Evaluation of the Police

INDEM Foundation offers a new approach to building classifications in sociological and marketing research. This approach is based on survey data and expert evaluations of conditional probabilities of choice of answers with typical representatives of different classes.

 

In the framework of this approach, we are constructing consistent estimators of class membership functions for all respondents. Besides, we are introducing characteristics of quality for different experts, and solving the problem of dimensionality reduction (the number of questions used for estimating). Public opinion surveys are one of the primary sources of information in sociological and marketing research. The task of dividing respondents into classes that reflect some number of their social and behavioral particularities is one of the most important. Examples of these particularities might be types of motivation, forms of relationships to the choice of product or to evaluation of events, electoral preferences, etc.

 

The traditional method of dividing respondents into classes is based on their responses to the questions. Respondents who respond equally to questions form a class. However, in this case some methodological problems arise.

 

Firstly, there is a problem of language ambiguity by which communication between social scientists and respondents is made. Secondly, there is a difference in understanding of the same words by respondents and social scientists. Thirdly, responses on the questions are motivated by several reasons or features, among which a researcher's interest is only one of them. Fourthly, this latent feature can be concerned with confidential spheres of respondents' life and in such situation direct questions in a questionnaire would be useless and inadequate.

 

Thus, in this method, respondents answer the range of questions. We intend that all these questions deal with the feature being investigated. It means that responses on these questions are determined to high extent by the degree of testament of this feature in respondents. Based on the range of respondents' answers we aimed to decide to what extent each of the respondents has this or that feature.

 

To solve the problem we have to fix a correlation between respondents' answers on the questions from the range and investigating feature. This correlation will be determined by experts' procedures. We will presume that the degree of testament of the feature being investigated will be described by fuzzy classification.

 

To attribute a feature to a respondent is to determine his function of attitude to the classes of classification. In doing so it is essential to determine the chances of typical representatives of different classes of fuzzy classification to choose this or that answers. Naturally, such a method solves some subsidiary problems. Among them is experts' evaluation and their selection, synthesis of different experts' estimations and evaluation of questions as indicators of the investigated feature.