worker participation and productivity in...

WORKER PARTICIPATION AND PRODUCTIVITY

IN LABOR-MANAGED AND PARTICIPATORY

CAPITALIST FIRMS: A META-ANALYSIS

CHRIS DOUCOULIAGOS*

Using meta-analytic techniques, the author synthesizes the results of43 published studies to investigate the effects on productivity of variousforms of worker participation: worker participation in decision making;mandated codetermination; profit sharing; worker ownership (employeestock ownership or individual worker ownership of the firm's assets);and collective ownership of assets (workers' collective ownership ofreserves over which they have no individual claim). He finds thatcodetermination laws are negatively associated with productivity, butprofit sharing, worker ownership, and worker participation in decisionmaking are all positively associated with productivity. All the observedcorrelations are stronger among labor-managed firms (firms owned andcontrolled by workers) than among participatory capitalist firms (firmsadopting one or more participation schemes involving employees, suchas ESOPs or quality circles).

Scientists have known for centuries that a single study will not resolve amajor issue. Indeed, a small sample study will not even resolve a minor issue.Thus, the foundation of science is the cumulation of knowledge from theresults of many studies.

Hunter and Schmidt, Methods of Meta-Analysis

I nterest in the effects of worker participa-tion on enterprise performance has

grown phenomenally. In general, the lit-erature can be divided into two camps.

Supporters of participation argue that itstrengthens workers' commitment to thefirm, reduces the need for costly monitor-ing, and increases work effort and hence

*Chris Doucouliagos is a lecturer at the School ofEconomics, Deakin University. This paper benefitedsubstantially from discussions with, and assistancefrom, George Tratolos, Phillip Hone, Douglas Kruse,Asraul Hoque, and participants in seminars at Deakinand Monash Universities.

All the data used in this study, as well as thecomputer program and a technical appendix, areavailable from the author at the School of Economics,Deakin University, 221 Burwood Highway, Burwood,Victoria, 3125, Australia.

Industrial and Labor Relations Review, Vol. 49, No. 1 (October 1995). © by Cornell University.0019-7939/95/4901 $01.00

58

WORKER PARTICIPATION AND PRODUCTIVITY 59

efficiency and productivity. The other campargues that various forms of participationreduce managerial power, obstruct man-agement decision making, waste valuableand scarce resources, and lead to free riderproblems. Regardless of the theoreticalarguments, however, "the proof of the pud-ding is in the eating." Accordingly, in thispaper I apply meta-analysis to 43 studiesthat together represent a large proportionof the published empirical work on thesubject, and by this means I estimate theaverage correlations between productivityand various forms of participation (workerparticipation in decision making, profitsharing, worker ownership, and collectiveownership) found by other researchers.

Meta-analysis, a set of techniques for dis-tilling a single estimate from a number ofstudies, is widely used in psychology andsociology and is beginning to be applied inmanagement studies. Unfortunately, witha few notable exceptions (Jarrell and Stanley1990; Weitzman and Kruse 1990), it has notbeen adopted in economic analysis. Onereason for its disuse is that few researchers,apparently, areyetfamiliarwith it (Wanous,Sullivan, and Malinak 1988); another isthat varying methodological quality anddifferences in the measurement and speci-fication of both dependent and indepen-dent variables across many empirical econo-metric studies make those studies difficultto compare. The few serious meta-analysesin applied economics have demonstratedits potential for synthesizing empirical eco-nomic results (Stanley and Jarrell 1989).For example, a meta-analysis by Weitzmanand Kruse (1990) has been widely acclaimedand very widely cited as evidence of thepositive association between profit sharingand productivity.

This paper offers the first meta-analysisof the effects on productivity of variousforms of participation in labor-managedfirms (LMFs—worker-owned firms in whichlabor exercises ultimate and democraticdecision making power, with one vote perperson) and participatory capitalist firms(PCFs—firms adopting one or more par-ticipation schemes involving employees,such as ESOPs, quality circles, gainsharing.

profit sharing, and autonomous workgroups). It is also the first meta-analysisexamining how productivity is affected byworker ownership—employee stock owner-ship or individual worker ownership ofthefirm's assets—and collective ownership—collective ownership of reserves over whichworkers have no individual claim. (I do notlook at participation through union repre-sentation.) Previous studies have eitherignored LMFs (Miller and Monge 1986;Wagner and Gooding 1987) or groupedthem together with PCFs (Weitzman andKruse 1990).

I am interested in two broad questions.First, what productivity effects are associ-ated with the various forms of participationexamined here—and, in particular, whatare the relative effects on productivity ofparticipation in ownership, decision mak-ing, and profit sharing? And second, arethose effects sensitive to organizational set-ting? That is, do the effects of various formsof participation on productivity differ ac-cording to whether the firm is labor-man-aged or participatory capitalist?

The Meta-Analysis Methodology

Meta-analysis is essentially a techniquefor combining results across studies, withthe objective of reaching conclusions aboutthe overall association among variables(Rosenthal 1987). Because meta-analysiscondenses numerous studies into one study,it can greatly reduce the onus on scholarswho need to digest the empirical literatureon a given subject. Thus, it is easier to referto, say, Weitzman and Kruse (1990) than itis to refer to a list of different studies, withdifferent sample sizes and different results..Apart from narrative and vote countingreviews (comparisons ofthe number of sig-nificant and insignificant findings), meta-analysis is the only technique available forthe cumulation of results from differentstudies. Further, and more important, twogreat advantages of meta-analysis over asimple narrative review are that it allowsquantifiable assessment of the empiricalliterature and hypothesis testing of the re-lationships under investigation. Narrative

60 INDUSTRIAL AND LABOR RELATIONS REVIEW

reviews and vote counting reviews are noto-rious for erroneous conclusions (Hunterand Schmidt 1990, Chapter 1).

The meta-analysis undertaken in thisstudy is concerned with four issues. First,are there any relationships between pro-ductivity and various forms of worker par-ticipation (participation in decision mak-ing, profit sharing, and ownership)? Sec-ond, what is the magnitude of each of theserelationships (known as the effect size orcorrelation)? Third, is the estimated effectsize statistically significant? And fourth,what is the variance of the observed effectsize from each individual study around theoverall estimated mean effect size of allstudies combined? These four issues havebeen called "bare-bones meta-analysis"(Hunter and Schmidt 1990).

A fuller meta-analysis would investigateand correct for study artifacts (for example,measurement error, restriction of range,and construct validity). Data limitationsrule out those further steps in the presentanalysis.' A fuller meta-analysis would alsoexplore the effects of moderator variables(for example, the size of the firm, its geo-graphical location, product market condi-tions, and years of operation). Although Iundertake some moderator analysis (forexample, cqmparison of LMFs and PCFsand the impact of codetermination laws)and I test for the need for other moderatoranalysis, most analysis of that kind is setaside for further research (and awaits fur-ther data). This study should therefore beseen as an initial exploration.

Nevertheless, this study is more compre-hensive than previous meta-analyses on thissubject (Miller and Monge 1986; Wagnerand Gooding 1987; Weitzman and Kruse1990). The often-cited Weitzman and Kruse(1990) study only reported and cumulatedt-statistics and, hence, only focused on sig-nificance testing and did not investigatevariance and sampling error.^ Further,

'Unfortunately, this problem is common in meta-analytic work, Wagner and Gooding (1987), Goodingand Wagner (1985), and Schmitt, Gooding, Noe, andKirsch (1984) all faced similar problems,

Weitzman and Kruse attempted to estimate theeffect size between profit sharing and productivity

Weitzman and Kruse (1990) and Levineand D'Andrea Tyson (1990) did not com-pare LMFs and PCFs, but grouped themtogether. As will be shown below, thatfeature of their studies is a major limita-tion. Wagner and Gooding (1987) notedthat the study by Miller and Monge (1986)was biased because it included laboratorystudies and studies without any objectivedata (percept-percept studies). Wagnerand Gooding (1987) only included studiesfrom the United States, did not include anystudies from economics, and only exploredthe association between participation indecision making and productivity. Finally,previous studies have not compared therelative effects of the different participa-tory variables. For example, the productiv-ity effects of profit sharing have not beencompared with those of ownership.

Procedure

Study Selection

In order to cumulate findings across stud-ies, it is necessary to compile a comprehen-sive set of published studies adopting acomparable methodology. The estimationtechnique chosen for this study is econo-metric analysis, because of its familiarity ineconomics, its reliability and rigor, and itswide application to the estimation of theassociation between participation and pro-ductivity. An extensive manual and com-puter search identified 43 studies that re-ported regression results relevant to vari-ous forms of participation and productiv-ity. Employee participation was defined asjoint decision making or influence sharingbetween employees and managers; profitsharing was defmed as group-based com-pensation of any form, including Scanlonplans, Rucker plans, and Improshare;worker ownership was defined as employeestock ownership or individual worker own-ership of the firm's assets (but excludingmanagerial stock ownership); and coUec-

(1990:138), but because they did not use effect sizesderived from meta-analysis, they could only estimatean effect for 12 of their 16 studies.


tive ownership was defined as collectivelyowned reserves over which workers have noindividual claim.

