Police Compulsory Arbitration: A Review of the Research By Brian R. Johnson and Greg L. Warchol

Brian R. Johnson is currently an instructor with the School of Criminal Justice at Grand Valley State University, Grand Rapids, MI. Johnson holds an MS in Criminal Justice and an MLIR from Michigan State University. Currently, he is completing his Ph.D. in Criminal Justice at Michigan State University, concentrating in police-labor relations. Johnson is a former police officer who worked for the Combined Locks, Wisconsin Police Department. 

Greg L. Warchol is an Assistant Professor in the Department of Criminal Justice at Bemidji State University in Bemidji, Minnesota. Prior to his academic career, he was on the staff of the U.S. Attorney's Office in Chicago, Illinois where he was supervisor of the Asset Forfeiture Support Unit.


During the last twenty-five years, researchers and authors interested in police labor relations have provided policy makers and administrators a greater understanding of police compulsory arbitration. Although this research provided a sound foundation for police compulsory arbitration, much of it is now dated; it has not kept pace with methodological developments in labor relations.

This creates a disturbing situation for the labor relations professional. How effective are contemporary police compulsion arbitration policies if they are based on dated, limited and statistically questionable research methods?

In an attempt to correct this deficiency in labor arbitration research, we will review and classify the methods used in police compulsory arbitration research. Through this review of the strengths and weaknesses of each method, the authors hope to expand and enhance the existing theoretical foundation for police compulsory arbitration, and thus assist the labor relations professional in developing sound policies.


To gain a complete understanding of the different methods and classifications, we performed content analyses on all documents related to police compulsory arbitration (including published articles, research notes and texts). Content analysis, according to Hagan (1982),* is the systematic analysis and classification that is used for comparative and historical studies. Likewise, Fitzgerald and Cox (1987) indicate that it is a research method that "systematizes the use of documents by providing a predetermined coding scheme and categories for tabulating the contents of the documents" (p. 110).

In the context of this research, we selected three primary categories and their subgroups as the classification schema. They are as follows: 

  • exploratory, 
  • descriptive, and
  • multivariate.

* The Bibliography at the end of this article contains the full citation for each study referred to in the text.

Although these classifications are not mutually exclusive and exhaustive (many researchers combined classifications), we classified the data according to the most powerful or advanced method used.

Introduction to Exploratory/Descriptive Research

One of the most basic methods used by researchers is to explore and describe an existing condition, phenomenon or event. Much of the compulsory arbitration literature contains both exploratory and descriptive research techniques or methodologies. There are, however, differences between these research methodologies. Babbie and Maxfield (1994) write that "descriptive studies are often concerned with counting or documenting observations while exploratory studies focus more on developing a preliminary understanding about a new or unusual problem" (p.71). Although these categories or approaches are not mutually exclusive and exhaustive, it is possible to differentiate between them in the existing literature.


Exploratory research explores the nature or frequency of a problem; it is used when a practitioner knows little about a subject or policy, or when there is a policy change (Babbie and Maxfield, 1994). In the context of this article, exploratory research is best illustrated by the research that explored the underlying influences and dynamics that led to the passage of compulsory arbitration statutes. As exploratory research is the beginning of social inquiry, it may also be conducted to develop robust research methods for future research and statistical analysis (Babbie and Maxfield, 1994). Most of the existing exploratory research explains the arbitration processes in various states in the United States, and will be classified as Site-Specific Exploratory research.

DiTolla (1993) provided a historical analysis of compulsory interest arbitration in the state of New Jersey by examining the hearings, commission reports, judicial reviews and public hearings that led to the eventual passage of the state's compulsory arbitration law. Similar to DiTolla, Gilbert (1987) examined factors that influenced mediation, fact-finding and interest arbitration in Ohio. Also, Chvala and Fox (1979) investigated the use of mediation-arbitration in Wisconsin by providing a historical review of the legislation, negotiation and mediation procedures used, and the scope of bargaining under final offer arbitration. Other authors, such as Kochan (1978), examined the history of New York's Taylor Law in the context of political influences that led to the passage of compulsory arbitration; and Anderson and Krause (1987) discussed New York's system as an alternative to the strike. Howlett (1984) examined the processes and procedures, standards and history of interest arbitration in Michigan. Other research examined specific municipalities, as illustrated by Fyfe's (1985) analysis of six municipalities in California that had compulsory arbitration clauses in their city charters.

