Begin typing your search above and press return to search. Press Esc to cancel.

Predicting future scientific success


Decisions to hire, promote, and fund a scientist depend on the future prospects of the candidate. Until my article Future impact: Predicting scientific success, Nature,  was published, most of these decisions were based on measures of past success. These measures included citations, number of publications, and the hindex. My article showed that these measures alone are poorly predictive of future performance, specially for long-term horizons such as 10 years. This result produced many policy-relevant changes in the way in which academic institutions think about their candidates. It also spurred a large number of research projects aimed at developing better measures of future performance.

Predict your future h-index

These are approximate equations for predicting the h-index of neuroscientists in the future. They are probably reasonably precise for life scientists, but likely to be less meaningful for the other sciences. Try it for yourself online at go.nature.com/z4rroc.

Try online equation

The accuracy of future h-index prediction decreases over time, but the Acuna et al. formula predicts future h-index better than does current h-index alone. The contribution of each factor to the formula accuracy also changes over time. Shading indicates 95% confidence error bars.

What's next

A life-time index of academic productivity

The problem with several bibliometric measures of publication success is that they are heavily dependent on the scientific age of the research. For example, it is absurd to compare the h-index of a young researcher and an old researcher because obviously the old researcher has had more time to publish and get to be known. In this project, we propose to estimate the life-time impact of a scientist, which will directly correct for differences in career stages.

Combination of bibliometric indices

There are dozens—if not hundreds—of scientometric indices: h-index, m-index, ar-index, and so on. All of them have different aims and need different features. In this project, we aim at using all these measures and estimate how well we can predict all the others. By doing variable selection, we will be able to see which is the minimum set of variables that predict all measures the best. The goal is to provide one final index that is the best at predicting everything.