STERNBERG-WAGNER THINKING STYLES : A RESEARCH TOOL IN SOCIAL SCIENCE DIDACTICS

Due to its nature, the teaching of social sciences and the way of thinking of teachers and students maintain a complex, if not delicate relationship. Personal viewpoints and affinities on issues like gender, how things must be done or arranged in class could have a strong influence on the development of a subject. To understand this in a given context, we propose Sternberg-Wagner test analysis and data mining as class tool to identify student behavior patterns and its influence on selected variables (scores, gender, choice of small group for class tasks). As the study shows, significant differences have been detected between sexes as to the form they prefer to do a task, and the way they conceive it. Knowing this can help to better class planning, insofar identifies strengths and weaknesses that otherwise might not have been detected.


Introduction
This article offers a hybrid technological tool designed to collect useful data for teachers looking for assistance to take decisions about how to plan or improve their didactic strategies (Niehaus, Reading & García, 2018) in Social Science contexts.In few areas of education, the interest in knowing the impact of the teachers' and students' way of thinking is relevant in Social Sciences (Jamil-Asghar, 2016;Smetanová, Drbalová & Vitáková, 2015;Selesho, 2014;Valli, Perkkilä & Valli, 2013;Keirn & Luhr, 2012).
There are two large groups of reasons lie at the basis of this fact.First of all, the plural nature of the epistemology of this discipline (Bautista, 2015;Gerring, 2012;Beltran, 1985) should be accommodated to extensive subjects as History, Sociology, Anthropology, and even Geography in countries like Spain or France.Thus, classical approaches as quantitative or qualitative analysis are accompanied by methodologies including critical-rational, historical, or comparative analysis, to name just a few.Here, the difficulty is about quality considerations on taught science.
Secondly, social thought covers rational and irrational thinking dispositions, following Perkins' theory (Bouhnik & Carmi, 2012;Perkins, Jay & Tishman, 1993).The latter implies beliefs, feelings, and collective/self-identity questions (Taylor, 2017;Christiansen & Koeman, 2015;Hocutt, 2005).Therefore, picked ways of teaching could be considered, even if not alleged, indoctrination.Here, the difficulty lies in the ability to not project teachers' viewpoint as unquestionable standard.It is important, then, to distinguish between facts and personal valuations, beliefs, or affinities.Pre-service teachers need to be aware of their future responsibilities and challenges, especially nowadays, where learning, technological, and sociological dynamics are faster than in past decades (Cartelli, 2012).In Spain, a didactic background is compulsory to graduates who wish to work in secondary education or vocational training, through one-year master's degree, since 2011.During that course, they acquire basic knowledge about Pedagogy, teaching methods, and educational legislation among other contents.However, the brevity of the formative period not always grants they take on appropriately the second fact: as far as own ideology is a decision to be taken in academic contexts, it is considered part of the intellectual debate.This explains how teachers see, sometimes, themselves as owners of the ultimate truth for their students, instead focusing on their autonomous mind development (De Robertis, 2017;Midhatovna-Galimullina & Nikolaevna-Korshunova, 2016;Rockenbach, Mayhew, Wolniak, Terenzini, Seifert, Bowman, et al., 2016).
Sternberg-Wagner questionnaire is among most reputed tools to explore and classify thinking styles in education (Yuan, Zhang & Fu, 2017;Canbolat, Erdogan & Yazlik, 2016;Fan, 2016;Zhu, 2013;Fink & Garner, 2008).It covers five categories, disaggregated in thirteen styles, each of them checked through eight questions in a mini-test.In exchange for the time required to gather, to process and to analyze it, we obtain a pretty accurate image of the intellectual structure of people's mind.This is the reason why the present work has two main objectives: to identify the behavior patterns present in class, as learning and analysis tool, insofar each way of thinking has its strengths and weaknesses; and to explore if they have any influence in the academic performance, social or gender behavior within the class.

Methodology
The target population was 29 students of the subject Initiation to Research and Educational Innovation in Social Sciences within the Master Degree in Secondary Education, during 2016-2017 course in University of Málaga (Spain).The students did several individual tasks, and one in small group that included a practical exhibition of it.The latter was assessed through 360º methodology (Martins & Martins, 2006;Brutus & Gorriti, 2005) using CoRubric, a federated online rubric tool (Martínez-Romera, Cebrián-Robles & Cebrián- de-la-Serna, 2016;Stevens & Levi, 2013).Seven students left the course before it was finished.To avoid personal influences, the project aims were explained only after data gathering.
Firstly, we used the Sternberg-Wagner test about thinking styles (Sternberg, 1999).It was needed to detect and to classify the patterns of self-concept about teaching dimensions present in class.Its 104 questions are arranged in octets, one for each style that implies a specific psychological tendency, as summarized in Table 1.
Then, we analysed if those patterns had any influence with: 1. Group election (social behavior).

