There is no reason to fear that higher rates of unionization will impede efficiency or labor productivity. Notes 1. Doucouliagos, Christos, and Patrice Laroche. A meta-analysis. OECD estimates of labour productivity for September System error. Please try again! How was the reading experience on this article?
The text was blurry Page doesn't load Other:. Details Include any more information that will help us locate the issue and fix it faster for you. Thank you for submitting a report! Submitting a report will send us an email through our customer support system. Submit report Close. Recommended Articles Loading A Review Article.
Interior Mines. Surface Mining of Coal. Evidence from Coal Mining. Automobile Industry: Comment and Discussion. Engineering Firms. For example, theory identifies factors such as closed-shop arrangements, recognition of unions by firms, existence of participatory mechanisms, existence of multi- unionism, competitive pressures, establishment size, and industrial relations climate as important in moderating the impact of unions on productivity.
The aim of meta-regression analysis is to identify moderator variables asso- ciated with the empirical literature and to explore the impact of specifica- tion differences. Unfortunately, many of the factors that theory identifies as important cannot be investigated. The meta-regression analysis is restricted to data drawn from the studies themselves.
For example, there are not enough observations to explore the impact of closed-shop arrangements and multiunionism. Equation 1 presented earlier also includes variables that can be quantified the K values. These are variables for which authors present the average value and include the establishment size, the capital-to-labor ratio, and union density. Unfortunately, too few studies provided information on establishment size and the capital-to-labor ratio.
Hence the impact of these variables could not be tested. However, 24 studies did provide information on average union density. A separate meta-regression analysis was under- taken with union density as an additional explanatory variable. With only 73 observations it is necessary to limit the number of potential moderating variables. The procedure adopted was as follows: First, after assembling all the available information from each of the published studies, a list of potential moderator variables was constructed.
Second, the simple correlation coefficients were calculated for each of the potential moderating variables and the union-productivity effect. This helped to identify which potential moderator variables could be omitted as candidates on the basis of very small simple correlation coefficients. The simple correlation coeffi- cients of all pairs of variables also were examined to detect the existence of high collinearity among potential explanatory variables.
This process 13 An additional issue concerns the possibility of selectivity. For example, unions may choose to unionize more productive firms see Chezum and Garen This issue is not explored in the meta-analysis. The definitions of these variables are presented in Appendix A. For example, 75 percent of the studies used U. The third step involved using the 31 potential explanatory variables in meta-regression analysis. The results are presented in column 2 of Table 4. This is our starting meta-regression model.
Not surprisingly, many of the variables in the base or general model are not statistically significant.
This process of sequentially eliminating statistically insignificant variables is known as the general-to-specific modeling strategy see Hendry The final or specific 14 In order to test the sensitivity of the results to the exclusion of a number of variables, meta- regression analysis also was undertaken with the addition of the excluded variables to the models presented in Table 4.
In all cases, statistical tests show that these variables can be excluded from the meta-regression analysis. These are mainly similar to those presented for total productivity effects; e. Naturally, the partial correla- tions and the associated productivity effects are highly correlated the simple correlation coefficient between the two is 0.
In order to test the sensitivity of the meta- regression analysis, the specific model was reestimated with some of the outliers removed. First, we removed 5 percent of the largest positive and largest negative total productivity effects.
Second, we removed 5 percent of the smallest and 5 percent of the largest sample size studies. These results are pre- sented in column 4 of Table 4, with the later estimates presented in brackets. We concur with the body of literature that argues for the adoption of a production function as the preferred framework for investigating the impact of unions on productivity e. Unions affect the production process, and hence empirical investigations should attempt to model this process.
In particular, a sound modeling strategy should at least control for capital stock. In addition, it is desirable to control for the qual- ity of the labor input and, where it is appropriate, to allow for technical change in the production process.
Capital stock, labor quality, and technical change are all likely to vary in response to cost-minimizing responses by firms to unionism. Accordingly, the meta-regression analysis was carried out separately only on those studies which adopted a production process framework and which controlled for capital stock. This reduces the sample from 73 to 53 studies. In some ways this group can be regarded as a sort of best-practice group of studies.
The meta-regression analysis was undertaken by first estimating a base model and then reducing this sequentially.
The results of the final reduced model are presented in column 5. The meta-regression results presented in column 6 relate only to studies that used union density as the measure of union presence. This allows inves- tigation of the hypothesis that higher union density is associated with lower estimated union-productivity effects.
One of the findings in some of the individual studies is that the association between unions and productivity is moderated by union density. That is, the association may be different at low levels of union density compared with high levels of union density. This reduces the sample to only 24 studies and hence limits this analysis. Studies using higher levels of trade union density tend to find lower union-productivity effects.
The first group of variables explores industry-specific effects through the CONS, MANUF, 18 Many of the studies using union density did not report the mean density, and this information is not available from alternative sources. None of these is statistically significant.