Fifteen of the studies refer to LMFs and28 to PCFs. Tables 1 and 2 list the studies.'These 43 studies are the population of allrelevant published studies. Meta-analysisfocusing on published data is the norm(Wagner and Gooding 1987); unpublishedworks (doctoral dissertations, mimeo-graphs, and working papers) are excludedon the grounds that they may be method-ologically weak.*

It may be objected that relying solely onpublished data biases the results. In par-ticular, it may be argued that studies re-porting positive fmdings (that is, findingsof statistically significant positive correla-tions between the variables of interest) arefar more likely to see publication than arestudies reporting negative findings. Thatargument, however, is premised on the as-sumption that negative results are sup-pressed by either authors or journals. AsRosenthal (1987:223) pointed out, this ar-gument, at the extreme, implies "that thejournals are filled with the 5 percent ofthestudies that show type I error, while the filedrawers back at the lab are filled with the 95percent of the studies that show nonsignifi-cant results." In the present case, given thelively debate in the economics literatureabout the merits or otherwise of various

'Information on the studies can be obtained di-rectly from the studies themselves or from an appen-dix available on request to the author.

••Meta-analysis can be applied to unpublished work,but corrections should then be made for method-ological weaknesses. For example, a lower weight canbe attached to working papers than to journal ar-ticles. Because ofthe subjective nature of such weights,I have excluded unpublished studies, like most otherresearchers doing meta-analysis on this subject (al-though Weitzman and Kruse used three unpublishedstudies). Hunter and Schmidt (1990:509) noted thatthe publication "bias" could be in favor of method-ologically stronger research studies. Reviewers areoften selected byjournal editors based on theirjudgedmethodological expertise, and it is therefore to beexpected that their evaluations will focus heavily onthe methodological quality ofthe study. Many meth-odological weaknesses have the expected effect ofartifactually reducing the expected study effect size.

forms of participation, such a pattern seemshighly unlikely. For example, balancedagainst the many writings advocating LMFsis a large body of theoretical literature ar-guing that LMFs are inefficient (for re-views, see Doucouliagos 1993 and Bonin,Jones, and Putterman 1993). Indeed, itmay even be that the mainstream of theeconomics profession is of the latter view.It is difficult to believe that negative theo-retical findings are published while nega-tive empirical findings are censored. Inany case, the current meta-analysis doesinclude negative findings, so publicationbias is unlikely to be significant (if it existsat all) . Further, a number of investigationsinto publication bias have found no suchbias (Bullock and Svyantek 1983; Barrickand Alexander 1987).«

I excluded six bodies of literature fromthis meta-analysis:' (a) studies exploringthe effects of unions on productivity, orestimates of the interaction among unions,various forms of participation, and produc-tivity; (b) case studies that did not providequantifiable relationships; (c) studies thatdid not analyze the relationship betweenparticipation and productivity (for example,studies focusing solely on information shar-ing and productivity without participation,or the effects of the industrial relations

*A referee pointed out that publication bias is aproblem when research reveals inconclusive and in-significant results, and authors omit these from theirpublished work. While positive and negative findingsmay be published, insignificant findings may not.Such publication bias is a major problem for meta-analysis, but it is also a problem for narrative reviewand any other cross-study analysis.

A review of the unpublished literature is beyondthe scope ofthe present paper, but the meta-analysisresults are consistent with this literature. See, forexample, Kruse (1993) for a review of this literaturerelating to profit sharing.

'A full list of the excluded studies and reasons forexclusion is available on request to the author. Sev-eral studies presented problems with classification,for example, studies of Scanlon plans and studiescontaining both LMFs and PCFs. Works in the psy-chology and organizational change literatures wereparticularly likely to be screened out by the studyselection procedure. The technical appendix (avail-able from the author) explains how such studies werehandled and why.


Table 1. The Associations Between Productivityand Various Forms of Participation in Labor-Managed Firms.

Study Sample Size

Average rDecisionMaking

Average rProfit

Sharing

Average rWorker

Ownership

Average rCollective

Ownership

Berman and Berman 1989Lee 1989Sterner 1990Jones and Svejnar 1985Estrin 1991Jones and Backus 1977,

Jones 1982, Estrin et al. 1987(U.K. data)

Jones 1993Defourny et al. 1985,

Estrin et al. 1987, andEstrin and Jones 1992 and 1995

Jones 1987Espinosa and Zimbalist 1981Bellas 1972

14415057

63184

146181

1064503518

-0.22**-0.21**+0.02+0.03+0.05

+0.07+0.09

+0.13**+0.30**+0.38**+0.59**

n.a.+0.26**

n.a.+0.34**

n.a.

+0.27**+0.16

+0.23**+0.19

n.a.n.a.

n.a.-0.08

n.a.+0.13**

n.a.

-0.05+0.09

+0.15**-0.29**

n.a.n.a.

n.a.n.a.n.a.

-0.16**n.a.

-0.11+0.29**

+0.04-0.13

n.a.n.a.

Note: "r" = correlation.Definitions of independent variables: Worker Participation in Decision Making is joint decision making or

influence sharing between employees and managers; Profit Sharing is group-based compensation of any form,including Scanlon plans, Rucker plans, and Improshare; Worker Ownership is employee stock ownership orindividual worker ownership of the firm's assets, excluding managerial stock ownership; and CollectiveOwnership is collectively owned reserves over which workers have no individual claim.

*Statistically significant at the .10 level; **at the .05 level (two-tailed tests).

climate on productivity; studies comparingthe productivity performance of LMFs withthat of PCFs; studies analyzing participa-tion but not productivity; and studies ana-lyzing the effects of, say, profit sharing onabsenteeism and defect rates, but not onproductivity); (d) studies based on labora-tory tests; (e) studies in which both thedependent variable and the independentvariable measured perceptions rather thanobjective data; and (f) studies based onsample means. With regard to (f), differ-ences in sample means compare the pro-ductivity levels of one group to those ofanother, yielding data different from effectsize estimates derived from econometricwork. For example, by comparing the pro-ductivity of one type of firmi with that ofanother we get an idea of the net differ-ences in productivity. Suppose, though,that participation in decision making isassociated with increased productivity butcollective ownership is negatively associ-ated with productivity, and the latter offsetsthe former; the finding that the LMF has

lower productivity than the PCF will thenmask the positive association between par-ticipation in decision making and produc-tivity.

Multiple Correlations

Most econometric studies provide sev-eral estimates—for example, results fromboth Cobb-Douglas and Translog produc-tion functions. Often, several independentvariables may be included. The standardprocedure, which I follow here, is to in-clude all the available and relevant esti-mates in the meta-analysis. Thus, effectsizes are derived from all specifications ofestimated production functions and fromall independent variables proxying for vari-ous forms of participation. The only esti-mates that are usually excluded from meta-analyses are those acknowledged by theauthors of a study to be unreliable andreported as a contrast to the main results orout of curiosity. The inclusion of suchestimates would bias cumulation and are


Table 2. The Associations Between Productivityand Various Forms of Participation in Participatory Capitalist Firms.

Study Sample Size

Average rDecisionMaking

Average rProfit

Sharing

Average rWorker

Ownership

FitzRoy and Kraft 1985, 1986,1987, and 1992

FitzRoy and Kraft 1993Svejnar 1982Katz et al. 1985Kruse 1993Cooke 1994Mitchell et al, 1990Cutcher-Gershenfeld 1991Conte and Svejnar 1988, 1990Katz etal, 1987Conte and Tannenbaum 1978Cooke 1989GAO 1987Cable and FitzRoy 1980Rosenberg and Rosenstein 1980Kumbhakar and Dunbar 1993Wadhani and Wall 1990Kruse 1992Blanchflower and Oswald 1988Schuster 1983Schuster 1984aSchuster 1984bJones and Kato 1993a and 1993b

12322437464

4672841886368155332087471266886197

20720948404202495543

-0,23**-0,12*-0,060,00

+0,01+0,02+0,08*+0,10**+0,19**+0,27+0,27+0,28**+0,35**+0,36**+0,43**n,a.n,a.n,a.n,a.n,a.n,a.n,a.n,a.

+0,26**n,a.n,a.n,a.

+0,02+0,12**+0,10**n,a.

+0,14*n,a.n,a.n,a.n,a.

+0,15n,a.

+0,14**+0,18*+0,02**+0,05+0,22*+0,25**-0,03n,a.

+0,31*n,a.n,a.n,a.

+0,01n,a.

+0,09*n,a.

-0,03n,a.

+0,54**n,a.

