As outlined in the Discussion of Chapter II, the diverging results could not be attributed to random error across the experiments. Therefore, one needs to look for systematical sources of variance. Proceeding from a mere estimation of the effect size to a more detailed analysis of explanatory third variables will also lead to a deeper understanding of the psychological mechanisms underlying the effect of too much choice and it will provide a necessary basis for establishing a process model that will allow us to predict when and why the effect will occur. As a first step toward this goal I will start by introducing a conceptual differentiation of third variables into moderator and mediator variables.
According to Baron and Kenny (1986), a moderator is a third variable that affects the direction and/or strength of the relation between an independent and a dependent variable. With regard to the effect of too much choice, a moderator would be a variable that specifies the appropriate conditions or boundaries in which the number of options within an assortment (=independent variable) affects the motivation to choose and/or satisfaction with the chosen option (=dependent variable). As such, there need not be a direct influence of the moderator on the dependent variable; only the effect of the interaction between the independent variable and the moderator matters. Besides this, no assumptions about the underlying processes or mechanisms are made. This is conceptually different from the notion of a mediator variable, which tabs directly into the intervening entities or processes. In the latter case, the independent variable is assumed to cause the dependent variable through the mediator and therefore the mediator needs to be related to both the independent and the dependent variable.
Searching for mediators explicitly aims to flesh out underlying psychological processes. Therefore, the approach seems to be generally more appropriate for the goal of testing and developing theories. Yet, as pointed out by Baron and Kenny, search for mediation is best done in the case of a strong relation between the independent and the dependent variable. If this relationship is weak or inconsistent, a search for moderators is recommended. Once moderators are identified that help to establish a consistent relationship between independent and dependent variables, the ground is prepared for testing intervening mediators.
In the remainder of this chapter, I will systematically identify potential moderator and mediator variables that might help to explain why some studies, including my own, did not find any effect of too much choice while others reported strong effect sizes. I will then strive to test some of the most promising of these moderators and mediators in a series of controlled choice experiments. In these experiments, I will specifically focus on the influence of option complexity, the number of options in the large choice condition, the average attractiveness of the small assortment, cultural differences, and individual differences in personality, expertise, and search behavior, all of which will be outlined in more detail below.
This study was designed to test the influence of three variables, namely, the moderating effect of option complexity, the moderating effect of option attractiveness, and the perceived variety as a mediator.
As mentioned above, scholars have argued that choice overload might not occur for trivial decisions among very simple options (Dhar, 1997). With regard to the difference between Iyengar and Lepper’s (2000) chocolate study and the jelly bean study, I argued that a choice between a piece of chocolate and a jelly bean is of equal (un)importance. However, one could conjecture that the stimuli at least differed slightly in complexity. Jelly beans only differ in color and taste, whereas Godiva chocolates also differ in texture, shape, and filling. The higher option complexity of the chocolates is also reflected in their names. While the chocolates had rather long names (e.g. “Grand Marnier Truffle”), the jelly beans could sufficiently be described with one or two words (e.g. “Strawberry”).
To test the possibility that option complexity is indeed a moderator of too much choice, the present study explored the possibility that the options have to exceed a certain level of complexity and that they have to vary on several possibly conflicting attributes in order to elicit choice overload. A potential moderating role of option and assortment complexity would link choice overload to previous research on the related concept of information overload. I will come back to this relationship in more detail in the general discussion in Chapter V.
In the present study I also tested the hypothesis that a too-much-choice effect is more likely to be found if the average attractiveness of the large set is lower than the average attractiveness of the small set, which can happen if the small set only consists of very attractive options. I raised this possibility earlier in Chapter II when discussing the composition of the small assortment in the jam study. There, I conjectured that Iyengar and Lepper might have found the effect because by chance they ended up with a sub-sample of very attractive jams in the small assortment.
In an extension to the work of Kahn and Wansink (2004), who found that it is the perception of variety rather than the mere number of options, Broniarczyk, Hoyer, and McAllister (1998) found that the perception of variety mainly depends on how much physical space an assortment takes up on the shelf of a store and not so much on the actual number of different options. As a consequence, in a series of studies in the lab and in the field, the removal of up to 25% of the options (they used microwave popcorn as stimuli) did go unnoticed as long as shelf space was held constant and the most popular brand was available. Broniarczyk et al.’s results again show that the perception of variety does not solely depend on the mere number of different options but also on other, maybe less obvious characteristic of the environment as well as the expectations and prior experiences of the individual decision maker. As such, the results stress the need to control for perceived assortment size as a potential mediator of choice overload.
In line with this, Huffman and Kahn (1998) argued that perceptions of high variety do not necessarily depend on the structure of assortment but can also be determined by how selective people are in their perceptions. As an example, one might think of an expert who, within his or her field of expertise, can process more options than a layperson (Chase & Simon, 1973).
To test for the influence of option complexity, in the experiment on hand, participants had to choose among options that were described on several attributes and thus constituted a more multifaceted selection than, for example, jelly beans or jams. Furthermore, in the experimental design on hand subjective perception of variety was controlled for and, as I will outline in more detail below, the setup also allowed me to test if the effect depends on the attractiveness of the small choice set.
Participants were asked to browse through a list of restaurant descriptions. For each restaurant, participants were asked if they knew the place and if they had eaten there before. Next, participants were entered into a lottery with a 1 in 40 chance to win. They were instructed that they had to choose whether they would want to receive 30 euros in cash ($36 at that time) or a restaurant coupon worth 40 euros ($48) if they won. Participants were further told that those who chose the coupon had to pick one restaurant from the list, for which the coupon would be issued. The percentage of participants choosing the restaurant coupon over the cash was used as a measure of motivation.
The main independent variable was the number of restaurants presented to the participants. In the large choice condition, 30 restaurants were presented, spread equally over the following five cuisines: Italian, Asian, German, French, and International. The small choice set consisted of 5 restaurants, one of each cuisine type.
The restaurant descriptions were taken from a recent restaurant guide for the city of Berlin (Marcellino’s Restaurant Report 2005). Only restaurants in the district Berlin Mitte, which marks the center of Berlin, were selected. The average price for a dish in all restaurants ranged between 10 and 20 euros ($12–24). As Marcellino’s Restaurant Report often lists more than five restaurants per cuisine type for this district, restaurants were chosen to ensure an equal distribution of food prices among the different cuisines.
To ensure a certain complexity of the options, each restaurant was described on a small sheet of paper using a short narrative description and a numerical rating for the quality of food, drinks, service, and atmosphere, all taken from the restaurant guide. The participants were also told the name, address, and type of cuisine for each restaurant. The descriptions were assembled in booklets in two different random orders, together with a cover sheet reading “Restaurant Descriptions.”
To test whether choice overload is more likely if the small set consists of very attractive options, I ran the large choice condition first to find out which restaurants were chosen most often (and thus were most attractive). The small choice set was then constructed from the most attractive restaurant for each cuisine. Thus, it had a total of five presumably attractive restaurants to choose from. Given that in my previous studies I did not find any too-much-choice effect, I hoped that these manipulations would increase my chances of finding the effect.
To control for perceived variety, participants were asked to judge the assortment size of the offered restaurants on a Likert scale ranging from 1 (“very little choice”) to 7 (“a lot of choice”). The middle of the scale was described as “average number of choices.” Participants who preferred a restaurant coupon over a cash coupon were also asked how difficult it was to choose their one restaurant on a scale of 1 (“very easy”) to 7 (“very difficult”).
To control for the factor that the attractiveness of the restaurant coupon might decrease the further people lived from the center of Berlin, participants were asked in which part of the city they lived and in which part they worked or studied.
In total, 80 people participated in the restaurant study, 40 in each choice-set-size condition; most of them were students from a local university. The mean age was 25 years, ranging from 19 to 33 years with no significant differences between the two experimental conditions. Within both conditions, half of the participants were male and half were female.
