People's recognition of political decision arguments

313 A cta Psychologica 68 (1988) 313-327 North-Holland PEOPLE’S DECISION RECOGNITION ARGUM ENTS OF POLITICAL Irmtraud N. GALLHOFER, W illem E. SARIS and M aarten SCHELLEKENS * Sociometric Research Foundation, Amsterdam, The Netherlandr In a previous research 235 political decisions were studied based on documentary material. One of the results of this study was that, given a decision maker’s description of the problem, (i.e., the alternatives, probabilities and values of the outcomes), his choice could be predicted, even though he did not state the decision rule he used. Due to this incompleteness there is a question about whether the decision makers and their audiences really applied the decision rules which have been predicted. Therefore, this study investigated whether subjects in general understood the policians’ problem structure and recognized the expected decision rule when presented with the information as the politicians provided. The results of the study showed that logically compelling choices were recognixed by the subjects and that in clear situations the majority of the subjects specified the same rule as the one that was predicted. But if information was available which was not strictly necessary to derive the correct choice, they often ignored this information and specified a simpler rule. zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA Introduction In a long term study (Gallhofer, Saris and M elman 1986) 235 real-life political decisions were studied based on documentary material such as meetings of the Council of ministers, reports of governmental officials etc.. The decisions related to Dutch foreign policy from 1900 to 1955. In order to derive the choice rules a text analysis instrument was developed which we call the Empirical Decision Analysis approach (EDA). One of the results of this study was that, given a decision maker’s description of the problem, i.e., the alternatives, probabilities and values of the outcomes, his choice could almost perfectly be * W e wish to thank Hem-y M ontogomery for his useful comments. M ailing address: I.N. Gallhofer, Sociometric Research Foundation, Blauwburgwal 20, 1015 AT Amsterdam, The Netherlands. OOOl-6918/88/$3.50 0 1988, Elsevier Science Publishers B.V. (North-Holland) 314 I.N. Gdhofer et al. / Poliiical decision arguments zyxwvutsrqponmlkjihgfedcbaZYX predicted, even though he did not mention the decision rule used. It seemed that the rule to be applied, given the specified structure, was expected to be self-evident and was therefore not mentioned. Due to this incompleteness the problem remains, whether the decision makers and their audiences really applied the decision rules which we have predicted. In order to investigate this, a new study was carried out. Since the politicians could not be asked about the decision rules they had used, some of the problem structures of arguments indicated by the politicians for specific choices were presented to a random sample of the Dutch population. The questions of interest were: zyxwvutsrqponmlk First, do people in general arrive at the same choice of alternatives as the politicians, given the same description of the problem? If the majority of subjects chooses the same alternatives as the politicians, we have an indication that they understand the politicians’ problem structure and that they are able to draw the same inferences from the given data regarding which alternative to choose. Second, do people in general recognize the expected decision rule, when they are presented with the same description of the problem as provided by the politicians? If the majority of subjects explains in their own words their choice relating to a specific decision problem in terms of the decision rule we had predicted by our approach, we have some indication that the rules we derived are generally understood by human decision makers. In the following, the decision rules and the method of detecting them with the EDA approach will be briefly summarized. (1) Decision rules were classified into four groups according to the precision of information about values and probabilities they require. Table 1 summarizes the four classes of decision rules. The table shows that the repertoire of decision rules of the politicians was the following: (a) The Subjective Expected Utility model states that a decision maker should choose the strategy with the highest expected utility. Values and probabilities with intensities are required in order to test the fit of this model. (b) The Risk-Avoiding rules developed by Gallhofer et al. (1986): the choice rule consists of selecting the alternative with the highest probability of positive outcomes, or, which amounts to the same thing, since 315 zyxwvutsrqp I.N. Gallhofer et al. / Political decision arguments Table 1 Relationship between decision rules and the precision of the information about values and probabilities. Probabilities with intensities without values with values intensities intensities SEU model Risk-Avoiding rules I II Dominance rule, Simon’s rule, without Lexicographic rule, Reversed Simon rule intensities A ddition of Utilities rule Probabilities III IV zyxwvutsrqponmlkjihgfedcbaZYXWV the sum of the probabilities is assumed to be one, of selecting the alternative with the lowest probability of negative outcomes. Probability statements with intensities and value statements without