A comparison of interest arbitrator decision-making in experimental and field settings.
Two research strategies have been followed in recent studies of this topic. In one design, data on actual arbitration decisions are used to estimate the mean and variance of the distribution describing the fair award beliefs of arbitrators. The parameters of this distribution are estimated using data describing factors believed important in wage determination: the characteristics of the disputing parties, the bargaining environment, and the parties' positions on the unresolved issue. Aside from the parties' position on the unresolved issue, the actual information communicated by the parties to the arbitrator during the hearing is typically not known. The second, more popular, strategy is to ask arbitrators to decide one or more hypothetical disputes in which the facts and offers are experimentally varied.
Arguments can be made for and against either approach. On the one hand, field data on arbitration decisions are often difficult to collect and analyze, and some field studies have yielded inconsistent results, probably because of the omission of important variables influencing arbitrators' beliefs. Experiments can be designed with extensive controls that guarantee a level of internal validity unobtainable in field research. On the other hand, the obvious differences between experimental and field environments cast doubt on the external validity of any experimental results.
In this study we provide evidence on whether and how the results from field and laboratory research on arbitrator decision-making differ by comparing similar experimental and field decisions made by a common group of arbitrators. Specifically, the decisions made by 19 arbitrators who participated in an experiment are compared with the field decisions these same arbitrators made in teacher disputes under Wisconsin's mediation-arbitration statute.
Field Versus Experimental Settings
The popularity of the experimental approach to the study of arbitrator decision-making is a function of the difficulties associated with collecting and analyzing field data. One problem is that field data are complicated by the existence of different types of arbitration. Studying conventional arbitration decisions is difficult for two reasons: first, there is frequently no documentation of participants positions presented to the arbitrator, and second, the potential effect of offers on arbitrators' beliefs must be controlled when evaluating the effect of the offers on awards (Ashenfelter and Bloom 1984; Bloom 1986).
Although the use of field data to study decision-making under final offer arbitration does not suffer from these problems, the empirical results have sometimes been difficult to interpret. In Ashenfelter and Bloom (1984), the coefficients on the variables that were thought to influence arbitrators' views of an appropriate award were very unstable across the three years under study. For two of the four variables, the estimated parameters were statistically significant in each year, but were of different signs from one year to the next. The signs for the other two variables were the same across the three-year period, but the estimates for one year were about half the size of those for the other years. As the authors note, this variation in results across years may be due to important variables influencing arbitrator beliefs that were not included in their analysis.
An experiment can overcome the short-comings of field data. The inherent control in the experimental design maximizes internal validity by ensuring that there are no omitted characteristics of the dispute or simultaneity between awards and offers that might bias parameter estimates. Such a design has special appeal in the case of arbitrator decision-making, since theory has failed to suggest a clear set of factors that influence arbitrator preferences. In addition, orthogonal construction of the facts can facilitate inferences about which factors are dominant in determining arbitrator preferences.
Balanced against these advantages are concerns over the external validity of simulations. Although the experimental studies have tended to use actual decision-makers (that is, arbitrators) rather than college students, these studies possess many of the same characteristics that have led some to question the generalizability of experimental research on negotiation and arbitration (Anderson 1981; Gordon, Schmitt, and Schneider 1984). One criticism is that such studies lack "ecological validity" (Locke 1986). Reading and deciding cases based on short scenarios differs sharply from hearing testimony during the arbitration process. In an experiment, the scenarios present "facts" to the decision-maker as if the parties had stipulated them. For example, the composition of the outside wage comparability group is taken as undisputed by the parties. Such agreement often does not obtain in actual disputes. When the "facts" are in dispute, the arbitrator must either resolve the evidentiary dispute or discount the facts when deciding on the appropriate award.
Second, decisions in an experiment are made without the pressure to make an "acceptable" award that is found in actual cases. In an experiment an arbitrator knows the decisions have not consequences for his or her future arbitration practice and, as a result, may have little or no concern about the decisions' acceptability to the parties.