The size of the large assortment was perceived as rather high (4.9 on a scale from 1 to 7; SD=1.4) and larger than the size of the small assortment (4.9 vs. 3.1), t(78)=5.61; p<0.01. Thus, the experimental manipulation was successful in terms of perceived set size. Participants who chose from the large assortment also reported slightly more difficulty in making that choice (3.5 vs. 2.4 in the small assortment), t(78)=1.9; p=.066.
Because I used descriptions from real restaurants, I wanted to know how familiar participants were with my stimuli: In the large choice condition, the participants had never heard of 80% of the restaurants on average (24 out of 30; SD = 4.2) and had never eaten at 95% of the restaurants (28.5 out of 30; SD = 2.9). In the small choice condition, on average participants had never heard of 75% of the restaurants (3.7 out of 5; SD = 1.2) and had never eaten at 86% of the restaurants (4.3 out of 5; SD = 0.8).
However, the number of recognized restaurants (the familiarity with the choice set) was not related to the probability of choosing a restaurant coupon, nor correlated with the difficulty of choice or the perceived assortment size. And so the familiarity with the choice set did not seem to have an influence on the dependent variable.
For the main dependent variable, the number of people who preferred a restaurant coupon over a cash coupon, there was hardly any difference between the large and the small choice set: From the large set, 14 out of 40 participants (35%) chose a restaurant coupon while from the small set 12 out of 40 (30%) chose a restaurant coupon. Thus, if anything, people were more likely to choose a coupon from the large assortment. According to Cohen (1977), the difference between these proportions corresponds to an effect size of d=−.14.
Across both set sizes, participants who chose a cash coupon did not perceive the assortment of restaurants as larger compared to those who chose a restaurant coupon (3.9 vs. 4.3), t(78)=1.02; p=0.31. This indicates that the subjective perception of variety also did not seem to influence people’s propensity to make a choice.
As in the experiments reported in Chapter II, I did not find any effect of too much choice on the likelihood of making a choice. The result is especially surprising as this time the setting was well controlled, the set-size manipulation was successful, the options were complex, the decision was not trivial, and the small set was highly attractive—all measures that should have increased the chances of finding the effect. Again, the question arises as to what distinguishes my experiment from those where an effect was found.
The data on hand indicates that the restaurants I used in my experiments were mostly unknown and the familiarity with the choice set did not influence the effect. Nevertheless, choosing a restaurant is not that uncommon in daily life and therefore participants might still have had specific prior preferences on what types of restaurants or cuisines they like in general. As mentioned above, it could be that these prior preferences enabled them to engage in a decision process in which they simply matched their preferences. Also, it could be that the lottery at the end introduced an additional source of error due to individual differences in risk-taking behavior. Finally, as mentioned above, thus far there is no definition of what constitutes too much choice and as a consequence it could be that 5 different options in the small set were already sufficient to induce choice overload or that 30 different restaurants are not enough. To rule out these explanations, I conducted a series of experiments that involved concrete choices among a wider range of options that are less common in everyday life, namely, public charity organizations.
The present study marks the first in a series of three experiments that aimed to rigorously test the effect of varying assortment sizes on the motivation to make a choice. All experiments involved real choices in which participants could either donate a certain amount of money to a charity organization or keep the money for themselves. In contrast to the selection of a restaurant, choices among different charity organizations are much less common. To further control for the moderating effect of clear preferences or evaluation standards prior to choice, the awareness level of the charity organizations within the choice set was subject to experimental manipulation.
To make the choice real, participants who had come to our lab to participate in other experiments received a 1-euro coin and a sheet with charity organizations listed in alphabetical order. Participants could choose if they wanted to donate the money to one of the organizations on the list or keep the money themselves. If they decided to make a donation they had to check the name of the organization that the money should go to. To make the choice reasonable and to ensure a certain degree of complexity, each organization on the list was described by its name and a list of keywords indicating its mission. The charities were sampled from the population of all 180 German organizations that complied with the standards of the German charity seal (Deutsches Spendensiegel), an association that certifies trustworthy charities. From this set, I created the large-assortment list from the 30 leading (and presumably most well known) charities according to the amount of money they collected in 2004. To ensure that the small-assortment list would be small enough to prevent overload, it consisted of just the 2 biggest charity organizations. To further control for prior preferences and to alternate the set sizes, I created an additional large list that consisted of the 40 smallest (and presumably least known) charity organizations and a corresponding small list consisting of the 5 smallest organizations.
To rule out demand effects, participants made their decision anonymously. Anonymity was ensured by putting the instructions and the list of charities into an envelope. Participants in the experiment were asked to open the envelope and to follow the instructions in a separate booth that was set up in a corner of the lab. After completion (including indicating their sex and student status), they were instructed to put the list, and if they had made a donation also the euro coin, back into the envelope, seal it, and throw it into a “ballot box” within the booth. The envelopes with the different lists were mixed and indistinguishable from the outside so that the experimenter who gave out the envelopes was blind toward the experimental condition. At the end of the study, I transferred the money donated by the participants to the charities that they had indicated.
In total, 120 people participated in the study, 30 in each of the four conditions (with three sheets not filled out completely). There were 68 women and 49 men and a total of 90 participants (78%) were enrolled as students.
In the conditions based on the lists of leading charities, 28 out of 30 participants who saw the large-assortment list (30 charities) chose to donate (93%). Out of the 30 participants who saw the small-assortment list (2 charities), 25 chose to donate (83%). With an alpha level set at 0.05, the comparison of the two proportions revealed no statistically significant difference, t(58)=-1.2; p=.235; Cohen’s d=-0.53.
In the conditions based on the least known charities, 20 out of 28 participants who saw the large assortment list of 40 charities chose to donate (68%). Out of the 29 participants who saw the small assortment list of 5 charities, 19 chose to donate (66%). As in the previous conditions, the comparison of the two proportions revealed no statistically significant difference, t(55)=-0.18; p=.855; Cohen’s d=-0.17. If anything, in both sets of charities, participants were more likely to donate when facing the large assortment. There was an effect of the type of charity such that well-known charities received more money (88%, vs. 68% for the low-profile charities), t(115)=2.7; p=0.08. Figure 5 gives an overview of the main results.
To test if participants simply ignored the charity organizations at the end of the list, thus effectively shielding themselves from having too much choice, I correlated the position of the organization on the list with the number of donations it received. A positive correlation would indicate that charities on the top of the list would have a higher probability of receiving money. Yet for both the long list with well-known charities and the long list with least-known charities, the correlation was virtually zero.
|Figure 5: Proportion of participants who gave to charity depending on assortment size|
The experimental design on hand ruled out demand effects and it involved real choices among authentic and fairly complex options. Also, it was not an everyday choice, the number of options varied widely, and at least for the condition with least-known charities it was unlikely that participants had prior preferences. Despite all these measures, again I did not find any effect of assortment size on choice motivation. That there was an effect of the type of charity such that more people decided to give to leading charities suggests that people did not choose randomly and that the decision task was reasonable. Given the vague definition of what constitutes too much choice, it could still be the case that when it comes to choices between charities 40 options are not sufficient to elicit the effect. To test for the possibility that the number of options needs to be increased even further, I conducted another charity study in which I doubled the number of options in the large assortment.
The purpose of the study on hand was twofold. First, following up on the discussion of the charity study above, there is a necessity to explore even larger assortment sizes. This is because, as mentioned in Chapter I, there is no clear definition of what constitutes extensive choice and how many options are needed to elicit the effect. Because of this, the second charity study circumvents this problem by exploring a wider range of assortment sizes. Second, I aimed to test if the effect of too much choice depends on cultural differences between Germany and the United States. As I will outline in more detail below, there are a number of differences between the two countries with regard to choice and the perception of assortments that might mediate the effect.
