WU Di( )， PENG Rui( )， SUN Dongbai()， XU Wenchao()， QIAO Lili()

1 Shool of Management, Xi’an Jiao tong University, Xi’an 710049, China2 Donlinks School of Economics &Management， University of Science & Technology Beijing， Beijing 100083， China3 National Center for Materials Service Safety， University of Science & Technology Beijing， Beijing 100083， China

Abstract： Research infrastructure is crucial for development of research， and thus the evaluation of its performance is important. However， existing researches mostly focus on its past observations， lacking of a prediction for future. In this paper， procedures are proposed to predict the distribution for the number of papers published in a certain future year. The publication reliability， which is defined as the probability that the number of published papers in the future year is bigger than a pre-specified number， is evaluated. Illustrative examples are proposed to show the applications of the model.

Key words： research infrastructure; performance; prediction; reliability

Introduction

Research infrastructure is significant crucial for the development of fundamental research. Typical research infrastructure includes variety of experimental devices， databases and materials[1]. The performance of research infrastructure comprises of many aspects[2-3]， and the most important and direct reflection of performance is the number of publications. Other indicators include approved patents， cultivated talents， products，etc. In recent years， many researchers have studied the performance of research infrastructure[4]， however， just based on past observation and thus lack of prediction for future performance.

Without loss of generality， the number of papers is used as indicator of research infrastructure， and procedures based on historical data are proposed to predict future publication. Actually， for many research infrastructures， publications which are based on researches done with the infrastructure are publicized on related websites. Therefore， people will feel easily to obtain the historical data in practice. More specifically， the distribution for the number of papers published in a certain year can be estimated， based on which the probability that the number of papers is higher than a pre-specified requirement can be obtained. This probability is termed as the publication reliability of research infrastructure[5]， the concept of which is borrowed from engineering field.

The rest part of the paper is formed as follows. Section 1 makes a formulation of publication reliability. Section 2 presents the illustrative example. Section 3 makes a conclusion and points out some future works.

1 Formulation of Publication Reliability

Consider a research infrastructure which has already run forNyears. We use the parameterOito denote the number of papers published in each year. The publication reliability is defined as the probability that the number of papers published in yearN+Mwill be bigger than a pre-specific requirementDas

R=P(ON+M>D).

(1)

The relationship between the numbers of papers published in adjacent years is assumed to observe the following relationship

(2)

Actually this assumption is widely used to describe the relationship between two adjacent events， that is to say， the stock prices in adjacent days and the profit of a company in adjacent years[6].

Based on the distribution of formula(2)， the number of papers published in a certain year dividing the number of paper published in the year before the certain year follows the logarithmic normal distribution. By employing the historical data of each research infrastructure， we can obtain the expectation and the standard deviation of the logarithmic normal distribution. Since ln(ON+M)-ln(ON) represent the sum ofMnormal variables， it approximatively follows the normal distribution， and the relationship is

(3)

Based on the axiom that logarithmic function is monotonous，R=P(ln(ON+M)>ln(OD)) is obtained. As a result， we can solve the publication reliability based on the distribution of ln(ON+M). Since ONis given and ln(ON) can be calculated， ln(ON+M) also follow the normal distribution and

(4)

2 Illustrative Examples

Here we obtain the data from Chinese Academy of Sciences， in particular， the sharing service platform of CAS large research infrastructures. Since it only recorded the number of published paper from 2010 to 2015， we use these years’ data as illustrative examples except for 2015. Because some research infrastructures haven’t been built so long (Remote Sensing Aircraft) or have already finished its task (Land Observation Satellite Data Receiving Station)， we eliminate them in order to get enough data to analyze. The data of 2015 are eliminated because the service platform where we obtained the data have’t collected the complete number of published paper of each infrastructure in 2015. Note that the numbers here include all types of paper published， which combine not only SCI or EI but also CSCD， international conferences and domestic conferences. The specific number of published paper from diverse infrastructures these years are shown in Table 1. We choose four first-level infrastructures and four second-level infrastructures where first-level infrastructure serves for research whereas second-level infrastructure for public service. The diverse levels of infrastructure are shown after their names.

