This article describes research into software that supports Family Law mediation. Most divorcing couples enter into mediation to resolve the decisions in who is allocated items from the common pool of assets. AssetDivider supports this task by asking parties to assign ratings to the items in question. The software takes this information and from it develops a list of allocations to each party. This list is developed with knowledge of an ideal “percentage split” that has been set by mediators. The system has been tested informally by our contacts at RAQ, and we now look forward to extensive testing and evaluation by mediators at RAQ in the near future.
The focus of this research is in extending our work in interest-based negotiation to
developing research into systems for use in mediations. We have developed several
Negotiation Support Systems (NSS) including DEUS, Split_Up and Family_Winner
[
Negotiation is a process by which two or more parties conduct communication or
conferences with the view of resolving differences between them [
AssetDivider’s predecessor is Family_Winner [
Family_Winner takes a common pool of items and distributes them between two
parties based on the value of associated ratings. Each item is listed with two ratings
(a rating is posted by each party), which signify the item’s importance to the party. A
rating in Family_Winner is a number in value from 0- 100 (0 being of no importance;
100 to signify absolute importance). The algorithm to determine which items are
allocated to whom works on the premise that each parties’ ratings sum to 100;
thereby forcing parties to set priorities. The program always checks this is the case,
and if not, it realigns ratings to ensure all sum to 100. The basic premise of the
system is that it allocates items based on whoever values them more. Once an item
has been allocated to a party, the ratings of the remaining items are modified
(according to the actions of trade-offs) to ensure the items (and their associated
ratings) are ready for the next round of allocation [
Family_Winner was evaluated by a number of family solicitors at Victoria Legal
Aid (VLA). Whilst the solicitors were very impressed with the way Family_Winner
suggested trade-offs and compromises, they had one major concern – that in focusing
upon negotiation, the system had ignored the issues of justice [
Section 2 will detail this new theory of negotiation support, while Section 3 will discuss the presentation of a family law case to AssetDivider. We are in the process of organising the evaluation of AssetDivider at RAQ, and expect this to occur in the near future.
Early decision-support negotiation systems primarily used Artificial Intelligence
techniques to model negotiation. LDS [
NEGOPLAN [
Negotiation Support Systems (NSS) were primarily responsible for tracking past
preferences and informing disputants about progress being made towards a solution
to a conflict. We refer to these systems as template systems. Template systems
assume disputants take on a passive role after the initial intake of preferences and
issues, since they fail to implement any strategies that incorporate change. Modelling
the dynamic properties of negotiation infers the incorporation of decision support into
a traditional negotiation support system. DEUS [
We are mostly interested in extending the primary role of a template based NSS to a system capable of providing decision support. We have classified these as Negotiation Decision Support Systems (NDSS). A Negotiation Decision Support System (NDSS) supports negotiation by modelling the properties of a template NSS as well as applying functions to interpret the goals, wants and needs of the parties to provide advice on how disputes can be settled.
Our earliest NDSS was Family_Negotiator [
AdjustWinner [
Family_Winner is a negotiation decision support system that allocates items to one of two parties in the dispute. Family_Winner’s method of decision support involves a complex number of techniques, including the incorporation of an Issue Decomposition Hierarchy, a Compensation and Trade-off strategy, and an Allocation strategy. The trade-offs pertaining to a disputant are graphically displayed through a 1 Open textured legal predicates contain questions that cannot be structured in the form of production rules or logical propositions and which require some legal knowledge on the part of the user in order to answer series of trade-off maps, while an Issue Decomposition Hierarchy enables disputants to decompose issues to any required level of specification.
Mediator, Persuader, NEGOPLAN and Family_Negotiator are considered to be intelligent systems since they can generate solutions using the system’s internal knowledge as well as users input. All incorporate some level of negotiation support, together with the ability to provide users with a resolution to the current problem. Artificial Intelligence techniques such as case-based, rule-based and hybrid reasoning have had mixed degrees of success in providing negotiation support. The Mediator proved quite successful in its retrieval and adaptation of previous cases. NEGOPLAN used rule-based reasoning to successfully model Canadian industrial disputes, while PERSUADER successfully modeled US industrial disputes through the use of a hybrid case and rule-based methodology. Family_Negotiator however, did not perform to its initial expectations, primarily due to its relatively simple modeling of the domain.
Apart from AdjustWinner, most of the systems surveyed above do not make
allowances for measuring the fairness or justness of the settlement. Further, most of
the systems discussed are rarely based on theories derived from practice or empirical
studies. For example, INSPIRE [
AssetDivider is our latest development in negotiation support systems. It extends on Family_Winner by modifying its’ decision making theory to provide advice based on interests and the monetary value of items. Family_Winner provides advice based only on interests (known in the system as ratings). The rest of the paper will discuss the architecture and theory behind Asset Divider and in Section 3 we will illustrate how AssetDivider operates though an example.