As mentioned in Chapter I, Iyengar and Lepper (2000) defined extensive-choice conditions as “reasonably large, but not ecologically unusual” (p. 996). These terms are difficult to pin down: How does one define “ecologically usual” for cases that are seldom encountered in everyday life? The same difficulty holds true for the vague term “reasonably large.” In any case, it can be assumed that the usual sizes widely differ depending on the context.
Almost all studies that report an effect of too much choice were conducted in the United States, whereas all my studies reported so far were conducted in Germany. Both countries are highly developed market democracies, but there are also a number of differences between them that might explain the diverging results.
The only scholars who found an effect of too much choice outside the United States were Reutskaja and Hogarth (2005). As outlined in Chapter I, they studied hypothetical choices among different numbers of gift boxes that were displayed on a computer screen. What is noticeable about their study is that they collected data in Spain but also in two countries in Eastern Europe, namely, Belarus and Ukraine. Due to the difficult economical situations, consumer choice in these countries is noticeably smaller than in Western Europe and in the decades prior to 1990, choices were even more scarce. What Reutskaja and Hogarth found was that on average, more options were needed to elicit choice overload in the Eastern European sample as compared to the Western European sample. This suggest that in environments in which choice is scarce, people might be more excited to have large assortments, while in environments in which variety is extensive, people might become less and less attracted, saturated, and maybe even irritated by large assortments.
On the other hand, a recent study on how people evaluate variety (Rozin, Fischler, Shields, & Masson, 2006) suggests that at least in the comparison between the United States and Germany it might be the other way around. In that study, a representative sample of 1,450 participants in the United States and 851 participants in Germany were asked whether they would prefer an ice cream parlor with 10 different flavors or 50. The results showed that 56% of the Americans would rather go to the parlor with 50 ice creams, as compared to only 33% of the German sample. When asked about their expectations of the selection of dishes at a top-class restaurant, 36% of the Americans answered that they would expect a large choice with numerous different dishes rather than a small number of suggestions from the chef. Among the Germans, the percentage of people expecting a large choice was only 22%. Based on their results, Rozin and colleagues conjectured that at least in the food domain, there is a preference for quantity over quality in the United States, while in Germany it might be the other way around.
Again, it is not obvious how these findings translate into different motivations with regard to too much choice. On the surface, one would rather expect that Americans, who seem to value large assortments, would be more likely to make a choice from a large assortment.
To further explore the potential moderating effect of cultural differences on the effect of too much choice, I also ran an experiment in the United States. To allow for a straight comparison to the results in Germany, the U.S. experimental setting closely resembled that of the charity study conducted in Berlin.
As in Germany, the U.S. study was administered following other, unrelated psychological experiments ranging in duration from 15 minutes to 1 hour. Similar to the German study, participants received an envelope containing instructions, a list of charities, and a 1-dollar bill. Participants could either donate the dollar to one of the charities on the list or keep the money for themselves. The decision was made in private, and after completion participants dropped the envelope into a sealed box. The list of charities contained either 5, 40, or 792 different organizations.
The organizations on the list were sampled from the U.S. website charitynavigator.com, a nonprofit organization similar to the German Charity Seal that lists and evaluates U.S.-based charities that have operated for at least 4 years, that have been granted tax-exempt status, and that make their accounting information publicly available. At the time I conducted the study, charitynavigator.com listed a total of 204 organizations that had a national scope of work related to either animal, educational, or environmental issues. From this sub-set, I drew a stratified sample of 80 organizations (27 environmental, 27 educational, and 26 animal charities). From this large sample I then drew random sub-samples for the 5- and the 40-option condition such that the proportions of environment, educational, and animal charities were equal. As in the German study, the charities were listed alphabetically. Besides their name, they were also described by one sentence about their mission.
A total of 112 envelopes were administered, 36 in the 5-option condition, 37 in the 40-option condition, and 39 in the 79-option condition. Of the participants, 58 were male, 51 were female, and 3 did not indicate their gender. Participants’ mean age was 20 years, ranging from 18 to 28 years. From the total number of 112 participants, 95 (85%) decided to give to charity. Of those who made a donation, 14 (15%) indicated that they had heard the name of the organization of their choice before, which suggests that the vast majority of participants were indeed unfamiliar with the choice set and thus could not engage in preference matching.
In the 5-option condition, 29 participants (81%) gave to charity, in the 40-option condition, 32 participants (87%) donated, and in the 79-option condition, 34 (87%) gave to charity. Thus, if anything, an increase in the assortment led to more choice. For the comparison between 5 and 40 charities, the effect size is Cohen’s d=-0.16. For the comparison between 5 and 79 charities, the effect size is Cohen’s d=-0.18.
The probability of choosing a certain charity did not depend on the position of the charity in the list. The correlation between the number of donations a charity received and its position on the list was virtually zero for the lists of 40 and 79 options.
The results closely resemble those found for the first charity study that was conducted in Germany. Again no effect of assortment size on choice motivation was found, and if anything, the propensity to choose increased with assortment size. Also, charities that were at the beginning of the list were not more likely to receive a donation than charities at the end. This suggests that participants indeed looked through the whole list and deliberately selected an option. These data provide some preliminary evidence against the idea that participants sequentially went down the list and chose the first satisfactory option that they came across, as a satisficing decision strategy would suggest. As outlined in Chapter I, applying a fast and frugal heuristic such as satisficing should shield people from being overloaded with choice. If anything, not using such a satisficing heuristic should increase the vulnerability to choice overload. Perhaps if the list had been even longer, such a search strategy might eventually have been observed.
Participants in the present study were undergraduate students at a U.S. university and thus presumably similar to the participants in most of the previous studies outlined in Chapter I that successfully found an effect of too much choice. Given that I did not find an effect in such a sample, the role of culture or population differences as a moderator is called into question.
Based on the charity paradigm outlined above, the present study was designed to further increase the power of finding the effect by asking people to justify their choice. Furthermore, to allow for a direct comparison to the U.S. data, the experiment on hand also includes an extended set size of 80 charities.
From a social perspective, active choices are often more difficult to justify than omissions (Ritov & Barron, 1990). As a consequence, decisions may at times be avoided due to an anticipation of blame. It seems likely that an increasing number of alternatives also increases the difficulty of justification because the decision made has to be defended against more alternatives. Furthermore, Fasolo, Huber, Hertwig, and Ludwig (2007), examining real-world assortments in actual stores, showed that options get more similar to each other as the assortment size increases. It is possible that any justification of choosing a single option would become even more difficult because it is more likely to apply to more than one alternative and thus will not be sufficient to single out one solitary option. At the same time, the difficulty to justify no choice or the choice of a default option remains the same.
For example, in a choice between a red and a green apple, red color would serve as a distinctive reason to choose one apple over the other. As the number of apple types increases there will eventually be two red apple varieties and one would need a more sophisticated reason to justify the choice. On the other hand, the justification of no choice would be unaffected by an increase in assortment size (e.g. one could always argue that one prefers oranges, irrespective of the number of apples). Taken together, people might be more likely to resign from the choice if they know that that they will be asked to justify their choice, which should increase the chances of finding the too-much-choice effect.
The experimental design closely resembled that of the former charity studies. As before, participants received an envelope with a list of charity organizations in alphabetical order and were asked if they wanted to donate 1 euro to one of the organizations on the list. Again, charities were sampled from the German charity-seal-approved set. Participants were randomly assigned to one of three different conditions with either 5, 40, or 80 of the smallest-sized and presumably least-known charity organizations listed by Deutsches Spendensiegel.
To include a justification of choice, participants who gave to charity were asked to write down a short statement explaining the reason why they chose this organization in particular and not another one. To make the choice even more real, the study was attached to an unrelated experiment on Bayesian reasoning for which participants received 26, euros as compensation. After receiving the money, the envelope that contained the instructions and the charity list was handed to them. If participants decided to give 1 euro to charity, they had to take it from what they earned in the previous experiment. Likewise, making no donation maintained the status-quo of keeping the money and can be reasonably interpreted as no choice.
