system at the locality. The log of the rate of casing hits was significantly associated with the number of evidence casings entered into NIBIN at the locality but was not associated with any other predictor. In particular, the association between the log rate of hits and the number of nonevidence casings in the system was negative, albeit not statistically significant.

We examined the fit of the linear model by inspecting residuals and estimating the degree of multicollinearity among predictors. All of the predictors in the model were positively correlated, which might partially explain the lack of statistically significant associations between the response variable and the predictors. A reasonable (43 percent) proportion of the total variability observed in the log probability of a casing hit was explained by the predictors in the model, and no outliers were detected when inspecting the standardized residuals from the regression. However, patterns in the standardized residuals plotted against the observed log probabilities of hits on casings do suggest that other, potentially important predictors are missing from the model.

The most basic interpretation we draw from this analysis, despite its limits, is the same reached by the Inspector General audit: the probabilities of getting a hit on either bullets or casings depend vitally on the number of entries entered into the NIBIN system at each locality. We observed the strongest connection to be with the counts of bullets or casings entered as evidence, whereas hit probabilities were negatively (but not significantly) associated with the number of nonevidence (test fire) samples entered into the system. This finding suggests that agencies might be better served by prioritizing entries so that evidence samples are entered into NIBIN most promptly.

The existing state reference ballistic image databases in New York and Maryland operate using the same IBIS computer and microscope image; their networks and correlation servers are entirely distinct, however, thus complying with the current prohibition on noncrime-gun evidence in the NIBIN database.

As part of a larger gun legislation package, the Maryland Responsible Gun Safety Act of 2000 established a statewide database of images of cartridge cases test fired from every handgun sold, rented, or transferred by manufacturers in the state or whose products are sold in the state, under the premise that handguns are most frequently used in crimes. The database is known as Maryland-IBIS (MD-IBIS), was established under the Maryland