This section will discuss the theory used to develop AssetDivider. The main
principles behind AssetDivider were derived from theories developed and
implemented in Family_Winner. [
Family_Winner takes a list of issues (items for distribution between two parties) and allocates them based on ratings given by the parties in dispute. Two sets of ratings are provide, one for each party in dispute. This rating (a numerical value between 0 and 100) does not represent the monetary value of the item, instead it symbolises how important the item is to the party. We assume a party wants to keep an item they feel is important to them.
AssetDivider accepts a list of items together with ratings (two per item) to indicate
the item’s importance to a party. In addition it also accepts the current monetary
value of each item in dispute. We assume this dollar value has been negotiated (if
necessary) before AssetDivider is used3. Hence, only one dollar value is entered per
item. The proposed percentage split is also entered; this reflects what percentage of
the common pool each party is likely to receive in the settlement. The system is not
capable of determining the percentage split; this figure has to be derived from the
mediator’s knowledge in past cases or from computer systems such as SplitUp [
AssetDivider’s output consists of a list of items allocated to each party. All of the items (except one) on the allocation lists were provided in the intake screen by the disputants. The additional item is a “payout” item, which reflects the amount of money a disputant would need to pay the other party for the items they have been allocated and collectively are valued greater than the percentage split offers them. For example, party A have been allocated a total value of $100,000 in assets, and party B $115, 000. Under a 50/50 % split, party B will need to pay $15,000 to party A to satisfy the percentage split.
The order by which issues are allocated is of paramount importance in a negotiation. Professional mediators have indicated issues attracting little disputation should be presented foremost for allocation, so as to help foster a positive environment in which to negotiate. By summing the ratings of issues to 100, the level of discourse surrounding an issue can be measured by calculating the numerical difference between the ratings of an issue assigned by each of the parties. For example, if two parties assign the same high rating to an item, then it is expected the level of disputation surrounding the issue to be substantial (because both parties want the item), whereas large differences between the ratings of parties indicate the issue will be resolved much more quickly. AssetDivider uses this strategy in deciding the order by which items are presented for allocation.
Asset_Divider allocates items to parties according to whoever values them the
most. Once an item has been allocated to a party, the remaining ratings (of items still
in dispute) are modified by trade-off equations. These modifications attempt to
mimic the effect losing or gaining an item on the rest of the items still in dispute.
The equations directly modify ratings by comparing each against that of the item
recently lost or won (each party’s set of ratings are modified as a result of an
allocation). The equations update ratings based on a number of variables - whether
the item allocated was lost or gained, the value of the allocated item in relation to
items still in dispute and the value of the item whose rating will change as a result.
The allocation strategy described above is similar to that implemented in
Family_Winner. It describes the extent to which ratings were modified as
3 Sometimes the parties cannot agree on the monetary value of the item. In this case,
mediators would reference standard objective tables and the like to reach a consensus. For
example, if parties are arguing over the value of a car, then mediators may access websites that
gave independent valuations, such as redbook.com.au.
determined through an analysis of data we collected from mediation cases provided
by the Australian Institute of Family Studies. These are detailed in [
AssetDivider’s allocation strategy works by provisionally allocating an item to the party whose rating is the highest. It then checks the dollar value of items it has allocated previously (that is, their current list of items), the dollar value of the item presently allocated and the dollar amount permitted under the percentage split given by mediators. If by allocating the item in question the party exceeds its permitted amount, the item is removed from its allocation list and placed back into negotiation. In this case, the item has not been allocated to a party. If the dollar value of the item was within the limits of the amount permitted under the percentage split rule, then the allocation proceeds. Once an allocation has occurred the ‘losing party’ is compensated by the trade-off equations modifying ratings.
The equations used to modify ratings depend on a number of variables. One of these is the rating of the issue allocated. The following table (Table 1) lists the ratings and corresponding the equations that apply. The following pseudo code gives the reader an understanding of the equations fired and under what conditions. Where RR = Rating(issue in dispute) – Rating(issue lost). if party has lost the issue
If issue's rating was <= 10 then /* graphlose0 */
If RR between -10 and 0 then %change is 0.5* RR + 5 if RR is between 0 and 10, then %change = 5 If RR is between 11 and 25 then %change = -2/15*RR + 6
If RR is between 26 and 100 then %change = -5/75*RR + 7 Endif if issue's rating was between 11 to 20 then /* graphlose1 */
If RR is –20 to 0 then %change = 5 If RR is between 0 and 89, then %change = -5/89RR + 5
Endif if issue's rating was between 21 and 35 then /* graphlose2*/ if RR is between –40 and –10, then %change is -5/30 *RR + 3 if RR is between –10 and 0 then %change is 5/10RR + 10 If RR is between 0 and 15 then % change = -5/15RR + 10 If RR is between 15 and 44 then %change = -5/29RR + 8 Endif if issue's rating was between 36 and 55 then /*graphlose3*/ If RR is –55 and –25, then %change = 15% If RR is between –25 and –20 then %change = -RR -8 If RR is between –20 and 0 then %change = 5/20RR + 15
If RR is between 0 and 70, then %change = -15/70 + 15
if issue's rating was above 55, then /*graphlose4 */ If RR is between –100 and 0 then %change is 15%.
