Renooij, S.; Witteman, C. L. M.: Talking probabilities: communicating probabilistic information with words and numbers. International Journal of Approximate Reasoning 22 (1999), pp. 169--194.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....warrant the study of the possible query optimization. Research in this area is currently underway. Eliciting probabilities from experts has always been part of the stochastic modeling repetoire. While there are a variety of techniques for doing so by hand or via online tools (see, for instance, [21, 22, 16, 24]) our application demands more than any of these methods delivers. The rst diculty in building an advisor network is the sheer volume of probabilities needed. There are three to seven or more values for each variable (from Pass Fail NotTaken to A B C D E NT W, plus possible information about ....

Silja Renooij and Cilia Witteman. Talking probabilities: communicating probabilistic information with words and numbers. International Journal of Approximate Reasoning, 22(3):195-215, 1999.

....techniques [10] to betting. Morgan and Henrion [5] indicate that many experts from whom they elicited probabilities felt more comfortable providing direct probabilities or even drawing distribution functions then using less direct techniques like betting. Research by van der Gaag and Renooij [11, 9] found that for some groups of people, most notably, medics, the best way of specifying probability was not numerical but rather verbal, with words such as possible , likely or unusual being consistently associated with particular numerical values. In [11] they have studied the (Dutch) words ....

....be elicitation of such individual distributions. The interface should provide the opportunity not only to enter individual values for each probability in the distribution, but to change the shape of the entire distribution. ffl Not everyone likes numbers. Research by Renooj and van der Gaag [11, 9] on human perception of probabilities and verbalization of uncertainty suggests that for many people it would be easier to specify uncertain information in terms of words such as usually , possibly or unlikely , than in terms of numbers. While exact correspondence between these words and ....

Silja Renooij and Cilia Witteman. Talking probabilities: communicating probabilistic information with words and numbers. International Journal of Approximate Reasoning, 22(3):195--215, 1999.

....and three parent nodes, there are 5 3 probability distributions for that one node. We will be using a variety of techniques, including decision trees, pattern copying, and verbal numeric scales, for reducing the number of queries and making the elicitation process simpler for the users [10, 11, 8]. 4.2 PLANNING Given a Bayes net model of the university program in question, and given a full specification of the goal state (in terms of university and departmental requirements, including number of credit hours) the Advising Tool will have two functionalities: it will be able to compare ....

Silja Renooij and Cilia Witteman. Talking probabilities: communicating probabilistic information with words and numbers. International Journal of Approximate Reasoning, 22(3):195--215, 1999.

....application of the decision analytic techniques for probability elicitation rapidly becomes impractical if not impossible for network quantification. For probability elicitation for probabilistic networks, therefore, supplementary techniques are being sought [Druzdzel and Van der Gaag, 1995, Renooij and Wittman, 1999, Van der Gaag et al. 1999] To conclude our brief discussion of sources of probabilistic information, we would like to note that, although tempting, combining information from di#erent sources in a single probabilistic network can be risky and can in fact lead to incorrect results [Druzdzel and ....

....reliable. Elicitation using graphical means that allow an expert to directly manipulate a pie chart or a bar graph o#ers more support to the expert and is likely to lead to numbers with higher accuracy. Probabilities can also be related to verbal descriptions such as very likely and improbable [Renooij and Wittman, 1999]. Although verbal descriptions of probabilities are known to be context sensitive and can describe wide ranges of numerical quantities, the use of both words and numbers in probability elicitation can result in reasonable assessments [Van der Gaag et al. 1999] Rather than pushing an expert to ....

S. Renooij and C.L.M. Witteman. Talking probabilities: communicating probabilistic information with words and numbers. International Journal of Approximate Reasoning, vol. 22, 1999, pp. 169 -- 194.

....a di#erence in presentation mode, that is, either verbal or numerical, did not a#ect our subjects decisions. We would like to note that the four studies included subjects as well as examples from the field of medicine. For further details of the studies, we refer the reader to an extended paper [Renooij Witteman, 1999]. Because people may have di#erent preferences in di#erent situations, we decided to include both the numerical and the verbal anchors on our response scale. Since the verbal probability expressions were explicitly not intended as translations of the numerical probabilities, we decided to position ....

S. Renooij and C.L.M. Witteman (1999). Talking probabilities: communicating probabilistic information with words and numbers. International Journal of Approximate Reasoning, vol. 22, pp. 169 -- 194.

....indicated that the decisions made were independent of whether the probability information was expressed numerically or verbally. We would like to note that the four studies included subjects from the eld of medicine. For further details of the studies, we refer the reader to an extended paper [Renooij Witteman, 1999]. The fact that the subjects in our studies interpreted the verbal probability expressions as intended, motivated us to further elaborate on a scale with both numerical and verbal anchors for use as an aid for probability elicitation. Since the verbal probability expressions were explicitly ....

S. Renooij, C.L.M. Witteman (1999). Talking probabilities: communicating probabilistic information with words and numbers, submitted for publication.

....the thousands of probabilities required. For this purpose, we used an elicitation method tailored to obtaining a large number of judgemental probabilities in little time. At the heart of this method lies the idea of presenting experts with a separate figure for every probability to be assessed [2]. Figure 2 shows, as an example, the figure pertaining to the conditional probability Pr(Invasion = T2 Shape = polypoid, Length 5cm) for the oesophagus model. On the left of the figure is a fragment of text that transcribes the probability under consideration, thereby circumventing the need ....

S. Renooij, C.L.M. Witteman (1999). Talking probabilities: communicating probabilistic information with words and numbers. International Journal of Approximate Reasoning, vol. 22, pp. 169 -- 194.

No context found.

Renooij, S.; Witteman, C. L. M.: Talking probabilities: communicating probabilistic information with words and numbers. International Journal of Approximate Reasoning 22 (1999), pp. 169--194.

No context found.

S. Renooij and C.L.M. Witteman. Talking probabilities: communicating probabilistic information with words and numbers. International Journal of Approximate Reasoning, vol. 22, 1999, pp. 169 - 194.

No context found.

S. Renooij and C. Witteman. Talking probabilities: communicating probabilistic information with words and numbers. International Journal of Approximate Reasoning, 22:169 -- 194., 1999.

No context found.

S. Renooij and C. Witteman. Talking probabilities: communicating probabilistic information with words and numbers. International Journal of Approximate Reasoning, 22:169 -- 194., 1999.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC