THE DISCRIMINATION METHOD FOR THE TRANSFORMATION PROCESS OF RESULTS OF RESEARCH PROJECTS INTO PRACTICAL APPLICATIONS

Received: 28 August 2011 Abstract Accepted: 8 January 2011 The purpose of this paper is to discuss the usage of discrimination tools in R&D activities especially in Innovative Programme. Based on these tools, in this paper is drawn up a method for project evaluation and their analysis. A case study has been adopted based on projects realized by ITeE-PBI in Radom within the PW-004 Multi-Year Programme. The main obtained results are following: selection of innovation project characteristics, evaluation of rank of project characteristics, prediction of launch data for a innovative solutions, sensitivity analysis. The presented results indicate that this mathematical tools can be very useful for innovation project analysis.


Introduction
Over recent years innovation policy has risen to the fore.It is now well recognized that among other things competitiveness are dependent on industrial innovation.Government initiatives to support innovation have proliferated.These initiatives named here as a Innovation Programme take many forms supporting R&D units [1].In Poland, the most popular form of Innovation Programme were Multi-year Programmes, in which the linear model of innovation was implemented.
Scientific units participated in such Innovation Programme usually ended its activity by making products or services with new or different characteristics without their commercialisation.
It means, that only part of innovation chain was realized in such programmes.Lack of realization such phases of chain as a product commercialization, mar-ket ready product and market entry1 indicate to need construction of supporting tools.
These tools should support of science units activity to conversion obtained results into innovation (practical application).It means that adoption of such tools as Performance Measurement System (PMS) in science units is needed.
The most relevant questions that have been related to PMS of R&D are [3]: • how to design a PMS that fits the characteristics of R&D activities, • how the PMS should be implemented and which are the major effects of its use, • how metrics and indicators should be related in an uncertain and unpredictable context like R&D.
The set of metrics and indicators may help to answer the following research question: which features of research project decided of its market success.
Selecting proper dimensions of performance and operative indicators to be used in a PMS for R&D units is a challenging task because of the high level of uncertainty, the relevance of intangible factors, the complexity and the low standardization of R&D activities [3].However, literature2 has deeply investigated the problem and gives us several suggestions but only for R&D units of enterprises.
This paper is focused on the last above mentioned question 3 .For this purpose a discrimination method has been proposed.The application of the methodology based on discrimination methods is carried out in a leading Multi-years Programme PW-004.
Results of the empirical research consist of: • evaluation of rank of project characteristics; • prediction of date launch; • making sensitivity analysis.
Last section concludes and outlines some directions for future research.

Innovation Programme PW-004
During 2004-2008 in Institute for Sustainable Technologies-National Research Institute (ITeE-PIB) 4 in Radom was realized the multi-year programme PW-004 entitled "Development of innovative systems of manufacturing and maintenance 2004-2008 ".
The Programme passed by the Council of Ministers on 21 st April 2004, as an executory element of strategic governmental programmes for innovativeness development, was generated within the realization of the economy growth policy of the country presented in the programme document "Enterprise-Development-Labour " [7].
The Programme was focused on the development of new solution in strategic research fields, in which Poland has a scientific potential, including a laboratory base, technological infrastructure and the possibilities of international co-operation.
Projects realized during this Programme are concentrated on 5 areas such as Over 70 scientific units participated in realization of the PW-004 Programme and brought exceptionally extensive scientific and application achievements.
The effects of the Multi-Year Programme include 327 new solutions ready to be commercially implemented (59 of which have already been patented) among which are: • 142 original devices (test apparatus mainly); -111 production technologies;, • 74 system solutions particularly for the support and automation of production processes.Possibilities of practical application of developed products and technologies were verified in 60 model implementations in real maintenance conditions [8].
As a part of PW-004 programme a research task entitled "Methodology of the transformation of results of scientific research into practical applications" was also realized.One of among other effects was a quantitative effectiveness analysis by using discrimination tools.The analysis mainly concerns the effectiveness of research realization and the possibilities of their implementation [9].

