Modelling approach to estimate pertinent human criteria for a selection and orientation process of a technical profession.

Popescu, Catalin ; Boussier, Jean Marie ; Boussier Ion, Luminita 等

Abstract: What we are trying to accomplish may be considered a career "guide" for the graduating students of technical profiles. This methodology is a hybrid coupling between a marketing technique called "declared preferences" and an approach for process optimisation named "Design of Experiments". It is a flexible approach that allows testing conjointly the effects of various criteria and interactions between criteria in order to point out the abilities, skills or human values of a future engineer. We proposed to test it in two parallel universities: UPG, Ploiesti (Romania) and EIGSI, La Rochelle (France) with the complementary objective to compare and to support the efficient integration of students during international exchanges between technical universities.

Key words: creativity, education, orthogonal array, Doehlert matrix

1. ABOUT AN ENGINEER'S HUMAN VALUES

Except the technical competences of an young engineer, the objective judgment of an evaluator/ consultant is based on diverse criteria such as creativity, responsibility, efficacy, ethics: these values are frequently neglected in a teaching process or, when there are pointed out, each one uses a "fuzzy" perception of these values which is unable to assist a student for his career orientation or to help an evaluator in the selection process of candidates (figure 1). Furthermore it is well known that the lack of a preliminary experience of students in R&D domain or in management field has drastic consequences in their orientation.

Specialists in sociology, in pedagogy or in psychology proposed interesting methods to evaluate human values of an engineer (Pais, 2004) which are frequently judged as being too "sophisticated" or not enough "scientist" by both partners as well in an academic formation (student and mentor) as in a negotiation process for integrating a company (young engineer and manager). Generally this kind of problem is solved by an interview where the evaluator wants to "trap" a candidate, who carefully prepared "eligible" answers.

In response, approaches based on multi-criteria analysis were developed by specialists in operational research (Grabitch, 2003); (Roy, 2006) to conduct such kind of questionnaire, but two important problems were pointed out.

[FIGURE 1 OMITTED]

Firstly, what kind of criteria must be included to define a quality of an engineer (such as creativity or responsibility)? Our point of view is not to search to design "universal" models, but to give a large flexibility to select particular criteria in agreement with specificities of each university or company. Secondly, how to take into account possible dependences between criteria (Rico, 2005)? We do not speak about "structural" dependencies (such as the candidate is "young" and "without a great experience", certainly correlated), but about dependencies between "preferences", very difficult to perceive and to capture in a model. Just an example: two candidates must be classified for a job. A questionnaire step by step (each question tests only one criterion) shows the following evaluations (scores in range 1-5): for technical competences [S.sub.T1] = [S.sub.T2] = 2; for experience level [S.sub.E1] = 5: [S.sub.E2] = 3; for facility to communicate [S.sub.C1] = 2; [S.sub.C2] = 5.

Now suppose that the evaluator preferred candidate 2. Who can conclude that if [S.sub.T1] = [S.sub.T2] =5, the choice could be the same??? This means that the effect of a criterion is depending on the value of another criterion; in this case a dependence (interaction) exists having a positive or a harmful effect (sometimes more drastic than criteria!). The solution could be to test jointly several criteria at several levels (values) in order to capture the interaction effects.

2. COULD HUMAN VALUES BE "MODELLED"?

In order to obtain the maximum information with such a questionnaire, a full matrix with all possible combinations of criteria and levels is necessary. Let us suppose a questionnaire in order to test the effects of the variation of five criteria, each one having two levels. A full factorial array using 32 questions will be strongly inadequate for this kind of analysis. Design of Experiments method uses an orthogonal factorial array that is a subset of a complete array (Taguchi, 1987). Take an example conjointly studied in UPG (Ploiesti) and EIGSI (La Rochelle) where specialists in automatic and in computer sciences decided to elaborate a model for "creativity" in order to help students for orientation in R&D domain. Conjointly, the set of tested criteria was the same: N (Innovator character of a solution), C (critical analysis of existent solutions), F (feasibility degree), P (estimated cost). Two levels were affected to each criterion (low, high) and respondents placed in hypothetical scenarios done by L8 (2) (7) factorial array must give an answer on the range from 1 to 8. A part of this orthogonal array is presented in Table 1 (8 questions for 16 possible scenarios).

The questionnaire is convivial and without "trap". For the 5th scenario: if the idea is "original", "without" critical analysis of literature, the feasibility degree is "great" and the cost is "high", the score is "4". An additive matrix model based on ANOM (analysis of means [Vigier, 1988]) allows us to estimate the scores for all unobserved scenarios. Reduced models are obtained by Analysis of Variance and contain only effects of pertinent criteria and interactions. A 9th question (out L8 (2) (7) array) served to validate the models.

Models, which are presented below, show the teacher's team perception for the creativity of an engineer:

For UPG Ploiesti:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (1)

For EIGSI La Rochelle:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (2)

where

[a.sub.i] = [s.sub.mean] ([A.sub.i])-[S.sub.mean]

are matrix elements of mean effect of criteria A at level i,

[a.sub.i][b.sub.j] = [s.sub.mean] ([A.sub.i], [B.sub.j])-[s.sub.mean]-[a.sub.i]-[b.sub.j] (4)

are matrix elements for mean effect of AB interaction for A at level i and B at level j with [S.sub.mean]--mean value of all scores; [S.sub.mean](Ai)--mean value of scores if A is at level i; [S.sub.mean]([A.sub.i], [B.sub.j])--mean value of scores if A is at level i and B at level j.