+0,05-0,11n,a.

+0,09**n,a.

+0,02**-0,02n,a.n.a.n,a.

+0,05

Note: "r" = correlation.*StatisticaUy significant at the ,10 level; **at the .05 level (two-tailed tests).For definitions of independent variables, see text or note to Table 1,

accordingly not included in this meta-analysis.

Some of the multiple estimates are statis-tically independent and others are not.The procedure for averaging multiple esti-mates from a single study is outlined inHunter and Schmidt (1990, Chapter 10). Ifa study reports fully replicated design (forexample, a production function is estimatedfor the footwear industry and is then repli-cated for the construction industry), theneach estimate is statistically independent.Such fully replicated estimates are aver-aged and the sample size is the sum of thesample sizes across the different organiza-tions or industries. If a study applies differ-ent functional forms or different indepen-dent variables to the same data set, however(making it a "conceptual replication"), theestimates are not statistically independentand are averaged with the sample size for

the average being the sample size for thestudy.

Heterogeneous Measures,Methods, and Data

Studies generally differ in how they mea-sure and specify the dependent aiid inde-pendent variables and in the methodologythey employ. Even when studies are identi-cal in those respects, they are likely to differin data quality. This heterogeneity, whichat first blush might appear to be a problem,is in fact a good thing for the field and thevery raison d'etre for meta-analysis. First, itcontributes to the advance of understand-ing because it provides many cross-checkson results. As Weitzman and Kruse(1990:137-38) noted,

if all studies used identical methods and data,the results would be more suspect (since they


could all share common defects). The variety ofspecifications employed, as well as the diversityof data sources, lends greater credibility to thefindings.

Second, it is because all studies are notexact replications and do not arrive at thesame conclusions that meta-analysis isneeded and is valuable. Hunter andSchmidt (1990, Chapters 11 and 13) pointedout that meta-analysis does not mix appleswith oranges: it cumulates and averagesstudy results (numbers), not studies them-selves. The meta-analysis undertaken inthis paper does not combine different in-dependent variables (for example, it doesnot combine the association between profitsharing and productivity with the associa-tion between participation in decision mak-ing and productivity), but different mea-sures of the same variable. Similarly, withrespect to the dependentvariable, in all theincluded studies it is defined as productiv-ity, even though it is measured in differentways.*

Meta-analysis is well suited to test theuniversality of a relationship. For example,the data used in this study are from variouscountries. Meta-analysis can, and in thisstudy does, detect differences across differ-

^Even if theory offered guidance to researchers,data limitations would still restrict the way productiv-ity and participation variables are measured. Some ofthe variables used are as follows. (The full list isavailable from the author.) Productivity was typicallymeasured as value added, value added per worker,real sales, total factor productivity, measure of above-average performance, and net sales revenue per em-ployee. Participation was defined by a dummy forparticipation plans or as the proportion ofthe boardof management who are worker-members, propor-tion of workers who are members, members as aproportion of the total labor force, or index of par-ticipation. Profit sharing was measured by a dummyfor profit sharing or as total surplus distributed toworkers, profits per worker, or dividend distributedper member. Worker ownership was measured by adummy for the existence of an ESOP or as total sharecapital owned by workers, average capital stake perworker-member, individually owned capital, finan-cial input per employee, or the proportion of stockowned by ESOP. Collective ownership was measuredas collective reserves as a percentage of net assets andtotal assets.

ent countries in the relationship betweenvarious forms of participation and produc-tivity. If statistically significant variance isdetected, then the hypothesis of homoge-neity of effect sizes is rejected. Data permit-ting, moderator analysis can then be un-dertaken. For example, we can then test tosee whether there are any differences be-tween developed and underdevelopedcountries in the association between vari-ous forms of participation and productiv-ity. Meta-analysis can also be used to deter-mine whether differences in effect sizes aredue to real moderator variables or simply tothe way variables were measured.

Multiple Authorship

A substantial proportion of the studiesemanated from the same author (see Tables1 and 2). This is a problem that is commonin all cross-study comparisons, includingliterature reviews. While we can assumestatistical independence for studies withdifferent authors who used differentsamples, we cannot make the same assump-tion for studies in which the sameauthor(s) used the same data set. Forthis meta-analysis, studies that were notstatistically independent were averagedand treated as "conceptual replications."Thus, the number of studies was reducedfrom a total of 43 to 34, with studiesrelating to LMFs reduced from 15 to 11and studies relating to PCFs reduced from28 to 23.

The number of studies examined herecompares favorably with the number inother meta-analyses. Bullock and Tubbs(1990), for example, looked at 33 case stud-ies; Weitzman and Kruse (1990), 16 studieson profit sharing (14 of these are includedin the present meta-analysis but 2 remainunpubhshed); Miller and Monge (1986),25 studies; and Scott and Taylor (1985), 21studies. It should also be noted that a smallnumber of studies in cells is also commonin meta-analysis. While 34 statistically inde-pendent studies may appear to be a smallnumber, it is the total sample size fromthese studies that is important (see Tables3-6 for the relevant total sample sizes).


Estimation Procedure

The first step in the estimation is toderive the relevant t-statistics from eachstudy. The corresponding effect size isthen calculated from the t-statistic. Severaleffect size statistics are available (Hunterand Schmidt 1990).' The effect size adoptedhere, because of its familiarity and generaluse, is the correlation coefficient.'" Notethat the resulting correlations are partialcorrelations, and not zero order (simple)correlations. This means that the estimatesmay change according to which other vari-ables are included in the primary study'smultiple regression estimation. Differentvariables may change the partial correla-tions, but I assume that the reported esti-mates are the best estimates from that dataset (the most methodologically and theo-retically sound, produced using robust esti-mation techniques).

The econometric literature cautionsagainst the use of simple correlations andrecommends use of partial correlations.For example, simple correlations may re-veal, say, a positive association when theoryand (well-constructed) econometric esti-mation yields a negative association. Un-like the simple correlation, regression co-efficients and partial correlations have thesame sign, a very important piece of infor-mation when cumulating associations thatmay, from a theoretical perspective, be ei-ther positive or negative. Simple correla-tions may fail to reveal the underlying rela-tionship; partial correlations give a betteror purer approximation of the true associa-tion (Johnson 1984:82)."

The next two steps are to calculate anoverall weighted average correlation acrossall studies and then estimate the varianceof the population correlations (the vari-ance in observed correlations not due tosampling error). The latter involves esti-mating the weighted average squared erroracross all studies and estimating the sam-pling error variance. The sampling errorvariance is subtracted from the observedvariance and the difference is called theremaining (or corrected) variance. If atleast 75% of observed variance is due tosampling error variance, then it is likelythat the remaining variance is due to arti-facts not accounted for, such as measure-ment error (Hunter and Schmidt 1990).

However, the 75% rule may not be pow-erful enough in cases where only a fewstudies are available. Accordingly, the U-statistic is also estimated. The U-statisticprovides a test for the presence of modera-tor variables. A chi-square test is then per-formed on the U-statistic to determinewhether any remaining unexplained vari-ance is statistically significant. (This test isgiven in Marascuilo [1971], is recom-mended by Rosenthal [1987] and Spectorand Levine [1987], and has been applied inseveral studies, for example. Fisher andGitelson [1983] and Scott and Taylor[1985].) The existence of a moderatorvariable is indicated by (a) the 75% rule,(b) the U-statistic, and (c) credibility inter-vals. A large or zero inclusive credibilityinterval indicates the presence of distinctsubpopulations.'^ Separate meta-analysis

'All of these are transformations of each other, sothe conclusions drawn from one effect size will beconsistent with those drawn from another. That is,the results of meta-analysis do not depend on thechosen effect size. For an extended discussion of thisissue, see Rosenthal (1987), Partial correlations andt-statistics are monotonic transformations of eachother and hence it makes no difference which is used(Chow 1983:71),

'"A technical appendix is available from the au-thor with a comprehensive list of all the formulasused in this meta-analysis,

"If data permitted, a meta-analysis of simple cor-relations could be conducted and its results could be

compared with the results from the meta-analysis ofthe partial correlations,

'^In meta-analysis, credibility intervals are con-structed on the basis of 1,96 corrected standard devia-tions (the square root of the corrected variance, or thevariance remaining after removing sampling errorvariance) from the point estimate of the populationcorrelation. Confidence intervals are constructed onthe basis of 1,96 standard errors from the point esti-mate of the population correlation. There are twoways to derive the standard error, one appropriatewhen moderator variables exist and one appropriatewhen they do not. For a full discussion, see Schmidt,Hunter, and Raju (1988), Whitener (1990), or thetechnical appendix available from the author.


(if possible) is then undertaken. The exist-ence of moderator variables is confirmedby (a) point estimates differing among sub-sets and (b) a corrected aferagevariance inthe subsets that is lower than the variancefor the data as a whole (Hunter and Schmidt1990:112). In all cases, moderator analysisshould be guided by theory.