intensities are required in order to test the fit of these rules. This means that values must be indicated by the decision maker in terms of good or bad and probabilities in terms of certain or uncertain. (c) The Dominance rule states that the selected alternative must be better than the other(s) on at least one dimension and not worse than the other(s) on the remaining ones. To test the fit of this rule the same aspects or dimensions of outcomes must be systematically considered across alternatives by the decision maker and values must indicate at least a rank-ordering (intensity statements) while probabilities without intensities are required. (d) The Lexicographic rule suggests that the aspects should first be rank ordered in importance and then the strategy which is most attractive with respect to the most important aspect should be selected. In order to test the fit of this rule, aspects of outcomes must be systematically considered across alternatives; a rank ordering of aspects in importance is required and values must contain intensity statements. (e) The Addition of Utilities rule states that the values of all aspects should be summed up for each alternative and then the alternative with the highest total value should be selected. It will be clear that again a systematic use of aspects across alternatives is required, intensity statements of values are necessary and they must be formulated as a summary of all the aspects of one alternative. (f) Simon’s Satisfycing rule (Simon 1957: 248) suggests selecting the (first) alternative detected which only leads to satisfactory outcomes. 316 I. N. GalIhofer et al. / Political decision arguments (g) The Reversed Simon rzde zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLK covers political situations for which no satisfactory alternative is available. It consists of excluding all the alternatives which lead with certainty to only negative outcomes, as long as there is another strategy available which might lead to a positive outocme (Gallhofer et al., 1986). To test the fit of these rules, values must be indicated by the decision maker in terms of good or bad and probabilities in terms of certain or uncertain. (2) In order to determine the decision rule which should be applied in a specific case, the structure specified by the decision maker has to be classified into one of the four classes table 1 indicates. For the precise steps of analysis we refer to Gallhofer et al. (1986). Since different rules could often be used to predict the choice of a politician, we needed a procedure to choose a rule. We assumed for this purpose that the decision maker used all the given information in the argument. (3) Finally, the alternative obtained by a specific decision rule of the appropriate class was compared with the alternative actually chosen by the decision maker. If the alternatives were identical we said that the rule fit or predicted the choice. If this was not so, the rule could not predict the choice. We could predict almost perfectly the choices with one of the above rules, i.e., the seven rules correctly predicted 225 decisions and only in ten cases could no fitting decision rule be found. In the next section the research design will be elaborated in detail. Thereafter the decision problems used in the experiments will be described. Then the data collection and the text analysis for the verbal protocols will be briefly explained. Finally the results of the experiments will be presented and conclusions will be drawn based on the results. zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA Methodology Research design Thirteen decision problems were selected from our foreign policy data-collection relating to five decision rules. The choice of the five decision rules was mainly based on their frequency of occurrence in our data set. The assignment of the number of decision problems relating to each rule was based on structural characteristics of the decision I.N. Gollhojer et al. / Poliiicol decision arguments 317 zyxwvutsrqpo problems. For some rules several slightly differing structures were selected withinthe subset of data relating to a specific structure. This was done in order to test whether subjects recognized these structures as similar. For the Risk-Avoiding rules five decision problems were selected, two for the Reversed Simon rule, for decision situations which could either be resolved by Simon or Reversed Simon three problems were selected and for the Simon, Dominance and Lexicographic rule one decision problem was selected for each. The SEU rule and Addition of Utilities rule were excluded from the experiment, because of their infrequent occurrence. The next step consisted of choosing the form for presentation of the problem to the subjects. In think-aloud experiments (e.g. Huber 1982; Svenson 1988; Ericsson and Simon 1984) abstract and concrete presentations are used. Based on the results of a pilot study where we explored both modalities, it was detected that when presented with concrete problems subjects frequently preferred to discuss political details or speak in terms of norms rather than explain their choice in terms of decision-making concepts, which is the basis for detecting the decision rule they used. It seemed that their attitude toward the political problem had an impact on how they explained their choice. Given the assumption that with concrete decision problems subjects might