The idea of the discrimination method
According to [10], discrimination methods require information about the form of projects completion.It was suggested that the effect of the implementation of the projects is expressed by time needed to implementation them.
In the set of projects were distinguished ones with results: • already implemented or these which implementation is sure (success), • having a chance for implementation within one year (partial success), • having a chance for implementation within at least two years or without this chance (failure), The above definition of implementation effects, allowed to divide populations of research projects into separate classes (groups) with the different degree of advancement of implementation potential.
Every project was considered as multidimensional object described by set of 37 chosen features.Each of them potentially affect the result of implementation.
The usage of discrimination method depends on distributing the elements of a set of observation among K of classes (groups) so that every of them is characterized by a maximum homogeneity.The elements of these classes are known.In literature such set of observation is called as a learning set.
The distribution of population into classes were possible through defining effects of projects implementation.So, three classes of projects were distinguished (K = 3): 1. first class which contains 17 projects with results implemented or those which implementation is highly probable at the moment of the finalization of R&D works, 2. second class which contains 14 projects with results that have a chance to be implemented within one year, 3. third class which contains 21 projects with results that have a chances to be implemented within at least two years.
Using discrimination methods, some features of projects were selected according to following criteria: the most similarity of features of projects within groups and maximum differentiation of them for projects which belong to other groups.
So, selected features influence on project results implementation, are significant from statistical point of view and indicate the success or the failure of the final result.Finally, the main results of discrimination methods are following: 1. Selection of features which significantly influence on the level of advancement of the implementation; 2. Assessment of the importance of features; 3. Construction of classification functions, which enable to make a prediction of degree of advancement of implementation potential and to make a sensitivity analysis.

Assumptions of the analysis
Presented analysis utilizes the database filled by author.Both final reports of every research project and interviews with thematic area managers (from ITeE-PIB) were employed in data collection.
Every task was described by 82 features6 .Among of them were both quantitative and qualitative characteristics of project 7 .The latter were expressed by binary variables what let the author realize their aggregation.Due to that fact, the number of variables was reduced to mentioned earlier 37 and used in further analyses.Additional, to identification of degree of advancement of results, so called grouping variable y was defined: • y = 0 for the first group of task already implemented; • y = 1 for the second group of tasks that have a chance to be implemented within one year; • y = 2 for the third group of tasks that have a chance to be implemented within at least two years.

Results of the discrimination analysis 8
As mentioned earlier, every project is described by 37 variables.The literature [13] suggests, that these variables can be classified at least into following major categories: 1. the financial performances of the project; 2. the project's value; 3. the technical performances of the project; 4. the efficiency of R&D activities of the project; 5. the capacity of fitting in the estimated cost and duration of activities; 6. the degree of integration between the R&D and production activities; 7. the degree of integration between the R&D and marketing activities.By the discrimination analysis, x variables were selected, which significantly differ the projects located in different groups and differ not significantly in the same groups.These variables (see Table 2) influence on implementation time of projects results.Finally, 11 from among 37 of x significant variables are selected which influence on the time of the project results implementation 10 .The classification matrix, given in Table 2, prove the good quality of chosen x variables.1, relatively well identify (recognize) effects of the projects results implementation.These variables represent features of the projects which significantly influence on implementation time.
In the next step of analysis, two essential canonical variables U1 and U2 were received.They are linear combinations of x variables and diversify tasks to groups (Table 3).The first canonical variable U1 diversifies in 66% (cumulative probability), whereas the second variable U2 in 34% (cumulative probability -66%).It means that in further analysis it should be taken both these variables.In another words, 11 of x variables given in the Table 1, can be replaced in analyses by two canonical variables.On the basis of results presented in table 3, it was derived the following form of canonical variables U1 and U2: where ẋ means the standardized x variable.Coefficients value in (1)-( 2) indicate rank of x variables in project classification process.average values of above variables were calculated for each group (Table 4).Significant difference between average values of the variable U1 in groups G 1:0 and G 3:2 (1.33-(-1,66)) indicates that this variable can diversify projects to group of projects already implemented (G 1:0) and to group of projects with the implementation time at least 2 years (G 3:2).However, taking into account average values of the variable U2 in group G 2:1 (1,49), notice that the second variable can fulfill the variable U1, distinguishing projects with the implementation time within one year.The confirmation of this fact is graph 1 of objects (projects) in the space of canonical variables.This graph presents groups of objects in the space of the variables U1 and U2.
The main conclusions from this graph are following: • x variables with positive coefficients of the first canonical variable U1 (Table 3, column first) represents these features, which may increase the time of implementation ("moving" from the group G 1:0 towards groups G 2:1 and G 3:2). • x variables with positive coefficients of the second canonical variable U2 (Table 3, column second) represents these features, which "move" projects to group G 2:1, with the implementation time within one year.
It is necessary to consider interpretation of the influence of x variables simultaneously in the context first as well as of the second canonical variable.
According to above consideration, the following kind of conclusions are possible to obtain: simultaneous appearance of features represented by x11 and x5 variables (positive values of coefficients of U1 and negative value of U2 respectively) can increase the time of implementation to at least 2 years (direction to the group G 3:2).On similar principles, other conclusions about the influence of individual x variables on the time of implementation might be formulated.The next result of discrimination method is a importance (rank) of x variables.It may be described on the basis of so called structural matrix (see Table 5), which contains values of correlation coefficients between individual x variables and canonical variables.For instance, r 11 means correlation coeffi-cient between the x0 variable and the first canonical variable U1 and equals -0.37, whereas r 21 = 0.01 means correlation coefficient between the x0 variable and the second canonical variable U2.The highest and the smallest (in the absolute value sense) values of correlation coefficients are marked with bold font.