Tbe next step is to test the statisticalsignificance of tbe average correlation sta-tistics. Tbe statistical significance of a par-tial correlation is based on tbe statisticalsignificance of tbe t-statistics from which itwas derived (Chow 1983:71). Several testsare available for testing tbe significance oftbe average t-statistic."

The final step is construction of confi-dence intervals. Confidence intervals pro-vide information on tbe uncertainty sur-rounding point estimates. Unfortunately,reporting of confidence intervals is not acommon practice (for example, Weitzmanand Kruse [1990], Wagner and Gooding[1986], and Miller and Monge [1985] offeronly point estimates). Confidence inter-vals can also be used to test wbetber tbeeffects of different forms of participationon productivity are similar—a conditiontbat would be indicated, for example, byoverlapping confidence intervals.'"'

Averages derived from meta-analysisshould be weigbted, but caution must beexercised wben a single study dominatesand reverses results. In such cases separate

"These include the Edgington test, the Winermethod, the Stouffer method, the weighted Stouffermethod, and the mean p-value test (see Rosenthal1987 for formulas and a discussion). The estimated t-statistics and tests for statistical significance are avail-able from the author on request. All confirmed theresults and conclusions reported in Tables 3-6.

'••An alternative estimation procedure is meta-re-gression analysis. This procedure involves making thecorrelations the dependentvariable, and making studycharacteristics the explanatory variables, which canalso proxy for moderator variables (see Stanley andJarrell 1989 for full details). Unfortunately, datalimitations preclude a full meta-regression analysis atthis time; because the large number of independentvariables results in very low degrees of freedom, noneof the meta-regressions produce statistically signifi-cant variables.

meta-analysis should be conducted and tberesults compared.

A major limitation of tbis estimation pro-cedure is tbat it assumes tbat studies are ofequal quality and reliability. Sucb an as-sumption is not necessary in all meta-analy-sis; it is possible to grade or weigh studiesaccording to some quality or reliability cri-terion. I did not attempt to do so in tbisstudy because of tbe subjective nature ofsucb grading and because tbe literatureprovides little guidance in assessing re-searcb quality.

Various Forms of Participation andProductivity in LMFs and PCFs

Tbere are several theoretical reasons wbyLMFs and PCFs may differ in tbe channelstbrougb wbicb various forms of participa-tion work and in tbe productivity effectstbey bave. First, governance structures dif-fer. In LMFs tbe locus of strategic decisionmaking power rests witb, and is diffusedtbrougbout, tbe entire niembersbip. Work-ers, at least in tbeory, control tbe LMF. InPCFs, ultimate and strategic decision mak-ing power lies witb tbe owners and seniormembers of tbe bierarcby. Tbus, tbe ex-tent of participation in decision making ispotentially greater in LMFs than in PCFs, adifference tbat can be expected to result indifferent productivity effects from partici-patory programs.

Second, workers' control can lead toendogeneity. Tbat is, tbe degree of partici-pation in decision making, profit sharing,and ownership is under workers' controland, bence, so too is tbe impact of tbesevariables on productivity (Jones and Svejnar1985:458).

Third, supporters of participation arguetbat its presence in decision making, profitsbaring, and worker ownersbip of tbe firmbas positive effects on tbe firm. Participa-tion in tbose tbree program types, bow-ever, is more extensive and more inten-sively adopted in LMFs tban in PCFs. Tbatis, various forms of participation may attaina critical mass in LMFs. Furtber, LMFs andPCFs differ in incentives and monitoringmecbanisms. LMFs can be described as


firms characterized by high trust, with highautonomy, high discretionary decisionmaking power, and low direct supervision(Fox 1974). The monitoring of work inLMFs tends to be undertaken with peergroup monitoring as a substitute for formalmonitoring (Bonin, Jones, and Putterman1993), a mechanism with cost savings forthe firm and without necessarily any loss inmonitoring intensity.

Labor-Managed Firms

The results for the eleven statisticallyindependent studies of LMFs that were in-cluded in this meta-analysis are shown inTable 1. Democratic worker participationin decision making (one vote per person)in LMFs was positively but not always statis-tically significantly correlated with produc-tivity in seven (64%) ofthe eleven studies,was negative and statistically significant intwo studies (18%), and was close to zero intwo others (18%). All six studies that lookedat profit sharing and productivity reporteda positive association between the two, andin four of the six cases the association wasstatistically significant. Three of the sixstudies that looked at worker ownershipfound that it was negatively related to pro-ductivity, but in only one case was the asso-ciation statistically significant. Two of thestudies found a positive and statisticallysignificant association between worker own-ership and productivity. Collective owner-ship is associated with reduced productivityin three of the five studies that estimatedthis relationship, but the negative associa-tion was statistically significant in only one

Participatory Capitalist Firms

Three (20%) ofthe studies found a nega-tive association between worker participa-tion in decision making and productivity

"'The estimate for collective ownership in Table 1combines the U.K. estimates in Estrin et al. (1987)with the U.K. estimates in Jones and Backus (1977).The results are nearly identical if the reverse proce-dure is adopted.

(all three relate to the West German expe-rience with legislation imposing WorksCouncils); in two of these cases the associa-tion was statistically significant. Of the 13studies investigating profit sharing, onlyone (8%) found that it had a (nonsignifi-cant) negative association with productiv-ity. Ofthe 11 studies estimating the rela-tionship between worker ownership andproductivity, three (30%) found a negativeassociation, in all cases nonsignificant.

Meta-Analysis of Various Forms ofWorker Participation and Productivity

Worker Participation in DecisionMaking and Productivity

Table 3 presents the meta-analysis re-sults for the association between workerparticipation in decision making and pro-ductivity. Four studies relating to LMFswere excluded from the final meta-analysis.(The first numerical data column in thefirst panel of Table 3 shows the impact ofincluding and excluding these studies.)Three of these (Berman and Berman 1989;Sterner 1990; Estrin 1991) pooled data fromLMFs and PCFs and also used dummies toproxy for the cooperative form. The studyby Lee (1989) focused more on quasi-LMFsthan on LMFs (Lee noted that "there isoften no one-member one-vote principle"[1989:14]). The use of pooled data anddummy variables may, in this case, biascorrelation estimates because, as a proxyfor the cooperative form, dummy variablescapture participation in decision making,profit sharing, and ownership. Dummiesfor cooperatives capture the net effect ofthese variables and not the individual ef-fects in which I ani interested.

The population correlation coefficientbetween worker participation in decisionmaking and productivity in LMFs is highlyunlikely to be zero (see the 95% confi-dence interval, column 3). Removing theFrench data (42% ofthe total sample size)does alter this result, but there is no statis-tical reason to exclude those data (Table 3,column four). We can conclude that workerparticipation in decision making in LMFs


Table 3. The Association Between Worker Participation in Decision Makingand Productivity in Labor-Managed and Participatory Capitalist Firms.

Labor-Managed Finns

MeasureAll Data,

K= 11

Without Lee andPooled Data,

K= 7

Without Lee,Pooled Data,

and French Data,

Sample SizeMean rWeighted Mean rU-Statistic

Remaining Variance95% Credibility Interval95% Confidence Interval

2,560+0.11+0.06

40.15**d.f. = 10

0.011 (72%)-0.14 to +0.27-0.01 to+0.14

2,125+0.23+0.1014.13*d.f. = 6

0.003 (46%)0.00 to +0.21+0.05 to +0.16

1,061+0.24+0.0812.76*d.f. = 5

0.005 (49%)-0.06 to +0.230.00 to+0.17

Participatory Capitalist Firms

MeasureCodetermination

Only, K = 3

WithoutCodetermination,

K= 12

WithoutCodetermination and

Kruse, K= 11



Measure



721-0.14-0.112.98

d.f = 20.0 (0%)

-0.11 to-0.11-0.04 to-0.18

Full Sample

All Firms,K= 19

9,492+0.21+0.06

104.07**d.f = 18

0.005 (72%)-0.08 to +0.20+0.02 to +0.09

7,367+0.20+0.0491.6**

d.f = 110.005 (75%)

-0.09 to +0.180.00 to +0.09

Labor-ManagedFirms, WithoutLee and Pooled

Data, K=7

2,125+0.23+0.1014.13*d.f = 6

0.003 (46%)0.00 to +0.21+0.05 to +0.16

2,695+0.21+0.11

35.86**d.f = 10

0.007 (65%)-0.06 to 0,27+0.04 to+0.17

ParticipatoryCapitalist Firms,

WithoutCodetermination,

K= 12

7,367

+0.20+0.0491.6**

d.f = 110.005 (75%)

