The study noted the widely varying results of prior research, the importance of specification problems, and the need to use procedures to reduce specification bias and alleviate concerns expressed in the National Academy of Sciences' review of deterrence research. This study conducted the Granger test on pooled city and State data for more than 20 years, ending in 1992 for most of the data and in 1993 for the State crime statistics. The data came from 49 States and 56 cities. The multiple time-series design differed from other pooled designed in that it used only time-series variation. Crime, police, and other continuous variables were divided by population. A crucial feature of the research was to use year dummies, unit dummies, unit trends, and lagged dependent variables as proxies for otherwise omitted control variables or controls with inaccurate data. Results revealed Granger causation in both directions. The impact of crime on the number of police was slight, but the impact of police on most types of crime was substantial. This latter result was stronger at the city level than at the State level. Findings indicated the importance of addressing specification problems and demonstrated that police levels and crime rates clearly affect each other. Tables, footnotes, and 93 references