Based on values of correlation coefficients it is possible to determine:
• The importance of every variable x on account of the first canonical variable, i. e. distinguishing implemented projects (G 1:0) from other (G 2:1 and G 3:2); • The importance of every variable on account of the second canonical variable , i. e. distinguishing projects with the 1 year implementation time (G 2:1) from other (G 1:0 and G 3:2); • average importance on account of both canonical variables.
The importance of j-th variable on account of the first canonical variable was calculated using the following formula: where The importance of j-th variable on account of the second canonical variable was calculated using the following formula: where Average importance was calculated as a weighted average: where weight λ i means the percent in which the ith canonical variable explains the changing of variables x 11 .Such obtained importance are presented in Table 6.For instance, the x8 variable has the highest importance equals 70.83% and so the strongest influence on changing of U1 and U212 .

Prediction of implementation time
Prediction of the implementation time is possible by usage so-called classification functions which are linear combinations of x variables.These functions indicate one of three above mentioned groups to which the project should be classified 13 .
The number of functions is the same as the number of groups 14 .Values of coefficients of classification functions are shown in Table 7. Therefore we have: -classification function dealing with the first group: -classification function dealing with the second group: -classification function dealing with the third group: The algorithm of prediction which use above functions may be realized according to the following stages: Stage 1: to calculate value of every classification function ( 6)-( 8) by inserting appropriate x values that characterize new project; Stage 2: the function with the greatest value indicate the group, into which project should be classified.
The example of the prediction is presented below.

The example of the prediction
A few projects were excluded from presented here discrimination analysis because of the lack of infor-mation about their degree of the implementation maturity (x0).One of these projects was selected to implementation time prediction.Characteristics of this one shows Table 8.

Table 8
Characteristics of the project.

X0
The assessment of implementation maturity (SDW) Lack of information For the above values of x variables and for different value of x0 three classification functions were calculated.
Table 9 summarizes the obtained results of prediction process.We can conclude that the predicted implementation time changing from two years to the immediate implementation with the changing of x0 from 5 to 6.

Sensitivity analysis
In Table 2, the set of variables was given, which significantly influence on the implementation time of the projects results.Among of them are variables, which changing after the completion of the project isn't possible (for instance x6) and variables which changing is possible only after the project completion.The latter group of variables is presented in Table 10.For instance, the x0 variable determining degree of the commercial maturity may be changed during additional works realized after the completion of the project.The main goal of sensitivity analysis is shown what change of these variables make change of implementation time.
Sensitivity analysis can be realized by using classification functions according to following algorithm: 1.To choose the project to analyses.2. To calculate the value of all three classification functions for analyzed project.
The classification function having the greatest value indicates the group, to which project should be classified.3. To change the value of the some x variables chosen from table 10 in classification functions and check, according to the rule given in point 3, how it will influence on the time of project implementation.Results achieved in the point 4 of above algorithm can suggest, what actions 15 should be taken after the completion of the project, in order to increase the certainty (or speed up) of their implementation.

