U.S. Academic Scientific Publishing
6.0 Variable Transformations: Inflation Adjustment and Lagging
This section describes variable transformations that were used in the analyses. Monetary measures were deflated using the GDP implicit price deflator. Measures of research resources were lagged relative to publication date to account for the time necessary to conduct research, submit a journal manuscript, and have the manuscript published. Citations were given negative lags relative to publication dates, to account for the time difference between publication and citation in later articles. However, the effect of different choices for lags was modest suggesting that our analyses were robust.
To adjust for the effects of inflation, all monetary variables were deflated using the GDP implicit price deflator. Since academic R&D expenditures are used for a variety of purposes, and inflation rates were not available for salaries of faculty, non-faculty research personnel, and S&E postdoctorates, nor for material costs associated with research, we cannot determine whether this inflation adjustment was successful in equating real purchasing power.
Because there is a delay between receipt of funds for R&D research, the conduct of that research, the publication of results in journals, and the citation of other publications, we examined the correlations among lagged versions of monetary measures, personnel measures, publication measures, and citation measures. In general it appeared that publications in a particular year were most highly correlated with personnel measures from the previous year, monetary measures from two years previously, and citation measurements two years in the future. Consequently in subsequent analyses we modeled publications in a given year as a function of personnel measures from the previous year and monetary measures from two years before. We modeled citation measurements as a function of personnel measurements from three years previous and monetary measurements from four years previous. However, we note that the correlations did not change dramatically with different lags than the ones that we chose, and consequently our choice probably had little effect on the results of subsequent analyses.
The correlations among the publication and forwarded citation measures (i.e., citation counts "lagged" by 2 years) were quite high. The correlations among the four publication measures ranged from 0.982 to 0.996. The correlations among the four citation measures ranged from 0.993 to 0.996. The correlations between the publication and citation measures ranged from 0.764 to 0.857. The correlations of fractional count publications in the fixed journal set had correlations between 0.764 and 0.798 with the citation measures; the other publication count measures had correlations between 0.804 and 0.857 with the citation measures.
A factor analysis on publication counts and forwarded citation counts (i.e., citation counts "lagged" by 2 years) aggregated to the institution-year level yielded a primary factor accounting for 95.8% of the variance. That is, it is possible to calculate an index using the different publication and citation measures, which is very highly correlated with those measures, and consequently, summarizes the variability in those measures across institutions and years. This index is called the first principal component. The first principal component was essentially the average value of the normalized publication and citation measures, and the correlations of the first principal component with the eight publication and forwarded citation measures ranged from 0.931 to 0.970. This analysis was repeated for each of the field groups with essentially identical results.
 For example, publication counts in 1990 were correlated with citation measures in 1992.