This paper empirically investigates wage inequality within the group of skilled workers in the recent four decades in the USA using CPS data and finds evidence that the trend of wage growth of the top and bottom 10th percentile of skilled workers significantly diverged starting from 2000. Using a task-based framework of occupation, I find that the changing trend of wage inequality was entirely driven by one category of occupation, namely the nonroutine analytic occupation. Then, I consider in a model task reallocation between two broad task categories, namely the routine and abstract task, induced by an ongoing investment-specific technical change. In my model, the labor in the routine task is replaced by cheaper machines due to investment-specific technical change, then workers that are less productive in the abstract task enter abstract occupations. As a result, the wage inequality in the abstract task widens because of the reallocation of less productive workers from the routine task to the abstract task, that is, the “composition effect.” In addition, since economic agents tend to postpone the investment in machines after the ongoing investment-specific technical change takes place for a while, the expansion path of wage inequality is not linear but features an acceleration of wage dispersion in the middle of the technical change. The quantitative results suggest that the model is able to provide a well-matched timing and magnitude of the nonlinear expansion path in wage inequality that is observed in the data.

The presumption that R&D is a key driver of economic growth is difficult to reconcile with empirical evidence. For example, in most studies, which identify technical change with total factor productivity (TFP), the link between TFP and measures of knowledge is found to be weak.

This paper shows that a reconciliation may be possible in a model where R&D contributes to growth through investment-specific technical change. Such a model predicts that the empirical link between knowledge and productivty would be weak even if the generation of knowledge is the predominant factor of economic growth. The paper also shows that estimates of the production function for knowledge using patent data may be biased.