Example of the sensitivity analysis
Sensitivity analysis was made for the project with characteristics shown in Table 6.This analysis relies on examining how an implementation time will be changed as a result of changes of the variables value chosen from Table 10.The sensitivity analysis was made for two levels of SDW: 5 and 6.These results are presented in Table 11.The current state of the analyzed project is distinguished by bold font.
The main conclusion of sensitivity analysis in the case of SDW = 6 (see Table 10, column 4) are following: • decision related to realization of more further works relying on the making an application for the goal project (x29) can increase the implementation the time to at least 2 years.• decision related to realization of more further works relying on the making an application for a developmental research project (x28) should not influence on the time of implementation; • increment in the ratio market value per unit of project costs (x15) over 10% can influence on increment in the implementation time.Source: self study. 15Represented by x variables from Table 10.

Conclusion
Realization of product, technology or service innovation projects is sometimes connected with problems of implementation their results for practical applications.In the case of Poland that kind of problems may be arisen during execution of long-term Innovation Programme.Polish typical Innovation Programme is considered here programme PW-004.Similar problems exist in the situation of other types of research and generally deal with the building a analytical decision support tools with multi-criteria selection projects in enterprises with R&D activities [15].
The purpose of the present paper was description of quantitative tool which usage may answer the following question: what should be done to the implementation results of project with the success?
Discrimination methods were proposed as a tools of multidimensional statistics.These methods assume the existence of objects (here projects) described by many characteristics.Additional, discrimination method allows to make evaluation based on both quantitative and qualitative criteria 16 .The main goal of usage of these method is selection such characteristics, which in the best way, identify kind of project success.
In this article were obtained answer following research questions: • What appropriate features of project should be selected to supporting decision about practical applications of this results?• Which from these features are the most important?• What would happen with the implementation when we change the conditions of the realization of projects?• Will realized project finish with the transformation of their results into practical applications?A reason of usage of discriminatory methods is their exceptional effectiveness in the case of the forecasting bankruptcy of enterprises by well known Altman model [16] and [17].In their case the objects are enterprises described by many financial ratio and their division into groups is determined by degree of bankruptcy risk.
Therefore, it seemed natural that in case of the innovation projects, analogical approach is most recommended.Whether effective?It would be subject of future analysis based on projects realized later within Programme PW-004.

The author gratefully acknowledges the many helpful suggestions of Professor Andrzej H. Jasiński from Faculty of Management University of Warsaw, during the preparation of this paper.
The author also wishes to thank the Professor Adam Mazurkiewicz, the Chairman of ITeE-PIB in Radom, for facilities of information about research projects realized in his Institute.

Fig. 1 .
Fig. 1.Research projects in canonical variable space.Source: self study.Results of the STATISTICA.

x17 1 x19 1
Projects sale value of the license during one year expected by thematic area managers 0 -lack of expectations x28 Making an application for a developmental research project 0 -no x15 Approximate market value established by thematic area managers / costs of the task 0 Value of realized services dur-

Table 1
Results of choice of x variables.The assessment of implementation maturity (SDW 9 ): {1,. . .,10} where SDW=1 means that the essence of action was identified and described, SDW=10-means that the final product was checked with the possibility of producing them in the replication version.
Tolerance express the redundancy of variables, as a consequence of mutual doubling.The lower tolerance indicate variable which is unnecessary.Level p < 0.1 indicates the x significant variable.Source: self study.Results of the STATISTICA.

Table 3
Standardized coefficients for canonical variables.
Source: self study.Results of the STATISTICA.

Table 4
Average values of canonical variables (so-called Centroidy group) in individual groups.
Source: self study.Results of the STATISTICA.

Table 5
Structural matrix of r ij .
Source: self study.Results of the STATISTICA.

Table 6
The importance of x variables.
Source: self study.Results of the STATISTICA.

Table 7
Values of coefficients of classification functions.

Table 11
Results of the sensitivity analysis.