Delphi Forecasting

Author: Dr. Brian Slack

---

1. Introduction

Delphi forecasting is a non-quantitative technique for forecasting. Unlike many other methods that use so-called objective predictions involving quantitative analysis, the Delphi method is based on expert opinions. It has been demonstrated that predictions obtained this way can be at least as accurate as other procedures. The essence of the procedure is to use the assessment of opinions and predictions by a number of experts over a number of rounds in carefully managed sequences.

One of the most important factors in Delphi forecasting is the selection of experts. The persons invited to participate must be knowledgeable about the issue, and represent a variety of backgrounds. The number must not be too small to make the assessment too narrowly based, nor too large to be difficult to coordinate. It is widely considered that 10 to 15 experts can provide a good base for the forecast.

2. Procedure

The procedure begins with the planner/researcher preparing a questionnaire about the issue at hand, its character, causes and future shape. These are distributed to the respondents separately who are asked to rate and respond. The results are then tabulated and the issues raised are identified.

The results are then returned to the experts in a second round. They are asked to rank or assess the factors, and justify why they made they their choices. During a third or subsequent rounds their ratings along with the group averages, and lists of comments are provided, and the experts are asked to re-evaluate the factors. The rounds would continue until an agreed level of consensus is reached. The literature suggests that by the third round a sufficient consensus is usually obtained.

The procedure may take place in many ways. The first step is usually undertaken by mail. After the initial results are obtained the subsequent round could be undertaken at a meeting of experts, assuming it would be possible to bring them together physically. Or, the subsequent rounds could be conducted again by mail. E-Mail has greatly facilitated the procedure. The basic steps are as follows:

1. Identification of the problem. Researcher identifies the problem for which some predictions are required, e.g. what is the traffic of port x likely to be in 10 years time. Researcher prepares documentation regarding past and present traffic activity. Questionnaire is formulated concerning future traffic estimates and factors that might influence such developments. A level of agreement between the responses is selected, i.e. if 80% of the experts can agree on a particular traffic prediction.
2. Selection of experts. In the case of a port scenario this might include terminal managers, shipping line representatives, land transport company representatives, intermediaries such as freight forwarders, and academics. It is important to have a balance, so that no one group is overly represented.
3. Administration of questionnaire. Experts are provided with background documentation and questionnaire. Responses are submitted to researcher within a narrow time frame.
4. Researcher summarizes responses. Actual traffic predictions are tabulated and means and standard deviations calculated for each category of cargo as in the case of a port traffic prediction exercise. Key factors suggested by experts are compiled and listed.
5. Feedback. The tabulations are returned to the experts, either by mail or in a meeting convened to discuss first round results. The advantage of a meeting is that participants can confront each other to debate areas of disagreement over actual traffic predictions or of key factors identified. The drawback is that a few individuals might exert personal influence over the discussion and thereby sway outcomes, a trend that the researcher must be alert to and seek to mitigate. Experts are invited to review their original estimates and choices of key factors in light of the results presented, and submit a new round of predictions.
6. These new predictions are tabulated and returned to the experts either by mail or immediately to the meeting, if the level of agreement does not meet the pre-determined level of acceptance. The specific areas of disagreement are highlighted, and the experts are again requested to consider their predictions in light of the panel’s overall views.
7. The process is continued until the level of agreement has reached the pre-determined value. If agreement is not possible after several rounds, the researcher must terminate the process and try to pinpoint where the disagreements occur, and utilize the results to indicate specific problems in the traffic prediction process in this case. This method could be applied in a classroom setting, with students serving as ‘experts’ for a particular case study. The traffic at the local airport or port might be an appropriate example. On the basis of careful examination of traffic trends and factors influencing business activity, the class could be consulted to come up with predictions that could then be compared with those of some alternate method such as trend extrapolation.