| With the development of network information technology,research collaboration is gradually free from the constraints of time and space and has diversified into cross-territory,cross-disciplinary and multilateral collaboration,but it also brings about the problem of the efficiency of collaborative output.Although some research results have been conducted on this issue,most of them are based on correlation analysis or regression analysis.On the one hand,most of the existing studies explore the impact of cooperation from a correlation perspective,and a strong correlation between two variables may not lead to a causal conclusion due to sample selection bias and confounding factors between variables.On the other hand,most of the existing studies do not consider the common support domain of the control variables and the equilibrium of the control variables,which leads to unrobust or inaccurate results.As causal inference methods have been intensively investigated,the application of matching methods to various domains has made considerable progress.To this end,matching methods are applied to this thesis to investigate the impact of the collaborative model on the quality of papers.The ’two-step estimation’ of matching methods allows us to examine the problem of common support domains and to ensure that the control variables are balanced in the treatment and control groups.The main findings of this thesis are as follows:This thesis examines papers to identify the impact of author collaboration patterns on the number of citations based on the Propensity Score Matching(PSM)method.The PSM is used to analyse and compare the differences in citation frequencies generated by four groups of collaboration patterns: international versus domestic collaboration,international multilateral versus international bilateral collaboration,domestic inter-organizational versus domestic intra-organizational collaboration,and domestic multi-author versus domestic single-author collaboration.Specifically,we conduct this analysis with publications from three computer science subfields in the Web of Science(Wo S)database.In addition,after obtaining causal effects,this thesis also considers the issue of hidden bias caused by unobservable factors,using Mantel-Haenszel bounds for sensitivity analysis of the matching process.Based on the above research,this thesis further proposes a research proposal to identify authors’ propensity to collaborate on their paper citations based on the Generalised Propensity Score Matching(GPSM)method with authors as the research target.We use GPSM to obtain non-linear results for three sets of continuous disposition variables to systematically assess the heterogeneous effects of individual authors’ propensity to collaborate internationally,propensity to collaborate between domestic organizations and propensity to collaborate multilaterally internationally on paper citations.Data from 201,726 scientists in the field of machine learning from 2005 to 2019 are used to construct social networks and use their network metrics as covariates,retaining authors who published at least three papers over a 15-year period for GPSM analysis.In addition,as GPSM cannot address confounding bias caused by unobservables that do not vary over time,we use a two-way fixed effects model to control for both time and individual effects to further validate the robustness of the results.The results of the PSM experiment show that the presence of some form of collaboration in a paper generates more citations than the absence of collaboration.Single-author papers are very few and least influential in the overall sample,with international multilateral collaborations having the highest impact.Unexpectedly,we find that the average frequency of citations of collaborative publications within national organizations was higher than that of collaborative publications between national organizations in the three subfields.The GPSM results show an inverted ’U’ shaped correspondence between the degree of international collaboration of authors and the average citations per paper,with a maximum at moderate levels of international collaboration.Among scientists involved in international collaboration,there is also an inverted U-shaped relationship between the degree of multilateral international collaboration and their citations per article,with the maximum number of citations per article being reached at lower levels of international multilateral collaboration tendencies.In addition,for scientists not involved in international collaborations,the degree of inter-organizational collaboration has an inverse ’N’-shaped relationship to the frequency of citations. |