Gender (sex influence).
Finally, we briefly explore some interdependencies among selected variables through associative data mining (Algarni, 2016;Koedinger, D'Mello, McLaughlin, Pardos & Rosé, 2015;Papamitsiou & Economides, 2014;Ahmadi & Ahmad, 2013) with WEKA (University of Waikato Environment for Knowledge Analysis).We chose Apriori, a very well know algorithm for boolean association rules (Tanna & Ghodasara, 2014) whose key concepts matches the behavior of our population: it requires itemsets with internal frequencies in its variables (denoted as Lk), its subsets must be also frequent (downward closure property) and it implies a join operation between candidates in order to generate the rules.The algorithm uses a two-steps logic to find association rules: the join step, were is set the k-itemsets of candidates (denoted Ck); the prune step, determines if Ck is a superset of Lk.Ck tends to growth fast, so Apriori uses two properties to refine its results: itemsets that are not frequent cannot be subset of frequent k-item sets; every subset of candidate k item not present in Lk-1 is removed as it cannot be frequent.Results are ranked by its support, percentage of itemsets covered by each rule, starting at 100% and decreasing it in steps of 5% until confidence is lower than 0.9 or 10% bound is reached.
Due to the amount of potential rules derived from de analysis, Apriori establishes four alternative ways for ranking rules (Frank, Hall, Witten, 2016): Confidence, determines the proportion of examples covered by the premise also covered by the consequent; Lift, result of dividing confidence by the support; Leverage, proportion of new examples covered by both premise and consequent, considering them statistically independent; and Conviction, criterion defined by Brin, Motwani, Ullman and Tsur, (1997) but not specifically documented by WEKA.

Analysis
For the first step, the data gathering of the questionnaire gave us a total of 3,016 items to work with.We considered maximum frequency as modal style, and we used 'Mixed' if tie happened.Summarized in this way, we obtained the first approximation about thinking dynamics in class, as shown in Figure 1.
Despite statistical highest frequency on each style defines the class as Mixed, Hierarchic, Local, External and Liberal (MHLEL for short) in its way of thinking, the minimum and maximum absolutes within each category are relatively close, which implies high internal diversity.Differences were less than 10% in Functions & Levels, and 25% in Scopes & Leaning; leaving Forms as the less diverse one, with a gap of 52%.This makes clear the need for more detailed analysis.
In order to do so, we went through each student's questionnaire.The result, as shown in Table 2, were joined with sex and group variables, and the three main scores considered here, for further analysis: FM as final mark assessed by teacher; PInd as peer assessment on each member of the group exposition and PGrp as peer evaluation of the group, done only by audience students.The last two variables are expressed as arithmetic mean.Peer assessment reveals that best individual average is that of the group G, best group average is for the group A, and best FM point out to F. On the opposite side, we found the group B in all three cases.Square of Pearson product-moment correlation coefficient shows a strong link between PInd and PGrp (79.40%), but not with FM (35.29% and 36.70%respectively).As shown in the Figure 2, the gap between teacher and students criteria ranges from almost 1.5 points in D to barely 0.2 points in F: Counting internal group coincidences aggregations (∑Clusters), and the total number of individuals included (∑Students), the best categories for clustering are, in decreasing order: Leaning, Scopes, Forms, Levels and Functions.As far as large aggregations ( 3) are concerned, the best options are, in decreasing order: Forms, Scopes, Leaning, Levels and Functions.
Data mining was focused on association rules (Scheffer, 2001;Liu, Hsu & Ma, 1998;Agrawal & Srikant, 1994).In order to do so, discretization of numeric variables (step needed to convert them to a qualitative form) was done on PInd, PGrp and FM.Three score intervals were defined on them: low, medium, and high.The Table 4 summarizes best rules found:  On the one side, Apriori reveals that D and G are the less coherent groups, as far as only one rule could be established within them.On the other side, A and C cover 71.4% of the all rules found, pointing out them as groups with strong and diverse internal bonds.Finally, B and F had two members each so, despite their size, must be considered the strongest one too.If fact, those groups are the only ones with perfect rules, affecting all members.For that analysis the two main criteria considered were standard support and confidence > 0.8, as they were explained in Methodology.Further settings were discarded due to the small size of the itemsets and our interest in discovering general rules (common, or very common subsets) and not marginal ones (what would only be of interest in samples of greater size).

Discussion
Social behaviour is foundational in this classroom.Group C shares two thinking styles among all its members (Function & Leaning); Group A, one (Form); Groups B and F, small ones, shares 2 and 3 styles respectively; E has one couple with strong bonds (sharing 2 thinking styles); Group G, despite having only one rule, has two members sharing Function and Leaning styles; and only Group D seems to be a group with no social patterns, as far their only sharing items are mark ranges.Being acquaintances from previous degree or time shared on the actual one become very relevant and emphasized intellectual thinking as academic affiliation strategy.
On the one hand, peer review does not show clear patterns.Internal gap between PInd and PGrp affects almost all teams, with D and B as the bigger ones.In fact, only Group F shows a convergence criterion among PInd, PGrd, and even FM.So, in this case, possible antagonistic or dissenting ways of teacher thinking do not seem to influence the criteria as assessors.The only verifiable point here is that teacher marks in all cases, except Group F, is the lowest, by far.The common strategies could explain that a clear goal of the students, insofar, it to pass the subject.
On the other hand, Sex has been shown by itself as a very interesting criterion to classify thinking styles in class, for two reasons: what they do choose and what they do not.Women picked massively Judicial or Legislative forms, but no one of them picked Anarchic.Also, women in class were clearly Liberal in their leaning.Men did not show a clear tendency in forms, with frequent ties (mixed forms); and no one identify himself as Oligarchic and leaning almost split them in half.The rest of styles, despite internal diversity, does not reflect sex patterns.
Internal group coherence tends to be strong in all team groups, showing that pre-service teacher students rely on their concept of teacher to choose their work group.Even if they lack of knowledge about it, as long as they created the groups before the theoretical explanation and the Sternberg-Wagner test.
A big study carried out in China (Yuan et al., 2017) over larger samples confirm thinking styles as powerful tool to predict stress-coping and participation in classroom.But, as we observed here, further studies (Song, 2018) have shown there is no clear relationship between thinking styles and learning outcomes.Other independent variables, like gender and age still remain without clear relationship explanation.With that in mind, our results about gender thinking styles must be taken prudently.

Conclusion
Data collected and analysed this way could be very useful for teachers in the long term.It allows better didactics strategies, regarding both theoretical and practical tasks, as far as it points out students' strengths and weaknesses.
For example, in this case, it can be foreseen that women probably will not show interest on activities that could seem chaotic or unstructured at first (Anarchic form), meanwhile men will not be attracted to proposals that imply doing several things at once (Oligarchic form).Also, both sexes will tend to prefer collaborative tasks (External scope) over individual ones, so teacher could think about leveraging this, working on the inherent benefits of the Internal approach.
Sternberg-Wagner test is a well-known instrument, data gathering can be easily done in one hour and, even, it is possible to do that through online-forms.So, its impact on a course schedule is minimal.Also, its statistical analysis is reasonably fast, with free or open source tools, as shown above, and grants its access to every teacher with a computer and internet access.Thus, its results can be applied to class plan one or two weeks after data gathering.
However, we can detect several weaknesses like the need of a team-work subject oriented, as it makes much easier identify patterns.In the same way, it is an uncommon approach due to its strong mathematical component, so it requires a different way of doing things as teachers.A clear threat to this is the teacher background on statistics and ITC, as far as it is therefore essential that to have certain knowledge for those methods and tools.
Finally, we think that this class-research tool requires more research to clarify and optimize, also bigger study groups and coordinated researches in different subjects and education levels.But even without it, this approach has proven its usefulness in a practical class context.Current lines of work about thinking styles as predictor for student's performance (Matošková & Kovářík, 2016) or creative thinking (Chegeni, Darabi & Niroomandi, 2016) are good examples of it and of its utility to continue exploring the benefits of this approach for the improvement of teaching in the field of Social Sciences, and represent the evolution of this scientific approach since the first systematic international studies on Sternberg's intelligence theories (Sternberg, Castejón, Prieto, Hautamäki & Grigorenko, 2001).

Table 1 .
Conservative Do things in a proven and real manner and follow traditions.Thinking styles dimensions and meanings according Sternberg-Wagner test ScopesInternal Work alone; focus on the inside and to be independent.External Work with other people; focuses on the outside and to be interdependent.LeaningLiberal Do things in a new manner and deviate from traditions.

Table 2 .
Individual thinking styles.Paired with sex, picked group and marks Figure 2. Marks, ordered by FM resultsConsidering Sex, males tend to be Mixed (53.3%) or Judicial (27.7%) on category Functions, meanwhile females are mostly Judicial (42.85%) or Legislative (28.57%).On category Forms, both are clearly Hierarchic, 73.3% for men and 57.14% for women, but with cluster exclusions: Oligarchic is not present on men as it happens with Anarchic on women.Something similar happens in category Levels, with both males and females choosing Local over Global (60% and 57.14% respectively).Also, with category Scopes, were External (60% and 71.42% respectively) is the dominant option.Finally, we found another clear choosing difference in category Leaning, as men chose almost fifty-fifty between Conservative and Liberal (46.66% each), but women were clearly Liberal over Conservative (71.42% and 28.57% respectively).If we repeat this analysis, focusing this time on the influence of each thinking style category inside group elections, we gather 40 packs of 2 or more people, as shown in Table3:

Table 3 .
Group election by category and its thinking styles cluster size

Table 4 .
Association rules selection in each group