behave differently, and given the number of decision problems to be presented, it was decided to run two experiments. In the first experiment ten decision problems were presented only in abstract form. In the second the remaining three problems were presented in both abstract and concrete modalities and the attitude towards three political situations was measured in terms of agreement/ disagreement with the politicians. Instead of doing a think-aloud experiment we chose a fully automatic interviewing technique, the so-called Tele-interview approach (Saris and De Pijper 1986; De Pijper and Saris 1986). Non-response did not occur. The subjects were 59 members of Dutch households based on a random population sample: 21% were younger than 20 years, 39% were between 20 en 40, 29% were between 40 and 60 and 11% were older than 60 years; 48.6% were male and 51.4% female; 25% had only a lower education, for 47% the educational level was medium and 28% received higher education. The subjects received on their screens at home the information with respect to the decision problem and they were asked to make a choice. Based on the results of this question it can be determined to what extent other persons arrive at the same choice as politicians. It was assumed that in the abstract form the agreement w ill be higher than in the concrete form, depending on respondents’ attitude towards the political decision problem (see also Bonham and Shapiro 1982, Walker and Murphy 1982). Attitude was measured by another question in the interview. The next question-an open-ended one-required that respondents explain their choice, taking both the rejected alternative(s) and the chosen one into account. When performing this task the information of the decision problem could always be brought back to the screen. The choice of this technique, which corresponds with retrospective verbal protocols (Ericsson and Simon 1984; Svenson 1988), was appropriate for the second purpose of our study, namely, to test if subjects explained their choices in terms of the predicted decision rules. It w ill be investigated here whether decision problems, for which our approach predicted the same rule, are perceived as the same type of decision by the subjects. 318 I.N. GaIlhoferet al. / Political decision arguments zyxwvutsrqponmlkjihgfedcbaZY For groups of decision problems which are perceived as the same, it wiI1 be studied whether subjects explained their choices in terms of the predicted decision rules. For choices which are explained differently than expected, it w ill be investigated whether or not the explanations used by the subjects are correct with respect to the presented information or that they restructured the information in different ways. Finally, efforts w ill be made to explain differences between the explanations of subjects and the predicted rules by background variables. Description of the decision problems The first problem which had to be resolved was the transformation of concrete forms into abstract ones without losing the essential characteristics of the problem. This was done as follows: we first indicated the alternatives by a number and assigned capital letters to outcomes. All different outcomes were labeled with different letters, even if they related to the same aspect or dimension. In order to reduce further the complexity of the representations we combined outcomes of an alternative if the values and probabilities were the same. With respect to the sequence of presentation of alternatives we followed the presentation of the decision maker. In our data set decision makers most frequently presented first the rejected alternatives and then the chosen one. In the following we w ill describe the decision problems presented to the respondents. They w ill be discussed with respect to the predicted decision rule and other decision rules which would be normatively correct w ill be indicated. Decision problem relating to the dominance rule One decision problem relating to the dominance rule was presented to the respondents. Its abstract form is as follows: Alternative 1 and alternative 2 both lead to outcome A. Outcome A is negative. Alternative 1 also leads to outcome B. Outcome B is very negative. Alternative 2 also leads to outcome C. Outcome C is less negative than outcome B. Based on this information (values with intensity statements, probabilities without intensity statements but indicated as certain) we predicted a dominnnce rule which corresponded with the choice of alternative 2 of the political decision maker. Since it is conceivable that respondents may produce other decision rules based on the presentation of this problem, it seems useful to determine whether there exist other rules which are correct. A rule is correct if the required information for that rule is available in the decision problem presented and the chosen rule leads to the same conclusion. Given this problem structure the SEU model could be applied, but in a very trivial way since both probabilities can be considered as 1.0 Respondents would then have to indicate that they combined values and probabilities with intensities. Simon’s rule, the Reversed Simon rule and the Risk-Avoiding rule are not correct, because subjects would have to restructure the information entirely. The Lexicographic rule is not correct, because no importance of outcomes is mentioned. I.N. GaIlhoferet al. / Political decision arguments 319 Decision