Finally, the practice of orthogonal construction of the independent variables may actually impair external validity. For example, in Bazerman and Farber (1985), where the offers were constructed without regard to the facts of the case, as substantial minority of conventional arbitration awards were made that were equal to or outside the boundaries defined by the positions of the parties. Such awards, which are rare in actual disputes, could indicate that the orthogonal construction of facts offers affected decisions in all the experimental disputes and biased the results (Bloom 1986; Olson 1988). Surely arbitrators expect the position of the parties to be influenced by the factors presented in an arbitration hearing, and they may alter their decision rules when that expectation is not borne out.
Although these limitations call into question the external validity of laboratory research, the assessment of external validity is more an inductive than a deductive exercise (Locke 1986:5). Lab settings do differ from those found in the field, but empirical evidence is needed to determine whether, and how, results from field and laboratory research on arbitrator decision-making differ. This study provides such evidence.
The Experimental Data
Twenty-two arbitrators where asked to decide 32 hypothetical interest disputes during the fall of 1986. The 22 arbitrators were selected to participate in the study because they had made numerous decisions under Wisconsin's two interest arbitration statutes.(1) Nineteen of the original 22 subjects were used in this study. (One subject in the original study refused to allow his or her data to be used in research beyond the initial study, and two arbitrators in the experiment did not decide any teacher disputes during the period studied.) The 32 hypothetical disputes were between a city and police union and dealt with only a single issue, the hourly wage for police officers. The nine variables describes in Table 1 were manipulated in the disputes. The variables fell into three broad categories; wage levels and wage increases for "comparable" bargaining units, changes in the cost of living, and financial characteristics of the employer.
The 32 scenarios were constructed using a two-step process designed to minimize the correlation between the treatment levels. First, the values for each of the nine treatment variables were divided into "high" and "low" groups, with four values in each category. The criteria were then combined based on the two categories according to a 1/16 fractional replication of a 2. factorial design. This procedure resulted in 32 scenarios in which the two ordinal levels on each of the nine factors were orthogonal. For each factor shown in Table 1, the values in the first row were the "low" values for that factor and those in the second row were the "high" values. Because of variation within the "high" and "low" categories, the procedure minimized but did not eliminate the correlation between the treatment levels. Table 2 shows the correlation between the interval values of the variables show in Table 1 for the 608 scenarios (19 arbitrators X 32 scenarios). Most of the correlations are less than .10, and the largest correlation - between cost of living changes and the percentage increase in comparable police units - is .39. The cases were presented to each arbitrators in an identical but randomly selected order.
[TABULAR DATA OMITTED]
The experiment differed from the Bazerman and Farber (1985, 1986) studies in several important respects. First, the data were collected in face-to-face interviews by the second author, permitting greater control than is possible in a mail survey. Second, the use of a single occupation covered by statutory decision criteria that were familiar to all the participants increased the realism of the design. Third, both the current wage ($10/hour) and the offers of the parties ($10.30 and $10.80/hour) were constant across all disputes.(2) Fourth, the subjects made only one award for each case, and this award was made under the final offer procedure of Wisconsin's police and fire-fighter interest arbitration law.
Following previous research, the arbitrator decision model was estimated using a standard probit model. Let [W.sub.A,] the arbitrator's view of a fair percentage wage increase, be a linear function of the nine variables:
(1) [W.sub.A] = X[BETA] + [epsilon], where epsilon N(O, [delta][.sup. 2]
The probability that the union's offer is chosen is the probability that [W.sup.A] is greater than or equal to the mean of the final offers (MOFFER), 5.5%:
(2) Pr(U == 1) = 1 - [THETA]((5.5 - X[BETA])/[sigma]
where [THETA](.) is the CDF for the standard normal distribution. Previous studies have been able to estimate [sigma] because of variation in the final offers of the parties. We cannot do so using these data because the offers were identical in all the cases.
Column 1 and 3 of Table 3 report the probit estimates using two model specifications. The first column reports the results for a model that includes the nine manipulated variables, and column 3 reports results for a model that includes dummy variables for each arbitrator. The latter specification permits E([W.sub.A]/X[Beta]) to vary across the arbitrators while constraining the effect of the exogenous parameters to be the same for each arbitrator.
Six of the nine variables had a significant impact on decisions. The probability that the union's offer was chosen was positively related to wage increases in comparable police bargaining units (CPERC), employee working conditions as measured by the crime rate (CRIME), the cost of living (COL), and the share of the community's budget allocated to police protection (EXP). The arbitrators were also sensitive to local property tax rates (TAX) and were less likely to select the union's offer when tax rates were high. The subjects also favored wage increase that compressed wages relative to comparable units. The negative coefficient on CWAGE shows that the probability the union's offer was chosen declined if the city's wage level was high relative to comparable police bargaining units. The three variables that were not significant were wage levels and wage changes in other bargaining units within the local gtovernm,ent: arbitrators were not influenced by the firefighters' wage level (FWAGE), the increase received by firefighters (FINC), or the increase agreed to between the city and its other unionized employees (OTH-WAG).
[TABULAR DATA OMITTED]
The coefficients from the arbitrator dummies are not reported in Table 3, and as a set they were not significant at conventional levels using a likelihood ratio test ([X.sup.2] 18 df = 21.382, p = .216). Moreover, as the third column of Table 3 shows, the parameters on the other variables changed only slightly when mean shifts in arbitrator preferences were permitted. Although the dummy variables were not significant, some of the differences between individual arbitrators were substantial. For example, one arbitrator was significantly more likely (at the .10 significance level) to select the union's offer when compared to 11 of the remaining 18 subjects. Figure 1 shows the distribution of arbitrators based on the percentage of the 32 cases that were decided for the union. Since all the subjects decided the same set of cases, the variation in this figure reflects differences in their decision models.
The Field Data
Nineteen arbitrators who participated in the experiment were also decision-makers in teachers disputes under Wisconsin's mediation-arbitration statute. During the first 8.5 years of the statute (January 1, 1978, through the 1985-86 school year), complete data on offers and the hypothesized determinants of [W.sub.A] were available for 289 cases, of which 208, or 72%, were decided by participants in the experiment.(3)
Although the experimental and field data involved different occupations and laws, key features of the data were similar. Both were characterized by final offer by package schemes, and the statutory criteria arbitrators were to consider when making a decision were nearly identical in the field and experimental cases.(4) The major difference between the laws was that under the mediation-arbitration law the individual appointed as the arbitrator was required to mediate the dispute before initiating arbitration. Although this procedure may have influenced what disputes went to arbitration, it is unlikely to have had an effect on [W.sub.A] different from the effect of the arbitration hearing alone.
The variables determining the arbitrators' views of an appropriate award included recent and expected changes in the cost of living, comparable wage levels, wage increases in other relevant bargaining units, and local property taxes. Changes in the cost of living were proxied by the average annual change in the cost of living index (C.P.I. - All Urban Consumers) for the first 12 months of the arbitrated contract. Because of time delays in the arbitration process, awards covering a school year were usually not made until most of the year was completed.(5) Thus, arbitrators had nearly perfect information about changes in the cost of living for the first year of the contract when making a decision. Community fiscal support for education (willingness to pay) was measured using the school property tax rate in year t - 1.
The athletic conference has become the comparison group used by arbitrators when evaluating the offers of the parties in teacher disputes, because districts within an athletic conference are of comparable size and in close proximity to one another. A study of written awards in teacher disputes from 1978 to 1984 found arbitrators used athletic conference wages as a comparison when making an award in 71% of the cases (Halm 1985). Therefore, two measures of comparable wage settlements were constructed based on the district's standing in its athletic conference. The first comparability measure was the district's position in the conference wage distribution for the preceding school year (CWAGE), computed as the district's wage divided by the conference mean.