The next set of variables involves specification issues, most of which relate to characteristics of production functions. Controlling for labor quality tends to increase the reported union-productivity effect. This is consistent with the argument advanced by Wessels that unionized firms do not hire better-quality workers.
The DENSITY variable was included to capture any differences between studies that used union density to measure union pres- ence and those which used a union dummy. Some studies measured union presence as union recognition for collective-bargaining purposes.
These were grouped together with the union dummy studies. The way unioniza- tion is measured does not appear to make any difference to the estimated effects. Studies that use firm-level data as opposed to data aggre- gated at the industry level report lower total productivity effects. Studies using U. This is consistent with the meta-analysis presented in Table 2 and is at odds with some of the qualitative reviews. JAPAN emerges with a statistically significant nega- tive coefficient in the production function group of studies.
This association is revealed to be negative. While it does have the hypothesized posi- tive coefficient, it is not statistically significant. As indicated earlier, the existence of publication bias can be explored through meta-analysis. Publication issues are examined through several variables.
In unreported results, dummies were used for individual author effects, testing, for example, whether publications by Allen, Clark, and Kleiner generate distinct estimates. There is no evidence that this is so. However, Table 4 shows that three publication-related variables are important. Papers published in management journals MANJ tend to find positive productiv- ity effects. After controlling for other study characteristics, the variable for JLR has a negative coefficient.
While it is true that a number of authors have associated with each other, there is no reason to believe that their research is biased in any systematic manner. The existence of a cross-author effect is investigated through the INFLUEN variable, which attempts to capture the influence of any author over another. Naturally, it is possible for authors to be influenced by other researchers whom they have not acknowledged.
Such effects cannot be tested. Not necessarily. The results for JLR could, for example, reflect a self- selection process by the authors and possibly even more carefully prepared studies. The results are nevertheless interesting. If we accept that studies published in other journals, such as the Industrial and Labor Relations Review and Industrial Relations, are also of high quality, then the negative coefficient on JLR may indicate publication bias.
After controlling many other aspects of the studies, JLR is the only publication outlet with a negative coefficient. A similar conclusion 22 Sixty-four percent of the studies published in this journal reported a negative productivity effect. The use of these two separate variables produced unstable parameter estimates. Time effects are explored through four year dummies with as the base. YEAR is included to capture the fad effect. Certain types of research become fashionable in journals at certain periods of time, and this variable is designed to capture this.
There is no evidence that this is the case. Of the time dummies, only the variable was significant, suggesting that stud- ies that used data relating to the s found favorable productivity effects compared with other years. It is not clear why this is the case.
A number of measurement variables are included. Output measurement issues are investigated by the inclusion of VALUE, comparing these studies with those which used a physical measure of output. A priori, the use of valued-added measures can be expected to overstate the union-productivity effect.
Value-added measures of output may be influ- enced by the impact of unions on wages, which are then passed onto con- sumers through higher prices. Interestingly, after controlling for other study characteristics, the coefficients on VALUE are negative.
Studies measuring output as value added report lower productivity effects. A similar effect emerges with respect to studies that use time-series data. The other variable appearing in the regressions is UFOCUS, which is added to capture any differences between studies whose primary focus was estimation of the union-productivity effect versus those studies which included unionization merely as a control variable.
It may be the case that studies that set out to investigate the impact of unionization may give more thought to the relevant issues and hence may produce different results than studies that include unions merely as a control variable. This variable is not statistically significant. Diagnostic tests were conducted on each of the mega-regression analysis models in order to test their reliability.
These are large-sample tests and hence should be interpreted with caution. Nevertheless, the diagnostic tests indicate that the meta-regression models appear to be free from het- eroscedasticity and mispecification. This is comforting because it is important to pay particular attention to heteroscedasticity in meta-regression analysis. What do the estimated models presented in Table 4 tell us about the union-productivity effect? First, it is clear that at least some of the variation in published results is artifactual.
That is, it is the product of measurement, data, and specification differences rather than differences in the underlying union-productivity effect. We conclude that the way models are constructed and especially the type of data used does systematically influence the reported union-productivity effects. Second, the statistical significance of the USA and JAPAN variables shows that some of the variation in published results derives from real economic forces rather than the way the studies are conducted.
Third, many control variables do not affect the union-productivity association. This does not mean that they should not be included in empirical investigations, only that the union- productivity association appears to be insensitive to them. Fourth, there is a tendency for positive results to be published in some parts of the literature and negative results in others, suggesting the possible existence of some publication bias in this literature.
The results are suggestive only, but do raise the specter that at least part of this literature may present a misleading picture of the union-productivity effect. Clearly, the issue of publication bias warrants further investigation. After reviewing the literature, we offer a number of comments regarding methodology and reporting of results. First, specification of the production function itself is of some interest.
Most studies used the Cobb-Douglas specification. It is well known that the Cobb-Douglas specification is a very restrictive functional form. However, only a handful of studies reported either in the text or in the footnotes actually testing the appropriateness of this specification versus more flexible specifications, such as the translog. Second, a production-function approach offers the best approach for modeling the production process and hence the role of unions within that.