problems relating to Simon’s rule zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONML One abstract decision problem relating to Simon’s rule was presented: Alternative 1 leads certainly to outcome A. A is negative. Alternative 2 leads possibly to outcome A or B. A is negative and B is positive. Alternative 3 leads possibly to outcome B or C. B and C are both positive. Simon’s rule predicted the choice of alternative 3. A correct rule would be what we call the Dominance principle. Subjects would have to state that the outcomes of alternative 3 are better than the outcomes of the other alternatives. The SEU-model would also be correct, because outcome C of alternative 3 is more positive than outcome A of alternative 2, while these alternatives do not differ in outcome B. The chance of a better outcome would be higher with alternative 3 than with the other ones. The Risk-Avoiding rule is also correct, because there is more chance for positive outcomes with alternative 3 than with the other ones. The Reversed Simon rule is of wurse not correct, because, although we can in this way eliminate alternative 1, this rule does not lead to the choice of one single alternative. Decision problems relating to the Reversed Simon rule Two decision problems of this kind were presented to the respondents: (1) Alternative Alternative (2) Alternative Alternative Alternative 1 2 1 2 3 leads leads leads leads leads to outcome possibly to to outcome to outcome possibly to A. Outcome A is negative. outcome A or B. B is positive. A. A is negative. B. B is negative. outcome C or D. C is positive, D is negative. In the first problem the political decision maker chose alternative 2 and in the second alternative 3. Since values and probabilities were without intensities and only one alternative led possibly to a positive outcome, the Reversed Simon rule was predicted. The Risk-Avoiding rule would be correct because there is more chance of a positive outcome if there is any chance than if there is no chance at all. Our approach did not predict this rule because the original decision maker did not indicate probabilities with intensities. The Dominance principle and the SEU model are only wrrect for the first presentation because there it is known that the negative outcomes are the same for both alternatives. Simon’s rule is not correct because of the absence of an alternative which leads to positive outcomes only. Decision problems relating to both Simon’s rule and the Reversed Simon rule Three decision problems relating to these rules were presented. They had the following forms: (1) Alternative 1 leads to outcome A. A is negative. Alternative 2 leads to outcome B. B is positive. 320 I.N. Gallhofer et al. / Political decision arguments zyxwvutsrqponmlkjihgfedcbaZY (2) and (3) Alternative 1 leads to outcome A. A is positive Alternative 2 leads possibly to outcome B or C. B is negative and C is neutral. The only difference between the last two problems was the sequence of presentation of alternatives. The values and probabilities of these problem structures are all without intensities. In the first example the political decision maker chose alternative 2 and, in the second example, alternative 1. We predicted both Simon’s rule and the Reversed Simon rule, because, when using all the available information, the choice can be derived by either formulating that one selects the first alternative which only leads to positive outcomes or one rejects all alternatives which lead to negative oucomes while there exists an alternative where a positive outcome is available. Not predicted, but indeed correct, is the Dominance principle, because a positive outcome is always better than a negative or a neutral one. The SEU model would also be correct when a respondent indicated that in some way he combined values and probabilities with intensities. Also the Risk-Avoiding rule would be correct, because the probabilities of obtaining positive outcomes differ for the alternatives (namely certain against certainly not), although this kind of reasoning seems very unlikely. Decision problems relating to the Risk-Avoiding rules Five decision problems relating to the Risk-Avoiding rule were presented to the respondents in abstract form and two of them also in concrete form. We present the abstract forms below. The concrete forms, which related to the presentations 4 and 5, are omitted for the sake of conciseness and are in the same line as shown before. (1) Alternative 1 leads to outcome A. A is negative. Alternative 2 leads probably to outcome B. B is positive. However, there is also a small chance that A w ill occur. (2) Alternative 1 leads to outcome A. A is negative. Alternative 2 leads probably to outcomes B and C. B is positive and C is acceptable. (3) Alternative 1 leads to outcome A. A is negative. Alternative 2 leads probably to outcome B. B is positive. However, there is a small chance that C w ill occur. C is negative. (4) Alternative 1 leads probably to outcome A. A is positive. There is also a small chance that B w ill occur. B is negative. Alternative 2 leads probably to outcome C. C is negative. There is also a small chance that D will occur. D is positive. (5) Alternative 1 leads probably to outcome A. A is negative. There is also a small chance that B w ill occur. B is positive. Alternative 2 leads with certainty to outcome C. C is negative. Alternative 3 leads with certainty to outcome D. D is