The second comparability measure was the average percentage wage increase in the current year for districts in the conference that reached a settlement without resorting to the mediation-arbitration process (CPERC). Bargaining pairs that settled without petitioning for arbitration reached agreement before arbitration awards were made because of the time delays in the arbitration process. Therefore, using the mean settlement in districts that did not resort to mediation-arbitration avoids the potential simultaneity between negotiated awards and arbitrator decision that is created when parties petition for arbitration but eventually settle after observing what arbitrators have done within the athletic conference.
The first column of Table 4 reports the standardized probit results; each reported coefficient is [BETA.sub.i/sigma.sub.f] and the coefficient on the mean offer of the parties is equal to - 1/[sigma.sub.f]. The second column reports the unstandardized parameters, which are interpreted as the effect of the independent variables on [W.sub.A.sup.6] Two parameters are significant at the .05 level (one-tail). The district's wage position within the district's athletic conference (CWAGE) had a substantial impact on [W.sub.A.] As in the experiment, arbitrators' view of [W.sub.A] depend on how the district's wage compared to wages in its group of comparable districts. If a district's wage was 5% above the athletic conference mean, [W.sub.A] was 2.4 percentage points lower than a comparable districts 5 percentage points below the conference mean. Wage settlements within the conference (CPERC) also had a significant effect on arbitrator beliefs about the appropriate award. The coefficient, .96, is significant greater than zero and not different from one (p = .916), indicating a one-to-one relationship between the average percentage increase in negotiated settlements within the athletic conference and [W.sub.A]. The parameter on local property taxes (TAX) is negative but not significant at conventional levels. The cost of living (COL) is positive and significantly different from zero (p = .045), suggesting that cost of living had only a modest direct effect on arbitrator beliefs over the 8.5 years.
[TABULAR DATA OMITTED]
The standard deviation of [sigma] was 8.4% and statistically significant at the .05 level. This result implies a substantial amount of uncertainly about what is viewed as a fair award even when the facts of the case and the weight given to the facts are known. The predicted fair wage increase for an average case was 8.23%, but the estimated error variance implies there was a 50% probability arbitrators believed the appropriate settlement was less than 2.56% or greater than 13.89.(7)
When the arbitrator dummies are added, the fit of the model improves only slightly. The hypothesis that the appropriate settlement given the facts of the case is the same across these arbitrators cannot be rejected using a likelihood ratio test ([X.sup.2]17 df = 13.208, p = .722])[.sup.8]. Again, however, some of the differences between individual arbitrators were significant. Finally, the identity of the arbitrator had virtually no effect on the error variance. Uncertainty declined only slightly to 7.l7% when the arbitrator dummies were added.
A Comparison of the Decisions in
the Two Samples
The test used to evaluate decision model consistency across the two samples is based on the simple premise that if the decision models are comparable, the variables common to both data sets should have a similar impact on [W.sub.A.] Note, however, that there are five variables in the experimental data (CRIME, FINC, FWAGE, OTHWAG, and EXP) hypothesized to influence arbitrators in police disputes that are not relevant in teacher disputes. These variables, which measure other wage settlements within the government and the budget share devoted to police, have no counterpart in teacher bargaining. A review of written teacher awards by the authors shows that these factors are not explicitly raised by the parties or the arbitrator. Blue-collar units in school districts tend to follow the settlement reached by teachers, and since school districts in Wisconsin are independent government units, all of their budget is devoted to education.
Two strategies exist for dealing with the unique variables in the experimental data. First, the variables unique to the experiment could be included in the analysis and only the effect of variables common to the two designs could be constrained to be the same. The second strategy - the one we employed - is to use only the variables common to both data sets and simply ignore information on the five criteria that were unique to the experiment. Although this second strategy ignores information available to the arbitrators in the experiment, the nearly orthogonal relationship between the criteria across the experimental scenarios ensures that any bias introduced by excluding the five unique variables is slight. As shown in the second and fourth columns of Table 3, the parameters on the remaining three variables change only slightly when the five variables are excluded. The virtue of the second strategy is that it makes decision consistency in the two samples easier to evaluate.