As compulsory arbitration became a more common phenomena, researchers developed new Multi-State Exploratory classifications. One of the earliest multi-state studies was conducted by Wortman and Overton (1973), who explored and compared all state arbitration statutes that existed in 1970. Bent and Reeves (1978) also researched which states had arbitration, what their statutes covered, and the type and scope of arbitration. Then in 1981, Kruger and Jones reviewed all existing arbitration legislation in the United States, as well as constitutional issues, and provided recommendations for improving the existing dispute resolution techniques.

Other research can be classified as Historical-Exploratory, which emphasizes the social, political and legislative processes that lead to the creation of compulsory arbitration statutes. Nolan and Abrams (1983a, 1983b) conducted one of the most comprehensive reviews of the evolution of arbitration. They provided an extensive historical analysis of the development States by examining the factors that eventually led to both private and public sector arbitration. Similarly, LaRue provided an historical review of interest as it pertained to legislation, Governmental intervention in the private and public sectors, and controversies in public sector labor arbitration.

Other authors examined and classified Legal/Constitutional issues pertaining to police labor arbitration. The William and Mary Law Review (1977) provided an extensive multi-state review of constitutional issues. Hyman (1983) and Staudohar (1976) provided a review of compulsory arbitration legislation and court decisions in various states. Kanowitz (1987) also provided a legal/constitutional analysis of interest arbitration; he examined the implementation of arbitration legislation to see if it served the public interest. Other research (see Craver 1980) was more specific, concentrating on post-arbitration judicial enforcement and review of awards by courts within various states that have arbitration statutes.

Process classifications pertain to various arbitration mechanisms and operations. Some early examples include Olmos (1974), Feuille (1979) and Murray (1982), who discussed general issues involving the benefits and controversies in arbitration. More recently, McGinnis (1989) discussed types of arbitration, and Samavati, Haber and Dilts, (1991) studied the uses and limitations of comparing economic issues. Also, DiLauro (1991) provided an overview of types of arbitration, arguments related to its chilling effect, and constitutional issues; while LaVan (1990) examined arbitration in the context of police and teachers, including a description of the effects of various state statutes on the arbitration process. Years before LaVan's study, Grodin (1976), citing case law related to arbitration, provided an analysis of the political aspects of interest arbitration as they relate to salary and non-salary issues, and of the use of tribunal panels in arbitration hearings. Gallagher (1982) also presented the strengths and weaknesses of compulsory arbitration, including its effectiveness and alternatives to it. 

Some organizations have researched and written about the strengths and weaknesses of compulsory arbitration. The Police Officers Association of Michigan (POAM, 1980) wrote an evaluation of, and made recommendations regarding, arbitration in Michigan. The Public Service Research Council (1978) analyzed the impact of different types of impasse procedures, strikes, and attitudes toward compulsory arbitration (1980). In addition, the Contract Research Corporation (1975) evaluated the existing literature and the effectiveness of different arbitration methods on police labor relations.


Accompanying and closely related to exploratory analyses are other research efforts that have quantified and investigated police compulsory arbitration through a variety of basic statistical analysis techniques. This approach, and its subsequent methodologies, can best be generalized as Descriptiveclassifications. As Simon (1969) said, "in the beginning, there is description" (p.52). Like exploratory research, when someone does not know anything about a problem or variable, it must be understood in a general context. But unlike exploratory research, when researchers attempt to gain an understanding through descriptive analyses they begin by "standardizing the data and separating it into convenient or interesting categories" (Simon, 1969, p.52).

