Decision making styles and the use of heuristics in decision making

In a world of limited knowledge, resources, and time, the concept of rational decision making is not always a suitable model to describe and explain human behavior. Research indicates that individuals, due to a lack of time and commitment, tend to rely on shortcuts when making decisions (Payne et al. 1993). As a consequence, the concept of fast and frugal heuristics has gained attention as an important concept to describe human judgments (Gigerenzer et al. 1999). The term heuristic stems from Greek, meaning to find, and it is used in cognitive psychology as a “useful shortcut, an approximation, or a rule of thumb for searching through a space of possible solutions” (Hoffrage and Reimer 2004, p. 439).

The literature on fast and frugal heuristics pioneered by Gigerenzer et al. (1999) has identified several different types of heuristics that can be used in decision processes. In this paper, we particularly focus on two heuristics from the “adaptive toolbox” proposed by Gigerenzer et al. (1999), which are widely discussed in literature and which can be seen as particularly closely related to consumer decision making (Hauser 2014, p.1692), the recognition heuristic and the take-the-best heuristic. The recognition heuristic has been related to consumer choices in several empirical studies (Marewski et al. 2010; Hauser 2011; Thoma and Williams 2013; Hilbig 2014). Take-the-best is among the most studied heuristics, and since it refers to potentially multiple attributes of alternatives, it is considered a plausible model of boundedly rational decision making (Martignon and Hoffrage 2002). We experimentally study the use of those heuristics in comparison to a more rational, but also cognitively more demanding decision process, as well as an affective decision process guided by emotions.

The recognition heuristic as an efficient rule of thumb can be described as a frugal inference strategy relying on only one piece of information, namely the recognition cue (Hilbig and Pohl 2008), and ignoring other information (Hilbig 2014). According to Gigerenzer and Goldstein (2011), if a decision maker has to make a choice between two objects and recognizes one of them, he or she infers that the recognized object has the higher value on a given criterion and thus will prefer it over the unrecognized object. Empirical evidence demonstrates that this one-cue non-compensatory heuristic can be surprisingly accurate (Hilbig 2014), resulting in a remarkably high number of correct inferences by individuals (Borges et al. 1999) and in better predictions regarding consumer behavior than compensatory models (Marewski et al. 2010). Relying on recognition can be useful, particularly considering the fact that recognition lasts long and is easily accessible. While it is not limited to the case of just two alternatives, its application is limited to problems involving only a small number of alternatives (Marewski et al. 2010), and some experiments have shown that alternative compensatory models account better for aggregate data patterns (Hilbig 2014). The recognition heuristic has been widely studied, but has received only mixed support in literature on probabilistic inferences. However, very few studies focus on the application of the recognition heuristic in preferential decision situations such as consumer choice (Hilbig 2014). Marewski et al. (2010) propose that recognition is used to form consideration sets of alternatives, so that unrecognized alternatives are ranked behind recognized alternatives.

The take-the-best heuristic takes into account that decision makers might possess several clues about the objects they have to decide on, but it proposes that decision makers use only one single cue in the decision making process. For example, when estimating the size of a city, such cues could be whether the city has an airport, an opera house, etc. (Reimer and Rieskamp 2007). Decision makers then would assume that the city that has such a facility (while the others do not) is also larger. The take-the-best heuristic tries cues in order of validity, one after the other, and stops at the first cue that discriminates between the objects under consideration (Gigerenzer et al. 1999). Take-the-best is considered to be a fast heuristic due to the fact that it does not require complicated computations and it is frugal because the stopping rule limits the search for cues, sometimes even to just one cue.

Fast and frugal heuristics like recognition or take-the-best still resemble rational theories of decision making in that they apply (although only limited) information on the available alternatives to evaluate the alternatives and ultimately make a choice. Depending on the heuristic, information can be processed in a compensatory or non-compensatory manner. However, humans sometimes approach decision problems also in a very different way and rely on emotions in decision making. Gigerenzer (2001, p. 6–7) points out that “theories of decision making have often neglected emotions, and sometimes even cast them as the very opposite of rationality.” Nevertheless, emotions are a crucial factor when being engaged in decisions and can support the process of decision making (Gigerenzer 2001). In this context, emotions can be classified as behavioral theories and are sometimes also referred to as affect heuristic (Slovic et al. 2007).

Empirical literature has shown that heuristics are not universally used: they are used only by a fraction of subjects, and only in certain situations. For example, Newell et al. (2003) found that roughly one third of subjects were actually using take-the-best. Similarly, in the experiments of Bröder (2000) and Bröder and Schiffer (2006), between 28 and 53 % of the subjects were classified as using take-the-best. Other studies found even lower numbers, for example Glöckner and Bröder (2011) found that less than 10 % of their subjects used recognition, and only about 15 % used take-the-best. Concerning the recognition heuristic, Pachur et al. (2008) found that about half of their subjects used recognition in a strict sense (without incorporating additional information).

This limited extent of actual use of heuristics has led researchers to study factors that could influence the application of heuristics. Gigerenzer and Goldstein (2011) thus noted that decision makers seem to apply different decision strategies, but little is known about the factors that influence their choice. One framework to explain the choice of different approaches to decision making is the effort-accuracy framework by Payne et al. (1993). According to this framework, a decision making strategy is selected by considering the trade-off between the cognitive effort needed to make a decision, and the quality of the decision, since methods leading to better decisions typically also require higher effort. However, since the fast and frugal heuristics we consider here are all characterized by very little cognitive effort, this framework cannot be applied to the choice between these heuristics.

Two types of factors can be distinguished: factors that refer to individual characteristics of the decision maker, and factors related to the decision problem or the environment. Research on individual factors has indicated that intelligence moderates the choice of decision strategies such as heuristics (Bröder 2003). Researchers also studied the impact of the Big Five personality characteristics on the use of heuristics and found that persons high in neuroticism are more likely to use the recognition heuristic (Hilbig 2008). Furthermore, additional knowledge (beyond recognition) a decision maker has about alternatives has some impact on choices (Pohl 2006), although its impact seems to be limited (Hilbig et al. 2009).

The type of environment in which a decision is made has also an impact on the decision making strategy (Bröder 2003). In particular, a high amount of uncertainty present in the environment seems to increase reliance on heuristics like take-the-best (Hogarth and Karelaia 2006; Newell et al. 2003). The choice of strategies also depends on the availability of additional cues, their validity and the cost at which such information can be obtained. If the validity of the primary cue is high, take-the-best is more often used (Newell et al. 2003), while higher validity of other cues reduces its usage (Newell and Fernandez 2006). Some authors also report that the cost of additional information increases the use of heuristics (Newell et al. 2003), while others report that choices can still be better explained by alternative models even if information acquisition has high costs (Ayal and Hochman 2009).

Another important factor that influences the use of heuristics is time pressure. A positive effect of time pressure on the use of the recognition heuristic was found by Pachur and Hertwig (2006). Furthermore, Hilbig et al. (2012) found that time pressure increases reliance on the recognition heuristic even when violating time constraints does not have a negative impact on the decision maker. It thus seems that the mere presence of time pressure has a strong effect on the choice of a decision strategy. Time pressure was also used as an experimental factor to induce heuristic choice behavior in consumer decision experiments, like for example Langner and Krengel (2013).

Since the presence or absence of time pressure is likely to induce different levels of use of heuristics for the same decision problem, we employ it as an experimental factor in the subsequent empirical study to trigger the use of heuristics in decision making.

In contrast to factors related to the choice problem, research on individual factors that might trigger the use of heuristics is so far limited. However, decision making behavior in general has been studied more intensively, and different classifications of decision making styles can be found in the literature. Many of them refer to specific settings like shopping situations (Sproles and Kendall 1986). For this paper, we use a more general classification of decision making styles developed by Scott and Bruce (1995). They distinguish five dimensions of a decision making style:These five styles represent different dimensions of a decision making process. The instrument of Scott and Bruce (1995) does not classify decision makers uniquely into one style, but assigns each decision maker a score in each dimension indicating how much that decision maker’s individual style resembles each of the five prototypical styles.

The instrument developed by Scott and Bruce (1995) has been extensively used in studies on decision making behavior. Among them, and in direct relation with the present paper, Crossley and Highhouse (2005) used a variant of the instrument to classify different approaches applicants use in searching for jobs. Curseu and Schruijer (2012) studied the relationship of decision making styles to actual decision performance and found that subjects scoring high in the rational dimension were less affected by decision biases, and subjects scoring high on the avoiding style were more indecisive. Similarly, de Bruin et al. (2007) found a positive relationship between rational decision making style and decision competence. Gettinger et al. (2013) used the instrument to classify users of a decision support system and their preferences for different graphical information representations.

Although many of the items used in the instrument by Scott and Bruce (1995) describe decision making behavior that either reflects the use of heuristics (e.g., “When making decisions, I rely on my instincts”, item I1 for the intuitive style), or reflect the avoidance of heuristics (e.g., “My decision making requires careful thought”, item R3 of the rational style), to the best of our knowledge the relationship between decision making styles and the use of fast and frugal heuristics has not been studied before. We therefore in the next subsection formulate research questions to explore this relationship. There we also consider time pressure as a problem characteristic that can trigger the use of heuristics in general.

The relationship between time pressure and the use of heuristics has been previously studied in literature (Hilbig et al. 2012), but for the sake of completeness, we still formulate it as a research question:

Bröder and Schiffer (2006) test the effects of the presence of verbal information on the use of several compensatory (Franklin’s Rule, Dawes’s Rule) and one non-compensatory (take-the-best) heuristic. They find that cognitive load enhances the usage of take-the-best irrespective of the stimulus format. As take-the-best may be considered the easiest way to reach a decision that is at least partly rational, we expect rational decision makers to switch from a thorough analysis of the problem to a take-the-best heuristic under time pressure.

Marketing scholars suggest that recognition based decision rules are particularly applicable for purchase decisions regarding frequently purchased product categories (Hauser 2011). Recognition based decision rules can be considered a comparatively easy way to reach a decision, and do not require acquisition of further information on product attributes. Since avoidant decision makers tend to postpone decisions, it is reasonable to assume that they will also try to circumvent the initial pre-decision step of acquiring additional information. Therefore, it can be assumed that avoidant decision makers will make more use of the recognition heuristic than decision makers who score low on the avoidance dimension, and that this relationship will hold both under time pressure and without time pressure. The same might hold for dependent decision makers, since well-known products are also those products which are bought by many other consumers.

Spontaneous decision makers tend to make fast and spontanous decisions, but do not necessarily rely as much on intuition and gut feeling as intuitive decision makers. Thus, they will likely use an approach that is based on some, but only limited information about the alternatives. Their decision making behavior will therefore more likely match the take-the-best heuristic, which bases a decision on factual information about one attribute, rather than, for example, the recognition heuristic, which does not directly consider any attribute of the decision alternative. We also consider it unlikely that spontaneous decision makers will engage in the heavy cognitive processing required to find an objectively best alternative.

We conducted two identical experiments, one in Vienna, Austria, in May 2014, and the other one in Madrid, Spain, in October 2014. Students and staff from the Faculty of Business, Economics and Statistics of the University of Vienna (Austria) and from the Economics and Business School of the Complutense University of Madrid (Spain) were asked to voluntarily participate in the study as well as spread the questionnaire link among their family and friends in order to get a wider sample.

The Austrian sample consisted of 178 respondents, thereof 63 % females. Participants’ age ranged from 13 to 69 years with an average of 28 years. The sample contained 48.9 % persons in education, 33.7 % white-collar workers and employees working in public institutions, 3.9 % blue-collar workers, 5.6 % self-employed, 0.6 % retirees and 7.3 % unemployed. Participants were recruited for the experiment via two channels. On the one hand, respondents were recruited through social media platforms and e-mail, on the other hand, students participated in exchange for course credits.

An average of 5.57 for product liking can be reported on a seven point scale. The 81 respondents who were randomly assigned to the time-pressure condition, spent on average 36.57 s on the respective page. In contrast, the 97 respondents in the non-time pressure condition spent on average 75.05 s. With respect to the frequency of product purchase, almost 40 % of respondents purchase eggs a number of times per month. Approximately 30 % specified a purchasing frequency of once a week, followed by 12 % of respondents, who purchase the respective product only once a month. Almost 8 % of participants buy egg cartons less than once a month, 4.5 % several times a week and 3.4 % never. Respondents were also asked to indicate the frequency of their product consumption. The majority of participants (36 %) consume eggs several times a week, followed by 28.1 %, who indicated to consume eggs once a week. 24.2 % of participants specified a consumption frequency of a number of times per month.