positive. In the first three decision situations the political decision maker chose alternative 2, in the fourth decision situation alternative 1, and in the fifth alternative 3. We I.N. Gallhofer et al. / Political decision arguments 321 zyxwvutsrqpo predicted for all five problems the Risk-Avoiding rule because values are indicated without intensity statements, while probabilities are indicated by intensity statements. For the first three presentations the Reversed Simon rule is also correct, because only one alternative leads possibly to positive outcomes. We did not predict this rule here, because this rule does not make use of all relevant information (i.e., intensity statements of probabilities), altough this information is not necessary to make the choice. The Dominance principle and the SEU model are only correct for the first and the second decision problem, since only in these cases is it known that the negative outcomes of the two alternatives are the same and therefore equally negative. Simon’s rule is only correct for the fifth problem, because only in this case is there one alternative which leads to positive outcomes only. Decision problem relating to the Lexicographic rule One decision problem of this kind was presented: Alternative 1 leads to outcomes A and B. A is positive, B is negative. A is the most important outcome. Alternative 2 leads to outcomes C and D. C is positive, D is negative. Because it is indicated which outcome is the most important one, we predicted the Lexicographic rule, which leads to the choice of alternative 1, No other rules are appropriate here; the lexicographic rule is not in agreement with the SEU model or the dominance principle and the other rules mentioned in this study cannot handle importance-criteria. Given this overlap of rules we do not expect the predicted rule always to be used but we w ill also have to take into account the correct rules. Data collection Two questionnaires were constructed according to the Tele-interview approach (De Pijper and Saris 1986). For the first experiment it consisted of a short introduction telling the subjects that 10 choice problems (first experiment) relating to political choices would be presented. They were instructed to read the information carefully and, on the basis of this, make their own choice. They also were informed that within each choice problem the information could always be retrieved on the screen. They were further instructed that they first had to select an alternative and thereafter they would have to explain why they chose the specific alternative and not the other one(s). In order to put them at ease they were told that neither their intelligence nor their political opinion was being tested and that the purpose of this questionnaire was just to study how other persons make decisions: writing errors would not matter. In order to give them some training two exercises were given: first a concrete decision problem was presented which we discussed in terms of alternatives, outcomes, probabilities and values. Thereafter it was transformed into abstract form and we asked them to select an alternative and to give their explanation for the choice. For answering this question approximately 28 lines on the screen were reserved, which proved to be more than they needed since the median of the number of lines used was eight. The second exercise 322 I.N. Gatlhoferet al. / Political decision arguments was only abstract and served as additional training. Then the ten decision problems were presented in an arbitrary way. For the second experiment the procedure was similar but subjects were no longer obliged to do the exercise examples since they had done them already for the first experiment. The second questionnaire consisted of three abstract and three concrete decision problems as well as three questions relating to the attitudes of the subjects with respect to the concrete decision problems. Since the second experiment was carried out during a holiday period the number of subjects was not 59 but 50. Coding of the verbal protocols A coding instrument was developed based on verbal protocols of a pilot study. Since the explanations of the choice were (similar to think aloud protocols) not formulated by well-formed sentences and punctuation marks, this instrument had to make use of semantic rather than grammatical units. The EDA approach was not applicable because it focusses on the construction of decision trees. The coding instrument split the decision rules into elementary units, such as parts relating to the rejected or chosen alternative or to a comparison of both. Within these units, segments relating to probability and value statements had to be detected. Also, the coding instruction specified which units were minimally necessary to detect a specific rule and which elements in addition would give a maximal formulation of a specific decision rule. For more detailed information of this instrument we refer to Gallhofer and Saris (1988). Two coders independently coded 740 protocols; their inter-coder reliability was 0.86 (Cohen’s kappa). However, in order to resolve differences between them, the coders came together to compare their results and to arrive at a common solution by discussion. zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA Results The first question was whether the subjects from our sample arrived at the same choice as the politicians, given the same