-0.09 to+0.180.00 to +0.09

Note: "r" = correlation; K = number of studies.Worker participation in decision making is defined as joint decision making or influence sharing between

employees and managers.*Statistically significant at the .05 level; **at the .01 level (chi-squared tests).

bas a small, positive, and statistically signifi-cant association witb productivity, reject-ing the traditional view that democratic

management of the firm is associated withreduced efficiency (Alchian and Demsetz1972; Jensen and Meckling 1979). How-


ever, the unexplained remaining variance(46% of observed variance), the statisti-cally significant U-statistic, and the zeroinclusive credibility interval all suggest thepresence of moderator variables.'®

For participatory capitalist firms. Table 3reports separate meta-analysis witbout tbetbree studies relating to codetermination."Meta-analysis confirms tbat worker partici-pation in decision making imposed by gov-ernment decree is negatively associated witbproductivity (r = -0.11, and the confidenceinterval does not include zero, column 2).This finding is consistent with tbe tbeoreti-cal arguments against codetermination(Furubotn 1978).'« Tbe weigbted averagecorrelation is positive if tbe codeter-mination studies are excluded (+0.04, col-umn 3). However, even after tbe codeter-mination studies are removed, the confi-dence interval suggests the possibility of noassociation, and tbe U-statistic and cred-ibility interval indicate tbe existence ofmoderator variables. Excluding tbe Krusestudy (58% of tbe sample size) results in astronger and statistically significant asso-ciation. Tbere is, bowever, no apparentreason to exclude tbat study.

Tbe average correlation across all stud-ies—tbat is, witb studies of LMFs and PCFscombined (but witb Lee, tbe codeter-mination studies, and the studies usingdummy variables excluded)—is +0.06, but

' ^ h e remaining variance in observed correlationsmay be entirely due to artifacts not accounted for,such as measurement error, or it could reflect theinfluence of moderator variables. However, the smallnumber of studies (notwithstanding the large size ofthe combined sample) precludes moderator analysis,principally because the studies do not include a con-sistent set of moderator variables, for example, sepa-rate regressions for small and large cooperatives.

"The Cable and FitzRoy (1980) study analyzedparticipation schemes other than codetermination inGerman firms.

'^Benelli, Loderer, and Lys (1987) found negative(but statistically insignificant) effects ofcodetermination on dividend payments, firm lever-age, firm profitability, and investment policies.Gurdon and Rai (1990) used a non-parametric testand found that revenue per unit of labor declined asa result of the 1976 West German codeterminationlaw.

among LMFs it is +0.10 and aniong PCFs itis +0.04 (tbird panel of Table 3). Wbiletbere is some overlap in tbe confidenceintervals, analysis of variance suggests tbattbe association between worker participa-tion in decision making and productivity inLMFs differs from tbat in PCFs. Tbe aver-age corrected variance for LMFs and PCFsis less than tbe corrected variance fortbe entire data set ((0.003 + 0.005)/2 =0.004 < 0.005). We can conclude tbat tbetype of firm acts as a moderator variable intbe association between worker participa-tion in decision making and productivity.Tbe average correlation between workerparticipation in decision making and pro-ductivity is likely to be smaller in PCFs tbanin LMFs, possibly because tbis form of par-ticipation occurs to a greater degree inLMFs.'9

Profit Sharing and Productivity

Meta-analysis confirms tbat tbe sbaringof profits in LMFs is positively associatedwitb productivity; r = +0.26 and tbe 95%confidence interval is strongly statisticallysignificant (Table 4, column 3). Moreover,tbat association exceeds tbe positive pro-ductivity association of worker participa-tion in decision making in LMFs. Tbeweigbted average correlation is larger forprofit sbaring, and tbe confidence inter-vals for profit sbaring and participation indecision making do not overlap. Meta-analysis also indicates an absence of mod-erator variables, as only 30% of observedvariance remains unexplained after sam-pling error is removed, tbe U-statistic is notstatistically significant, and the credibilityinterval does not include zero. Tbis pat-tern suggests tbat profit sharing in LMFsalways has a positive association with pro-ductivity and that this association in LMFsis not moderated in any way (for example,by tbe size of tbe profit sbare). Removing

"This conclusion is unchanged if all the studiesare grouped together (including codeterminationstudies. Lee, and pooled data) and then all relevantstudies for LMFs and CMFs are meta-analyzed sepa-rately.


Measure

Table 4. The Association Between Profit Sharing and Productivityin Labor-Managed and Participatory Capitalist Firms,

Full Sample

ParticipatoryLabor-Managed Capitalist Firms,

Firms, K= 6 K= 13All Firms,

K= 19



32,752+0,16+0,05

181,12**d,f, = 18

0,005 (89%)-0,08 to+0,19+0,02 to +0,09

Part Samples

2,222+0,24+0,269,23

d,f, = 50,001 (30%)

+0,20 to +0,32+0,21 to +0,31

30,530+0,12+0,04

59,56**d,f, = 12

0,002 (79%)-0,04 to+0,11+0,01 to +0,06

Measure



Labor-ManagedFirms, WithoutFrench Data,

K = 5

1,158+0,24+0,296,51

d,f, = 40,001 (24%)

+0,22 to +0,35+0,23 to+0,35

ParticipatoryCapitalist Firms,Without Kruse

1992 and 1993,K= 11

5,138

+0,14+0,11

26,59**d,f, = 10

0,003 (60%)0,00 to +0,22+0,07 to +0,15

Note: "r" = correlation; K = number of studies.Profit sharing is defined as group-based compensation of any form, including Scanlon plans, Rucker plans,

and Improshare,*Statistically significant at the ,05 level; **at the ,01 level (chi-squared tests).

the French data does not alter these con-clusions.

Profit sharing in PCFs has a small andstatistically significant association with pro-ductivity. The 95% confidence interval forall PCF studies suggests that the populationcorrelation may be close to zero (columnfour). Including or excluding the Krusestudies does not alter the positive and sta-tistically significant association betweenprofit sharing and productivity in PCFs.

These results, in conjunction with the U-statistic, credibility interval, and averagevariance of subpopulations, suggest thatprofit sharing has a smaller association withproductivity in PCFs than in LMFs. That

conclusion highlights the importance ofconducting separate meta-analysis for LMFsand PCFs. "

"A referee suggested that the estimation tech-nique may deflate the correlation. Most of the studiesfor PCFs use dummy variables for the existence ofprofit sharing, and this practice may account for thelarge difference in correlations between LMFs andPCFs, The average correlation among studies notusing a financial variable was +0,03, compared to+0,26 among studies using some financial measure,Meta-regression analysis did not support this hypoth-esis; the correlations were regressed against samplesize (t = -1,41, p = 0,18), adummy for the use of profitsharing dummies (t = -0,67, p = 0,51), a dummy forthe type of firm (t = +1,54, p = 0,14), and a dummy forthe use of production functions (t = +0,17, p = 0,87),


Table 5. The Association Between Worker Ownership and Productivityin Labor-Managed and Participatory Capitalist Firms,

Measure



Full Sample

All Firms,

31,323+0,06+0,03

60,90**d,f, = 16

0,001 (72%)-0,05 to +0,10+0,01 to +0,05

Labor-ManagedFirms, K= 6

2,222-0,01+0,10

18,73**d,f = 5

0,006 (70%)-0,05 to +0,25+0,03 to +0,18


K= 11

29 101+0,09+0 02

29,42**d,f = 10

0,001 (62%)-0,03 to +0 070,00 to +0,04

Part Samples

Measure

Labor-ManagedFirms, WithoutFrench Data,


Without Kruse

Sample SizeMean rWeighted Mean rRemaining VarianceU-Statistic

95% Credibility Interval95% Confidence Interval

1,158-0,04+0,06

0,008 (65%)13,93**d,f = 4

-0,12 to+0,23-0,04 to +0,15

3,709+0,11+0,05

0,004 (64%)24,98**d,f, = 8

-0,07 to+0,180,00 to+0,11

Note: "r" = correlation; K = number of studies, 'Worker ownership is defined as employee stock ownership or individual worker ownership of the firm's

assets, excluding managerial stock ownership,*Statistically significant at the ,05 level; **at the ,01 level (chi-squared tests).

Worker Ownership and Productivity

The weighted average correlation is posi-tive for the association between workerownership and productivity in LMFs. The95% confidence interval for the true popu-lation correlation indicates statistical sig-nificance, but does not rule out a near zeroassociation (Table 5, column 3). The lowassociation between worker ownership and

Caution should be exercised in evaluating these re-sults because of the small sample size andmulticollinearity among the dummy variables. Theresults do suggest, however, that it is the type of firmrather than the estimation technique that moderatesfor profit sharing.

productivity is not surprising, consideringthat LMFs often pay shareholders either nodividends or below market (opportunitycost) returns. If the French sample is ex-cluded, the association may become nega-tive (see 95% confidence interval, column2), but there is no apparent reason to ex-clude those data.

Most of the observed variance is not dueto sampling error. The statistically signifi-cant U-statistic and the very wide credibilityinterval strongly indicate the presence ofmoderator variables. Possible moderatorsare capital starvation and inadequate ac-cess to finance, problems that have beenidentified with LMFs (Craig and Pencavel1992); the size of worker ownership hold-


Table 6. The Association Between Collective Ownershipand Productivity in Lahor-Managed Firms.

Measure

Sample SizeMean rWeighted Mean rRemaining VarianceU-Statistic

95% Credibility Interval95% Confidence Interval

All Data,

2,072-0.01-0.01

0.015 (86%)37.09**d.f. = 4

-0.25 to +0.22-0.13 to+0.10

Without FrenchData, K = 4

1,008-0.03-0.07

0.025 (86%)31.17**d.f. = 3

-0.38 to +0.24-0.24 to +0.10

Without French andPolish Data, K = 3

827

-0.13-0.15

0 (0%)0.31

d.f. = 2-0.15 to-0.15-0.08 to -0.22

Note: "r" = correlation; K = number of studies.Collective ownership is defined as collectively owned reserves over which workers have no individual claim.*Statistically significant at the .05 level; **at the .01 level (chi-squared tests).

ings; and any risk aversion associated withindividual share holdings because of work-ers' inability to spread risk (Buck 1982).These factors may moderate and, indeed,offset any incentives to work arising fromworker ownership. Some of the studieswith positive correlations are also the oneswith the highest proportion of the firm'sassets owned by workers.^'

The point estimate ofthe degree of asso-ciation between worker ownership and pro-ductivity is substantially smaller in PCFs (r= +0.02) than in LMFs (r = +0.10), and theconfidence interval suggests a near zeroassociation. (Removing Kruse [1992,1993]does not alter the results.) A possible ex-planation for this result may be that in PCFsworkers typically own a small proportion ofassets. Also, where workers in PCFs have

^'For example, the correlation of-0.05 from Jones(1982) was derived from cooperatives in which theshare of assets owned hy workers was around 4% to7%. The Italian cooperatives (r = +0.13) and theFrench cooperatives (r = +0.15), however, had sharesof around 11% and between 10% and 38%, respec-tively. In contrast, in the Polish sample only 0.3% ofthe total assets were owned by workers, but a positiveassociation between worker ownership and produc-tivity was established. Removing the Polish sampledoes not alter the point estimates or confidence inter-vals for the association between participation in deci-sion making, profit sharing, or worker ownership.Results are available from the author. The Polish datado, however, influence the results for collective own-ership of assets; see Table 6.

little input into decision making, financialparticipation may be ineffectual. The non-significant finding may also reflect the weak-ness of the link between work effort andshare price. '

The confidence intervals overlap andthe average corrected variance for the sub-populations is greater than the variance forthe entire data set. This finding indicatesthat the type of firm may not moderate theassociation between worker ownership andproductivity. Rather, factors such as thesize of workers' capital stake and the pro-portion of total assets owned by workersmay be more important moderators. Fur-ther research is clearly needed to deter-mine what aspects of worker ownership, ifany, are conducive or nonconducive to pro-ductivity.

Collective Ownership and Productivity

Because of the small number of studieslooking at collective ownership, the resultsfor collective ownership are less conclusivethan those for the other forms of participa-tion (see Table 6). When all available dataare subjected to meta-analysis, the associa-tion between collective ownership and pro-ductivity is negative, small (r = -0.01), andstatistically nonsignificant. One of the twostudies reporting a positive association (Es-

anonymous referee pointed this out.


trin, Jones, and Svejnar 1987) relates toFrench cooperatives and the other to Pol-ish cooperatives (Jones 1995). If the Frenchand Polish samples are excluded, theweighted average correlation increases to -0.15 and becomes statistically significant,and there is no remaining variance. Sepa-rating the French data is justified because,as noted by Estrin et al. (1987) and con-firmed by analysis of the other studies, theFrench cooperatives have a relatively lowproportion of collectively held assets. Jones(1993) noted that the Polish results may beinfluenced by the impact of underdevel-oped capital markets in that country. Meta-analysis suggests that the ratio of collec-tively owned assets to individually ownedassets may be a moderator variable in non-socialist countries. Some small degree ofcollective ownership, as in the French data,may not be negatively associated with pro-ductivity, perhaps because it encouragessolidarity.

The meta-analysis point estimates, corre-lations, and confidence intervals presentedhere have the same sign as in previousmeta-analyses, but generally are lower inmagnitude. That is an important differ-ence, because with the exception ofWeitzman and Kruse (1990), the other meta-analyses have used entirely different datasets (that is, have looked at different stud-ies). For example, in relation to workerparticipation in decision making and pro-ductivity in PCFs, Miller and Monge (1985)found an overall weighted mean correla-tion of +0.15, and Wagner and Gooding(1987) found an average correlation of+0.12; Bullock and Tubbs (1990) derivedan average correlation of +0.21 forgainsharing plans; and the Weitzman andKruse (1990) study yielded an average cor-relation for profit sharing of+0.04 (author'sestimate from Weitzman and Kruse).

No previous meta-analyses have investi-gated the productivity effects of workerownership or collective ownership. Noneof the previous meta-analyses or narrativereviews constructed confidence intervals.Some meta-analysis results have differedfrom results derived from narrative review.For example, in their review of the litera-

ture, Bonin, Jones, and Putterman(1995:1305) concluded that the empiricalresults linking collective ownership to pro-ductivity "seldom yield significant resultsfor any country." The results from thepresent meta-analysis suggest otherwise.

Concluding Remarks

This study is the first meta-analysis exam-ining the productivity effects of workerownership and collective ownership. It isalso the first meta-analysis to examine theproductivity effects ofthose and other formsof participation separately for labor-man-aged firms (LMFs) and participatory capi-talist firms (PCFs). The results suggestseveral conclusions.

First, all the average correlations aresmall, although many are statistically sig-nificant. Second, with the possible excep-tion of collective ownership (workers' col-lective ownership of reserves over whichthey have no individual claim), the variousforms of participation do not hinder pro-ductivity. There is, however, a negativeassociation between firms operating undercodetermination laws and productivity.

Third, organizational setting appears tomake a difference to the productivity effectof profit sharing and participation in deci-sion making, but not to the productivityeffect of worker ownership. In all cases thepoint estimates are higher for LMFs thanfor PCFs and, in general, so too are theconfidence intervals. Fourth, contrary tothe belief of many observers, democraticgovernance in LMFs is not negatively corre-lated with productivity.

Fifth, in LMFs profit sharing is morepositively related to productivity than isworker participation in decision making.This finding lends some support to the view(for example, Locke et al. 1980) that work-ers are more motivated by remunerationthan they are by participation in decisionmaking. However, meta-analysis qualifiesthat view: profit sharing does not appear tobe more important than participation indecision making in PCFs. These divergentfindings for LMFs and PCFs may be ex-plained by the fact that worker-owners asso-


ciated with LMFs are actually worker-entre-preneurs (labor hires capital), so that fi-nancial participation can be expected toboost productivity more than does partici-pation in decision making. That is, worker-entrepreneurs are likely to be more inter-ested in profit and the firm's survival thanordinary employees would be. Workers inPCFs, in contrast, who operate in an envi-ronment in which capital hires labor, maybe more interested in participating in deci-sion making concerning issues that affectthem as workers than in furthering whatare primarily the owners' interests.

Five important qualifications should benoted. First, the meta-analysis assumedthat all studies were of equal quality. Be-cause of that assumption, together with theheterogeneity of estimates, the results mustbe interpreted with a "broad brush." Sec-ond, any review, qualitative or meta-ana-lytic, is limited to the available data. As inall studies, the results from this meta-analy-sis have to be tested against the results offuture studies. Third, despite my occa-sional use, for convenience, of the phrase"productivity effects," it should be borne inmind that the results do not indicate cau-sality. Analysis of causality is clearly needed.

with, for example, the application ofGrangian causality tests to time series data.Some time series studies (Schuster 1983,1984a; FitzRoy and Kraft 1993) suggest thatvarious forms of participation do causechanges in productivity, but the majority ofstudies only establish an association.

Fourth, interactions among the variousforms of participation, and between thevarious forms of participation and otherrelevant variables, such as unionization,were not subjected to meta-analysis. Fifth,the meta-analysis presented in this paperpresents only part of the picture of theanalysis of LMFs. In order to get a fullerpicture of the economic impact of LMFs,meta-analysis needs to be applied to studiesinvestigating the impact of workers' con-trol on the firm's survival rate, employmentcreation, probability of unemployment, andnon-pecuniary benefits.

Finally, although this analysis has identi-fied some moderator variables, it has yieldedevidence of the presence of others thatremain unidentified. Identification of thevariables that moderate the association be-tween various forms of participation andproductivity will assist the formulation ofpolicy relevant to participation.

