Practices:Delphi survey

From FORwiki

Revision as of 01:26, 29 April 2010 by Dan Grosu (Talk | contribs)
Jump to: navigation, search

Delphi method is a version of survey analysis that involves repetitive questioning of respondents. Delphi researchers aim to predict and explore alternative future developments, the probability of their occurrence and their desirability. Delphi studies are action-oriented meant to affect actions or thoughts of decision makers. Herman Kann developed the Delphi Survey Method within RAND Corporation in early 1960s.


The FOR-LEARN Guide to Expert Panels

This is a summary of the article on the Delphi Method from the FOR-LEARN guide. To read the full article go here.

Overall description

The Delphi method is based on structural surveys and makes use of information from the experience and knowledge of the participants, who are mainly experts. It therefore yields both qualitative and quantitative results and draws on exploratory, predictive even normative elements. Delphi is an expert survey in two or more 'rounds' in which, in the second and later rounds of the survey the results of the previous round are given as feedback.

Delphi studies tackle issues formulated in statements about which uncertain and incomplete knowledge exists. They involve making judgments in the face of uncertainty by large number of experts selected based on knowledge and experience. The assumption is that self-fulfilling and self-destroying prophecies will be thus developed, thus shaping or even 'creating' the future. A Delphi study usually involves experts from business, government, research associations and any other stakeholders of a specific field under debate.

When is this method appropriate?

A Delphi study is usually organized when there is a need to organize a debate, to collect and synthesize opinions and to achieve a degree of convergence. This is the case when there is not a lot of evidence about possible developments, or when long-term issues are involved. Also, common incentives for using the method are the intention to produce statistical significant results, or the will to involve a large number of people in processes.

Step by Step guide

Developing a classical Delphi study usually involves running the following steps:

  • Selection of the subject to forecast (one or more thematic fields);
  • Definition of the procedure;
  • Formulation of the statements and questions;
  • Administration of the questionnaire;
  • Analysis of responses. In the analysis of data are used descriptive statistics (median, inter-quartile range, etc.) in order to quantitatively summarize the set data and to anticipate possible developments of the characteristics / variables measured.

Resources needed

The costs depends on the number of experts, the length of questionnaire, type of technology used. A Delphi can take between three weeks and 3-4 months. The organizers need management skills, neutrality and to be open to creativity.

Pros and cons

Supporters of the Delphi study underline certain advantages in using the method:

  • It’s a credible and popular approach for policy-makers.
  • It forces people to think about long term issues.
  • It highlights clearly whether there is a consensus on an issue or not.
  • The judgment allows for analyses, rankings and priority–settings.

At the same time, there are limits to using the method. Certain statements or forecasts cannot be assessed even by accomplished experts, making the Delphi study pointless. At the same time, if the Delphi is not well designed it will produce poor quality information and might compromise the entire Foresight activity. Other criticisms are aimed at the fact that Delphi studies are time-consuming, labour intensive and require expert preparation, or that many participants might drop-out during the process.

Complementary methods

The Delphi method implies identifying statements (topics) that are relevant for the future. Therefore, creativity methods (e.g. brainstorming, 6-3-5 or others), scenarios or key technology can be used in the preparatory phase to define these statements. Data from desktop studies: literature research, patent analysis or bibliometrics can be added. In the analytical phase, different modelling or statistical methods (calculation, rankings, correlations) or the re-building of scenarios as well as pseudo-roadmaps can be used. For comments or additional explanations, qualitative analyses are necessary. A SWOT analysis can be based on the results.

Delphi Method


Delphi is a particular collaborative process that is designed to improve group communications about a complex problem or topic. In the past, it was largely done by paper and pencil communications and is now often done on the Web. The Delphi Method has the following properties:

  • The gathering of what might be a very large group of participants to consider a complex problem (tens, hundreds, or thousands).
  • The participants usually number about five people in each area of special knowledge or expertise needed to present and share information about the problem and various solutions to it.
  • A knowledge structure allows the participants to place their comments, insights, and concerns in the appropriate location so a large involved discussion is easy to follow.
  • There is also the ability to vote on contributions so the group can determine what specific things they agree or disagree on.
  • Individual participants are usually anonymous when authoring items and when voting.
  • Since the computer process or paper process keeps track of the contributions, what each individual has contributes, what they have read or seen, every participant can participate asynchronously at a time and place convenient for them.

