|
RESEARCH REPORT: DETERMINATIONS OF THE
EMPLOYMENT RELATIONS AUTHORITY
|
 |
Some of the data presented above showed perceptible patterns in the relationships between some variables. However, variables can have more than a random relationship without having a direct causal relationship. A more sophisticated regression analysis allows for some tighter and more validly grounded speculation about which factors are demonstrably associated with adjudication or investigation outcomes.
It is safe to assume - and important to acknowledge - that, by a wide margin, the major determinants of adjudication or investigation outcomes are the merits of the cases decided. But of course those are largely locked in place by the time the adjudication or investigation process begins.
What is examined in this part of the paper, using regression analysis, is whether there is evidence of any associations between investigation or adjudication outcomes, in the Authority and the Tribunal respectively, and any of the variables defining case types, or the parties, or the decision process.
A regression analysis is a statistical technique that can divide the sample of decision outcomes (such as all grievant wins, or all compensation awards) first according to the variable (whether type of case or occupation of the grievant, or whatever) that is statistically most strongly associated with the outcomes.
The analysis then goes on to separate each sub-sample created by that first division into still smaller sub-samples according to the variable that is statistically next most strongly associated with the outcomes in each sub-sample. The process continues until all variables associated with the outcomes to a statistically significant degree have been recognized.*
We look first at the full samples of adjudicated decisions in the Tribunal and determinations in the Authority, before narrowing to the personal grievance outcomes as the bulk of the substantive decision making work of both institutions.
Figure One illustrates the primary relationships associated with adjudication outcomes across the full range of the 1,566 decisions issued by the Employment Tribunal in the 18 months to September 30, 2000. Employees prevailed in 885 of the 1,566 decisions (57 percent) and lost in 597 ((38 percent), while there was no clear advantage to either party in 84 cases (5 percent). The case variable most strongly associated (p=<.0001, where .05 or less indicates statistical significance) with win-lose outcomes was employer representation. In other words, the statistical package first separated the full sample of 1,566 decisions into bundles or sub-samples of different win-lose ratios on the basis of employer representation.
So, again across the full range of decisions, applicant employees were most likely to be successful (91 percent) where employers were not present or represented at hearing. Employee applicants were progressively less likely to be successful where the employer self-represented (73 percent employee success), where the employer was represented by an advocate (60 percent employee success), and where the employer was represented by a lawyer (49 percent employee success). Again, the regression analysis establishes that these differences represent more than mere chance distributions, and have (or had) some predictive value in terms of Tribunal decision outcomes.
Figure One shows, also, some of the second-tier factors associated with Tribunal decision outcomes in some of the sub-samples generated on the basis of employer representation. There were no further explanatory factors in the sub-sample of decisions in which the employer self-represented. For the sub-sample of cases where the employer was represented by a lawyer, the factor next most strongly associated with outcome (p<.0001) was the nature of, or primary issue in the case.
The success details for the various case types have not been reproduced in Figure One. They largely mirror the numbers presented in Table Two, though with more specifics and with some consolidation of case types. For two bundles or sub-samples of case types - the first including disputes, jurisdictional issues, and practice and procedure questions, and the second including some dismissal types, including misconduct and incapacity - a third predictor of likely outcomes emerges as groupings of Tribunal adjudicators.
Groupings of adjudicators as predictors of outcomes is not uncommon in statistical analyses of Tribunal decisions. What it means, in the case cited immediately above for example, is that the 28 or so Employment Tribunal adjudicators can be sorted into five distinct groups according to their different likelihoods that employees will have been successful in winning cases before them on disputes, jurisdictional issues and practice and procedure matters. The three case types - disputes, jurisdictional issues and practice and procedure matters - are unrelated, except in that the statistical package has identified that the same adjudicators have the same decision profiles across the three types of cases.
In the cases of dismissals for misconduct and incapacity (again, two different case types linked only by a common predictor), adjudicator identity is again the predictor, although the groupings are different from the previous grouping, and in fact adjudicators are sorted into only three groups representing three different decision profiles (22 percent employee success, 49 percent employee success, and 76 percent employee success).
Again, however, it is important to caution that whatever differences in values, or in the exercise of discretions, or in mere circumstances or case allocations that lead adjudicators into predictable groups, they pale in significance relative to the merits of cases heard and decided as predictors of decision outcomes. Any factors identified here are very much secondary to case merits as predictors of decision outcome patterns.
That said, of the variables tested, employer representation was the most significant predictor of outcomes for the Tribunal sample.
For the sub-sample in which the employer was represented by a lawyer, case type and, for some case types, adjudicator identity were secondary, but still statistically reliable predictors of likely outcomes, or rather of win-lose ratios over time.
For the separate sub-samples of decisions in which the employer was represented by a lay advocate, and again where the employer was neither present nor represented, adjudicator identity was in each case the factor next most strongly associated with win-lose ratios, much more strongly (p<.0001) where the employer was represented by an advocate than where the employer was not present or represented (p<.05). We will examine the significance of this factor as a predictor of decision outcomes more fully below in the discussion of factors associated with personal grievance outcomes.
A regression analysis was also run on the full sample of 624 determinations
issued by the Employment Relations Authority in the 18 months to April 2, 2002.
There is less to be said about that. The factor most strongly associated with
outcome in the model was case type, the figures again being along the lines
of those presented earlier in Table Two, but with somewhat more detail. Secondary
predictors were thrown up by the statistical model as being significant for
several case types. These were, however, mainly personal grievance types, and
these will be dealt with below.
* For purposes of this paper I have used AnswerTree, a statistical analysis technique that creates classification systems displayed in decision trees. CHAID (Chi-squared Automatic Interaction Detector) is a highly efficient statistical technique for segmentation of sample populations, and is the technique used in AnswerTree. Using as a criterion the significance of a statistical test, CHAID evaluates all of the values of a potential predictor variable. It merges the values that are judged to be statistically homogeneous (similar) with respect to the target variable and maintains all other values that are heterogeneous (dissimilar). It then selects the best predictor variable to form the first branch in the decision tree, such that each node is made of a group of homogeneous values of the selected variable. This process continues recursively until the tree is fully grown. The statistical test used depends upon the measurement level of the target variable. If the target variable is continuous, an F test is used. If the target variable is categorical, a chi-squared test is used. For this paper I have used Exhaustive CHAID, a more recent modification of CHAID developed to further refine the CHAID technique.
| Introduction | |
| The Authority and Tribunal caseloads | |
| The personal grievance profile | |
| Grievance outcomes | |
 |
Determinations and remdies: looking for explanations |
| Factors associated with grievance outcomes | |
| Grievance remedies | |
| Reduction of remedies | |
| Costs | |
| Summary conclusions |
|