Research Seminar Archive
Shubhabrata Das (Indian Institute of Management, Bangalore): Fuzziness in human response due to scale, respondent or attribute
When a product or performance is rated or a behavioural response is sought, the alternatives typically do not have well-defined and universally understood demarcations. The resultant fuzziness in response depends, among other factors, on the number of options available to the respondent and this vagueness is quantified here.
The measure, expressed as a percentage of the total variability, is arrived at by separating the fuzziness from the randomness present in the data, and it is used to provide a guideline regarding the appropriate number of levels in measurement. While the proposed measure can be computed under stringent assumptions, two approaches of estimating it under weaker assumptions are also discussed in this work. Numerical data analysis on a real as well as a simulated data set suggests that a response scale with five to seven options may be appropriate.
A second direction of fuzziness in Human can be pointed to the lack of sincerity and clarity in the thought process of the respondent as well lack of clarity of the attribute being evaluated. A formal measure of this aspect of fuzziness is formulated in this work on the basis of consistency of the respondent's response to the same question repeated in multiple scales. Discarding haphazard and insincere respondents can improve the quality of data resulting in more efficient survey analysis.
This may be achieved in the framework of statistical testing of hypothesis using the probability distribution of the proposed fuzziness measure. Similarly, an attribute can be fuzzy, and it may generate inconsistent response from many respondents. The application of the proposed methodology extends to identifying such fuzzy or unclear attributes. We will also discuss an algorithm for screening inconsistent respondents and fuzzy attributes simultaneously.
Time permitting, we would also look at a) objective determination of classes during a histogram construction b) some simple statistical inference problems work out in presence of such fuzzy data.
Related Links
Download this event in iCalendar format
