R & D经费结构与创新产出的差异化关联——基于2014~2023年省际面板数据

Differential Associations between R & D Expenditure Structure and Innovation Output—Evidence from a Provincial Panel, 2014~2023

摘要: R & D经费在基础研究、应用研究和试验发展之间的配置，可能与不同形态的创新产出呈现不同关联。基于2014~2023年中国31个省级地区的面板数据，以发明专利授权数和规模以上工业企业新产品销售收入分别衡量知识创新与经济创新，采用双向固定效应模型、Global Moran’s I和空间杜宾模型，并比较时间、区域及样本设定。相对于试验发展份额，应用研究份额与知识创新产出呈5%水平的负向条件关联，基础研究份额与经济创新产出呈10%水平的负向条件关联，其他基准核心系数未达常用显著性水平。两类产出均有正向空间相关；在陆地邻接设定下，知识创新中两项结构变量的本地直接效应区间不含0，间接效应证据有限。经济创新主方案中，基础研究份额的负向直接与正向间接效应均获区间支持，总效应区间包含0，间接效应的支持随样本变化。经济相似性矩阵下，部分空间效应的方向与统计支持发生变化。结果显示，经费结构的统计关联因产出维度和空间位置而异，部分估计对指标及样本处理敏感。

Abstract: The allocation of R & D expenditure among basic research, applied research, and experimental development may be associated with different forms of innovation output. Using a 2014~2023 panel of 31 provincial-level regions in China, this study measures knowledge innovation by invention patents granted and economic innovation by new-product sales revenue of industrial enterprises above designated size. It applies two-way fixed effects, Global Moran’s I, and a spatial Durbin model, with comparisons across time, regions, and spatial samples. Relative to the omitted experimental-development share, the applied-research share has a negative conditional association with knowledge innovation at the 5% level, while the basic-research share has a negative conditional association with economic innovation at the 10% level. The other core fixed-effects coefficients are not significant at conventional levels. Both outputs exhibit positive spatial correlation. Under land-contiguity weights, the confidence intervals for the direct effects of the two structure variables on knowledge innovation exclude zero, whereas evidence for their indirect effects is limited. For economic innovation, the main sample scheme supports the negative direct and positive indirect effects of the basic-research share; the total-effect interval includes zero, and support for the indirect effect varies across sample schemes. Under economic-similarity weights, some effects change in direction or statistical support. The associations therefore differ by output and spatial location, with some results sensitive to measurement and sample choices.