description of the decision problem. The results showed that when the information was presented in abstract form almost all subjects chose the same alternative as the politicians (in the first experiment 100% of the subjects chose the same alternative, and in the second experiment 98% of the subjects chose the same). We therefore conclude that people understood the arguments in the sense that they make the same choices as the decision makers. When presented with concrete information, however, the agreement with the choice of the politicians was lower. Examining the relationship between the attitudes of the subjects towards the concrete decision problems and their agreement with the politicians, we found a trend indicating that the more people agree with the politician with respect to the presentation of the problem, the more they choose the same alternative. The more they disagree the more they choose another alternative. It is clear that when they choose another alternative, they restructure the decision problem in their mind, based upon prior 323 zyxwvutsrqp I.N. Gallhofer et al. / Political decision arguments Table 2 Relationship between predicted rules and rules used by subjects (N = 59). Predicted Rules used by subjects rules a The same A lternative A wrong as predicted correct rule rule % % abs. Dominance rule 90 53 abs. % abs. 10 Simon/Rev. Simon rule 1 -lo 41 22 13 8 6 5 Risk-Avoiding rule 1, 2, 3 51 30 38 22 11 I Lexicographic rule 61 36 _ 39 23 Reversed Simon rule 1,2 63 31 28 17 9 5 Risk-Avoiding rule 4 60 35 _ 40 24 2 Simon group 70 41 27 16 3 Simon’s rule 46 21 46 21 8 5 Risk-Avoiding rule 5 10 6 86 51 4 2 Rev. Sii./Sim. rule 2,3 a The predicted rules are grouped according to the seven problem structures the subjects considered fundamentally different. zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIH beliefs and other information. When subjects had to give an explanation of their choice this became clear: frequently they added new information to the alternatives. We therefore conclude that concrete situations cannot be used for the test of drawing inferences from given problem structures. Secondly, we investigated whether subjects perceived different decision problems as different. This means, for example, that when we predicted for a given decision problem a Risk-Avoiding rule and for another a Reversed Simon rule, we wanted to know whether the subjects in general saw these situations as different and used different heuristics for these two types of problem structures. We first tested whether the distributions of the subjects for a specific group of decision problems like Risk-Avoiding or Reversed Simon/ Simon or Reversed Simon were the same. If they were the same within a specific group, then the subjects perceived the problem structures for these groups as similar. Chi * tests were done at the 1% level and we obtained nine instead of six groups of problem structures. In order to establish whether the distinction of these nine different groups of problem structures is meaningful, it was tested whether the distributions for each pair of the nine problem structures were significantly different from each other. The results of these tests showed that the subjects considered seven problem structures as fundamentally different from each other. This means that most of the problem structures which we predicted were also fundamentally meaningful to the subjects. After this first analysis we investigated whether the subjects recognized the expected decision rule. Table 2 summarizes the relationship between the predicted decision rules for the seven groups of problem structures and the rules used by the subjects when presented in abstract form. 324 I.N. Gallhofer et al. / Political decision arguments zyxwvutsrqponmlkjihgfedcbaZY As table 2 shows, subjects recognized the Dominance rule the best; 90% of them used the predicted rule. They did not use an alternative correct rule in this case. With respect to the wrong rules, some used a Reversed Simon rule, which means that they simplified the values. They used the Risk-Avoiding rules to a lesser extent, which implies that they transformed both values and probabilities. In some cases their explanations were unclassifiable. The problem structure relating to Reversed Simon/Simon I was also well recognized (70%). As an alternative correct rule again the Dominance principle was very often used. Also Risk-Avoiding occurred, which means that they indicated that there was more chance of a positive outcome. The wrong rules consist of unclassifiable explanations. The recognition of the Reversed Simon rule was also quite good: 63% of the subjects used this rule. With respect to alternative correct rules, they frequently used the Dominance princple, which means that they indicated that the chosen alternative was better than the rejected one(s). The Risk-Avoiding rule was also used often. The wrong rules do not give a clear picture: besides some other rules they also used their own heuristics. The Lexicographic rule was recognized by 61% of the subjects. Since there were no other correct rules, 39% of the subjects used inappropriate rules. Among these rules the transformation of the information into a Dominance structure was very common. Otherwise, the problem was mostly solved by their own heuristics. The Risk-Avoiding rule 4 was recognized by 60% of the subjects. Among the wrong rules Simon’s rule and the Dominance