In Tables 1-8, we use(1) to represent the infrastructure belongs to first-level while (2) represents the infrastructure belongs to second-level. Here, NSRL is National Synchrotron Radiation Laboratory; BEPC is Beijing Electron Positron Collider; SSTV1 is Ship Science and Techology V01; SSRF is Shanghai Synchrotron Radiation Facility; FAST is Five-hundred-meter Aperture Spherical Radio Telesope; BPL/BPM is Large Scale Scientific Facilities BPL/BPM; ODECAS is Center for Observation and Ddgital Earth Chinese Academy of Sciences; LMFST is Large Sky Area Multi-object Fiber Spectroscopic Telescope.

Table 1 Number of published papers for diverse

We also calculate the average value and the standard deviation of this sequence and demonstrate the result in Table 3.

InfrastructurelnO2011O2010 lnO2012O2011 lnO2013O2012 lnO2014O2013 NSRL(1)0.080.080.130.06BEPC(1)-0.020.260.180.40SSTV1(1)-2.481.700.440.39SSRF(1)1.360.140.47-0.23FAST(2)0.36-0.250.25-0.22BPL/BPM(2)0.080.360.16-0.20ODECAS(2)-0.130.23-0.54-0.60LMFST(2)-1.101.250.130.56

Table 3 Average number and standard deviation of the sequence

Basically the expected life length of infrastructure is around 30 years， and thus we assume 5 years as a phase or period. Now since we know the existing data conclude 5 years numbers of published number， we obtainO5+6，O5+11，O5+16as the expected number of published papers. In other words， we try to solve the publication reliability in 2020， 2025 and 2030 as the illustrative examples. With the in-depth study and the development of technology employed in these infrastructures， the number of published paper in a certain year is definitely larger than the previous year. Nonetheless， sometimes the popularity of some specific field may diminish respect to the time and thus the number may also decrease. Therefore， the pre-specific requirementDfor the year we study is assumed to be 0.9， 1.0， 1.1 times of the average number of 2010—2014， respectively. And these numbers represent three divergent cases each infrastructure may confront: positive， stable or negative. The values ofODand ln(OD) are presented in Tables 4 and 5.

Table 4 Values of OD and ln(OD) for different D when negative

Table 5 Values of OD and ln(OD) for different D when

The values ofM，D， publication reliability and the distribution of ln(ON+M)-ln(ON) in 2020 are shown in Table 6 with positive， stable and negative scenarios. The results of 2025 are given in Table 7 and predictions for 2030 are shown in Table 8.

In Tables 6-8, we use “1.” to represent the publication reliability of the specific infrastructure approximates to one.

Table 6 Values of publication reliability and the distribution of

Table 7 Values of publication reliability and the distribution of

Table 8 Values of publication reliability and the distribution of

In 2020， some infrastructures can definitely reach the pre-specific requirement we set， and NSRL is a great example of this type， while some may hardly achieve the goal like ODECAS.

In 2025， while some publication reliability goes down comparing with the probability we predicted in 2020， some values increase. Here we find out that the publication reliabilityR(1.1) is higher than that for 2020 for most scenarios.

In 2030， the phenomenon of fluctuation is getting much more obvious. The publication reliability infrastructure like SSTV1 is going down during 2020—2030 based on our estimation.Contrary from that， the same indicator of SSRF increases around 3%-4% since the difference between standard deviation. Note that the standard deviation of Ship Science in 2030 is 7.09， while the standard deviation of Radiation Facility is only 2.71. The other infrastructures follow the same rule because the higher the standard deviation is， the more unstable the trend will be.

3 Conclusions

In this paper， we proposed a model to predict the distribution for the number of papers published in a certain future year based on the previous data. Defined as the probability that the number of published papers in the future year is bigger than a pre-specified number， publication reliability is introduced and employed to predict the future performance of specific research infrastructures. We show eight infrastructures’ data after filtration from Chinese Academy of Sciences， Sharing Service Platform of CAS Large Research Infrastructures and use these data to run some illustrative examples. By calculating the standard deviation and average value of the logarithmic normal distribution， we obtain the estimated publication reliability for eight infrastructures in 2020， 2025 and 2030， which prove the practicability of our model. Comparing the different values in three diverse periods， we find out the relationship between the standard deviation and the publication reliability.

Future researcher can not only employ the published papers just like we used to evaluate the reliability of infrastructure， but also take patents， awards， talents and other potential indicators into consideration and come up with a more comprehensive estimation of infrastructure’s performance. Besides， new models can be proposed and compared with our model to find out which model is more appropriate.