REFERENCES

Alchian, Armen A., and Harold Demsetz. 1972. "Pro-duction, Information Costs, and Economic Organi-zation." American Economic Review, Vol. 62, No. 5(December), pp. 777-95.

Barrick, Murray R., and Ralph A. Alexander. 1987. "AReview of Quality Circle Efficacy and the Existenceof Positive-Findings Bias." Personnel Psychology, Vol.40, No. 3 (Autumn), pp. 579-92.

Bellas, Carl. 1972. Industrial Democracy and the Worker-Owned Firm: A Study of Twenty-One Plywood Companiesin the Pacific Northwest. New York: Praeger.

Benelli, Giuseppe, Claudio Loderer, and ThomasLys. 1987. "Laijor Participation in Corporate PolicyMaking Decisions: West Germany's Experience withCodetermination." Journal of Business, Vol. 60, No.4 (October), pp. 553-75.

Berman, Katrina V., and Matthew D. Berman. 1989."An Empirical Test of the Theory of the Labor-Managed Firm." Journal of Comparative Economics,Vol. 13, No. 2 (June), pp. 281-300.

Blanchflower, Daniel G., and AndrewJ. Oswald. 1988.

"Profit-Related Pay: Prose Discovered." EconomicJournal, Vol. 98, No. 392 (September), pp. 720-30.

Bonin,John P., Derek C.Jones, and Louis Putterman.1993. "Theoretical and Empirical Studies of Pro-ducer Cooperatives: Will Ever tbe Twain Meet?"Journal of Economic Literature, Vol. 31, No. 3 (Septem-ber), pp. 26-56.

Buck, Trevor. 1982. Comparative Industrial Systems:Industry Under Capitalism, Central Planning, and Self-Management. Hong Kong: Macmillan.

Bullock, R.J., and D.J. Svyantek. 1983. "Positive-Findings Bias in Positive-Findings Bias Research."Academy of Management Proceedings, Vol. 43, pp. 221-24.

Bullock, R. J., and Mark E. Tubbs. 1990. "A CaseMeta-Analysis of Gainsharing Plans as OrganizationDevelopment Interventions." Journal of Applied Be-havioral Science, Vol. 26, No. 3, pp. 383-404.

Cable,JohnR., and Felix R. FitzRoy. 1980. "Produc-tive Efficiency, Incentives, and Employee Participa-tion: Some Preliminary Results for West Germany."


Kyklos, Vol. 33, No. 1, pp. 100-121.Ghow, Gregory G. 1983. Econometrics. Auckland:

McGraw-Hill.Gonte, Michael, and Jan Svejnar. 1988. "Productivity

Effects of Worker Participation in Management,Profit-Sharing, Worker Ownership of Assets, andUnionization in U.S. Firms." Intemational Journal ofIndustrial Organization, Vol. 6, No. 1 (March), pp139-51.

. 1990. "The Effects ofWorker Participation inManagement, Profits, and Ownership of Assets onEnterprise Performance." In Katharine G.Abrahamand Robert B. McKersie, eds.. New Developments in theLabor Market: Toward a New Institutional Paradigm.Cambridge, Mass.: MIT Press, pp. 59-84.

Gonte, Michael, and Arnold S. Tannenbaum. 1978."Employee-Owned Companies: Is the DifferenceMeasurable?" Monthly Labor Review, Vol. 101 Huly)pp. 23-28.

Gooke, William N. 1989. "Improving Productivityand Quality Through Gollaboration." IndustrialRelations, Vol. 28, No. 2 (Spring), pp. 299-319.

. 1994. "Employee Participation Programs,Group-Based Incentives, and Company Perfor-mance: A Union-Nonunion Comparison." Indus-trial and Labor Relations Review, Vol. 47, No. 4 (July)pp. 594-609.

Graig, B., andjohn Pencavel. 1992. "The Behavior ofWorker Gooperatives: The Plywood Gompanies ofthe Pacific Northwest." American Economic Review,Vol. 82, No. 5 (December), pp. 183-1105.

Gutcher-Gershenfeld, Joel. 1991. "The Impact onEconomic Performance of a Transformation inWorkplace Relations." Industrial and Labor RelationsReview, Vol. 44, No. 2 (January), pp. 241-60.

Defourny,Jacques, Saul Estrin, and Derek G.Jones.1985. "The Effects of Workers' Participation onEnterprise Performance." Intemational Journal ofIndustrial Organisation, Vol. 3, No. 2 (June) pp197-217.

Doucouliagos, Ghris. 1993. "The Economics of Capi-tal Hiring Labour and Labour Hiring Gapital."Annals of Public and Cooperative Economics, Vol. 64,No. 2 (April-June), pp. 227-56.

Espinosa, Juan G., and Andrew S. Zimbalist. 1981.Economic Democracy: Workers' Participation in ChileanIndustry, 1970-1973. New York: Academic Press.

Estrin, Saul. 1991. "Some Reflections on Self-Man-agement, Social Ghoice, and Reform in EasternEurope." Joumal of Comparative Economics, Vol. 15,No. 2 (June), pp. 349-66.

Estrin, Saul, and Derek G.Jones. 1992. "The Viabilityof Employee-Owned Firms: Evidence from France."Industrial and Labor Relations Review, Vol. 45, No. 2(January), pp. 323-38.

. 1995. "Worker Participation, Employee Own-ership, and Productivity: Results from French Pro-ducer Gooperatives." In Derek G. Jones and JanSvejnar, eds.. Advances in the Economic Analysis ofParticipatory and Labor-Managed Firms. Greenwich,Gonn.: JAI Press, pp. 3-24.

Estrin, Saul, Derek G.Jones, and Jan Svejnar. 1987."The Productivity Effects of Worker Participation:

Producer Gooperatives in Western Europe." Joumalof Comparative Economics, Vol. 11, No. 1 (March), pp40-61.

Fisher, Gynthia D., and Richard Gitelson. 1983. "AMeta-Analysis of the Gorrelates of Role Gonflict andAmbiguity." Joumal of Applied Psychology, Vol. 68,No. 2 (May), pp. 320-33.

FitzRoy, Felix R., and Kornelius Kraft. 1985. "Union-ization, Wages, and Efficiency: Theories and Evi-dence from the U.S. and West Germany." Kyklos,Vol. 38, No. 4, pp. 537-54.

. 1986. "Profitability and Profit-sharing."/our-nal of Industrial Economics, Vol. 35, No. 2 (Decem-ber), pp. 113-30.

• 1987. "Cooperation, Productivity, and Profit-Sharing." Quarterly Joumal of Economics, Vol. 102,No. 1 (February), pp. 23-35.

. 1992. "Forms of Profit-sharing and Firm Per-formance: Theoretical Foundations and EmpiricalProblems." Kyklos, Vol. 45, No. 2, pp. 209-26.

. 1993. "Economic Effects of Godetermination."Scandinavian Joumal of Economics, Vol. 95, No. 3 pp365-75.

Fox, Alan. 1974. Beyond Contract: Work, Power, andTrust Relations. London: Faber & Faber.

Furubotn, Eirik G. 1978. "The Economic Gonse-quences of Godetermination on the Rate and Sourceof Private Investment." In S. Pejovich, ed.. TheCodetermination Movement in the West: Labor Participa-tion in the Management of Business Firms. Lexington,Mass.: Lexington Books.

Glass, Gene, Barry McGaw, and Mary L. Smith. 1981.Meta-Analysis in Social Research. Beverly Hills, Galif.:Sage.

Gooding, Richard Z., andjohn A. Wagner, III. 1985."A Meta-Analytic Review of the Relationship Be-tween Size and Performance: The Productivity andEfficiency of Organizations and Their Subunits."Administrative Science Quarterly, Vol. 30 (December)pp. 462-81.

Gurdon, Michael, and Anoop Rai. 1990."Godetermination and Enterprise Performance:Empirical Evidence from West Germany." Jouma/ ofEconomics and Business, Vol. 42, No. 4 (November)pp. 289-302.

Hunter,JohnE., and Frank L. Schmidt. 1990. Meth-ods of Meta-Analysis: Correcting Error and Bias inResearch Findings. Newbury Park, Galif.: Sage.

Jarrell, Stephen B., and T.D.Stanley. 1990. "A Meta-Analysis of the Union-Nonunion Wage Gap." Indus-trial and Labor Relations Review,Vol. A'i,No. 1 (Octo-ber), pp. 54-67.

Jensen, Michael G., and William H. Meckling. 1979."Rights and Production Functions: An Applicationto Labor-Managed Firms and Godetermination."Joumal of Business, Vol. 52, No. 4 (October) pp469-506.

Johnston, J. 1984. Econometric Methods. Auckland-McGraw-Hill.