If there was variation in the final offers in the experiment, testing whether the decision-making process was identical in the two designs is based on separate tests of the following two null hypothesis (where F and E denote, respectively, field and experimental parameters):
[H.sub.1]: [beta.sub.F] = [beta.sub.E] [H.sub.2]: [omicron.sub.F] = [omicron.sub.E]
[H.sub.1] tests whether the variables common to the two data sets have an identical impact on [W.sub.A]. Rejection of [H.sub.2] implies that the uncertainty associated with predicting decisions in actual cases. Without variation in the final offers, [H.sub.1] and [H.sub.2] can only be jointly tested. If the joint hypothesis is rejected, it could be because of differences across the two designs in the [beta], in the [omicron], or both.
The estimates that constrain all of the common field and experimental equation parameters to be the same are reported in columns 1 and 2 of the Table 5. Column 2 gives the estimates for a model that includes a dummy variable equal to "1" for the experimental data. To test [H.sub.1] and [H.sub.2] jointly, a likelihood ratio test was constructed using the sum of the values of the log likelihood functions of the unconstrained models (column 2 of Table 3 and column 1 of Table 4) and the values of the log likelihood function from the constrained model (columns 1 and 2 of Table 5). When the intercept is not allowed to vary across the two samples, the joint hypothesis defined by [H.sub.1] and [H.sub.2] is rejected by the data ([X.sup.2]5 df = 29.29, p < .0001). The joint hypothesis continues to be rejected when the intercept is allowed to vary across the two samples ([X.sup.2]4 df = 13.535, p = .009).
Table 5. Estimates of Arbitrator Decision Model for the Pooled Experimental and Field Data. (n = 816; Standard Errors in Parentheses) Variable (1) (2) Constant 1.764** 5.581** (1.199) (1.556) Mean of the Parties' Offers -.049** -.183** (MOFFER) (.016) (.038) Community Property Tax -.243** -.245* Rate (TAX) (.101) (.102) Wage Rate in Comparable -2.781* -5.588** Units (CWAGE) (1.252) (1.452) Wage Increase in Comparable .166** .164** Units (CPERC) (.020) (.020) Cost of Living (COL) .081** .101** (.016) (.017) Experimental Dummy Variable -1.289** (DESIGN) (.332) Arbitrator Dummies No No -log L 486.264 478.384 (*) Statistically significant at the .05 level; **the .01 level (one-tail tests).
Further Tests Comparing the
Two strategies were pursued to further investigate the differences between the two samples. First, we tested whether the relative weight placed on each of the "facts" within each data set was identical across the two data sets. Specifically, we tested the following restrictions:
[H.sub.3]: [beta.sub.F],tax/[beta.sub.F],cola = [beta.sub.E],tax/[beta.sub.E],cola [H.sub.4]: [beta.sub.F],cwage/[beta.sub.F],cola = [beta.sub.E],cwage/[beta.sub.E],cola [H.sub.5]: [beta.sub.F],cperc/[beta.sub.F],cola = [beta.sub.E],cperc/[beta.sub.E],cola
If we are unable to reject all three hypothesis, we can conclude that the weight placed on taxes, comparable wage increases, and comparable wage levels relative to the weight placed on cost of living was the same in the field and experimental data. Failure to reject these three hypotheses., and a finding that [BETA.sub.f,cola = [BETA.sub.E.cola]' would imply that the weight are the same in the two samples and only [sigma]'s differ. Since [BETA.sub.E,cola] is not identified in the experimental data, however, [H.sub.3] - [H.sub.5] tests only whether the ratio of the weights on the different facts is the same in the experimental and field data.
Table 6 reports the tests for [H.sub.3] - [H.sub.5]. The first two columns report the ratio of the weight placed on taxes, comparable wage levels, and comparable wage increases to the weight placed on inflation for each of the two sample. The last column reports difference between the ration of the weights for the samples. This column shows that the differences in the ratio are not statistically significant in the field and experimental data.
The rejection of [H.sub.1] and [H.sub.2] and the inability to reject [H.sub.3]-[H.sub.5] are consistent with the interpretation that the weights placed on the various facts are the same in the experimental and field data but the level of uncertainty about what the arbitrator will do is different. One reason for the differences in the actual variances between the two settings could be the composition of cases in the field data and the final offer by package feature of the Wisconsin law. As Olson and Jarley (1991) report, many of the teacher final offer decisions that include wages also include other economic and non-economic issues in the package. If these other issues are given some weight in the decisions of arbitrators but are ignored by the researcher, the estimate of the error variance will be biased. Olson and Jarley (1991) confirm the importance of this distinction by obtaining a much smaller estimate of [sigma[ when the sample of awards is restricted to those cases that involve only wages.
Table 6. The Differences in the Ratio of the Weights Placed on the Facts in the Field and Experimental Data. (Standard Errors in Parentheses) Field Experimental Field- Data Data Experimental [Beta]/[beta col] -4.634 -1.631 -3.004 (10.584) (.779) (10.612) [Beta cwage/ beta col] -7.729 -6.776 -.952 (6.411) (1.799) (6.659) [Betacprc/beta col] 2.967 1.172 1.794 (2.138) (.261) (2.153)
It is possible the inclusion of multi-issue cases in the field data in this study is responsible for the rejection of [H.sub.1] and [H.sub.2] because the experimental data involved only the single issue of wages. To test this possibility, the sample of field cases was restricted to cases that involved only wages. This restriction reduced the number of cases in the field data from 208 to 85. The first column of Table 7 reports the estimates of the fair wage beliefs of the arbitrators for these 85 cases. The key point to note is that the estimate of [sigma] declined from 8.4% to 2.45%.
Columns 2 and 3 of Table 7 show the estimates from pooling the experimental data and the 85 field observations. The results in column 3 constrain the constant to be the same for the two data sets, and column 2 includes an experimental dummy variable. The joint hypothesis described by [H.sub.1] and [H.sub.2] is tested by comparing the log likelihood values from columns 2 and 3 of Table 7 with the log likelihood values for the unconstrained model. The unconstrained value is simply the sum of the log likelihood values shown in column 1 of Table 7 and column 2 of Table 3. When a constant is not included in the constrained model, the hypothesis that the weights and [sigma] are the same in the experimental and field data is still rejected using the field disputes involving only wages ([X.sup.2.sub.5] df = 29.4 p < .001). Very different results are obtained for the model reported in column 2. The constrained model fits the data almost as well as the unconstrained model when the constant term is allowed to vary between the experimental and field data. The [X.sup.2] statistic comparing the constrained and unconstrained model is 3.54 and falls far short of conventional significance levels 9 [RHO] = .472).
Table 8 shows how well the constrained model with a design dummy variable predicts decisions in the field data. The top portion of Table 8 compares the actual final offer decisions in the wage only field data with the predicted decisions using the estimates from only the field data (estimates in column 1 of Table 7). The probit model correctly predicts 67.1% of the decisions (57 out of 85). The lower portion of Table 8 shows the same calculations using the estimates that constrain all of the coefficients except the constant to be the same in the two data sets (estimates in column 2 of Table 7). A slight decline in the percentage of correct predicted decisions is expected from the constrained model because it cannot fit the data any better than the unconstrained predictions shown in the top half of the table. As Table 8 shows, the percentage of correct final offer decisions predicted from the constrained model declined very slightly to 63.5%. These results imply that the weights arbitrators placed on the different facts and the level of uncertainty about which offer was to be chosen given the facts were the same in the experimental data and the wage only field decisions. [TABULAR DATA OMITTED]
The negative and statistically significant coefficient on the experimental dummy variable means arbitrators considered appropriate a lower fair wage for police disputes than for comparable teacher disputes. This finding is broadly consistent with the differences between arbitrations for police and firefighters and arbitrations for teachers in the state of Wisconsin. For example, in 1986-87 the average arbitration award for law enforcement and firefighter personnel was 5%, and that for teachers was 6.2% (WERC, undated). The lower average arbitration award for police and firefighters and the comparable union win rates across the two sectors suggest the fair wage beliefs of Wisconsin arbitrators were higher for teachers. The experimental dummy thus appears to be measuring real differences in arbitrator beliefs, not simply differences between the designs.
