In another study by Tversky & Kahneman the availability for construction was assessed by studying judgement of word frequency. The subjects were asked whether there are more words in the English language that begin with the letter k or have k in the third position. It was proposed that classes whose instances are easy to construct will be perceived as more frequent than classes of same size which are harder to imagine, thus revealing ‘a bias due to the effectiveness of a search set’ (Tversky &Kahneman, 1973; Tversky &Kahneman, 1974:1127). Therefore it was hypothesised that people judge the words beginning with the letter k more frequent, because it is much easier to retrieve words by their first letter than by their third letter. This is indeed what happened, even though the opposite is actually true - there are many more English words with k in the third position than in the first position (Tversky &Kahneman, 1973).
Different search sets are triggered by different tasks (Tversky &Kahneman, 1974). When rating the frequency of abstract words (thought, happiness) and concrete words (door, juice) in written English, people tend to search for a context where these words could appear, and it seems easier to retrieve contexts with abstract concepts (love in love stories) than concrete ones. This was observed in a study in which subjects mistakenly judged abstract words to be more numerous and to appear in a much greater variety of contexts than concrete words (Tversky &Kahneman, 1974).
It seems then that we employ a limited number of search strategies depending on the type of the task, and we seem to often hold on to the first satisfactory answer obtained through the use of heuristics, even though it may lead into serious error. Following these principles, we shall now investigate the availability heuristic with a similar word fragment judgement task to that of Tversky’s & Kahneman’s (1973) experiment. There were two conditions: Group A was asked: ‘ How many words of the form ‘----ing’ would you expect to appear in a 2000 word passage? Group B was presented with the same question, expect the word fragment was in a different form. It was ‘-----n-‘ instead of a ‘----ing’. Subjects were told to provide their answer within 30seconds. The frequency of the words should be mediated by an assessment of availability. It should be much easier to come up with examples of the ‘ing’ words than with the ‘n’ words and with these examples readily available people should estimate a higher frequency for the ‘ing’ word fragments, even though, naturally, the ‘n’ pattern is actually more frequent as it includes the ‘ing’ pattern and thus certainly contains more possibilities. Thus it was hypothesised that group A (‘ing’ words) will estimate that a higher number of these words would appear within 2000 word passage and that group B would generate a less numerous estimate for the ‘n’ word fragments.
METHOD
DESIGN
A between-groups design was employed to investigate the availability heuristic through a word fragment judgement task. The independent variable had two different conditions:
Group A was presented with a question: ‘How many words of the form ‘----ing’ would you expect to appear in a 2000 word passage? and Group B received the same question except the word form was different, that is ‘-----n-’ instead of ‘----ing’. The question of interest was whether there is any difference between the word frequency estimates generated in group A and B? The experimental hypothesis proposed that Group A (ing-words) would estimate a higher number of words than would group B (n-words).
SUBJECTS
This study had 60 participants, thirty per condition. There were 17 males and 43 females. The participants varied in age from 18 to 37, with a mean age of 23 years. All participants were psychology undergraduate students at the University of Glasgow, who took part in the study as a part of their second year curriculum.
MATERIALS AND STIMULI
The stimuli used in the experiment was either a puzzle containing the ‘----ing’ word fragment or the ‘-----n-‘ word fragment and the word fragment puzzles were administered to the participants on a sheet of paper with written instructions on it. Participants then independently completed the task by marking their answer on the paper with a pencil.
PROCEDURE
The study was held in the University of Glasgow’s psychology lab during a practical session on heuristics. The participants sat around 4 tables in small groups. The participants were told that they would be given a sheet of paper with puzzles in it and that their task was to engage in to independently solving these puzzles by following the written instructions on the sheet. There were two different puzzle sheets, but the participants were unaware of this. Puzzle sheet A contained the question: ‘How many words of the form ‘----ing’ would you expect to appear in a 2000 word passage? In puzzle sheet B the question was same but the independent variable was in a different form: ‘How many words of the form ‘-----n-’ would you expect to appear in a 2000 word passage? The whole class was administered the puzzle sheets at the same time and they were randomly given either the puzzle sheet A or the puzzle sheet B. The participants were instructed to generate their answer within 30 seconds. After The participants had completed the task they were de-briefed. Participants were told about the use of heuristics in problem solving and they were required to think back on how they solved each problem paying attention on the particular heuristics they employed in providing their judgements.
RESULTS:
It was hypothesised that group A who was presented with the question including the ‘ing’ word fragment would estimate that a higher number of these words would appear within 2000 words and that group B would generate a less numerous estimate for the ‘n’ word fragments.
The descriptive statistics for the two groups are as follows:
In group A (“ing” words) the mean number of estimates was 164 words (rounded) and in group B (‘n’ words) the mean was considerably less: only 34 words (rounded).
Let us illustrate these results in a clearer manner through a figure shown on table 1.
TABLE 1
It is evident from the boxplots that all the numbers of estimates lie below 100 for the group B (‘n’ words) and that the interquartile range for ‘n’ words is much narrower - approximately from 10 words to 50 words- than the interquartile range for the group A (‘ing’ words) as most of the estimates for ‘ing’ words lie between 100 and 200 words. The smallest estimate for ‘n’ words was 2 words and the largest 100 words, while as the smallest estimate for ‘ing’ words was much more: 50 and one outlier can be observed with the largest estimate of 400 words as well as long right tail. - Thus explaining the large difference in standard deviations between the groups, as the standard deviation for group A (‘ing’) was 86.3517 and for group B (‘n’) 26.3369.
The descriptive statistics show that the pattern of the results is clearly what was expected: Group A produced much higher word frequency estimates than did group B.
Let us now address the issue of interest formally by analysing the results with inferential statistics:
A two-tailed, independent samples t-test was used to compare the estimates generated by group A with those of group B. The difference between the means was statistically significant: ( t = 7.86, df = 58, p< 0.000). Thus the experimental hypothesis is confirmed: people tend to generate reliably higher estimates for the ‘----ing’ word fragment condition.
DISCUSSION
The results support the experimental hypothesis: people tend to generate reliably higher estimates for the ‘_ _ _ _ ing’ word fragment than they would if they were exposed to the ‘_ _ _ _ _ n _’ word fragment. Parallel with Tversky’s and Kahneman’s interpretation of similar studies, these results can be explained in terms of the availability for construction, which implies that the estimate of the frequency of the words was based on a quick count of examples and the ease with which these examples are retrieved provides the estimation for the final judgement (Tversky & Kahneman, 1973, 1974). In this case it is relatively easy to construct or imagine examples of the ‘ing’ words while as it is very difficult and effortful to generate examples from the ‘n’ words. Thus ‘ing’ words are falsely perceived much more frequent than the ‘n’ words because the ‘n’ words are less available, even though they are more numerous (Tversky & Kahneman, 1973). It seems that, as expected, people tend to rely on one type of search strategy (i.e.assessment of frequency through availability)instead of trying to look at the problem from different ankles. Thus, it seems, that in this case the error was due to what Tversky and Kahneman called the ‘bias due to the effectiveness of a search set’ (1974:1127). Another relevant type of availability bias is caused by the instances’ imaginability, in which case the judgement is not based on what we can remember but the ease with which particular instances can be constructed according to a given rule (Tversky & Kahneman, 1974).
It could be speculated that the method employed in this experiment had few weaknesses: It is possible that the subjects took longer than the instructed 30 seconds in providing their answer as the amount of time spent by the participants was not strictly measured and controlled. The fact that all the participants were psychology students could possibly indicate that the results were altered if the students were aware of the use of heuristics through their previous psychological knowledge. These factors in mind this experiment could be replicated by changing the circumstances accordingly. It would also be interesting to employ a control group in this experiment. In this condition the subjects would be exposed to both of the word fragments at once, therefore possibly eliminating the biased judgement?
Since there is no clearly direct way to examine the availability heuristic, how can we be sure that the the strenght of the association or the ease with which examples are retrieved is indeed what mediates the frequency judgements (Taylor, 1984)? This criticism is enforced by the fact that we are lacking an agreed-upon measure of availability (Taylor, 1984). Many studies, like the one employed in this paper, have used the speed of the information retrieval as a measure of availability. Others have looked at volume of instances recalled as infering the availability of information. These two measures have produced ambiguous results because they do not always correlate well thus hindering the empirical progress (Taylor, 1984).
The fact that our findings are consistent with Tversky’s and Kahneman’s empirical work on non-social judgements, supports their proposal that people employ a limited number of heuristics or shortcuts allowing as quick information processing as possible (Taylor, 1982). The extensive amount of laboratory research on availability heuristic has revealed the wide range of settings in which it is used (Halpern, 1984).Moreover, availability heuristic is used in establishing and maintaining social judgements, as research in the field of social psychology has suggested (Taylor, 1982). As emphasized earlier, availability biases due to the retrievability of instances, which is affected by the event’s familiarity and salience, can have an impact on our cognitive structures like schemata with negative social consequences, such as the formation of prejudice and stereotypes (Taylor, 1982; Tversky & Kahneman, 1974). A fascinating point is that the biases that occur through the use of cognitive heuristics reveal errors in cognition, which are independent from motivational factors and instead attributable to errors in information processing (Taylor, 1984; Tversky & Kahneman, 1974). Certainly motives do play an important role in our judgements, but it is interesting finding that major errors in cognition also exist (Taylor, 1984).
A natural response to these studies is the question whether something can be done in order to alleviate the biases in people’s reasoning? Statistical knowledge offers a route to more objective and accurate outlook on phenomena and is thus shown to diminish judgemental errors (Gleitman, 1999). An inapproprite coding prevents us from learning statistical principles from the situations we are exposed in our everyday life and we need training in order to understand the relation between sampling size and sampling variability, for example (Tversky & Kahneman, 1974). However, even highly educated and experienced researchers sometimes fall into erroneous reasoning in less transparent problems (Tversky & Kahneman, 1974). The inappropriate coding also underlies the fact that people rarely realize the biases in their judgements of frequency or probability (Tversky & Kahneman, 1974). However, one must stress the highly useful and efficient nature of the availability heuristic as it allows us to produce answers very quickly and these answers are more often right than wrong (Taylor, 1984). Taylor also emphasizes that ‘ whenever biases are uncorrelated over time, the process will begin to correct itself with repeated encounters.’ (1984:200). Through social interaction and communication our biases may be modified as well (Taylor, 1984). Sometimes, however, different biases may co-occur. Confirmation bias, is likely to lead into persistent misperceptions. For example if a prejudice was formed through a biased judgement, confirmation bias may maintain the prejudice through its strong resistance to any counterevidence (Taylor, 1984).
Future studies on heuristics are likely to focus on the circumstances where biases occur and a coherent theory of human reasoning may be obtained (Kahney, 1986). This kind of progress in our understanding of heuristics is crucial as serious negative consequences may occasionally occur due to availability biases and thus methods for the recognition and correction of these biases are needed (Taylor, 1984).
First of all, in order to gain evidence that the availability heuristic mediates frequency judgements through assessment of the ease with which relevant instances are recalled, studies are needed which are designed to examine availability per se, because, unlike in the cognitive experiments on availability, the social investigations were originally designed to examine some social phenomenon in general and availability heuristic was only one possible explanation for the results (Taylor, 1984). It would be interesting to carry out an experiment where the participants would speak out aloud the information processing sequences by which they draw their conclusions. However, judgements are always produced by what comes to one’s mind, thus revealing the conceptual ambiquity of the use of the term availability (Taylor, 1984). Thus clear distinguishable criteria of the use of the availability heuristic is needed (Taylor, 1984).
A study that examines the outcome of judgements when participants are instructed to evaluate their responses could reveal crucial information about the formation of biases as well as whether the participants would become aware of these biases when engaged into guestioning their values. Participants could generate answers to general guestions about their attitudes and many of the guestions should be about recent events that were widely discussed in media. After the participants had produced their answers they should be given another sheet of paper containing questions such as: ‘pause and think why you decided as you did? Is it because of the recency of the information? Who did you get the information from and why did they give it to you?’ (Changing Minds.org, 2003).
Nevertheless, the tendency in the field of social psychology to concentrate on what kind of information is available and why it is available should be accompanied by an effort to examine ‘how the availability heuristic ties the content of what is available to cognitive processing’ (Taylor, 1984: 199).
As a response to critique about the use of heuristics, experiments could be employed, which question whether heuristics truly are generally used with the expense of logical reasoning. Thus it would be interesting to test the outcome of the word fragment study employed in this paper if subjects were presented with both of the word fragments at once. The subjects would probably be likely to realise that the ‘n’ pattern includes the ‘ing’ pattern and consequently produce the right frequency judgement, even within the limited time of 30 seconds. In this case the way the word fragment puzzle was designed would perhaps trigger logical reasoning rather than the use of heuristics, evoking interesting questions about the frequency of the use of heuristics if alternative processing method is available, as well as the effect of the framing of the question has in the types of answers produced.