Should citations be field-normalized in evaluative bibliometrics? An empirical analysis based on propensity score matching

Field-normalization of citations is bibliometric standard. Despite the observed differences in citation counts between fields, the question remains how strong fields influence citation rates beyond the effect of attributes or factors possibly influencing citations (FICs). We considered several FICs such as number of pages and number of co-authors in this study. For example, fields differ in the mean number of co-authors (pages), and - on the paper level - the number of co-authors (pages) is related to citation counts. We wondered whether there is a separate field-effect besides other effects (e.g., from numbers of pages and coauthors). To find an answer on the question in this study, we applied inverse-probability of treatment weighting (IPW) which is a variant of the propensity score matching approach (an approach which has been introduced for measuring causal effects). Using Web of Science data (a sample of 308,231 articles), we investigated whether mean differences among subject categories in citation rates still remain, even if the subject categories are made comparable in the field-related attributes (e.g., comparable of co-authors, comparable number of pages) by IPW. In a diagnostic step of our statistical analyses, we considered propensity scores as covariates in regression analyses to examine whether the differences between the fields in FICs vanish. The results revealed that the differences did not completely vanish but were strongly reduced. We received similar results when we calculated mean value differences of the fields after IPW representing the causal or unconfounded field effects on citations. However, field differences in citation rates remain. The results point out that field-normalization seems to be a prerequisite for citation analysis and cannot be replaced by the consideration of any set of FICs in citation analyses. (C) 2020 Elsevier Ltd. All rights reserved.