principle were frequently used. The subjects who used Simon’s rule had weighted the high probability of the positive outcome as certain, which Tversky and Kahneman (1986) call a ‘pseudo-certainty’ effect. When using the Dominance principle they might have weighted the positive outcome by the probability without being explicit about that. The Risk-Avoiding rules I, 2, 3 were only recognized by 51% of the subjects. This results falls somewhat short of our expectation. The main reason for it is that subjects did not always make use of probabilities with intensities as presented in the information; they often reduced them to ‘possible’, which also is sufficient in order to resolve these problems. Thus the Reversed Simon rule was very frequently used as an alternative correct rule. Here again the transformation into a Dominance structure occurred frequently. Even an SEU type rule was used, which means that some subjects transformed the values in statements with intensities and combined them with the probabilities. The picture of the wrong rules is quite heterogeneous and no specific interpretation can be given to it. From the rules which fall under the Simon group, the Reversed Simon/ Simon structures 2, 3 were the best recognized by the subjects (70%). The most frequently used alternative correct rule was Dominance, but Risk-Avoiding and SEU also occurred. The wrong rules were mainly unclassifiable. The Simon structure was only recognized by 46% of the respondents, which is quite surprising since it is a very clear structure. A large part of the alternative correct rules were of the Simon type, which means that subjects did not explicitly mention that the chosen alternative led ‘only’ to a positive outcome. Among the other correct rules were Risk-Avoiding and Dominance. Subjects thus indicated for the chosen alternative that it was the best or that the chance of I.N. Gallhofer et al. / Political decision arguments 325 zyxwvutsrqpo positive outcomes was the highest. The wrong rules consisted of Unclassifiable and Reversed Simon. The Risk-Avoiding 5 structure was only recognized by 10% of the subjects. However, the percentage of correct rules the subjects produced was considerable, 86% . For this structure 52% of the correct rules consisted of Simon’s rule, which means that subjects resolved the problem without probabilities with intensities, which the political decision makers had indicated, which also is sufficient to arrive at a choice. This last result is analogous with the results of alternative correct rules obtained for the fist Risk-Avoiding group where it was also sufficient to resolve these problem structures without probabilities with intensities. In this latter case subjects very often used the Reversed Simon rule. The similarity in the use of decision rules between the Simon structure and the Risk-Avoiding 5 structure is mainly due to this frequent use of Simon’s rule. Risk-Avoiding structures 4 and 5 and the Simon structure also were presented in concrete form. The first test was whether these structures differed from the abstract ones according to the subjects. The results showed that the structures were considered as different from the abstract ones. W ithin the concrete problems the subjects discerned two groups, i.e., Risk-Avoiding rule 4 and a group consisting of Simon’s rule and Risk-Avoiding rule 5, which is in agreement with the findings of the abstract presentations. An analysis of the relationship between the predicted rules and the rules used by the subjects when the problems were presented in concrete form, showed that the recognition of the predicted decision rule was less high. The reason for these worse results for concrete situations lies (in our opinion) in the fact that subjects were very eager to restructure or add information to the presented information. This might be due to their familiarity with the topics, by which they were able to use their background knowledge (see also Huber 1982). Since for the abstract presentations the relationship between the predicted choice rules and the rules used by our subjects also was not perfect, we tried to explain the differences between the choice rules of the subjects and the ones we predicted in terms of background variables of the subjects. These variables were age, education, sex and interest in the topic. It turned out that only education had an effect. The result of a one-way analysis of variance for the 13 abstract decision problems showed that the average number of identical rules of lower educated subjects (x k 5.0) was significantly different ((Y = 0.01) from the medium (2 = 8.25) and higher educated (x = 8.07) subjects. There was, however, no significant difference between subjects with a secondary and higher education. This means that subjects with a lower education have fewer rules identical with the predicted ones. Based on this result it was investigated whether or not people with a lower education have a preference for a specific decision heuristic in order to cope with their task. T’he result of another analysis of variance showed that the average number of the use of the Dominance principle of subjects with a low education (x = 4.25) was significantly different (a = 0.01) from subjects with a secondary education (x= 2.25) and subjects with a high education (x= 1.85). There was again no significant difference between subjects with secondary and higher education. This means that Dominance is a very fundamental structure, especially to the less educated subjects, and they are more inclined to create it than other subjects, by restructuring the presented information. 