According to Tabachnick (1989), descriptive research "describes samples of subjects in terms of variables or combinations of variables" (p.9). Kachigan (1986) stated that descriptive analysis is concerned with exhaustively measuring the characteristics of a population or collection of objects. Simon (1969) indicated that descriptive research is often in the form of case studies, and is the starting point for research in new areas. Research of this nature generally applies fundamental descriptive or summary statistics that include frequency analyses and measures of central tendency. This type of research usually does not create laws or draw conclusions; instead, it provides suggestions for subsequent research.

Regardless of the variables under analysis, much of the existing descriptive research is inferential in nature. Inferential statistical analysis "is concerned with measuring the characteristics of only a sample from the population; and then making inferences, or estimates, about the corresponding value of the characteristics in the population" (Tabachnick, 1989, p.9). Rowntree (1981) stated that inferential statistics are used to generalize from a sample or to make inferences about a wider population. Thus, inferential analysis requires the sampling of subsets of populations. This sample is then used to draw conclusions about the population.

Much of the descriptive research has analyzed arbitration in specific locations over a fixed period of time. These research efforts may be classified as Longitudinal. Examples include the following research done in New Jersey: 

  • Tener (1982) analyzed the first four years of New Jersey's Police and Fire Arbitration Act, and reported the number of petitions filed, arbitrators assigned, and number of awards issued per year; 
  • Weitzman and Stochaj' (1982) reviewed the impact of arbitration in New Jersey from 1977 to 1979; and
  • Liebeskind (1987) examined the use of compulsory arbitration in New Jersey by examining the economic and non-economic factors of bargaining outcomes. 

In the State of Michigan, Stem researched arbitration from 1970-1974 to develop summary statistics on the parties involved with Act 312 arbitration. Also, in Florida, Magnusen and Renovitch (1992) examined public sector impasse procedures from 1980 to 1987.

A large number of descriptive research efforts were conducted in Ohio. Graham (1988) analyzed the following factors pertaining to the frequency of arbitration from 1986 to 1988: 

  • region of state, 
  • public safety organizations, 
  • unions involved, 
  • issues involved, and
  • results of interest arbitration
  • (e.g., win/loss; wage increases). 

In a 1987 Ohio study conducted by Graham (1987), he examined 60 agencies in Cuyohoga County to measure how often the following matters were raised: 

  • mediation, 
  • fact finding, 
  • arbitration, and
  • particular contract issues. 

Then in a 1993 study, Graham used a frequency analysis to examine the narcotic effect in Ohio.

Other states that have been examined include Iowa, where Gallagher and Pegnetter (1979) studied the arbitration process to discover the frequency that procedural steps were used in the public sector, for the periods from 1975-1976 and from 1976-1977. In Massachusetts, Somers (1977) examined final-offer arbitration where information on the petitioning parties included the following: 

  • an analysis of median percentage increases from 1973-1976
  • petitions for mediation and rates of change from 1970-1976, and
  • the rate of police and fire fighter negotiations from 1970-1976. 

Greer and Sink (1982) provided a review of Oklahoma's interest arbitration legislation from 1972 to 1981 by examining the stages in the process that settlements occur. They also conducted interviews with union and city officials in seven cities, asking them to assess their attitudes toward the existing procedures. In Michigan, Austermiller and Fremont (1985) studied factors related to arbitration from 1971 to 1984; Berrodin and Kurbal (1979) analyzed them from 1973 to 1979; and Kruger (1985) analyzed them from 1976 to 1983, and did a case analysis and history of arbitration in Detroit.

Longitudinal classifications also took on a national perspective. Petro (1992) researched the extent and distribution of public sector unionization and collective bargaining legislation throughout the United States, from its origins to 1982. Hirlinger and Sylvia (1988) investigated the frequency of work stoppages and the effectiveness of impasse procedures in all 50 states, from 1979 to 1980, to determine which procedures were most effective in avoiding work stoppages. Delaney and Feuille (1984), in their study of over 300 municipalities, reported on issues, changes in salary, and numbers of awards per state.

