A cognitive process of responding to a spatial survey: Textual analysis versus network analysis and spatial analysis for analysing an international corpus of co-citations about countries

Abstract : As demonstrated by the analysis of the general results of the EuroBroadMap survey on 18 countries and 42 places of survey (Didelon & al., 2011 ; Grasland & Beauguitte, 2012; Brennetot & al., 2013 ), the declaration by students of the countries and cities where they would like to live or not like to live is governed at aggregated level by some general gravity laws. Our contribution proposes three alternative approaches for the analysis of datasets containing explicit or implicit list of places where the order of the list is eventually important. Many researches are based on such type of data containing individuals characterised by list of places with an explicit or implicit ranking. The first hypothesis is the fact that the places given by the respondents are not randomly distributed at the individual level. It means that, even if we do not consider the answers ranking, an important part of the initial information is lost when we aggregate the list proposed by individual to upper levels of social or spatial aggregation. If we ask for example 6 students of a given country to quote 3 countries among seven proposed where they would like to live, and we obtain the set of answers (A,B,C), (A,B,C), (A,B,C), (C,D,E), C,D,E), (C,D,E), we will conclude at an aggregated level that C is the most appreciated country (100% of answers contain the country C) and that A,B,D,E are equally appreciated (50% of answers). But we will lose crucial information about the fact that, out of C, answers are strongly organised by the association of answers A-B on one hand, and D-E on the other hand. If we ask now to the same students to give 3 countries where they would not like to live, with answers (D,E,F), (D,EF), (D,E,F), (A,B,F), (A,B,F), (A,B,F), it will produce the same loss of information, but we will also miss the existence of interaction between answers to both questions. To demonstrate the existence of such effects, we propose to apply a set of methods from textual analysis where each place is considered as a word, each list as a sentence and each set of lists as a paragraph. These methods are able to take into account the contextual effects at different levels and can therefore help to validate or invalidate the existence of interactions between answers, not only at the level of individual lists but also between the different lists. The second hypothesis is the existence of a non-random distribution of ranks within answers. To illustrate this approach, let’s consider the following distribution of answers to the declaration of 3 countries among 8 countries where students would like to live : (A,B,C), (A,B,E), (A,B,F), (G,H,I),(H,I,G), (I,G,H). The aggregated approach will obviously miss a lot of information by concluding to the existence of only two levels of countries (A,B,G,H,I) and (C,E,F). Textual analysis will reveal the existence of two separated cluster of associated answers (A,B,C,D,E,F) against (G,H,I).. We propose here to apply firstly network analysis methods based on the analysis of successive couples of countries, based on ranks, in order to build an oriented network (A is followed by B which is followed by C, which defines only two edges A-B and B-C, optionally completed by A-C1). The network of followers in the list of countries reveals differences between countries that were not initially visible : the cluster A,B,C,D,E is based on a structure of “tree” when cluster (G,H,I) is rather like a “clique”. The third hypothesis is the fact that distance between places of a given list is not randomly distributed. We apply here spatial analysis method in order to verify if the choice of places rank is governed or not by rules of distance or other forms of proximity (common language, contiguity, colonial relations). For this, we compare the average distance between countries of the same list according to their ranks Applying this set of methods to the results of EuroBroadMap survey demonstrate first their greet interest but also their complementarity. We suggest in conclusion extending this kind of application to different data corpuses like co-citations of countries in press articles.
Type de document :
Communication dans un congrès
18th European Colloquium in Theoretical and Quantitative Geography (ECTQG), Sep 2013, Dourdan, France. 2013
Liste complète des métadonnées

https://hal.archives-ouvertes.fr/hal-01216591
Contributeur : Timothée Giraud <>
Soumis le : mardi 20 octobre 2015 - 09:29:21
Dernière modification le : mardi 26 septembre 2017 - 01:15:41

Identifiants

  • HAL Id : hal-01216591, version 1

Citation

Claude Grasland, France Guérin-Pace, Sophie De Ruffray, Timothée Giraud. A cognitive process of responding to a spatial survey: Textual analysis versus network analysis and spatial analysis for analysing an international corpus of co-citations about countries. 18th European Colloquium in Theoretical and Quantitative Geography (ECTQG), Sep 2013, Dourdan, France. 2013. 〈hal-01216591〉

Partager

Métriques