Description of the study
Annotation
Our Research and Study Group (RSG) is designed to substantiate and demonstrate the specific advantages of using methods for searching interaction effects in sociological studies in comparison with the traditional sociologists` approach to the analysis of data. By interaction, it is understood a combination of the values of the variables that determine the phenomenon interesting for a sociologist. Despite some methods for search these combinations of values, in research, most sociologists avoid searching them, thus, consciously or unconsciously simplifying the social reality in the models. Moreover, while remaining within the framework of classical models that are oriented toward a linear connection and do not consider interactions, the researchers are in a situation where the inherent in the method prerequisites are not adequate to the nature of the studied social phenomenon. Consequently, the results, first, are interpreted incorrectly, second, some of the conclusions remain outside the field of view of researchers. The evidence of this fact can be seen in researchers` final regression models with very low predictive power.
Our proposed methodological study is based on a secondary analysis of the data within the four thematic mini-groups headed by the project leader. In the framework of this secondary analysis, it is intended to consider and characterize the cognitive and prognostic capabilities of two broad classes of methods for searching interactions: logarithmic linear models and decision tree models. Both classes of methods are distinguished by their universality, simplicity, and accessibility.
There are at least three proofs of the need for the existence of study like that. Firstly, the results of the work of our RSG will be useful for specialists in quantitative data analysis in the social sciences and related disciplines. The developed recommendations of using methods for searching interaction effects will become a tool for a complete analysis of the nominal and ordinal variables that are so common in the social sciences and will be irreplaceable to build high-precision predictive models. Secondly, the planned scientific results will serve to eliminate one of the main reasons for the unpopularity of using these types of methods - the absence of theoretical and empirical grounds of the advantages of the use in the literature. Finally, results obtained by applying methods for searching interaction effects in specific thematic areas will be helpful to supplement the existing conclusions on relevant topics.
Novelty and perceptivity of research methods used in the study
Although considered methods are not new and most were developed in the 1960s and 1980s., the novelty of our work with them will be their application to specific sociological data from different thematic areas, as well as a comparison between the results obtained using classical and selected methods. Moreover, the planned attempt to apply the so-called ensembles of decision trees (an alternative to single trees) can be regarded as completely new, because nowadays, such algorithms are applied only in computer sciences. However, their application is of current interest for social sciences as well - to make the solution received on a single tree model and eliminate the model over-training. In addition, in our study, to obtain more reliable meaningful results, Bayesian analogs of some logarithmically linear models and their comparison with the corresponding traditional models are planned for applying. Comparison of the two selected classes of methods for finding interactions (logarithmic linear models and decision tree models) is also relatively new for research in the social sciences.
Perceptivity of applying considered methods can be seen in the following: firstly, as it was underlined above, both classes of methods are distinguished by their universality, simplicity, and accessibility; secondly, it is supposed to achieve improvement in the quality of predictive models.
Practical significance and application of the expected results of the study
The results of the work of our RSG will be useful for specialists in quantitative data analysis in the social sciences and related disciplines. The developed recommendations of using methods for searching interaction effects will become a tool for a complete analysis of the nominal and ordinal variables that are so common in the social sciences and will be irreplaceable to build high-precision predictive models.
In addition, the systematization of existing methods for searching interactions, describing the possibilities of their application to sociological research data (including advantages, disadvantages and limitations of methods, illustrated by specific empirical examples according to the thematic areas of the groups` work) can significantly supplement the material of advanced courses on methods of analyzing sociological data in universities.
Have you spotted a typo?
Highlight it, click Ctrl+Enter and send us a message. Thank you for your help!
To be used only for spelling or punctuation mistakes.