How important is each predictor in predicting whether a URL is available? This graph compares what portion of the overall deviance is explained uniquely by each predictor for each of the measured outcomes. A similar list of predictors (differing only in whether the first or last year a URL was published) without interaction terms was employed to construct 3 logistic regression models. The dependent variable for each of the outcomes under study (Live Web, Internet Archive and WebCite) was availability at the time of measurement. Unique deviance was calculated by dropping each term and measuring the change in explained deviance in the logistic model. Results were then expressed as a percentage of the total uniquely explained deviance for each of the 3 methods.