Accurate forecast of countries’ research output by macro-level indicators

There is a great variation of research output across countries in terms of differences in the amount of published peer-reviewed literature. Besides determining the causal determinants of these differences, an important task of scientometric research is to make accurate predictions of countries’ future research output. Building on previous research on the key drivers of differences in countries’ research outputs, this study develops a model which includes sixteen macro-level predictors representing aspects of the research and economic system, of the political conditions, and of structural and cultural attributes of countries. In applying a machine learning procedure called boosted regression trees, the study demonstrates these predictors are sufficient for making highly accurate forecasts of countries’ research output across scientific disciplines. The study also shows that using a functionally flexible procedure like boosted regression trees can substantially increase the predictive power of the model when compared to traditional regression. Finally, the results obtained allow a different perspective on the functional forms of the relations between the predictors and the response variable.