A total of 119 people participated in the study, 67 male, 47 female, and 5 who did not indicate their gender. In total, 72% of all participants were students. Of the 42 participants who saw the 5-option assortment, 37 (88%) gave 1 euro to charity. In the 40-option condition, 28 of 39 participants (72%) donated, and in the 80-option condition, 28 out of 38 participants (74%) donated. The difference in proportions between the small assortment of 5 and the large assortment of 40 options is 16% (88%-72%). Between the small assortment of 5 and the large assortment of 80 the difference is 14% (88%-74%). Under the null-hypothesis that no too-much-choice effect exists, the probability of finding differences of this magnitude or higher is p=.033 (z=1.8; Cohen’s d=0.4) for the comparison with the 40-option condition and p=.05 (z=1.7; Cohen’s d=0.37) for the comparison with the 80-option condition. Thus, for both comparisons there seems to be a small yet statistically significant (p<.05) effect of too much choice. As in the previous studies, the correlation between the number of donations a charity received and its position on the questionnaire was virtually zero in both large choice conditions.
The fact that an effect of too much choice occurred if participants had to give a reason for their choice supports the hypothesis outlined above that justification becomes more difficult when options become more similar and/or harder to distinguish from each other. In line with this interpretation, for those participants who gave a justification, the numbers of characters used to justify the decision were larger in the 40- and 80-option conditions as compared to the 5-option condition.
Of the 37 participants in the 5-option condition who gave to charity, 33 wrote down a reason for why they chose that particular organization. The mean number of characters used in these 33 justifications was 74 (SD=45). In the 40-option condition, 25 of the 28 participants wrote down a reason. Of these justifications, the mean length was 100 characters (SD=45). In the 80-option condition 24 of the 28 participants who donated wrote down a reason. Here, the mean length of all statements was 96 characters (SD=51). An analysis of variance with alpha set at 0.05 indicated that overall the differences are not statistically significant, F(79,2)=2.6; p=0.08, yet a post-hoc comparison of the small set with the two large sets indicates a statistically significant difference, t(80)=2.3; p=0.03.
Of course, the number of characters only provides a proximate measure of the difficulty of justification. Besides, even if the need for justification led to an effect of too much choice here, it is not obvious how this would explain the results of the studies by other researchers that did find the effect. In any case, none of these previous experiments asked participants for a reason or justification for their choice.
With regard to the moderate effect size of the main effect on choice probability, the positive result of the present study should be interpreted with caution. Given that I did not find an effect in any of the previous studies, the actual size of the too-much-choice effect in general—if it exists at all—might well be low. If so, the effect found in the present study might simply be due to random variation and the result would suggest a mediator that is in fact nonexistent. Figure 6 provides an overview of the results obtained in the Bloomington and the second Berlin study.
|Figure 6: Proportion of participants who gave to charity depending on assortment size in Bloomington and Berlin|
Taken together, the series of charity studies allowed for high experimental control and provided a realistic choice scenario, yet the data only reveal limited insight into the psychological processes underlying the choice. Apart from their age, we know little about participants’ personality traits, attitudes, and motives and it remains unknown how participants perceived the assortment, and how they derived their decision. As I will outline in more detail below, there is good reason to believe that these individual differences play an important role with regard to the effect of too much choice. To get a more detailed picture of the process and the decision makers themselves, I conducted another study that aimed to flesh out these variables.
The current study arose from the need for an experimental design that allowed for a more precise test of potential moderator and mediator variables on the effect of too much choice. Analogous to the previous studies, the main dependent variables in the study on hand were the degree of post-choice satisfaction, post-choice regret, and the motivation to choose.
Foreshadowing the difficulty of replicating the effect of too much choice, scholars in the past argued that the effect of choice overload is moderated by inter-individual differences in how people perceive a situation and how they go about making the choice. Among the most prominent of these personality constructs are the concepts of maximizing versus satisficing (Schwartz et al., 2002), and the need for cognition (Cacioppo, Petty, & Kao, 1984), both of which I will outline in more detail below.
The importance of individual differences with regard to the too-much-choice effect received empirical support in a recent study by Lin and Wu (2006), who found that participants with a low need for cognition (Cacioppo et al., 1984) were less confident about their decision when choosing from a large set of 16 options as compared to a small set of 6 options. In close resemblance to the study by Chernev (2003b) outlined above, in Lin and Wu’s between-subjects experiment, participants got to choose between different types of chocolate. Confidence, or preference strength as Chernev called it, was measured by participants’ propensity to switch their choice to another option that was recommended by the experimenters at the end of the study. Interestingly, people with a high need for cognition were less confident when choosing from the small set as compared to a large set. As a consequence, besides the interaction of need for cognition and assortment size, there was no main effect of assortment size on confidence. This suggests that in total across all participants, Lin and Wu did not find an effect of too much choice.
Lin and Wu explained the interaction effect between need for cognition and choice overload with differences in the way people adapt their decision strategy to the choice environment. They argued that people with a high need for cognition examine the options more thoroughly while people with a low need for cognition only do this if the assortment is small. For large assortments they expected the latter group to switch to a heuristic decision strategy such that less information is processed per option. They further argued that the heuristic leads to more uncertainty and hence a lower confidence.
However, this theory cannot fully account for their data as it does not explain why people with a high need for cognition were less confident when choosing from the small set as compared to a large set. A closely related but slightly more elaborate construct that has been connected to the effect of too much choice is the tendency to maximize or satisfice.
As mentioned in Chapter I, and in contrast to the hypothesis of Lin and Wu (2006), no effect of too much choice would be expected when people apply a fast and frugal satisficing heuristic (Simon, 1955) because in these cases, people would simply search for something that is good enough and the search would stop once an option is found that exceeds their level of aspiration. On the other hand, optimizing, the aim to find the best option, gets more difficult and involves more effort as the number of options increases. At the same time, the second-best, not-chosen option will be more similar to the chosen option, which implies relatively higher opportunity costs. The concept of opportunity costs originated from economic theory and describes the benefits that could have been received from choosing the most valuable forgone alternative. According to Schwartz (2000, 2004; Schwartz et al., 2002), an increase in opportunity costs will lead to increased feelings of regret as well as a decrease in satisfaction and the motivation to choose.
Schwartz proposed that decision strategies can be described on a continuum ranging from satisficing to optimizing (or “maximizing” as Schwartz calls it). He further assumed that decision makers differ in the degree to which they engage in one of the strategies and that this propensity is moderately stable across situations.
To measure people’s propensity to maximize, Schwartz et al. developed a 13-item scale that includes statements such as “I never settle for second best,” “When shopping, I have a hard time finding clothes that I really like,” or “Renting videos is really difficult. I’m always struggling to pick the best one.” All the statements have to be rated on a scale from 1 (strongly disagree) to 9 (strongly agree). Based on a total sample of N=1,747 that included diverse sub-samples of first year psychology students, nurses, and a convenience sample at a train station, Schwartz et al. reported scale reliabilities of Cronbach’s α = 0.71 for the maximization scale, ranging from 0.60 to 0.73 across different sub-samples. In a validation study, Schwartz et al. found that maximizers, defined as having a mean score above the middle of the scale, are more likely to engage in social comparisons than satisficers. They further found positive correlations between maximization, regret, and depression and they hypothesized that this relationship would be mediated by the presence of an overwhelming array of options. In a consumer context maximizers self-reported less positive feelings toward purchases and said they considered more products compared to satisficers.