Most studies have in fact used this approach and have allowed for the influence of labor and capital to be controlled for. However, given the substantial evidence that inefficiency exists in the production process Leibenstein , it seems inappropriate to assume that the production process occurs with full technical and allocative efficiency.
If there is inefficiency in the production process, then the use of OLS will tend to lead to inefficient estimates of the production function parameters see Fried, Lovell, and Schmidt ; Coelli, Rao, and Battese Most studies, however, have used OLS.
Indeed, it seems inappro- priate to investigate the impact of unions on productivity without allowing for inefficiency in specification of the production function. Only a small number of studies have allowed for this e. Since there are such few examples, we need to be cautious about reading too much into this. Finally, there is the issue of reporting standards.
Many studies failed to report important information, such as sample means. This restricts com- parisons between studies. With growth in both interest in the evaluation of a body of literature and techniques for doing so, authors should aim to increase the value of their work by providing enough information so that others can use their research in a manner that will facilitate the synthesis of research from various studies.
Concluding Remarks As in many areas in economics and industrial relations, theory does not establish an unambiguous association between unions and productivity. Importantly, the empirical literature has not resolved the conflicting theo- retical arguments.
Meta-analysis and meta-regression analysis offer one way of synthesizing the available empirical evidence and drawing statistically valid inferences from it. Meta-regression analysis is useful in quantifying the impact of differences in study characteristics on research findings. This has to be estimated using maximum-likelihood techniques. However, there exist country- and industry-specific associations between unions and productiv- ity.
A negative association appears for the United Kingdom and Japan, whereas a positive one exists for the United States in general and for U. The meta-regression results suggest that at least part of the variation in the estimated association between unions and productivity across studies is due to differences in study characteristics rather than to differences in the actual union-productivity effect. There are several issues that warrant further investigation. Additional empirical investigations are needed in countries other than the United States and the United Kingdom.
A clearer picture of the role of unions and productivity within a framework of inefficiency is also important. There is also need for additional investigations on the role of industrial relations climate and the role of establishment size and union density in moderating the union-productivity association. If the underlying relationship between unions and productivity is not universal and, for example, changes over time as well as across industry, then a change in research direction is war- ranted.
In particular, researchers need to allow explicitly for these changes to be captured in their modeling strategies. Furthermore, unions have an impact on areas other than productivity. There is a significant literature on the association between unions and employment, unions and profitability, and unions and productivity growth. Meta-analyses in these areas would offer additional insights into the net impact of unions on performance. A number of studies have found that unionization has a negative effect on stock prices, whereas others have found that it can be detrimental to productivity growth.
These are con- sistent with the results presented in this article. For example, unions can have a mild productivity-enhancing effect while also reducing profitability higher-priced union labor and productivity growth. The stock price effect is a reflection of profitability and concerns over future prospects rather than current productivity levels. Firms and financial markets are concerned not just with productivity levels but also with productivity growth. Indeed, the latter maybe given greater weight.
Additionally, firms are likely to compare the returns from unionization with those from other interventions, such as profit sharing and employee stock ownership. These interventions may make greater contributions to productivity than unionization.
These fac- tors could help explain, at least in part, the decline in union density and membership numbers, even if unions have a positive impact on productivity.
A Review Article. Allen, Steven G. Argys, Laura, and Rees D. Arthur, Jeffrey B. Baldwin, John. Caves, pp. Bairam, Erkin. Homogenous and Nonhomogenous Production Functions. Aldershot: Avebury. Bartel, Ann P. Batt, Rosemary. Belman, Dale. Voos, pp. Bemmels, Brian. Black, Sandra E. Boal, William M. Booth, Alison. The Economics of the Trade Union. New York: Cambridge University Press. Brown, Charles, and James L. Brunello, Giorgio. Bycio, Peter. Byrnes, Patricia, and Rolf Fare. Interior Mines.
Grosskopf, and C. Surface Mining of Coal. Card, David, and Alan B. Cavalluzzo, Linda, and Dennis Baldwin. Knox Lovell, and Shelton Schmidt, pp.
New York: Oxford University Press. Caves, Richard E. Caves and Lawrence B. Krause, pp. Washington: Brookings Institution. Efficiency in U. Manufacturing Industries. Chezum, Brian, and John E. Evidence from Coal Mining. Clark, Kim. Coelli, Tim, D.
Prasada Rao, and George E. An Introduction to Efficiency and Productivity Analysis. Boston: Kluwer. Connolly, Robert A. Hirsch, and Mark Hirschey. Conte, Michael A. Abraham and R. McKersie, pp. Conyon, Martin J. Blundell, D. Card, and R. Chicago: University of Chicago Press. Cook, Thomas D. Cordray, Heidi Hartmann, Larry V. Hedges, Richard J. Light, Thomas A.
0コメント