A total of 142 Spanish respondents participated in the replication of the experiment, consisting of 51 % females and 49 % males. With an average of 27 years, respondents’ age ranged from 19 to 71. In terms of occupation, the majority of the respondents were in education (73.9 %), followed by 16.6 % white-collar workers and employees working in public institutions, 2.6 % blue-collar workers, 2.6 % self-employed respondents, 1.3 % retirees and a minority of unemployed participants. Product liking, with an average of 5.83, is very similar to Austria.

The Spanish sample, similarly to the Austrian one, spent on average more time on the respective questionnaire in the condition without time pressure. Overall, 84 respondents were assigned to the condition without time pressure and 73 with time pressure. The majority of the Spanish respondents (46.5 %) purchase eggs once a week, 28.03 % a few times per month, and approximately 10 % once a month. Frequent egg purchasers (daily, a few times per week) and respondents who purchase eggs less than once a month or never constitute a minority. When comparing the egg consumption of Spain with Austria, it can be argued that Spanish respondents consume (and purchase) more eggs. Almost 60 % of the Spanish respondents consume eggs several times per week, approximately 20 % once a week and 15 % a few times per month. A \(\chi ^2\) test indicates that differences in egg consumption between the two countries are significant at p < 0.1 %, and differences in purchase behavior are significant at p < 5 %.

Since most groups except students and white-collar workers were only weakly represented in the sample, we decided to drop those groups altogether to obtain a more homogeneous sample. Furthermore, we dropped respondents who obviously were just clicking through the online questionnaire and thus we eliminated all respondents who had answered 7 to all four questions about attitude, and to more than ten questions of the decision style questionnaire. The resulting reduced samples were used for all further analyses. For the experiments in Austria, the reduced sample consisted of 143 respondents, 92 (64.3 %) of which were female and 51 (35.7 %) were male. The final sample from the experiment in Spain consisted of 128 respondents, out of which 66 (51.6 %) were female and 42 (48.4 %) were male.

Since we had to translate items for the decision making styles into German and Spanish, respectively, we first performed an exploratory factor analysis to analyze the factor structure of responses. Because decision making styles are not necessarily orthogonal, we used an oblimin rotation for this analysis.

Table 1 shows the factor loadings for data obtained in Vienna. For clarity, this table not only shows the factor loadings exceeding the usual threshold of 0.4, but in parentheses also factor loadings exceeding 0.3. Most items load as expected, only two items had higher loadings on other factors than the expected ones. Item S5 “When making decisions, I do what seems natural at the moment” loads onto the factor associated with the intuitive style with a factor loading of 0.45 (above the 0.4 threshold), rather than the spontaneous style. However, it should be noted that also in the original paper by Scott and Bruce (1995), the spontaneous style was added at a later stage, and the item in question is somewhat ambiguous. Item D5 loaded on the factor associated with the rational style with a factor loading of 0.35, which is below the threshold. Its loading in the factor associated with the dependent style was not much lower at 0.28.

Similar results were obtained for the data from Madrid (Table 2). Again, item S5 does not exhibit a high loading on the factor corresponding to the spontaneous style, but loads (at 0.37) onto the factor corresponding to the intuitive style. Item I4 did not load on any factor with a loading of more than 0.3, its highest factor loading was still on the factor associated with the intuitive style at 0.25. Given that these deviations from the theoretical factor structure are quite small, we decided to retain the original scale of Scott and Bruce (1995).

The distribution of scores in the decision making styles for the subjects in the two experiments is shown in Fig. 2. The pattern of decision making styles is quite similar across the two experiments. Subjects on average score rather high with respect to the rational style, and low with respect to the avoiding and spontaneous styles. In both experiments, the avoiding style also shows the highest variance.

Table 3 presents an overview of product choices in the entire sample.

Figure 3 shows the distribution of choices across the two experimental conditions. In both experiments, applying time pressure led to a smaller number of subjects choosing the “cognitive” product, and an increase in the share of the products fitting the take-the-best and recognition heuristics. Under both experimental conditions, subjects in Vienna selected the “cognitive” and “recognition” product more often than subjects in Madrid, and the “take-the-best” product less often. The emotional and filler products were chosen only rarely in both experiments, and the filler product only under time pressure.