Jones, Derek G. 1980. "Producer Go-operatives inIndustrialised Western Economies." British Joumalof Industrial Relations, Vol. 18, No. 2 (June), pp. 141-53.


. 1982. "British Producer Gooperatives, 1948-1968: Productivity and Organizational Structure."In Derek Jones and Jan Svejnar, eds.. Participatoryand Self-Managed Firms. Toronto: Lexington, pp.175-98.

. 1987. "The Productivity Effects of WorkerDirectors and Financial Participation by Employeesin the Firm: The Gase of British Retail Goopera-tives. " Industrial and Labor Relations Review, Vol. 41,No. 1 (October), pp. 79-92.

. 1993. "The Productivity Effects of EmployeeOwnership Within Gommand Economies: Evidencefrom Poland." Managerial and Decision Economics,Vol. 14, No. 5 (September), pp. 475-85.

Jones, Derek G., and David K. Backus. 1977. "BritishProducer Gooperatives in the Footwear Industry:An Empirical Evaluation of the Theory of Financ-ing." Economic Joumal, Vol. 87 (September), pp.488-510.

Jones, DerekG., andTakao Kato. 1993a. "The Scope,Nature, and Effects of Employee Stock OwnershipPlans in Japan." Industrial and Labor Relations Re-view, Vol. 46, No. 2 (January), pp. 352-67.

. 1993b. "Employee Stock Ownership Plans andProductivity in Japanese Manufacturing Firms."British Joumal of Industrial Relations, Vol. 31, No. 3(September), pp. 331-46.

Jones, Derek G., and Jan Svejnar. 1985. "Participa-tion, Profit Sharing, Worker Ownership, and Effi-ciency in Italian Producer Gooperatives." Economica,Vol. 52, No. 208 (November), pp. 449-64.

Katz, Harry G., Thomas A. Kochan, and Jeffrey H.Keefe. 1987. "Industrial Relations and Productivityin the U.S. Automobile Industry." Brookings Paperson Economic Activity, Vol. 3 (Autumn), pp. 685-715.

Katz, Harry G., Thomas A. Kochan, and Mark R.Weber. 1985. "Assessing the Effects of IndustrialRelations Systems and Efforts to Improve the Qual-ity of Working Life on Organizational Effective-ness." Academy of Management Joumal, Vol. 28, No. 3(September), pp. 509-26.

Kruse, Douglas L. 1992. "Profit Sharing and Produc-tivity: Microeconomic Evidence from the UnitedStates." Economic Joumal, Vol. 102, No. 410 (Janu-ary), pp. 24-36.

. 1993. Profit Sharing: Does It Make a DifferencedKalamazoo, Mich.: W.E. Upjohn Institute for Em-ployment Research.

Kumbhakar, Subal G., and Amy E. Dunbar. 1993."The Elusive ESOP-Productivity Link: Evidencefrom U.S. Firm-Level H^Lt^L." Joumal of Public Econom-ics, Vol. 52, No. 2 (September), pp. 273-83.

Lee, Barbara W. 1989. Productivity and EmployeeOwnership: The Case of Sweden. Uppsala: ActaUniversitatis Upsaliensis.

Levine, David I., and Laura D'.Andrea Tyson. 1990."Participation, Productivity, and the Firm's Envi-ronment." In Alan Blinder, ed.. Paying for Productiv-ity: A Look at the Evidence. Washington, D.G.:Brookings Institution, pp. 183-244.

Locke, E. A., D. B. Feren, V. M. McGaleb, K. N. Shaw,and A. T.Denny. 1980. "The Relative Effectiveness

of Four Methods of Motivating Employee Perfor-mance." In K. D. Duncan, M. M. Gruneberg, and D.Wallis, eds.. Changes in Working Life. Ghichester:John Wiley & Sons, pp. 363-88.

Marascuilo, Leonard A. 1971. Statistical Methods forBehavioral Science Research. New York: McGraw-Hill.

Miller, Katherine I., and Peter R. Monge. 1986."Participation, Satisfaction, and Productivity: AMeta-Analytic Review." Academy of Management Jour-nal, Vol. 29, No. 4 (December), pp. 727-53.

Mitchell, Daniel J. B., David Lewin, and Edward E.Lawler III. 1990. "Alternative Pay Systems, FirmPerformance, and Productivity." In Alan Blinder,ed.. Paying for Productivity: A Look at the Evidence.Washington, D.G.: Brookings Institution, pp. 15-94.

Rooney, Patrick M. 1988. "Worker Participation inEmployee-Owned Firms." Joumal of Economic Issues,Vol. 22, No. 2 (June), pp. 451-58.

Rosenberg, Richard, and Eliezer Rosenstein. 1980."Participation and Productivity: An EmpiricalStudy." Industrial and Labor Relations Review,Vol. 33,No. 3 (April), pp. 355-67.

Rosenthal, Robert. 1987. Judgement Studies: Design,Analysis, and Meta-Analysis. Gambridge: GambridgeUniversity Press.

Schmidt, Frank L., John E. Hunter, and Nambury S.RaJu. 1988. "Validity Generalization and Situ-ational Specificity: A Second Look at the 75% Ruleand Fisher's z Transformation." Joumal of AppliedPsychology, Vol. 73, No. 4 (November), pp. 665-72.

Schmitt, Neal, Richard Z. Gooding, Raymond A. Noe,and Michael Kirsch. 1984. "Meta-Analysis of Valid-ity Studies Published Between 1964-1982 and theInvestigation of Study Gharacteristics." PersonnelPsychology, Vol. 37, No. 3 (Autumn), pp. 407-22.

Schuster, Michael. 1983. "The Impact of Union-Management Gooperation on Productivity andEmployment." Industrial and Labor Relations Review,Vol. 36, No. 3 (April), pp. 415-30.

. 1984a. "The Scanlon Plan: A LongitudinalAnalysis." Joumal of Applied Behavioral Science, Vol.20, No. 1, pp. 23-38.

. 1984b. Union-Management Cooperation.Kalamazoo, Mich.: W.E. Upjohn Institute for Em-ployment Research.

Scott, Dow K., and Stephen G. Taylor. 1985. "AnExamination of Conflicting Findings on the Rela-tionship Between Job Satisfaction and Absenteeism:A Meta-Analysis." Academy of Management Joumal,Vol. 28, No. 3 (September), pp. 599-612.

Spector, Paul E., and Edward L. Levine. 1987. "Meta-Analysis for Integrating Study Outcomes: A MonteGarlo Study of Its Suspectibility to Type I and TypeII Errors." Joumal of Applied Psychology, Vol. 72, No.1 (February), pp. 3-9.

Stanley, T. D., and Stephen B. Jarrell. 1989. "Meta-Regression Analysis: A Quantitative Method of Lit-erature Surveys." Joumal of Economic Surveys, Vol. 3,No. 2 (June), pp. 161-70.

Sterner, Thomas. 1990. "Ownership, Technology,and Efficiency: An Empirical Study of Goopera-tives, Multinationals, and Domestic Enterprises


in the Mexican Cement Industry." Journal ofComparative Economics, Vol. 14, No. 2 (June), pp.286-300.

Svejnar,Jan. 1982. "Codetermination and Productiv-ity: Empirical Evidence from the Federal Republicof Germany." In Derek Jones and Jan Svejnar, eds..Participatory and Self-Managed Firms. Toronto: Lex-ington, pp. 199-212.

U.S. General Accounting Office. 1987. Employee StockOwnership Plans: Little Evidence of Effects on CorporatePerformance. GAO/PEMP-88-1, October.

Wadhwani, Sushil, and Martin Wall. 1990. "TheEffects of Profit-sharing on Employment, Wages,Stock Returns, and Productivity: Evidence fromtJ.K. Micro-Data." Economic fournal. Vol. 100(March), pp. 1-17.

Wagner, John A. Ill, and Richard Z. Gooding. 1987.

"Shared Influence and Organizational Behavior: AMeta-Analysis of Situational Variables Expected toModerate Participation-Outcome Relationships."Academy of Management fournal. Vol. 30, No. 3 (Sep-tember), pp. 524-41.

Wanous,John P., Sherry E. Sullivan, andjoyce Malinak.1988. "The Role of Judgement Calls in Meta-Analy-sis." fournal of Applied Psychology, Vol. 74, No. 2(April), pp. 259-64.

Weitzman, Martin, and Douglas L. Kruse. 1990. "ProfitSharing and Productivity." In Alan Blinder, ed..Paying for Productivity: A Look at the Evidence. Wash-ington, D.C.: Brookings Institution, pp. 95-142.

Whitener, Ellen M. 1990. "Confusion of ConfidenceIntervals and Credibility Intervals in Meta-Analysis."Journal of Applied Psychology, Vol. 75, No. 3 (June),pp. 315-21.

worker participation and productivity in...

Documents