Supporting the idea that the propensity to maximize moderates the effect of too much choice, Haynes and Olson (2007, Study 1) found that maximizers experienced more difficulties and more frustration with the decision process when the choice set increased from 3 to 10, whereas for satisficers they did not find a difference. Likewise, maximizers experienced more dissatisfaction and regret when choosing from the large assortment of ten options as compared to the small assortment of three. For satisficers it was the other way around. Satisficers were less satisfied with the option chosen from the small assortment as compared to the choice from the large assortment. In the Haynes and Olson study, maximization was measured based on Schwartz’s maximization scale. Maximizers and satisficers were defined relative to the sample mean. Everyone above one standard deviation from the mean was denoted a maximizer and likewise a satisficer for scores below one standard deviation from the mean. The differences in satisfaction and regret for maximizers confirm Schwartz’s prediction. Yet the finding that satisficers were less satisfied when choosing from the small set runs somewhat counter to the prediction according to which satisficers should not be affected by the number of options. Also, with a Cronbach’s α of 0.54 the reliability of the maximization scale was rather poor.
In a recent study on job search, Iyengar, Wells, and Schwartz (2006) found that senior college students who described themselves as maximizers on a questionnaire made a greater effort to search for job offers and were able to get better paid jobs after graduation, but at the same time they were less satisfied with their choices as compared to satisficers. While in their study the reliability of the maximization scale was also rather low (Cronbach’s α = 0.6), the results suggest that even though on an objective scale maximizers might make better choices, they are subjectively less satisfied with them. As an explanation, the authors suggested that decision makers determine their satisfaction by trading-off the “cost” of search against the benefits of the job they found and that for maximizers this trade-off is less advantageous. What speaks against this interpretation is that once a job is found, search costs are fixed, whereas the benefits of the job continue for the time it lasts. Thus, as long as a decision maker does not quit, the benefits should eventually outweigh the costs. On the other hand, at the time Iyengar et al. measured people’s satisfaction, they had just stated their new jobs and might not have had many benefits yet. I will discuss the relationship between search costs and the effect of too much choice in more detail in the General Discussion in Chapter V.
An alternative explanation for Iyengar et al.’s results would be that maximizers set their aspiration level much higher than satisficers do. If so, maximizers would search longer because they also satisfice but while trying to match a higher aspiration. The better job would then simply be a consequence of the extended search and the relatively lower satisfaction could be explained by the fact that maximizers were still less likely to match their aspiration. This interpretation would also be closer to Simon’s (1956) original conception of satisficing as a search and decision rule by which an individual looks up information and chooses according to a predetermined aspiration level.
It should be noted here that this latter interpretation of maximizing as the propensity for high aspirations would not necessarily predict a too-much-choice effect. It would predict that maximizers will search longer, but as a consequence of the prolonged search they should be relatively more satisfied when choosing from a large assortment as compared to a small assortment. This is because the large assortment provides a higher chance to achieve (or at least get closer to) their aspiration.
Personality traits that relate to decision making, such as the tendency to avoid decisions or to seek risks, have been shown to be highly context dependent (Beattie et al., 1994; Hanoch, Johnson, & Wilke, 2006). Thus, the individual tendency to maximize might also depend on the situation. As an example, one might think of a person who tries to maximize outcomes within a job-related context but less so in private matters. Therefore, a test of the impact of maximizing on the too-much-choice effect should also incorporate a context-specific way to measure that trait.
Further building upon Simon’s (1956) and Schwartz’s (2004) notions of satisficing, it seems to be a necessary precondition of the too-much-choice effect that decision makers actually take the additional options in the large assortment into account. One reason why I did not find an effect in previous studies could be that people simply ignored the excess and by this shielded themselves from being overloaded with choice, a strategy that resembles the concept of a satisficing.
However, given that in the series of charity studies, organizations at the end of the list had the same probability of being selected as organizations at the top of the list, the selection process might have looked somewhat different. A concept that would be in accordance with the data from the charity studies and that relates to the notion of satisficing is the idea that decision makers use a two-stage process such that they first screen options to form a consideration set (sometimes referred to as an evoked or relevant set) from which they then choose (Hauser & Wernerfelt, 1990; Reilly, 1985). From this perspective, what eventually matters is not the total number of options available but rather the number of options that are seriously considered for the final decision. If so, it can be conjectured that the effect of too much choice is mediated by the size of that sub-set, and assessing the size of that sub-set could help to explain the previously divergent findings.
Testing for cultural differences between Germany and the United States as a moderator ideally builds upon a study design that can be used in the same manner in both countries. While the charity paradigm outlined above was a first step in this direction, its results cannot be compared directly because in each country different charity organizations had to be used. To obtain a more precise comparison between Germany and the United States, a study design in which the exact same stimuli could be used in both countries would be preferable.
To empirically test the effect of too much choice and the influence of the moderators and mediators outlined above, I implemented a computer experiment in which participants repeatedly listened to samples of classical music that they had chosen from different assortments beforehand. Because classical music is equally enjoyed across many cultures, the exact same choice sets could be used in Germany and in the United States with the only difference being the translation of the instructions. In Germany, the experiment was conducted at the Max Planck Institute for Human Development in Berlin, and in the United States participants were recruited at Indiana University in Bloomington.
So far, all experiments on too much choice, including my own, have adopted a between-subjects design such that the group of people facing the small assortment was different from the group facing the large assortment. Yet the results of my previous studies suggest that if anything, the size of the too-much-choice effect is probably much smaller than previously thought. To be able to even detect such small effects, a within-subject design in which each individual faces both a small and a large assortment is preferable because it comes with a larger power due to a decreased measurement error (Hunter & Schmidt, 1990). As laid out by Hunter and Schmidt, a classical between-subjects design assumes that the independent variable (the number of options in the present case) affects all participants equally such that there is no interaction between the independent variable and the individual. Yet, it could be that half the participants would show a too-much-choice effect while the other half would show the reverse effect. In a between-subjects design, these differences would go unnoticed and the conclusion would be that there is no effect—a conclusion that would be completely invalid. To circumvent these problems, the study on hand employs a within-subject design in which the same participant consecutively chooses from both a small and a large assortment.
To allow for choices among identical options in both countries, participants in the experiments got to choose among recently released CDs of classical music from the record label Deutsche Grammophon, an internationally renowned label that specializes in high-quality recordings of classical music. For the purpose of the study, I compiled two large sets of 30 CDs each from the most recent releases of Deutsche Grammophon: one set with vocal music and one set with orchestra music. From each of these two large sets, I then randomly selected one sub-sample with 6 CDs. Thus, there were four different sets of CDs that can be described based on the two orthogonal factors music style (orchestra vs. vocal) and assortment size (small vs. large).
The CD assortments were displayed on a 17-inch computer screen as a collection of thumbnails (36×36 pixels each) that showed the miniaturized pictures of the CD covers in a random order. The last name of the composer and the abbreviated CD title was displayed underneath each thumbnail. A detailed description of each CD could be retrieved by clicking on the thumbnails with the left mouse button. The detailed description consisted of a full-size picture of the cover, the full CD title, and the full names of the composer, the conductor, the orchestra, and the choir (for vocal music only).
Participants could browse through one of the assortments at a time and they could look up as many details about the available CDs as they wished. Finally they were asked to choose one single CD from each of the assortments. From this CD they got to hear a sound sample on their headphones that lasted for 2 minutes. The samples actually were the first two minutes of the first track of each CD that I had downloaded from iTunes and embedded into the experimental software prior to the actual experiment. To control for differences in the volume of the recordings, the volume of each track was normalized and participants got a slider on the screen to control the volume according to their own preferences.
Participants in front of the computer screen consecutively chose from one small and one large assortment. To control for potential order effects such that the experience of choosing from the first set influences the choice in the second set, participants saw the assortments in one of two different sequences. Half the participants first chose from a small assortment followed by a large assortment; the other half first chose from a large assortment followed by a small assortment. To motivate a repeated choice, one of the assortments consisted of classical orchestra music and one assortment consisted of classical vocal music. The order in which participants saw the type of music was counterbalanced with the order of the assortment size which resulted in four different experimental groups A-D (Table 1).
