3.2 Future challenges
4.1.2 Data acquisition with Inspire
The experiments on the above mentioned negotiation case were conducted with the negotiation support system Inspire (Kersten and Noronha, 1999b,a) developed by the InterNeg research centre5 at John Molson School of Business of the Concordia University in Montreal, Canada.
Inspire – an abbreviation for InterNeg Support Program for Intercultural Research – is a web-based negotiation support system developed for educational and academic purposes, i.e. to facilitate negotiation teaching and study intercultural electronic negotiations.
Inspire’s support functions base on a three phase model of the negotiation process, consist- ing of the phases: pre-negotiation, conduct of negotiation, and post-settlement (Kersten and Noronha, 1999b,a). In the pre-negotiation phase users study the case instructions in order to gain understanding about the open issues and options available for their settlement. Afterwards the preferences of the users are elicited by Inspire using a hybrid conjoint analysis approach
4.1. Input 75
(Kersten and Noronha, 1999a). The users first indicate the relative importance of the issues in dividing 100 points over the issues. In a next step the users indicate the relative importance of each issue’s options by assigning the maximum of the points, assigned to the issue in the former step, to the most preferred option, 0 points to least preferred option, and values between the issue’s maximum and zero to the remaining options. The system uses this information to calculate the overall utility of several packages for holistic evaluation by the user as a controlling device. If the calculated overall utilities for these packages do not reflect the actual preferences of the user he can modify this overall utility value for the package and the system decomposes the information to partial utility values by ordinary least squares regression – consult Kersten and Noronha (1999a) for a detailed description of Inspire’s utility elicitation approach. The preferences indicated to the system can be revised anytime during the negotiations if preferences change or initial errors are detected.6
u(X) =
n
X
i=1
wiui(xi) (4.1)
Although the hybrid conjoint analysis does not explicitly distinguish between issue weights and partial utility functions, it is equivalent to the additive multi-attribute utility function (4.1) proposed by Keeney and Raiffa (1993). In (4.1) X = (x1, . . . , xn) is the vector of proposed
options xi, for the n issues i = 1, . . . , n to be negotiated, which constitutes an offer. wi is the
weight as a measure of importance of issue i and ui(·) is the partial utility function of issue i.
Adding the partial utility values of the options in all issues yields the overall utility of the offer u(X). While the case descriptions for both parties clearly indicate the preference direction for all issues, e.g. that the seller favors higher options for the price, no specific trade-off values – i.e. no complete scoring scheme – is provided, so that the subjects in the experiments have to establish their own priorities and trade-offs within and across issues. The partial utility functions therefore could be linear as well as non-linear (Kersten and Noronha, 1999a) and as preferences about the negotiation object were by no means enforced in the experiments there were also cases with non-monotonic partial utility functions and even few cases with monotonic partial utility functions in the wrong direction – i.e. contradicting the case description (Vetschera, 2006). In the second phase the actual negotiation is conducted. The users can exchange free-text messages and structured offers. Each offer has to consist of options for all issues negotiated, which means that package offers are enforced by the system. The multi-attribute utility function of the user is applied for negotiation support in this phase. Inspire automatically calculates and provides utility values during the construction of offers, using the preferences indicated in the pre-negotiation phase, and also evaluates offers of the opponent. The history of the exchanged messages and offers is tracked during the negotiation and can be inspected by the user anytime. Moreover the utilities of all sent and received offers is graphically represented as a function of time. If the negotiators reach an agreement – and the users mutually accept this – negotiations enter its third phase; the post-settlement phase. Inspire then calculates Pareto-improvements to the current tentative agreement i.e. packages that offer to at least one negotiator a higher
6Therefore it is necessary to determine which preferences of the users to employ as input to the simulation. We
decided to use the latest preferences indicated by the subjects in the experiments as this preference information probably is free of errors and therefore representing the actual preferences of the user best.
utility without lowering the utility of the other negotiator. If such dominating packages exist they are presented to the negotiators and negotiations may be continued.
All the data generated during the course of a negotiation experiment with Inspire is saved in the Inspire database. The results of the utility elicitation procedure are saved in the table OPTRATE, with one row per user and utility elicitation, in which the partial utility values for the 15 options are stored. A further table OFFERS tracks all the offers exchanged during the negotiation experiment, and two additional tables save the answers to a pre-negotiation questionnaire and a post-negotiation questionnaire. While the pre-negotiation questionnaire deals with questions about the subject’s demographic data, negotiation experience, and expectation and reservation levels for the upcoming negotiation after the case description was read and before the negotiation experiment starts, the post-negotiation questionnaire asks the subjects for their satisfaction with the negotiation process and outcome, their perception of their own and their opponent’s negotiation behavior, and their attitudes toward the negotiation support system used. These last two tables are due to the focus of this dissertation not considered in this study.
Though the main purposes of the experiments on the Itex-Cypress case are negotiation teaching, as well as testing and further-development of the the negotiation support system Inspire, the data that form a by-product of this endeavor was analyzed in a multitude of studies not only in research on negotiation support systems but also in studies on negotiation in general and electronic negotiation in particular.7 However up to now – at least to the knowledge of the
author – this data was not used as input for simulations of automated negotiations.
From October 1996 to September 2004 a total of 2,990 negotiation experiments on the Itex- Cypress case have been set up in Inspire. The majority of the subjects in these experiments were students, participating in fulfilling course requirements of their studies. We selected those experiments in which the subjects remained constant during the whole period of the negotia- tion and where each party sent at least one offer so that negotiations are actually conducted. This reduces the sample of experiments to 2,065 negotiations. The experiment IDs of these negotiations are saved in the vector experiments to access them during the simulation. For these relevant experiments the negotiators’ (latest) preferences and the negotiation processes were queried from the Inspire database and saved to the tables PREFERENCES and PROCESS. We furthermore created an additional table OUTCOME where we stored if an agreement was reached in the experiment and in case an agreement was reached we additionally stored the utilities to the parties and whether the agreement is a Pareto-optimal solution to the negotiation problem or not. Furthermore we saved information on the integrativeness of the negotiation problem in this table as it is the only one that structures the data by the negotiation dyad (see Appendix A).