms extended the coverage of their programs over time. By 1993, practices introduced earliest such as information sharing or suggestion/complaint systems were used more intensively than practices introduced later such as opinion surveys or giving workers a role in designing work processes. Because firms that use rarer practices generally have the most common ones, the practices fall into a reasonably well-ordered single dimensional "EI" scale, per Guttman scaling or the more general Rasch-type models (Bartholomew, 1996).

To measure firm EI activity over time, we constructed a summated rating scale of employee involvement based on the existence, intensity and timing of the eight practices. We gave each practice a measure from 0 to 5, depending on its presence and extent, and then summed the eight measures. Firms that applied every practice to almost all workers in a particular year received the value 40. Firms with no practices received the value 0. If the practices fit a perfect (Guttman) scale, firms having the lowest scores would have the fewest/least intensive practices while firms with higher scores would have the practices/intensity of those with lower scores and then some. They fit this pattern well enough to make the scale meaningful (Freeman, Kleiner, and Ostroff 2000).

Because the survey asked for a range of years when the program was implemented, we cannot identify the precise years when the firm has a program and when it does not. Our solution is to approximate the existence of a program in a given year by assuming that the program had a uniform probability of being introduced in one of the years in the reported range. For a program introduced 3-5 years ago, this leads to a 1/3rd chance the program existed 5 years ago, a 2/3rds chance it existed 4 years ago, and a 1 chance that it existed 3 years ago. These figures estimate the probability that a firm had a program in year t. Because the survey asked for intensity of use only in 1993, we estimated the intensity of use in earlier years by exploiting the fact that intensity of use is highly correlated with the length of time a program has been in place. We regressed the intensity of use on the length of program life for all companies and programs taken together in 1993 (for the details of this estimation, see Freeman, Kleiner, and Ostroff, 2000) and used the estimated coefficients to predict the unobserved intensity in years prior to 1993.

Given the estimated probability that a firm had a particular EI program in a given year (p') and the estimated intensity of use (U') in that year, we calculated a scale of EI activity for each year and program as the product p'U' and then summed the values across programs for each firm in each year to obtain an EI variable that measures the number of programs and their intensity of use in a given year. Finally, we estimate the effect of EI using the production function:

(1) ln Q = a + b ln K + c ln L + d EI + YR + FIRM + u

where Q = sales; L = employment; K = book value of assets, from Compustat. FIRM is a vector of firm dummy variables; YR is a vector of year dummy variables; EI is our scale; and u is a residual. With firm and year dummy variables, identification of an EI effect comes from the differential variation of EI over time across firms with productivity.

Table 1 gives the results of our analysis using two different estimating procedures: ordinary least squares and median regressions. The OLS results in line 1 show little or no impact for EI. By contrast, there is a small noticeable effect of the EI scale on ln sales in the line 2 median regression, indicating that results are sensitive to the mode of estimation. These regressions make maximum use of our data but suffer from the possible problem that EI programs require considerable time to bear fruition. The regressions in lines 3-4 deal with this by focusing on the final year of our survey, 1993, by which time many of the programs should have been in a mature state. Here too the estimated coefficients on EI are negligible. Finally, in lines 5 and 6 we regress changes in sales on changes in inputs and the change in EI over the 1983-1993 decade, with the data transformed into average annual changes so that we could include firms for whom we did not have data going back to 1983. Put differently, in this regression we took the longest period for which we had data and annualized the changes. Again, we find no EI effects.

Many analysts believe that a linear specification of the EI effect is incorrect. Rather, EI has a substantial non-linear effect on productivity, so that firms that introduce one involvement activity may gain nothing or even lose because "a single tree does not a forest make" whereas firms that introduce a full spectrum of complementary policies gain from EI (Ichniowski, Shaw, and Prennushi,1997). We mined our data in search of non-linearities, but found little evidence that any such patterns are confounding the EI impact. In sum, our data show that EI has little or no effect on productivity, with a positive effect in only one median regression. Perhaps a larger data set might uncover something that our data fail to reveal, but recent work by Capelli and Neumark (1999) support our finding in a larger file.

Effects on Employees

To find the effects of EI programs on workers, we turn to the Workplace Representation and Participation Survey, WRPS, (Freeman and Rogers, 1999). This is a nationally representative survey of some 2400 workers in firms with over 25 employees that focuses on employee attitudes toward various labor practices. The WRPS is a cross section, so that we cannot follow workers from a firm with an EI program to another firm or conversely. But we can contrast EI participants to non-participants in companies with programs and to employees in companies without programs, and can determine how employees view their firms' EI program.

Table 2 presents cross-tabulations that summarize the responses of non-managerial employees by EI status to questions relating to employee decision-making at work and attitudes toward the firm, and presents the views of EI participants toward the program. More detailed analysis controlling for covariates give comparable results, justifying our focus on the cross-tabs. Panel A compares the percentage of workers who report "a lot" of involvement in the company decisions that affect their work life in six different areas. In each area, workers on EI committees report greater involvement on decisions than other employees. Averaging across areas, EI participants have a 14 percentage point edge over non-participants in firms with programs and a 17 point edge over employees in firms without programs. The similarity between the responses of non-EI participants in firms with EI programs and those in firms without any program indicates that the EI impact is not a simple "good company" effect. Note that the survey asked about workers' influence on the job prior to the module on EI, so responses are not affected by questions about EI. In some areas the differences are striking. A substantial proportion of EI workers have a lot of direct involvement in setting goals for their work group, deciding what training is needed, or how to work with new equipment or software.

Panel B gives the responses from questions relating to satisfaction or attitudes toward work. A much larger proportion of EI participants report themselves as very satisfied with the "influence (they) have in company decisions that affect (their) job or work life" compared to other workers. In addition, proportionately more participants look forward to going to work (as opposed to not caring one way or the other or wishing they did not have to go); are more loyal to their employers; trust that their company will keep its promises to them and other workers; and view their firms' program for dealing with the workplace problems as very effective.

Panel C examines the experiences of EI participants toward their firms' program. Over three quarters say that they personally benefitted by gaining greater influence on the job; over a third say they obtained better wages/benefits. Most important, the vast majority of EI participants said that getting rid of their firms program would have bad or very bad effects on them personally.

Interpretation

What explains the negligible effect of EI in our (and other) production function analyzes compared to the strong effect that workers report EI makes on their working lives?

One possible explanation is statistical. Firm-based studies of EI like ours may fail to find productivity effects because the "true" effects are too small for the production analysis to uncover with any degree of certitude given the sample size, unexplained variation in productivity or changes in productivity, and imperfect measures of EI. Consider, for example, the situation in which a correctly measured EI program raises productivity by 3% over, say, a decade. This is modest given the variability of productivity for a firm over time, but sufficiently large as to raise profits considerably if the gain accrued primarily to capital. In our sample the standard deviation of the residual of changes in ln sales regressed on changes in ln employment and changes in ln assets over the decade and industry dummies is around 0.46 across firms. Assuming that the true population variance in productivity is of comparable magnitude, the critical effect size for detecting the EI impact is about 0.065 (= .03/.46). Then, for a 5% two-tailed t test on the EI coefficient, with 95% power, we would need a sample of about 3,124 observations (interpolated from Kraemer and Thiemann, 1987, p 109). But this calculation assumes that EI is measured correctly. If the signal to noise ratio in the EI variable was non-negligible we would need a larger file to obtain a significant relation, and would have to correct for the measurement error to uncover the true effect of EI. Alternative models for the critical effect size will yield somewhat different results, but the general point is clear: if productivity varies a lot, as it does, and if EI effects are modest, which seems likely, and if EI is measured with error, which is surely true, we need more studies with better measures of EI. to get the right answer or enough studies to permit a careful meta-analysis.