Participants in group A first chose from a small assortment of orchestra music followed by a choice from a large assortment of vocal music. Participants in group B first chose from a small assortment of vocal music followed by a large assortment of orchestra music. Group C first chose from a large assortment of orchestra music followed by a small assortment of vocal music and group D first chose from a large assortment of vocal music followed by a small assortment of orchestra music. To get familiar with the experimental setting, all participants first chose a CD from a “training set” of 14 CDs with Deutsche Grammophon piano recordings. The piano CDs were compiled and displayed in the same way as the orchestra and the vocal CDs.
Sequence of choice sets
In line with previous studies on choice overload, the main dependent variables of the study on hand were the degree of post-choice satisfaction and post-choice regret with the chosen piece of music as well as the motivation to make a choice in the first place.
The dependent variables post-choice satisfaction and post-choice regret were assessed immediately after listening to each of the three sound samples. To decrease measurement error and to assess construct reliability, post-choice satisfaction and regret were both measured based on multiple items. All items called for answers on a 9-point Likert scale ranging from -4 (lowest rating) to +4 (highest rating) with textual anchors on both sides of the scale.
Post-choice satisfaction was measured based on six items: Participants were asked to rate their satisfaction with the chosen piece of music in comparison to the other pieces on the screen, in comparison to other pieces of the Deutsche Grammophon label, and in comparison to other pieces of classical music in general. They were further asked to rate how much they liked the chosen piece of music, how much they enjoyed listening to it, and how satisfied they were with their choice.
Post-choice regret was assessed based on three items asking participants to rate how likely they would be to choose the same piece again, if they thought another piece would be better, and how much they regretted their choice. As an additional measure, participants were also asked to state exactly how many of the options they did not choose they now thought were better than the one they actually chose.
In this experiment preference strength and choice motivation were assessed indirectly through people’s willingness to pay for the option of their choice. At the end of the experiment, participants were asked to state how much they were willing to pay for the orchestra and the vocal CD that they had previously selected. To make this task meaningful, they were informed that the three participants willing to pay the most would be allowed to buy the respective CD at the price they stated, a procedure commonly referred to as a sealed-bid first-price auction or discriminatory auction. To give people a sense of the actual market values, they were told that the CDs on display on average cost about 20, euros or 20 dollars, respectively. To make sure that participants remembered their selections, they were presented with the covers of their previously chosen orchestral and vocal CDs.
The difference in how much participants were willing to pay for an option chosen from a small set and an option chosen from a large set indicates how much one option is preferred over the other. As the willingness to pay more increases the probability of an actual purchase, the measure can also be seen as a proxy for the motivation to choose.
As mentioned in Chapter I, another potential mediator of the effect might be the perception of a large assortment as less attractive or overly complex or the experience of choice from a large assortment as being more difficult or less enjoyable. As for the main dependent variables, all items on mediators and moderators were assessed at the end of each sound sample and if not stated otherwise, items called for answers on a 9-point Likert scale ranging from -4 (lowest rating) to +4 (highest rating).
Perceived choice difficulty was assessed based on four items: Participants rated how hard/easy it was for them to make a choice (very easy/very hard), how exhausting it was to choose a piece of music (very exhausting/not at all exhausting), how much they deliberated about their choice (very little/very much), and to what extent they experienced the choice process as frustrating (very frustrating/not at all frustrating). To get a more complete picture of the process, participants also rated how much they enjoyed making a choice between the pieces of music and how much they were trying to select the best piece of music.
The current study set out to measure the size of the consideration set and the degree to which decision makers ignore the excess choices based on self-reports and based on behavioral measures of the amount of search and the number of options inspected prior to choice. Consideration set size was measured by asking participants to state the exact number of options they short-listed (“Wie viele CD’s kamen für Sie in die engere Wahl?”/“How many options did you short-list?”). As a behavioral measure of the individual search process, the number of CDs that were looked up in detail and the decision time was measured by tracking each mouse click on the screen. As this tracking was achieved in the background through the experimental software, it did not interfere with the decision task and it could not be noticed by the participants.
After the choice task on the computer screen, participants filled out a paper-and-pencil questionnaire that consisted of the personality scales for maximizing and satisficing. To validate the maximizing scale, participants also filled out Schwartz et al.’s (2002) regret scale, which is supposed to moderately correlate with the maximizing scale. Maximizing and regret were measured based on the scales published by Schwartz et al. (2002) and their validated German versions, respectively (Greifeneder & Betsch, 2006).
As a domain-specific measure of maximizing, participants also rated how much they were trying to select the best piece of music. As mentioned above, this item was asked after each sound sample alongside the other items.
At the end of the experiment, prior preferences and domain-specific expertise were assessed. Participants were asked to indicate on a Likert scale how often they listen to classical orchestra music (never–daily), how often they listen to classical vocal music, how knowledgeable they consider themselves with regard to these two styles of music (don’t know anything–know a whole lot) and how much they like these two styles (don’t like it at all–like it very much).
As further control variables, participants also rated how motivated they were to participate in the study (not motivated at all–very motivated) and how carefully they answered the questions (very carefully–not at all carefully). As compensation for their participation, participants in Berlin received 8. euros and participants in Bloomington received course credit.
In Berlin, 80 students from local universities participated in the study (20 in each condition). Of the participants, 49% were female; the average age was 26 years (SD=2.9 years). In Bloomington, 87 undergraduate students from Indiana University participated (22 in conditions A, B, and C and 21 in condition D). Of the participants in Bloomington, 69% were female and the average age was 20 years (SD=1.6 years).
There was no statistically significant (α=0.05) difference between the four experimental groups (A-D) in either country on any of the three main dependent variables: post-choice satisfaction, post-choice regret, and willingness to pay. Therefore, to test for main effects of assortment size within subjects, the data was collapsed across the order of music style (orchestra first/vocal first) and the order of assortment size (small set first/large set first), separately within each country.
To allow for a more concise report of results in the following paragraphs, data from the Berlin sample will be stated first, followed by data from the Bloomington sample, separated by a slash.
Participants in both countries were motivated to participate in the study and all reported that they carefully answered the questions.
On a scale from -4 (not at all attractive) to +4 (very attractive), participants in both countries on average perceived the large set as more attractive than the small set, which is in line with previous findings on too much choice. The average attractiveness rating was -0.3/0.3 (SD=2.0/2.2) for the small set and 0.8/0.7 (SD=2.0/2.1) for the large set. When compared within subjects, 55%/45% of the participants rated the large set as more attractive compared to 21%/31% who rated the small set as more attractive. In absolute terms, the attractiveness ratings vary around the center of the rating scale, which indicates middling attractiveness.
The complexity of the small assortments was rated as -1.6/-0.7 (SD=1.8/2.1) and the complexity of the large assortments as 1.5/0.7 (SD=1.9/2.1). Thus, in both countries the two assortments clearly differed with regard to their perceived complexity. Yet in absolute terms, especially in the U.S. sample, the large assortment was not rated as very complex. In line with this, choosing from the small set was perceived as easier (-1.5/-2.1; SD=1.4/1.7) as compared to the choice from the large set (-1.0/-1.8; SD=1.6/1.9). Of all participants, 65%/47% rated the choice from the small set as easier and 37%/32% rated the choice from the large set as easier. In absolute terms, the negative mean scores indicate that both choices were perceived as rather easy.
On average, participants looked up 5/5 different CDs (SD=1.6/1.4) or 89%/84%3 in the small assortments while in the large assortments they looked up 16/14 different CDs (SD=11.2/11.3) or 52%/47%. In total, 74%/68% of the participants looked up more options in the large set as compared to the small set and 20%/16% looked up more options in the small set. Thus, a higher absolute number of options was looked up in the large condition as compared to the small condition while at the same time the relative number of explored options was notably smaller in the large assortment.
