The Influence of Language on Ambiguity Aversion

The influence of language on ambiguity aversion

Vincent van Noort

349138

Contents

Introduction3

Sapir-Whorf Hypothesis4

Questionnaire6

Results8

Language9

Vienna versus The Hague13

Gender13

CRT Scores14

Age15

Degree16

Other Regressions17

Conclusions19

Discussion20

Bibliography22

Appendix A: Questionnaire23

Appendix B: Results28

Introduction

In a very well known experiment Daniel Ellsberg showed in 1961 that people dislike ambiguous situations. He proved that in some situations people have the tendency to behave irrational. Using data from responses under non-experimental conditions he showed that people violate the Savage axioms when they are dealing with uncertainties. These responses came on questions in which there are two urns; the first urn had fifty red balls and fifty black balls. The second urn also contained hundred balls but with unknown distribution between the two colours. Ellsberg showed that most respondents prefer the urn with known distribution, no matter on which colour the bet. This suggests that they think the second urn contains less red balls and less black balls than the first urn. Since the second urn also contains hundred balls this is impossible. Ellsberg explained this by saying that people prefer the risk of the first urn over the uncertainty of the second urn. He called this ambiguity aversion (Ellsberg, 1961).Ambiguity is now often defined as uncertainty in the relevant information or completely unknown relevant information. This information is often regarding the probabilities or payoffs. Ambiguity can however also refer to information that is misunderstood. As previous research has proven a lot of factors influence ambiguity such as the context or the way information is obtained. Three different kinds of ambiguity can be distinguished: uncertain probability, uncertain pay-off and situations were both are uncertain. This research will only focus on situations where the probability is unknown.

Since Ellsberg published his findingsthere has been a lot of research regarding ambiguity attitudes. Part of this research is that people are so averse to ambiguity that they would be willing to pay to avoid ambiguous situations if a real payoff was involved (Becker & Brownson, 1964). Besides the consequences of ambiguity aversion it has also been investigated what factors influence ambiguity aversion. According to Fox and Tversky (1995) ambiguity is also influenced by the context. Ambiguity aversion is strong in case of a comparative context but disappears in the absence of a comparison(Fox & Tversky, 1995). Others found that the way people get (ambiguous) information also influences decisions. People who see a representative sequence seem to be less averse to the ambiguity than when a verbal description is given (Bleaney & Humphrey, 2006).

Because there already has been a lot research regarding ambiguity aversion this is not just a research to show how peoplereact to ambiguity. It will be tested whether people have a different attitude to ambiguity in different languages. One of the reasons that it is possible language could change the attitude comes from Sapir-Whorf hypothesis. This hypothesis named after linguist Edward Sapir and his student Benjamin Lee Whorf states that the language that someone speaks influences our decision making. Another reason to believe that ambiguity attitude is affected by language is the fact that it has already been proven that language can influence decisions. It has for instance been tested that in a foreign language the framing effect disappears and that people tend to be less averse to lossesin a foreign language. Usually the framing effect makes it possible that the same people give different answers to the same problem but framed differently. Because most people are risk seeking for losses and risk averse for gains they answer differently when the problem is framed as losses compared to the same problem framed as gains. Most famous framing problem is the Asian disease problem. This problem is also used in a bilingual study. This experiment showed that in the native language framing does work, leading to different answers for the same problem. In a foreign language however this framing effect disappears. The same study also shows that bets with positive expected value are more likely accepted and people are less averse to losses in a foreign language than in the native language. The writers argue that this is because the native language causes more emotional reactions leading to biased decisions (Keysar, Hayakawa, & An, 2012).

Two questionnaires will be used to test whether this is the case with ambiguity attitudes. Those two questionnaires will be exactly the same except one will be in Dutch and one in English. In this questionnaire the respondents will be asked four questions regarding their attitude to ambiguity. These questions illustrate situations in which the probability of winning is unknown compared to situations where this probability is known. With the results from these questionnaires it will be tested whether there is a significant different between the answer from the Dutch questionnaire and the answers from the English questionnaire.

Sapir-Whorf Hypothesis

As said the ambiguity attitude can be influenced by for instance the context and the way of obtaining information. Another factor that couldperhaps influence ambiguity attitude is the language in which information is given. As said earlier ambiguity is also information that is misunderstood. Information given in a second language could be more easily misunderstood making people more averse to ambiguity. There is however a more important reason to believe why the language could influence ambiguity attitudes. Some linguistics believe that the language spoken influences our view of the world and more important our decision making. The hypothesis these linguistics believe in is called the Sapir-Whorf hypothesis(Kay & Kempton, 2009). This hypothesis consists of two parts:

1. Structural differences between language systems will, in general, be paralleled by non-linguistic cognitive differences, of an unspecified sort, in the native speakers of the two languages.
2. The structure of anyone’s native language strongly influences or fully determines the world-view he will acquire as he learns the language (Kay & Kempton, 2009).

For this research the focus will be on the first part. This part could mean for this research that different languages will be accompanied by different decision making. Meaning that the results from the Dutch questionnaire could be very different from the English questionnaire. The hypothesis does not state how each language influences the view of the world or decision making. Therefore the results can go both ways, the Dutch results can be more averse to ambiguity but it could also be that the English results show more ambiguity aversion. This is dependent on the actual influences both languages have. This first part of the Sapir-Whorf hypothesis is also called linguistic determinism. It can also be further divided in strong and weak determinism. Strong determinism is the believe that what is said is responsible for what is seen by the mind (Badhesha, 2002). In an Australian experiment with deaf children it is shown that strong determinism can hold. After a doll is put in a box with a marble the doll is removed first and after that the marble is removed. When asked where the doll will look for the marble children with parents fluent in sign language answered correct. The children growing up in a family with non deaf parents who are not fluent in sign language answered incorrectly (Peterson & Siegal, 2006). Although in this situation strong determinism holds most linguistics reject the view of strong determinism. The version with weak determinism however is a lot more accepted. Weak determinism still means that the language someone speaks influences our view of the world or our decisions. But were strong determinism says that language defines this strictly; weak determinism means that there are still other factors that influence our view or decisions (Badhesha, 2002). Therefore weak determinism is a lot more probable than strong determinism. But because weak determinism states that language certainly has an influence it could still lead to differences in our decisions. These differences can perhaps also occur between a native language in which people are fluent (Dutch) or a second language which is probably not spoken completely fluent (English). So it is very well possible that people will show different attitudes to ambiguity in different languages.

Questionnaire

To test whether language really affects ambiguity attitudes a questionnaire is used. Or rather two questionnaires: one in English and one in Dutch (Appendix A). The questions in both questionnaires are exactly the same and phrased in the same way to avoid other factors such as framing. At first friends, family and some colleagues were asked to fill in one of the questionnaires. Later also other students were asked to fill in one. This gives a group of respondents that differ a lot from each other. Although a lot of respondents are in the early twenties there are also respondents with an age in the thirties, forties or even higher. It also gives a lot of variety in the field of study or profession despite most of the fellow students study economics. Thanks to the other respondents there will also be enough differences here. It was decided that all the respondents should speak Dutch as their first language. Therefore all the respondents speak Dutch as their first language the Dutch questionnaire gives results for first language and the English gives the results for the second language. There are a couple of reasons why only Dutch respondents were used. First reason is that this way it is sure that there are roughly the same amounts of results for first language as there are for second language. Otherwise it could have happened that only people speaking English as their native language filled in the English questionnaire. Since fluency in the language can have an effect this could influence results. The second reason is that the switch from Dutch to English could be different than for instance from Italian to English. This could influence the results. Another important decision was to let every respondent fill in only one of the questionnaires instead of both. The main reason for this was to avoid the will to be consistent. Because the questions are exactly the same in bothquestionnaires respondents would probably want to be consistent and use the same answers without thinking about it.

Table 1: Question 1 (Red ball)

The first question in the questionnaires looks like the questions used by Ellsberg. There are two urns with black and red balls. The distribution in the first urn is known and the distribution in the second urn in unknown. Respondents have to pick an urn to draw a ball from after they have chosen a colour. Table 1 shows the table the respondents have to fill in if they chose a red ball. If a black ball is chosen the same table is given but with the colours switched. The second question also uses two urns but they are filled with five different colours. This gives the chance to see whether the ambiguity attitude stays the same when the probability of winning in Urn K gets lower. Questions three and four are also used to measure ambiguity attitudes. In both questions respondents have the choice between a certain amount of money or a bet on the weather in Vienna or The Hague respectively. With these questions it will also be tested whether a city far away (Vienna) changes the ambiguity attitude in comparison to a city nearby (The Hague). Table 2 shows the Vienna question.

Table 2: Question 3 (Vienna)

The questionnaire will also make use of a cognitive reflection test (CRT). Three questions will be used to measure the cognitive ability of the respondents. The same test that Shane Frederick used in 2005 to research the relation between cognitive ability and decision making is also used in this research. His research proved that cognitive ability influences the time and risk preferences. Respondents having a higher cognitive ability tend to be more patient, more risky for gains and take less risk when it involves losses (Frederick, 2005). At last there will be asked a couple of demographic questions. This is to make sure possible differences are not caused by factors such as gender, age or education. The native language of the respondents is also asked to make sure whether the results belong to first or second language. In the English questionnaire respondents are also asked to rate their English. This way it is also possible to remove certain results if their English is very bad and they do not seem to understand the questions.

Results

In the end a total number of 54 respondents were reached. This was divided in 30 for the English questionnaire and 24 for the Dutch questionnaire. However due to several reasons some answers could not be used for research. Therefore I ended up with 21 answers to the English questionnaire and 23 for the Dutch. First the answers given to the questions regarding the ambiguity aversion were given numerical values. If for instance one choose a red ball in the first question and switched between 30 red balls in Urn K and 40 red balls in Urn K this would be given a value of 35. This was done in a similar way for the questions with five different colours and the weather in Vienna and The Hague. Also every respondent was given a CRT score. This varies between zero and three depending on how many CRT questions were answered right. This research will mainly focus on possible significant differences caused by language. This can be because of the different languages or the proficiency in English. But it will also be tested whether other factors such as CRT scores, gender, age or degree has an influence on ambiguity aversion.

Language

Now the results can be analyzed. First the descriptive statistics in table 3 are examined. The statistics have been separated by the language of the questionnaire. In the question with two colours respondents with a value below fifty are averse to ambiguity. Higher than fifty means they are ambiguity seeking. It is remarkable to see that most statistics for this first question are the same in both languages. In line with ambiguity aversion the means (and median) of both groups are below fifty. However there are apparently some respondents who seem to be ambiguity seeking as can be seen from the maximum of 55. More surprisingly are the results from question with urns containing five different colours. In this case ambiguity aversion should be indicated by a value below twenty. The means are however both above twenty and in case of the Dutch questionnaire even above the thirty. If you take a closer look at the frequencies of this question you will see that in the English questionnaire 72.2% of the respondents is ambiguity seeking. In the Dutch questionnaire this percentage is even higher at 78.3% (Appendix B, Table 1). The questions about the weather in Vienna and The Hague however give the expected results. Whether it is ambiguity aversion or ambiguity seeking depends on the belief of the raining probability of each respondent. Because this is not asked it is impossible to know whether respondents are averse to ambiguity or seeking ambiguity. But at both questions a higher value means less aversion to ambiguity so it is possible to test whether language influences the attitude. It is remarkable though that both have a maximum of twenty (or nineteen in the Dutch The Hague question). This means there are respondents that prefer to win €20 only if it rains in one of the cities above the opportunity to win €20 with certainty. This is in contrast to every behavioural theory and more importantly also the opposite of common sense. Therefore it could be doubted whether this person read the question well or answered it seriously. At the last three questions there are slight differences in means between the languages. And in the last two questions the median differs also. However whether these differences are significant remains to be seen.