Similar models are in development: for ethics (criteria: conflicts of interests, confidentiality, professional probity) and for responsibility (criteria: responsibility assuming, loyalty and collegiality, respect for the authorities, procedures and rules).

During the training period, this kind of models offers a possibility to compare the mentor's exigencies with the student's perception for a possible orientation (R&D, production etc.) or for corrective actions. It is also possible to compare education exigencies for similar technical education systems. Presented examples show drastic differences: if the "originality of the idea" is the most pertinent criterion for both, the points of view are completely different concerning the second one and the interactions weights. To integrate students during international programs exchanges professors have to realise a supplementary work; the solution is not to find a "middle" way (model), but to develop sensibilities satisfying both the "cost" perception and the "feasibility degree" of a technical project.

3. HOW COULD HUMAN VALUES BE "SOLD"?

Within a negotiation process between employer and the candidate each side expresses requirements and/or preferences that in many cases are opposite. The search of an optimal solution in this problem consists in defining a combination of levels of variables for better satisfying the requirements and constraints of the ones (recruiters) and the preferences of the others (candidates). Doehlert (Doehlert, 1970) proposed an original construction of plans that consists in uniformly covering the experimental framework. The design of the matrix is based on the definition of an initial simplex. The operation of coding consists in transforming value vi corresponding to the variable V into a coded value [x.sub.i]:

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (5)

Levels tested are: 7 levels for salary [1000 3000] euros; 5 levels for time frame/day [member of] [8/12] hours; 3 levels for benefits.[5% / 15%]; 3 levels for career management [member of] [2/5] years. Thanks to a regression model, we can estimate the coefficients of a score function and residue value.

[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] (6)

Each answer [S.sub.i] (range 1-8) is transformed starting from a function of individual desirability [d.sub.i] whose nature depends on the objectives to research (to maximize the score for each negotiation partner).

[d.sub.i] = [s.sun.max] - [s.sub.i]/[s.sub.max] - [s.sub.min] (7)

Global desirability ([d.sub.i]g]) is designed for each scenario i by using geometric means of individual desirability. The model of desirability makes it possible to identify the levels of the variables that make it possible to research the maximum value of global desirability. Virtual negotiation was tested at EIGSI: the compromise solution, in agreement with evaluator exigencies and candidate desires, is: 2500 euros, 9 h, 7% benefits and career promotion after 3 years. It is also possible to imagine a weighted geometric mean (weights of partners are different) but is it an "ethic" attitude?

4. DISCUSSIONS AND PERSPECTIVES

This work proposes a methodology to evaluate human qualities or values (creativity, responsibility and so on) of young engineers preparing a career in technical fields. A convivial questionnaire based on orthogonal arrays can capture the effects of criteria and the interactions and can be used for several objectives: to orientate students for a particular option (R&D, production etc.), to compare the points of view of students and mentors from different universities which are concerned by programs such as Socrates or Erasmus, to negotiate the professional insertion for a better exploitation of technical and human qualities. At this moment, an algorithm based on Fuzzy Logic is designed in order to model the "affinity" for a job, by including creativity, responsibility and technical competences.

5. REFERENCES

Doehlert, D.H. (1970). Uniform shell designs. Journal of royal statistical society, serie C, ns19, 1970, pp. 231-239

Grabitch, M. & Perny, P.(2003). Agregation multicritere. B. Bouchon-Meunier, C. Marsala (eds). Logique floue, principes, aide a la decision (Fuzzy logic, principles, aiding decision-making), Paris, Hermes, pp.81-120

Pais, A. (2004). The role of the 21st Century Engineer, Available from www.iee.org/OnComms/Circuit/members /WATW_Adrian_Pais_21st_Century_Engineer.pdf Accessed: 2007-07-05

Rico, A. (2002). Modelisation des preferences pour l'aide a la decision par l'integrale de Sugeno (Preference modelling for decision-making aiding by Sugeno integral), Universite Paris I, these de doctorat (thesis)

Roy, B. (2006). A propos de la signification des dependances entre criteres (Meaning of criteria's linkage): Quelle place et quels modes de prise en compte pour l'aide a la decision? 7e Congres de la Societe Francaise de Recherche Operationnelle et d'Aide a la Decision, Lille, France, 6-8 fevrier 2006

Taguchi, G.(1987). System of Experimental Design. Dearbon: Don Clausing UniPub/ Kraus International Publications, Vol I & II, White Plains, NY, USA, 1987

Vigier, M.(1988). Pratique des plans d'experiences: Methodologie Taguchi (Taguchi Methodology), Les Editions d'Organisation, Paris.

Table 1. Example of [L.sub.8][(2).sup.7] array completed by an expert

Test N C F P Score

1 low without great low 4
... ... ... ... ... ...
5 high/original without great high 4