Endif endif /*if item was lost*/ elseif /*item was won*/ No change EndIf
We have tried to focus on good usability when designing the user interface of the software. Since the software will be used by non-technical and persons not directly involved in the project, it is important the screens are self explanatory, model actual decision making and are helpful. For instance, there is space on screen for users (we presume will be mediators) to enter additional information about the case. We have also added reporting services, which in one case, will print case details such as case identifiers (case number), initial ratings given by users, ratings upon allocation and a final summary of the solutions arrived at by the system. This summary will include, for each solution, the allocation list for each party and the monetary value of each ‘allocation list’.
The system has been designed so users can print a number of percentage split scenarios very easily. Once the information pertaining to a case has been entered, the user can press the back button on the screen to arrive at the screen where the user can change the percentage split, and then press the ‘allocate’ button on the next screen to see the results. A mediator from RAQ commented they would this a useful feature as it would allow clients to view allocation lists based on different percentage split scenarios.
This section will review the process and outcome of a Family Law case on AssetDivider. The aim of this exercise is to demonstrate AssetDivider’s operation in practice.
The case description of this real-life divorce scenario and the relative point
allocations have been extracted from [
There are both real estate and liquid assets to be divided. The Husband’s diamond business is not treated as marital property as its growth was primarily due to market forces, especially the diamond boom of the 1970’s. The children’s welfare is not included as an issue as they are no longer considered minors at the time of separation.
The next screen (screen 2) that appears lists the issues in dispute, their ratings and the allocation summary, which is filled in appropriately when the user clicks button “Calculate allocations”. In the Allocation Summary table, we can see that the ratings for Husband (party A) and Wife (party B) are scaled to add to 100 in columns ComputedValuePartyA and ComputedValuePartyB respectively. It is then these ratings that are used to drive the allocation.
Screen 2: Final screen of AssetDivider. It gives the user the allocation list for each party; which includes a payout figure allocated accordingly.
According to AssetDivider, the preferred outcome, taking into account each party’s’ priorities (ratings) and percentage split indicates as follows: In analysing the case, we can see that both parties wanted the Paris Apartment above all else; though Wife (party B) valued it more than the husband (Party A). As a consequence, both parties gave the rest of the items relatively low values. On the whole, both parties received the items they valued considerably (except for Party A’s loss of Paris Apartment to Party B – since she valued it much greater). The only item valued equally by the parties was profit-sharing plan (15). It was given to Party B. Party B also need to pay out Party A the amount of 170,217 to ensure the split is exactly 50%.
This article describes AssetDivider as a new Negotiation Decision Support System (NDSS) in family law mediation. The software is one of many developed by our lab at Victoria University, including Family_Winner. Family_Winner was developed from the theories in the author’s PhD, and AssetDivider represents an improved version, following advice from our industry partners, Relationships Australia (Queensland).
AssetDivider uses the interest (rating given to symbolise the importance of the item to the party) to temporarily assign the asset to a party. AssetDivider tests whether the asset’s dollar value exceeds their allowable amount (given by the percentage split set by the mediator).
We are currently assessing AssetDivider via the CCCF System Operational Context
Checklist [
The software can also be used to provide mediators with confidence to effectively mediate property-related issues. Most family law mediators have degrees in social work or law. Their expertise lies in mediating child-related issues such as visitation schedules, primary care and other child related issues. If AssetDivider were to be used in child related mediations, it is expected both child-related and property issues could be resolved in one set of session (with mediators); thereby reducing their reliance on lawyers and of course often exuberant associated costs.
AssetDivider has not been extensively evaluated at this point in time. It is expected mediators at RAQ will test and evaluate the system in the near future. We are expecting results from testing to indicate further improvements to the decision making module and in particularly to the user interface. Our research has revealed a lack of negotiation support systems used in family law. We hope our collaboration with RAQ will enable AssetDivider to be used in their organisation, being the first negotiation support systems to do so.