The second explanation is that in fact EI is an innovation whose economic gain accrues largely to workers (and managers) rather than the firm and shareholders. Why might this be? One possibility is that an increasingly educated and knowledgeable work force wants more independent decision-making at their job (even if their decisions will largely mimic those of a more authoritarian management). Since EI has no adverse effect/slight positive effect on the bottom line, firms will offer it to please their workers. Some may even be able to get a compensating differential for the practice, as do firms that offer PhD researchers greater opportunity for independent work (Stern,1999). Modern information technology may also be complementary with employee decision-making, so that the growth of EI reflects technological change embodied in workers rather than managerial innovation.

Whatever the explanation, since EI has at worst a non-negative effect on productivity and a positive effect on the lives of workers, it is a net benefit to the U.S. labor market.

Table 1: Regression Coefficients (Standard Errors on Production Function Estimates of Effects of EI Scale, Employment, and Assets on LN Sales

SOURCE: COMPUSTAT, Standard and Poors, various years.

1983-1993 regressions include all years for which we have data; 1993 regression covers 1993 cross section; 1983-1993 change includes all companies for which we could get at least 2 years, with changes calculated as average annual changes over the longest period for which data exists. The 1993 and 1983-93 regressions include 7 industry dummy variables, which have no noticeable impact on the results.

Table 2: Involvement in Decisions and Attitudes of Non-Managerial Workers, by EI Status

SOURCE: Tabulated from Worker Representation and Participation Survey, Detailed Tabulations, See http://www.nber.org/data/ (Freeman and Rogers, 1994).

References

Bartholomew, David. The Statistical Approach to Social Measurement, Academic Press, London, 1996.

Black, Sandra E. and Lynch, Lisa M. "How to Compete: The Impact of Workplace Practices and Information Technology on Productivity", NBER working paper 6120, August 1997.

Cappelli, Peter, and David Neumark, "Do "High Performance" Work Practices Improve Establishment-Level Outcomes?", NBER Working Paper 7374, October 1999.

Commission on the Future of Worker-Management Relations. Fact Finding Report. Washington, DC: U.S. Department of Labor and U.S. Department of Commerce, May 1994.

COMPUSTAT, Standard and Poors, various years.

Freeman, Richard, Morris M. Kleiner, and Cheri Ostroff, "The Anatomy and Effects of Employee Involvement" , Working Paper, Harvard University and the University of Minnesota, 2000.

Freeman, Richard and Joel Rogers, What Workers Want, Cornell University Press and the Russell Sage Foundation, Ithaca, New York, 1999.

Freeman, Richard and Joel Rogers. Worker Representation and Participation Survey (WRPS). http://www.nber.org/data/ 1994.

Ichniowski, Casey, Kathryn Shaw, and Giovanna Prennushi. "The Effects of Human Resource Management Practices on Productivity: A Study of Steel Finishing Lines," American Economic Review, June 1997, pp. 291-313.

Kato, Takao, and Motohiro Morishima. "The Productivity Effects of Human Resource Management Practices: Evidence From New Japanese Panel Data," Working Paper, December, 1996.

Kleiner, Morris, Jonathan Leonard, and Adam Pilarski, "Do Industrial Relations Affect Plant Performance?: The Case of Commercial Aircraft Manufacturing," NBER Working Paper 7414, November 1999.

Kraemer, Helena and Sue Thiemann. How Many Subjects? Newbury Park, CA: Sage Publications, 1987.

Stern, Scott, "Do Scientists Pay to Be Scientists?", NBER Working Paper 7410, October 1999.

1We thank Hwikwon Ham for his excellent research assistance.

* Professor of Economics, Harvard University and Program Director, Labor Studies, National Bureau of Economic Research, 1050 Mass. Ave., Cambridge, MA. 02138

*Professor of Public Affairs and Industrial Relations, University of Minnesota and NBER, Humphrey Institute, 260 Humphrey Center, University of Minnesota, 301 19^th St. South, Minneapolis, MN 55455