There are three major objectives of a Delphi Process:

  • Gathering the information which is needed to deal with the problem or topic and fill in the resulting knowledge structure.
  • Making sure this information can be understood by the many different backgrounds of the contributors.
  • Exposing agreements and disagreements and trying to come up with various recommendations for actions of various types.

There is a rich History of the Delphi Method and the authors understanding of it. There is also References for much more specific information and examples of the method.

Some Definitional Aspects

The basic Delphi concept is the design of a collaborative communication structure and process that is tailored to the nature of the problem and the nature of the group (Linstone and Turoff, 1975). Although it was used largely in the early days for predicting future technical breakthroughs, it has been used to address a wide range of complex problems that are often current and it has been used subsequently to try to understand the past as well as the future. There have, for example, been a number of examples of experts in a given field using a Delphi to establish the most significant contributions to their field. Anonymity of the responses is one fundamental property so that people will feel free to express themselves and to be able to expose ideas that could turn out to be stupid as well as brilliant. However, in some current online approaches it is possible to allow the participants, if they chose, to put in a comment with their true name, or when they want to be anonymous, or when they want to use a pen name. An advantage of the pen name is that they can develop a series of comments to express a coherent viewpoint. In some cases, the respondents are told who is participating so they will feel they are part of a peer group of people they would like to communicate with about the particular topic. Usually those acting as the design team will commit to the fact that who said what would never be divulged to the other participants or to the sponsor. Delphis that are well done usually try to capture and seed the process with the material that can be found in the literature on the subject, so that those participating realize that they are not being asked to educate the design team on what should be obvious. The material to be asked of the respondents is what would be difficult to find in the literature and what is not obvious. Too many poor Delphis have attempted to give people a blank piece of paper that says "tell me what I should know about this problem!" Associated with the above is that people have to be motivated to put in the effort to participate in a Delphi exercise. The sort of motivation factors that have to be considered and made clear to the participants are:

  • Is this an important problem that should be addressed by a larger group of experts who will all have an equal opportunity to contribute?
  • Is this the right group to undertake this effort?
  • Is someone or some organization going to make real use of the results of this effort?
  • Is it worth it for me to spend the necessary time to make a good contribution?
  • Will I learn things I should learn from those in other professional areas that are participating?
  • Is it clear to me what the process is and what I will be committing to in time and effort?
  • If some of the above is not true, am I going to be paid, and what is my time is worth to participate?

The typical view of Delphi is that it has a round structure and goes through at least three phases: 1. Exploring the problem and exposing new insights and additional relevant material. 2. Gaining a collective understanding of the material generated. 3. Evaluating the material and hopefully reaching a consensus. This is usually what leads to a three round exercise for Delphi processes done via pencil and paper. Sometimes it does suffer because the design may lead to a premature consensus when there is not an adequate structure to expose hidden disagreements. Sometimes the pressure is to get just quantitative subjective estimates of variables such as costs, likelihood of success, effectiveness, etc. without a sufficient design in the structure to expose hidden or underlying disagreements. Voting is often used as a conclusion rather than for its real purpose, which is to expose potential disagreements and get rid of possible ambiguities so that true uncertainties can be dealt with. This leads to a number of other requirements that when done with paper and pencil can require five rounds for the complete process. They add the following phases after phase one above. 1.1 Initial voting on generated material to expose disagreements. 1.2 Exploration of the underlying reasons for disagreements. Underlying the above is the requirement to have a morphological structure for the information that is contributed that allows the participants to input their knowledge into appropriate categories that will organize and cluster information. Today this is referred to as a knowledge structure and it is exhibited in many of the Delphis that deal with complex problems. Many of the concepts underlying the Delphi Process have been adopted in other related methods: Prediction Markets, Recommender Systems, Collaborative Tagging or Folksonomies, and other Collaborative Systems Examples of Delphi like processes. Over the past forty years, a number of specific Delphi Structures have been designed and are very popular in terms of successful usage. This includes the conditional forecasting of trends where the emphasis is generating the conditions that affect the trend forecast. A second is a problem solving Delphi structure to come up with an evaluated list of alternatives or options. The third is the Policy Delphi which is devoted to determining the alternative and complementary policy options to a policy issue and the arguments supporting each one. The fourth is the example of Cross Impact Analysis for building individual and group models of interaction among future events and scenarios. Example Delphi Structures The specific area of cross impact analysis is a foundation for the creation of a Delphi based Planning process The foundation of Cross Impact Analysis.