On average, participants’ reported consideration set size was 3/3 (SD=1.0/2.1) in the small assortment and 5/8 (SD=4/8.1) in the large assortment. Of all participants, 80%/71% formed higher consideration sets for the large assortment as compared to the small assortment, and 9%/19% formed higher consideration sets for the small assortment.
In the small assortment, it took an average4 of 29/24 seconds from the moment the assortment was displayed on the screen until the final choice of one particular piece of music (SD=18/17 seconds). In the large set, this search and exploration phase took an average of 64/44 seconds per person (SD=54/36 seconds). Of all participants, 79%/77% took longer to choose from the large set as compared to the small set. When choosing from the small assortment, on average, participants spent 5.4/4.7 seconds to examine the details of one single option before they went on to the next or terminated their search (SD=2.7/3.0 seconds). In the large assortment, the time it took to examine one single option was 5.1/4.0 seconds (SD=3.1/3.3 seconds).
In the following, I will report the effect of assortment size on the main dependent variables post-choice satisfaction, post-choice regret and willingness to pay.
Prior to testing the effect of assortment size on post-choice satisfaction the reliability of this main dependent measure needs to be assured. Cronbach’s alpha for the six-item post-choice satisfaction scale is 0.98/0.98 for both the small and the large assortment, which indicates a good reliability in both countries (Bortz & Döring, 2002).
On a scale from -4 (very unsatisfied) to +4 (very satisfied), the mean satisfaction with options chosen from the large set is 1.1/1.0 (SD=2.4/2.2) and from the small set it is 0.5/0.9 (SD=2.4/2.2). Of the 80/87 participants, 35/39 were more satisfied with the chosen option from the small set, 41/43 were more satisfied with the chosen option from the large set, and the remaining 5 were equally satisfied in both conditions. When subtracting the post-choice satisfaction score in the large set from the post-choice satisfaction score in the small set within each participant, the mean difference across participants is -0.5/-0.1 (SD=3.9/2.9), where a negative value indicates a higher satisfaction for options chosen from the large assortment (Figure 7). According to Cohen (1977) this translates into an effect size of d=-0.17/-0.05. By means of a t-test for paired samples, and an alpha level of 0.05, the null hypothesis of a zero difference can be rejected neither in Berlin, t(79,1)=-1.3; p=0.21, nor in Bloomington, t(86,1=-0.3; p=0.74.
|Figure 7: Individual satisfaction with the chosen option depending on assortment size. Data points below the diagonal indicate higher satisfaction when choosing from the small set as compared to the large set (=too-much-choice effect).|
For the three-item post-choice regret scale, Cronbach’s α=0.79/0.80 for the small assortment and 0.86/0.83 for the large assortment. Analogous to the post-choice satisfaction measure, this indicates a satisfactory reliability.
The mean post-choice regret with options chosen from the large set is 0/0.3 (SD=2.3/2.4) and from the small set it is 0/-0.2 (SD=2.5/2.3). Of all participants, 35/40 experienced more post-choice regret when choosing from the small assortment as compared to the large assortment, 39/42 experienced more regret in the large assortment, and 6/5 equally regretted their choice in both conditions. A t-test analog to the one applied for post-choice satisfaction did not reveal a significant difference in Berlin, t(79,1)=-0.04; p=.97; Cohen’s d=-0.01, or in Bloomington, t(86,1)=0.17; p=.87; Cohen’s d=0.04.
The mean amount participants were willing to pay is 6.3 euros/6.9 dollars (SD=€5.5/$4.9) for a CD chosen from the small assortment and 6.7 euros/7.6 dollars (SD=€4.5/$5.2) for a CD chosen from the large assortment. The corresponding effect size (Cohen’s d) is -0.08/-0.13. Of all participants, 40/40 stated a willingness to pay more in the large-choice condition and 29/28 stated a willingness to pay more in the small-choice condition. These data indicates that if anything, people were willing to pay more for an option chosen from the large assortment (Figure 8).
|Figure 8: Amount willing to pay for the selected CD depending on assortment size.|
In summary, when averaging across all participants within each country, no effect of assortment size on the three main dependent variables, post-choice satisfaction, post-choice regret, and willingness to pay, was found. Yet, as the data indicate considerable inter-individual differences in these dependent variables, it can be conjectured that an effect of too much choice might have occurred at least for some participants. In line with this, the three main dependent variables are highly correlated with each other. The Pearson correlation between the difference in post-choice regret and the difference in post choice satisfaction is r= -.90/-.85. For the difference in post-choice regret and the willingness to pay, the correlation is r= -.61/-.60; and for post-choice satisfaction and the willingness to pay, the difference is r=.68/.63).
Together, these correlations indicate that a person who is dissatisfied regrets the choice and is willing to pay less—and vice versa. Merging these three measures assumes that they reflect the same underlying construct. With regard to their intercorrelation, this assumption seems justified as Cronbach’s alpha is 0.87/0.83. Even though preliminary, this provides some convergent evidence for the existence of the effect for some of the participants. The question that arises is if these individual differences can be explained by one of the moderator or mediator variables outlined above. If so, this would lead to an important insight into when and why the effect of too much choice occurs.
I set out to test the potential moderators and mediators outlined above, namely, the personality traits maximizing and regret, as well as behavioral differences in the amount of search, the perception of complexity, the size of the consideration set, the perceived attractiveness of the assortment, the perceived difficulty of making a choice, the number of options participants thought would be better than the one they chose, and the expertise in the domain on hand.
To test if these factors matter, I split the sample into two separate groups for each country. The “overloaded group” consisted of participants who showed signs of choice overload on all three dependent variables such that they experienced the choice from the small set as more satisfactory and less regrettable and were willing to pay more for a CD chosen from the small assortment. The “scarcity group” consisted of those participants who showed the reverse pattern on all three dependent variables. There were 26/20 participants in the overloaded group and 28/30 participants in the scarcity group.
As an alternative way to split the participants into two groups, I could have used the continuous compound score on the three measures of choice overload as variable within a random effects model. While in the latter case the statistical power would have been higher, due to the more complex statistic, the presentation of the results would have been less comprehensible. As the results are similar for both statistical methods, I will report the results based on the comparison between the overloaded and the scarcity group.
In Berlin, participants in the overloaded group had slightly higher scores on Schwartz et al.’s (2002) maximizing scale (4.3 [SD=1.0] vs. 3.8 SD=[1.0] on a scale from 1 to 9) and the regret scale (4.5[SD=1.4] vs. 3.9[SD=1.6]). Yet the magnitude of these differences was rather small (Cohen’s d=0.5 for maximizing and d=0.4 for regret) and on an alpha level of 0.05 the corresponding F-tests were not significant, F(52,1)=3.0; p=.09 for maximizing and F(52,1)=2.2; p=.14 for regret. In Bloomington, participants in both groups have similar mean scores and variance on the maximization scale (5.2, SD=1.1) and with respect to regret, the overloaded group in Bloomington had a slightly lower average score (4.7 [SD=1.0] vs. 5.3 [SD=1.0]). Thus, there was no consistent difference between the two groups with regard to their general propensity to maximize or to regret their choices.
In Berlin, only 16 (20%) of the participants scored above the middle of the maximization scale and thus according to Schwartz et al. (2002) would classify as maximizers, while the majority of participants should be classified as satisficers. In the Bloomington sample, 51 (59%) of the participants had a maximization score above 5. To rule out that the high proliferation of satisficers in Berlin diminished the effect in the sample, I directly compared maximizers to satisficers. Yet with regard to the three main dependent variables, there is virtually no difference between the two groups in Berlin. Consequently, only 6 of the 16 maximizers were found in the overloaded group, which is about what would be expected by chance.
One reason for why the assessed personality traits can only partly count as explanation for the effect might be that their reliability is rather low. Cronbach’s alpha for the maximization scale is 0.62/0.70 and for the regret scale it is 0.76/0.62. While these low reliabilities make it difficult to interpret the personality construct underlying these scales, it should be noted here that these values are slightly higher than the scale reliabilities reported by Iyengar et al. (2006) and those reported by Haynes and Olson (2007, Study I).