Other researchers investigated specific municipalities. Pursell and Torrence (1983) examined the impact of arbitration in Omaha, Nebraska, to measure increases or decreases in municipal expenditures within various departments of the City, including public safety, from 1968 to 1979. Anderson (1982), to measure the success of the New York City Bargaining Law, conducted a 10-year review of the use, history, and procedures of the law and of interest arbitration in the City.

Attitudinal-Descriptive studies and classifications that rely upon descriptive statistics have also been conducted. Portaro (1986), who provided a review of compulsory arbitration statutes in the United States, conducted a survey of respondents in Ohio to find out how they would modify that state's compulsory arbitration law. Also, Herrick (1983) conducted an attitudinal survey involving arbitrators registered with the FMCS to measure their attitudes on 24 lickert-scaled questions related to the arbitration process. In addition, Helbsy et al. (1988) conducted 28 case interviews to provide an analysis of the attitudes of union and management representatives in Florida toward fact-finding. Prior to these studies, Kressel (1972) researched labor mediation by interviewing 13 labor mediators on processes, strategies, and whether mediation should be considered a profession. 

Comparative classifications were used by other researchers to explore differences in compulsory arbitration between two or more groups. Comparative classifications are developed using statistical methods and procedures such as t-tests, analysis of variance, and other probability-based measures. 

The t-test is one of the best known and widely used methods to determine differences between two population samples. Levin and Rubin (1991) write that the student "t" distribution is a family of probability distributions, used to measure the differences between two independent sample means to determine if there are statistically significant differences between the two groups. To determine differences between groups, researchers construct an interval of values (which may be estimates) that can state with a certain degree of confidence that the characteristics of the population fall (or are likely to fall) within that interval (Levin & Rubin, 1991; Kachigan, 1988).

Although the t-test is an effective measurement device for small samples (of less than 30), and its accuracy can be maintained even if (within limits) some of the assumptions of the t-statistic are violated (see Kolstoe, 1969; and Cohen, 1977 for other conditions), it has not been widely used in compulsory arbitration research.

There are, however, some examples of its use. Rueschoff (1988) tested public workers, public managers, the general public and legislators to discover their most frequent responses to statements about the following: 

  • participation in decision-making; 
  • bargaining for wages, hours and conditions of employment; 
  • whether public work was perceived to be essential; and
  • whether employees should be allowed to strike. 

Wheeler (1978) also performed t-tests to determine if there were differences between the management and union officials. He examined their positions before they reached an impasse and at the point of impasse. (In compulsory arbitration, the point of impasse is the gap between the union demand and the management offer.) In addition, Chelius and Extejt (1983) used t-tests to discover if attitudes could be generalized between groups of subjects in various studies.

Making generalizations from the existing research on arbitration may be limited because the structure and methods of arbitration vary from state to statewhich raises further issues related to the external validity of the research findings. Anderson (1981) addressed this problem when he found it was difficult to generalize because jurisdictions didn't have identical arbitration schemes.

Another possible drawback to the existing research is the difficulty in making valid inferences from the various data because the analysis procedures are "far too simplistic ... If we are to... develop adequate tests of our theories, we shall need to improve the quality of our data analyses" (Namboodiri, 1975, p.2).

Most of the authors used only categorical variables, whereas contingency tables or cross tabulations were the most powerful research methods used. Although they reported degrees of association between variables, the frequency, magnitude or strength of the relationship could not be measured. Hence, certain aspects of the data are summarized at the expense of others, and some data may mistakenly be misrepresented.

There is another drawback to this type of research: it uses methodologies and statistical techniques that examine only one variable at a timecalled a "univariate" method. But variables are often interrelated in complex ways. Thus, if a researcher simply examines one variable at a time, his research method may not be sensitive or robust enough to detect and measure the complex relationship or inter-relationships that exist between the variables (Tabachnick, 1989). As a result, these limited statistical techniques only report the following information on the variables examined: 

  • frequency distributions, 
  • measures of central tendency, 
  • measures of dispersion, and
  • range and standard deviations.

Another problem with the use of descriptive research is that no conclusions about causality can be safely made (Kachigan, 1986). Researchers using descriptive research techniques and statistics generally cannot manipulate the levels of a variable to obtain changes in the other variables (or variables of analysis) because the arbitration process lacks the interaction of multiple variables.


As a result of such limitations, and to find explanations for various dynamics in the compulsory arbitration process, other research efforts have used multivariate statistical techniques. Unlike univariate analyses (explained above), multivariate analyses are used to investigate the relationships and interactions of two or more variables over a set of objects (Kachigan, 1988). In measuring these objects, researchers construct theoretical models, and then test the strength of these models (and the variables) through advanced statistical techniques. Two techniques used in police compulsory arbitration research are regression analysis and logistic regression.

Regression Analysis

Regression analysis, or Ordinary Least Squares (OLS), is the investigation or analysis of relationships among variables. Although it is related to correlational analysis because it tests the strength of association between variables, it also reports the nature of the relationship if there is a decrease or increase in the relationship between the variables (Levin & Fox, 1988). In doing so, the relationship is expressed as an equation between a dependent variable and one or more independent variables. Hanushek and Jackson (1977) defined regression as a model that simplifies reality because it identifies key variables while making a specific prediction about the interaction of those variables. 

A simple regression equation has one dependent variable and one independent variable. A multiple regression equation has a dependent variable and multiple independent variables (Chattejee, 1991).

Like those who have done descriptive and exploratory research, researchers using multiple regression have also conducted multi-state studies. The largest research effort employing multiple regression was conducted by Feuille, Hendricks and Delaney (1983). They studied collective bargaining and interest arbitration in approximately 1,015 cities and 16 states, for the years 1971 to 1981. This study, which contains a large amount of exploratory and descriptive research, is one of the most comprehensive studies to date of the impact that interest arbitration statutes have on police collective bargaining. Using the same data set, these authors subsequently published research findings that examined the following: 

  • the salary leveling effect of arbitration (Delaney, Feuille & Hendricks, 1984); 
  • the association between interest arbitration and labor
  • contracts that are more favorable for unions (Feuille, Delaney & Hendricks, 1985); and the impact of arbitration on police salaries (Feuille & Delaney, 1986).

Other researchers using smaller samples have also used OLS to examine compulsory arbitration. Connolly (1986) did a multi-state study of the impact of arbitration on wages in the states of lllinois and Michigan. Bloom (1981) measured the effect of final offer arbitration on the salaries of police officers in the State of New Jerseyin the fiscal year 1978 to 1979. Benjamin (1978) conducted similar research into Michigan's Act 312 arbitration law. Schwochau and Feuille (1988) used regression analysis to investigate interest arbitrators and their decision-making behavior. Olson, Dell'Omo and Jarley (1992) sought to discover any differences in the decision-making abilities of arbitrators by comparing their decisions in laboratory or hypothetical settings with actual decisions they made in field settings.

Strengths and Weaknesses of Multiple Regression

One of the primary benefits of regression analysis is that it enables the effects of various factors to be evaluated from the experimental dataeven when the experiment does not follow a simple pattern, or the variables affecting the results cannot be controlled (Williams, 1959). Another strength of multiple regression is that the interaction of multiple variables can be observed at one time. Unlike univariate techniques that measure one variable at a time, multivariate techniques (e.g., multiple regression) allow the researcher to measure correlations and pooled standard errors between the variables in the model (see Kennedy, 1992). In doing so, the interactions of all of the variables can be analyzed.

There are, however, some drawbacks. One drawback to regression analysis is that the variables that comprise a model may not have consistent or accurate specifications. For example, to compare variables, the variables in the models must be linear. Unfortunately, many regression models under analysis are not linear, and may have a curvilinear relationship (defined below under Logistic Regression) or some other non-linear form. Likewise, the reliability of a regression line (or the overall strength of the model) is determined by the coefficient of multiple determination (i.e., the R2). This coefficient, however, can be quite misleading, as a high R2 does not necessarily mean a strong model. The R2 is influenced by the number of independent variables in the model, and more variables always create a higher R2 value (see Lewis-Beck, 1980).