326 I.N. Gallhofer et al. / Political decision arguments zyxwvutsrqponmlkjihgfedcbaZY Conclusions zyxwvutsrqponmlkjihgfedcbaZYXWVUTSRQPONMLKJIHGFEDCBA This study showed that when the decision problems were presented in an abstract form almost all the subjects of our study chose the same alternative as the politicians. The structures were clear and there was no doubt about the conclusion one should draw, although this conclusion is not necessarily correct in terms of dominance (see, e.g., the structures of Reversed Simon 2, Risk-Avoiding 3, 4, 5 and the Lexicographic case). In other words: logically compelling choices are recognized as such by the audience. When subjects were presented with concrete political decision problems it depended on their attitude toward the political topic whether or not they chose the same altemative as the politicians. Concrete situations are therefore not suitable for the research about drawing of inferences from given problem structures. Second, the study showed that subjects generally perceived different problem structures as different. The results for both abstract and concrete presentations were similar. This implies that most of the problem structures which we predicted with our approach were also fundamentally meaningful to the subjects. Third, in clear situations the majority of the subjects specified the same rule as the one we predicted. But, if information was available which was not strictly necessary to derive the correct choice, they often ignored this information and used a simpler rule. This phenomenon can be due to the fact that in our experiments the subjects did not generate the information themselves, in contrast with the politicians. Therefore it is possible that the subjects in our experiment only searched for the necessary information to make their choice. Furthermore, in the original argumentation of the politicians they could not be sure that the audience agreed completely with their arguments. To anticipate this, they could bolster their argumentation. Thus the findings of these experiments do not invalidate the prediction of the choice rules by the EDA approach. Another form of simplification can be found in the frequent use of the Dominance principle. We predicted a Dominance rule only once, but subjects frequently transformed the problem structures in such a way that they could arrive at the choice by a Dominance principle. People with a lower education used relatively more Dominance than subjects with a higher education. These results support Montgomery’s 327 zyxwvutsrqp I. N. Gallhofer et al. / Political decision arguments hypothesis (1983) that human beings are in search for a Dominance structure. Given that politicians are in general more similar to the higher educated subjects in our sample, we can expect that the politicians also would have recognized the decision rules we have predicted. Our overall conclusions are that in most situations the majority of subjects explained their choice in terms of the predicted rule. If we take into account permissible simplifications, 85% of the subjects chose a correct rule out of the seven specified by us. This implies that our approach can be used for the prediction of decision rules based on argumentation. References Bonham, G.M . and M .J. Shapiro, 1982. ‘The cognitive process approach and policy analysis’. In: G. Hopple (ed.), Biopohtics, political psychology and international politics. London: F. Pinter. pp. 210-235. De Pijper, W .M . and W .E. Saris, 1986. The formulation of interviews using the program interv. Amsterdam: Sociometric Research Foundation. Ericsson, K.A. and H.A. Simon, 1984. Protocol analysis, verbal reports as data. Cambridge, M A: M IT Press. Gallhofer, I.N. and W .E. Saris, 1988. Political decision making. Amsterdam: Sociometric Research Foundation. Gallhofer, I.N. W .E. Saris and M . M elman, 1986. ‘The empirical decision analysis procedure’. In: I.N. Gallhofer, W .E. Saris and M . M ehnan (eds.), Different text analysis procedures for the study of decision making. Amsterdam: Sociometric Research Foundation. pp. 53-65; Huber, O., 1982. Entscheiden ah Problem&en. M ontgomery, H., 1983. ‘Decision rules and the search for a dominance structure: process model of decision making’. In: 141-150. W ien: Hans Huber. P. Humphreys, 0. towards a Svenson and A. VBri (eds.), A nalysing and aiding decision processes. Amsterdam: North-Holland. pp. 343-370. Saris, W .E. and W .M . de Pijper, 1986. Computer assisted interviewing using home computers. European Research 14, 144-150. Simon, H.A., 1957. ‘A behavioral mode1 of rational choice’. In: H.A. Simon, M odels of man, social and rational. New York: W iley. pp. 135-152. Svenson, O., 1988. ‘On the collection and analysis of verbal protocols’. In: H. M ontgomery and 0. Svenson (eds.), Process studies of judgement and decision making. Chichester: W iley. (In press.) Tversky, A. and D. Kahneman, 1986. ‘Rational Hogarth and M .W . Reder’ (eds.), Rational choice and the framing of decisions’. In: R.M . choice, the contrast between economics and psychology. Chicago, IL: University of Chicago Press. pp. 67-94. W alker, S. and T. M urphy, 1982. The utility of the operational code in political forecasting. In: Political Psychology 3, 24-60.