Examples of Delphi like Processes

"Treason doth never prosper: what's the reason? Why if it prosper, none dare call it treason." -- Sir John Harington 1561-1612 The name of Delphi was not chosen by the inventors of the method at RAND (Olaf Helmer and Norman Dalkey) but by their fellow professionals, since it was commonly used for future predictions. The strange name affixed to the Delphi process has not been favorable for the spread of this method. What has happened as a result is that many of the premises of Delphi have been rediscovered or renamed under other methods to use group processes to try to obtain some level of collective intelligence. This is the concept that the group can reach a higher quality result than any individual in the group would have acting alone (Hiltz and Turoff, 1978). The most common Delphi Derivatives today are:

  • Prediction Markets
  • Recommender Systems
  • Collaborative Tagging or Folksonomies
  • Wikis, and Collaborative Systems for humans

The idea of using a consensus on creating index tags for various objects by anonymous agreement on words to represent the object has become quite popular and goes under the name of collaborative tagging or Folksonomies. The PhD thesis that created the field of collaborative tagging was "" which is still on the web and a wonderful experience to appreciate fully a two-person Delphi structure. A design for a recommender system for professionals in online "communities of practice" incorporating a dynamic Delphi structure for obtaining group preferences and using collaborative tagging was published in 2009 (Turoff and Hiltz). In many Delphis, it is desirable to break down the votes into subgroups by the characteristics of professionals or knowledgeable people involved. In the design presented in this recommender paper, we specifically call for rating the documents of interest to the given community through collaboratively tagging by the community as to the special topics the documents represent. The users should also be self-tagged by same index so that it is easier to make participants aware of what is a significant information and evaluation vote of interest to them on a personal basis. This is a desirable feature for any online continuous Delphi/Recommender system. In recommender systems like the product review system used for Amazon the product of concern is the tag that links purchasers and those who might purchase a given product. Norm Dalkey emphasized the concept that Delphis could ultimately produce very concise results that because of quantification would not suffer from ambiguity and the other problems facing face-to-face verbal communications. He would have viewed the prediction market as an excellent example of a type of Delphi process. Certainly, the prediction market has roots in the concept of financial markets, but what people tend to forget is that those markets are only as good as the people who are making the investments. In addition, there is a tremendous amount of qualitative material used by people who recommend to others what to do in such markets. They are anything but concise in their output of material, nor are they always accurate, especially in forecasting negative financial events of any size. Delphi processes are no better than the group that participates. The movie financial success prediction market is quite good since most of the thousands who play that market are extremely knowledgeable about movies. Another example is that NETFLIX can identify very small groups of individuals who make very similar movie choices, and use that to make recommendations among the members of that small group. This is a complete Delphi process with anonymity within the larger NETFLIX recommender system. They might someday decide to allow social networks to form within that context for those willing to reveal their identity to each other. Wikis started out to be completely free for anyone to rewrite anyone else's material. Clearly, this did not work for any subject where disagreements existed. They have now evolved to have strong editing approval procedures like journal editors. They have evolved to be much more Delphi-like than their original conception. One of the most interesting examples is Wikimapia, which allows anyone to place information into a geographical database. There are number of examples where local governments allow citizens to place information directly into a database for the local community. Most local governments do not have the resources or funds to do this when the data has to be maintained. In this case, the citizens can update entries when needed. This is used to create GPS databases relevant to emergency management and provide information on sites vulnerable to certain disasters and the locations of equipment that might be shared among the community in emergencies such as a contractor's earth moving equipment or possible shelter locations. The recent emergence of Social Networks that allow the users to form their own groups to share information of common interest to the group has lead to a large number of local community activities including sharing information relevant to an expected or ongoing emergency (Palen, Hiltz, and Liu, 2007, Vieweg et al, 2008, White, et al, 2008). The idea of groups within online Web systems goes back to the earliest days of Group Decision Support Systems on the Web (Hiltz and Turoff, 1978, 1993). The unfortunate situation with respect to Social Networks is that most of these systems are designed to serve a commercial objective and the functionality does not really include what could be designed to facilitate collaborative goals for applications like emergency management information systems, as has been demonstrated in the professional literature of that field (Turoff et al 2004).

See also

Environmental Scanning & Monitoring
System Dynamics
Structural Analysis
Agent Modelling
SWOT Analysis
Trend Intra & Extrapolation
Modelling & Simulation
Creativity Methods
S&T Roadmapping
Critical & Key Technology Study
Scenario Building
Morphological Analysis & Relevance Trees
Cross-Impact Analysis
Multi-Criteria Analysis

Personal tools