Summary and Conclusions
We have compared the actual decisions made by 19 practicing arbitrators over an 8.5-year period with the decisions they made in 32 simulated "paper and pencil" disputes that were decided in a couple of hours. The experimental study offers greater internal validity than is possible using field data, but the differences between the conditions in that study and in the real-life arbitrations are so substantial that it is reasonable to wonder whether the decision processes employed by arbitrators in the two environments are the same. In true interest arbitration cases, the parties present oral arguments and written exhibits over a half-day or more that are designed to convince the arbitrator of the fairness of their offer. The arbitrator also knows that his acceptability to the parties in the future depends on the quality of his or her decisions.
Despite the substantial differences between the two sets of conditions, our results show a high degree of consistency in the decisions made across the two environments when the matter for decision was the same in both. The evidence strongly suggests that the weights arbitrators attached to the various facts, and the level of uncertainty they had about which offer to choose, were not the same in the single-issue wage decision made in the experiment as in the multi-issue field decisions under final offer by package, but were substantially the same when the experimental data are compared to the sample of field cases in which the wage was the only issue.
Our findings suggest that arbitrators make decisions in multi-issue disputes under a final offer package scheme using a model that is different from the single-issue wage model used in the experimental setting. Faced with a more complex decision involving multiple issues, arbitrators look to an expanded set of factors in reaching a whole package decision. We believe that the experimental data look different from the data from the entire sample of field cases because the decision process was different in the multi-issue field cases and not because of inherent differences between the field and experimental environments. Clearly, great care must be taken when comparing "identical" field and experimental decisions. The observed differences in the taste and color of apples reported in an experiment and the taste and color of oranges reported by subjects in the field may have more to do with differences between the fruit than with differences between experimental and field environments!
(1) These 22 arbitrators decided 82% of all public sector interest arbitration cases between 1979 and 1985. See Dell'Omo (1987, 1989) for additional detail about the design of this experiment.
(2) This design feature complicates the comparison between the two data sets because the uncertainty around the arbitrators' wage settlement cannot be identified.
(3) The awards during this period totaled 395 (Babcock 1987). Only 289 awards were available for analysis in this study because some of the cases did not involve wage disputes and in some case it was not possible wage disputes and in some cases it was not possible to construct the final offers of the parties from the written awards.
(4) The statute covering nonuniformed, local government and police and fire employees requires that the arbitrator give weight to the following factors when making a decision: (1) legal authority of the employer, (2) stipulation by the parties, (3) welfare of the public, (4) comparisons with other public and private sector employees, (5) changes in the cost of living, (6) current wage levels and benefits, (7) changes in the preceding factors that occur during the arbitration process, and (8) other factors traditionally considered by collective bargaining. See 111.70(4)(cm) and 111.77(cm)(b), Wisconsin Statute.
(5) A survey of experience under the Mediation-Arbitration Law for the contracts in effect in 1983 showed that the average time from an initial petition to invoke the procedure to an arbitration award was greater than 320 days (Wisconsin Legislative Council Staff 1985).
(6) The likelihood function was [PI] (1 - [PHI]((MOFFER - X[beta])/[sigma])) [PI] [PHI]((MOFFER - X[beta])/sigma).
Union Wins Empoyer Wins
This model assumes the arbitration picks the offer closest to [W.sub.A] and is not biased toward one of the parties. See Ashenfelter and Bloom (1984) for a discussion of this point.
(7) The reason for this high estimate of [sigma] is discussed later in the paper. Also see Olson and Jarley (1990).
(8) One arbitrator was dropped from this analysis because his intercept term was not identified. All of his decision in the field data were for only party.
Anderson, John C. 1981. "The Impact of Arbitration: A Methodological Assessment." Industrial Relations, Vol. 20, No. 2, pp. 129-47. Ashenfelter, Orley, and David E. Bloom. 1984. "Models of Arbitration Behavior." American Economic Review, Vol. 74, No. 1, pp. 111-24.
Babcock, Linda. 1987. "The Role of Incomplete Information in Bargaining Under the Threat of Final Offer Arbitration: A Theoretical and Empirical Analysis." Unpublished paper, University of Wisconsin.
Bazerman, Max H. 1985. "Norms of Distributive Justice in Interest Arbitration." Industrial and Labor Relations Review, Vol. 38, No. 4, pp. 558-69. Bazerman, Max H., and Henry S. Farber. 1985. "Arbitrator Decision Making: When Are Final Offer Important?" Industrial and Labor Relations Review, Vol. 39, No. 1, pp. 76-89.
Bloom, David E. 1986. "Empirical Models of Arbitrator Behavior Under Conventional Arbitration." Review of Economics and Statistics, Vol. 68, No. 4, pp. 578-85.
Bloom, David E., and Christopher L. Cavanaugh 1986. "An Analysis of the Selection of Arbitrators." American Economic Review, Vol. 76, No. 3, pp. 408-22.
Dell'Omo, Gregory G. 1987. "Arbitration Application of Wage Criteria Under the Wisconsin Police and Firefighter Arbitration Act: An Empirical Analysis." Ph.D. diss., University of Wisconsin. ______. 1989. "Wage Disputes in Interest Arbitration: Arbitrators Weigh the Criteria." Arbitration Journal, Vol. 44, No. 2, pp. 4-13.
Farber, Henry S., and Max H. Bazerman, 1986. "The General Basis of Arbitrator Behavior: An Empirical Analysis of Conventional and Final-Offer Arbitration." Econometrica, Vol. 54, No. 4, pp. 15003-28.
Farber, Henry S., and Harry C. Katz. 1979. "Interest Arbitration, Outcomes and the Incentive to Bargain." Industrial and Labor Relations Review, Vol. 33, No.1 pp. 55-63.
Gordon. Michael, Neal Schmitt, and Walter G. Schneider. 1984. "Laboratory Research on Bargaining and Negotiations: An Evaluation." Industrial Relations, Vol. 23, No. 2, pp. 218-33.
Halm, Francis P. 1985. "Athletic Conference: The Question of Comparability." Wisconsin Public Employer Relations Association.
Locke, Edwin A. 1986. "Generalizing from Laboratory to Field: Ecological Validity to Abstraction of Essential Elements?" Generalizing from Laboratory to Field Settings. Lexington, Mass.: Lexington Books, D.C. Health, pp. 1-10.
Olson, Craig A. 1988. "Dispute Resolution in the Public Sector." In B. Aaron, J. Najita, and J. Stern, eds, Public Sector Bargaining. Washington, D.C. Bureau of National Affairs.
Olson, Craig A., and Paul Jarley. 1991. "Arbitrator Decisions in Wisconsin Teacher Wage Disputes." Industrial and Labor Relations Review, Vol. 44, No. 3, pp.536-47.
Wisconsin Employment Relations Commission. Undated "Digest and Report of Arbitration Awards Issued Under the Municipal Arbitration Law and the Municipal Interest Arbitration law and Section 111.70(4)(jm) During Fiscal Year 1988-89." Madison, Wis.
Wisconsin Legislative Council. 1985. "Analyses of Employer and Employee Experience Under Wisconsin's Mediation-Arbitration Law." Madison, W is., Oct. 15.
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||Studies of Grievance and Arbitration Processes|
|Author:||Olson, Craig A.; Dell'Omo, Gregory G.; Jarley, Paul|
|Date:||Jul 1, 1992|
|Previous Article:||Salary arbitration and pre-arbitration negotiation in major league baseball.|
|Next Article:||A simultaneous analysis of grievance activity and outcome decisions.|