Other drawbacks related to the accuracy or usefulness of model specifications include multi-colinearity, which occurs when the researcher uses two independent variables that are highly correlated with each other. This situation subsequently results in high correlations among variables which in turn affect the R2. As indicated by Hanushek and Jackson (1977), these problems in multiple regression cause the model to not be BLUEthe Best Linear Unbiased Estimate. Such problems can be corrected through proper model specification and a comprehensive understanding of the principles of multiple regression; or by the application of other types of multiple regression techniques, including hierarchical and stepwise multiple regression (Tabachnick & Fidell, 1989). 

Logistic Regression

One statistical tool used very little in compulsory arbitration research is the logistic regression model. Like multiple regression, the logistic regression equation relies upon theoretical models constructed by the researcher. There are, however, some fundamental differences between the two procedures.

One of the basic differences between linear and logistic regression is the parameter estimate. In linear regression, OLS is used to select the parameter estimates that minimize the errors (i.e., the sum of squared errors) to create the most suitable fit (the regression line) between the data and the model (Aldrich & Nelson, 1984). With an OLS model, it is assumed that the dependent and independent variables are linear-related, but the OLS estimates will only be accurate within the range of the data in the sample. The curvilinear (i.e., bound by curved lines, not straight lines; includes more because of broader parameters) model is not subject to the same limitations because it uses the Maximum Likelihood Estimation (MLE). The MLE determines values for unknown parameters; it provides those parameters that would most likely have created the observed data, but which cannot be found in the actual data (Hanushek & Jackson, 1977). This, according to Kennedy (1992), is "based on the idea that the sample of data at hand is more likely to have come from a 'real world' characterized by any other set of parameter values" (pp.20-21).

For the formulation of an analysis, unlike the multiple regression model that assumes a continuous scaled dependent variable, the logistic regression model uses a nominal level dependent variable that is dichotomousit has two opposing parts (Morgan & Teachman, 1988). In traditional linear regression, the dependent variable is assumed to be continuous and linear in nature. The logistic model, however, assumes that the dependent variable is not continuous but is dichotomous, or bound between the values of 0 and 1 (Aldrich & Nelson, 1984). Thus, linearity cannot be assumed. Further, Aldrich and Nelson (1984) indicated that if linearity is violated, none of the distribution properties associated with OLS hold. Because the values of the dependent variable are bound between the values of 0 and 1, a curvilinear relationship now exists between the explanatory variables and the dependent variable (Osgood & Rowe, 1994). Also, if OLS was applied in this situation, meaningless and unreliable values could result, for OLS does not constrain observed values to remain between 0 and 1. Thus, there is no guarantee that the expected values from the fitted least squares equation will always get numbers between 0 and 1 (Namboodiri et al., 1975).

Logistic regression is also necessary because of error term  
the deviation from the conditional mean. In OLS or linear regression, the most common assumption is that the error term follows a normal distribution with a constant variance (Hosmer & Lemeshow, 1989). Other assumptions also include: 

  • that the model has been properly specified, 
  • there is no measurement error, and
  • the error terms have an expected value of zeros, and are not correlated, constant or normally distributed (Lewis-Beck, 1980). 

In logistic regression, however, the error term can only take one of two values. As a result, the expected value of the error term is not independent of the values of the explanatory or independent variable, which causes the variance estimates to be biased (Hanushek and Jackson, 1977). Thus, linear regression techniques are inappropriate where there is a dichotomous dependent variable, since the model will no longer generate the best linear unbiased estimate.

There is some research in police compulsory arbitration that has used the logistic regression model. Ichniowski (1982) studied 600 municipalities, from the years 1972 to 1978, using logistic regression to determine if arbitration statutes reduced the propensity of strikes. Later, Ichniowski (1988) researched police strike activities and unionization rates for municipalities in states with and without unionization rights, to investigate which municipalities experienced recognition strikes more often. Kochan and Baderschneider (1978) also used logistic regression in their analysis of police and firefighter negotiations, from 1974 to 1976, to determine if arbitration increased the probability of impasse or affected the settlement.

Strengths and Weaknesses of Logistic Regression

As previously indicated, logistic regression is a preferred method of statistical research when the research question involves a dichotomous dependent variable. In addition, logistic regression is extremely flexible and can use simpler mathematical functions; other methods, such as Probit models, require more sophisticated mathematical equations (see Kennedy, 1992). It is also considered a more general procedure because it allows for both categorical and continuous independent variables (Tabachnick & Fidell, 1989). In addition, it is easier to interpret than other models which use dichotomous or polytomous dependent variables (Cox, 1970; SPSS, 1994).

There are some drawbacks to logistic regression. Logistic regression measures the probability of an event occurring or not occurring; or it analyzes how the dichotomous dependent variable influences the underlying probabilities (Hanushek Jackson, 1977). The probability of a dichotomous event is calculated using the odds-ratio, which determines the relationship between the dependent and independent variables, and defines the unit of change for the independent variable. This process, at times, may be difficult to calculate, especially when the researcher is using categorical data (Hosmer & Lemeshow, 1989).

Other Multivariate Methods

Although the field of police labor relations has not used them, there are other multivariate statistical procedures. For example, researchers could use factor analysis. Unlike multiple and logistic regression, factor analysis is used to develop and test theories where the researcher is interested in finding variables; that is, variables that form subsets of variables and that are independent of one another (Tabachnick & Fidell, 1989). To illustrate this concept in police compulsory arbitration, one could take a large sample of police and municipality representatives with the following information: 

  • personality characteristics, 
  • education history, 
  • experience in collective bargaining, and other variables. 

Each of the areas would be assessed by other variables, and then be entered into the factor analysis one at a time to study correlations among them. This analysis could reveal patterns of correlation and underlying factors that affect the collective bargaining and compulsory arbitration processes. This multivariate technique could be used to generate additional hypotheses and subsequent theories about the underlying process of compulsory arbitration.

Other multivariate methods that arbitration researchers may consider are Cononical Correlations, which analyze relationships between two sets of variables to determine the highest correlation between them. The sets of variables are combined on one side of the equation to produce a predicted value that has the highest correlation with the predicted value on the other side of the equation (see Tabachnick & Fidell, 1989). Other suitable multivariate techniques may include a time series analysis that looks for trends and fluctuations over a designated period of time (see Kennedy, 1992). These and other techniques may prove to be suitable for determining relationships among variables, thus enabling researchers to construct more effective compulsory arbitration models.


This classification review of police labor arbitration reveals that a great deal of research has been conducted in police compulsory arbitration, providing the labor researcher a firm foundation on which to use advanced methods in police-compulsory arbitration research. The vast majority of the existing research, however, has been exploratory and descriptive in nature. Compounded by the fact that arbitration legislation, processes and procedures vary among states, these research classifications have done very little to explain why arbitration occurs or what specific factors contribute to (or impede) the arbitration process. 

Consequently, without the application of sound statistical methods, the compulsory arbitration process may not be clearly understood, and there may be risks in relying upon these research findings to develop social policy. Because of the societal implications, it is vital to fully understand the police labor compulsory arbitration process. By examining the classifications already applied, labor practitioners can gain understanding and develop their own research agendas, basing such agendas on sound theoretical models and advanced statistical applications. 

It is imperative that labor arbitration research takes on a new paradigm and develops better constructed theoretical modelsmodels that can be used to investigate factors involved in police labor arbitration. One alternative is for researchers to use multivariate research techniques. As illustrated earlier, there are a large number of multivariate research techniques that can be applied to compulsory arbitration research. Depending upon the research agenda, available data, and theoretical model constructs, there are a host of powerful multivariate research methods for the labor relations researcher.