In order to obtain a first impression of the relationships between time pressure, decision making styles and the use of different heuristics, we performed a Multiple Correspondence Analysis (MCA). This method is aimed specifically at quantifying categorical data (Greenacre 2007), that is, assigning numerical scale values to the response categories of several discrete variables, with certain optimal properties. These scale values have been shown to have interesting geometric properties and provide maps of the relationships between variables. The ultimate aim of MCA is to produce a map of those relations in a similar way to principal components analysis, in that the total variance of the variables is defined and then this total is decomposed optimally along so-called “principal axes”. The total variance in MCA is measured by the so-called inertia, which is simply the usual Pearson chi-square statistic calculated on the cross-tabulation, divided by the total sample size n. It is this inertia which measures the degree of difference between the values of the variables that we are trying to represent optimally in the eventual map. For mapping purposes, it is usually hoped that a large percentage of the total inertia is accounted for by the first two principal axes, thereby allowing the relations to be visualized in two dimensions.

The application of MCA on our data showed that the percentage of total inertia explained by the first two dimensions is equal to 53.39 %. As 41.80 % is due to the first axis, this is the more explicative dimension. A visualization of the results is presented in Fig. 4. The first axis reveals an opposition between the two cities where the experiment was undertaken (Madrid and Vienna) while the second axis, which explains a lot less variability (11.60 % of the total inertia), specifies the time pressure of the experiment. Therefore the horizontal direction in the map separates respondents on the basis of the city. The vertical dimension differentiates respondents on the basis of time constraint.

The proximity between levels of different nominal variables means that these levels tend to appear together in the observations. Different colors have been used for each of the variables to facilitate readability. So the top-right quadrant of the plot clearly shows that the categories Time Pressure, White collar, Recognition and Spontaneous are associated. In the case of the same nominal variable the proximity between levels means that the groups of observations associated with these two levels are themselves similar, but also that a variable discriminates better to the extent that its category points are further apart. Proceeding clockwise, Vienna, Female, Cognitive and Avoiding and Dependant are associated. Time unlimited and In education are associated. Finally there seems also to be a relation between Madrid, Male, Take-the-best, Emotional and Filler and Rational.

The MCA clearly indicates that the two cities are a major factor in explaining the differences of outcomes. We therefore decided not to pool data from the two experiments, but analyze them separately. All following analyses will thus be performed separately for the two data sets.

Given that the emotional and filler products were chosen so rarely, we decided not to include them in the following analyses. In order to analyze our research questions, we estimated four nested multinomial regression models for each experiment. The first model (M1) contained only the control variables Gender and Age. The second model (M2) added our treatment variable Time pressure. In the third (M3) and fourth (M4) models, we added decision styles and their interaction with the treatment variable, respectively. In all these regressions, the “cognitive” product was used as reference category, so regression coefficients indicate how strongly each factor influences the switch from “cognitive” to the indicated other product. Tables 4 and 5 show the regression results for the two experiments.

In contrast, in the Vienna experiments, we mainly find effects of decision styles which are not influenced by time pressure. Surprisingly, subjects who score high on the intuitive style are significantly more likely to choose the “cognitive” product than either of the other two products, and subjects who score high on the rational style choose the “take-the-best” product more often, rather than the “cognitive” product. More in line with our expectations formulated in RQ4, we also find that subjects who score high in the spontaneous style are more likely to choose the “recognition” product instead of the “cognitive” one. In neither data set we find the relationships between avoidant and dependent styles and the recognition product to which RQ3 refers.

Finally, to take into account the differences between choices of the “take-the-best” and “recognition” products, we also estimated multinomial regression models using these products as reference categories. Figures 5 and 6 summarize the significant results of all estimations. Please note that arrows in these figures do not represent influences from one of the products to the other, but indicate that the factor indicated in the text on each arrow causes a significant shift in choices from one product to another.

For the Vienna data, there are no significant effects causing a shift between “take-the-best” and “recognition” products. In the Madrid data, time pressure in general causes a shift from “take-the-best” to “recognition”, which is partly offset for subjects who score high on the rational style.