As outlined above, the tendency to maximize might be domain specific, which is why I asked participants after each choice to specify to what extent they were trying to select the best piece of music. Yet for this domain-specific measure there is also no notable difference between the overloaded group and the scarcity group. There is also practically no correlation between the domain-specific tendency to maximize and the maximization scale, which further questions the validity of the maximization construct in the context on hand.
In both countries, the six items assessing prior preferences and expertise were highly intercorrelated, with Cronbach’s α=0.89 for the Bloomington data and α=0.81 for the Berlin data. Thus, I collapsed the items into an aggregated score for each participant for subsequent analyses.
Despite the good reliability of the measure, the degree of domain-specific expertise was virtually the same between the overloaded and the scarcity groups, which suggests that in the current study this factor did not directly influence the effect of too much choice. This is in contrast to the findings of Mogilner et al., (2006, Study 3) who found an effect of too much choice for people with little knowledge and experience but not for experienced participants.
Also, in Berlin and in Bloomington there were virtually no differences between the two groups with regard to the number of options that were looked up prior to choice, the perception of complexity, the amount of time spent searching (in seconds) and the amount of self-reported deliberation. In the Berlin sample, people in the overloaded group experienced the choice from the small assortment as easier (-1.8 vs. -0.9), F(52,1)=5.2; p=.03, and the choice from the large assortment as slightly more difficult (-0.7 vs. -1.5), F(52,1)=3.5; p=0.07, than the scarcity group. However, I found no such differences in the Bloomington sample.
In Berlin, but not in Bloomington, the overloaded group formed considerably larger consideration sets when choosing from the large assortment as compared to the scarcity group (six vs. four options) while for the small assortment there was no such difference. This might indicate that for some people in Berlin the effect of too much choice was moderated through an appetence conflict as a result of having too many attractive options. Narrowing down these options might have been perceived as difficult and frustrating, which in turn would affect the satisfaction with the finally chosen option. However, the fact that no such relationship exists in the U.S. sample suggests that this interpretation should be considered preliminary.
In both countries, compared to the scarcity group people in the overloaded group thought that the number of better options in the rest of the large set was higher (8/9 vs. 2/5), F(52,1)=18.4; p<.001 and F(48,1)=4.8; p=.03. In Bloomington, but not in Berlin, people in the overloaded group also rated the large assortment as less attractive as compared to the scarcity group (0.2 vs. 1.1), F(48,1)=5.3; p=.03. Yet the relationship between these latter two variables and the dependent variables might be due to the close resemblance of their theoretical underpinnings: The number of better yet forgone options clearly is an aspect of regret while the attractiveness of the assortment might be influenced by the attractiveness of the chosen option. Thus, while these differences reconfirm the dependent variables, they add little to their explanation.
As in my previous experiments, I did not find any effect of assortment size on either satisfaction, regret, or willingness to pay. This was despite the use of a within-subject design and a well-controlled experimental setup through which even small effects would have been detected. The strong correlation between three main dependent variables and the inter-individual variance with regard to these variables suggest that something like a too-much-choice effect might have occurred for about a quarter of the participants. However, this intraindividual variance could not be explained by any of the theoretically well-grounded moderators and mediators that I assessed. There are several possible explanations for these findings.
The reason that most suggests itself is that the differences in the dependent variables are simply due to random variation. Independent of assortment size, there certainly is variance in how much participants liked the option that they chose due to influences such as individual preferences for certain types of sounds, the quality of the recording, or different expectations, to name only a few. As I did not control or measure these influences they count as error variance. This error variance causes variation in people’s satisfaction, regret, and their willingness to pay. In addition, as in any experiment, these three main dependent variables themselves are measured with an error. Yet, as indicated by the high reliability of the multi-item measures for satisfaction and regret, this latter error is probably negligible.
Thus, under the null hypothesis that there is no effect of assortment size, participants would be classified in an overloaded group and a scarcity group just by chance, and as a consequence, there would be no variance left to be explained by potential moderators and mediators.
An alternative explanation for the lack of findings might be that overall the attractiveness of the assortment was rated as mediocre and that even though the choice was real, it did not have important consequences. This might have increased the chances that people did not care much about the outcome and that they satisficed rather than optimized. In line with this, the data on hand indicate that participants found it rather easy to make a decision and that they did not perceive the assortment as very complex. What speaks against this interpretation is the fact that the perception of the assortment and the self-reported decision strategy (i.e. how much subjects said they were maximizing vs. satisficing) did not have an influence on the main dependent variables. However, the data also indicate that the validity of the maximizing vs. satisficing personality measure was poor, which makes it difficult to interpret these results. Moreover, there is some evidence that people have little insight into their own decision strategies (Nisbett & Wilson, 1977) and thus asking them might not be an adequate measure in principle. I will discuss the relationship between decision strategies and assortment size in more detail in the General Discussion in Chapter V.
It can be speculated that the effect of too much choice depends on several necessary preconditions such that in order for it to occur, decision makers need to have a low expertise in a complex and unfamiliar domain and at the same time seek an optimal outcome. Yet, due to its highly explorative nature and the decrease in statistical power, I refrained from testing multiple causation models and higher-order interactions of that kind. Besides, the more preconditions that have to be met in order to elicit the effect of too much choice, the lower its generalizability and ultimately also its importance.
In a total of six experiments across various domains and contexts with a total of 595 participants in Germany and the U.S., I found no effect in five studies and only a medium effect of too much choice in one study. Summarizing my results, the effect does not seem to depend on cultural differences between Germany and the United States, nor on a further increase in the number of options in the large set (i.e. from 40 to 80 in the charity study), an increase in the average attractiveness of the small assortment, people’s tendency to maximize their outcome or their perception of complexity, the number of options people explore, or their domain-specific expertise.
Across all my experiments, the only case in which a small effect of choice overload occurred was when participants were asked to justify their choices. In this case, the effect could be due to the fact that it is more difficult to justify a choice from a large assortment as compared to a small assortment because in the small assortment the options are less similar. While the data on the number of characters used to justify the choice provided some weak support for this idea, the finding needs to be replicated before any firm conclusion can be drawn. Also, as mentioned before, it does not explain most of the reported occurrences of the effect.
In summary, the question remains of when and why the effect of too much choice occurs. As outlined in Chapter I, other researchers have found the effect in different contexts. With regard to the results of my own research it can be concluded that the effect is far less robust than suggested by its proponents.
As before, there are two possible approaches to joining these divergent findings into a coherent theoretical frame. First, it could be that choice overload is indeed widespread but that the effect is much smaller than previously thought. In this case, the question arises of how probable these divergent results are. Second, it could be that moderator or mediator variables that explain the diverging results actually do exist and remain to be discovered. In the latter case, the core question is what these variables are. These approaches will be followed up in Chapters IV and V, respectively.
2 Initially, I planned to have 80 organizations in the large set but due to a mistake in the layout, the name of one educational charity did not appear on the large-set questionnaire.
3 The differences between absolute numbers and percentages are due to rounding to full numbers
4 Following the recommendation of Wilcox (1998), mean statistics on decision times are 20% trimmed to control for outliers, which means that 10% of the largest as well as 10% of the smallest observations are discarded
|© Die inhaltliche Zusammenstellung und Aufmachung dieser Publikation sowie die elektronische Verarbeitung sind urheberrechtlich geschützt. Jede Verwertung, die nicht ausdrücklich vom Urheberrechtsgesetz zugelassen ist, bedarf der vorherigen Zustimmung. Das gilt insbesondere für die Vervielfältigung, die Bearbeitung und Einspeicherung und Verarbeitung in elektronische Systeme.|
|DiML DTD Version 4.0||Zertifizierter Dokumentenserver|
der Humboldt-Universität zu Berlin
|HTML-Version erstellt am: