nongeocoded

To demonstrate our imputation based approach and evaluation for nongeocoded addresses only cases with complete age, race, grade, and stage data were used, and focused on white and black cases only, in order to have sufficient sample size in smaller geographic areas to test the methods.

PLoS ONE Alerts: New Articles Frank C. Curriero et al. 2010

For discussion below consider a fictitious nongeocoded Maryland prostate cancer case: a white male between the age of 45 and

PLoS ONE Alerts: New Articles Frank C. Curriero et al. 2010

Knowledge of the imputation success rate for any individual or collection of nongeocoded case records (although informative) is therefore limited in such endeavors when that success rate does not infer any performance related properties for how the geocoding and imputed nongeocodes are used in analysis.

PLoS ONE Alerts: New Articles Frank C. Curriero et al. 2010

Boundary files for all Census geography were also obtained and used in the evaluation phase for imputing nongeocoded addresses.

PLoS ONE Alerts: New Articles Frank C. Curriero et al. 2010

Against this, the existence of a successfully geocoded case at a particular location would make it less likely that a nongeocoded case (that needs to be imputed) would be nearby.

PLoS ONE Alerts: New Articles Frank C. Curriero et al. 2010

In a rural zip code with farms surrounding a central area where everyone has mail delivery, the geocoded cases will tend to be in the central area while the nongeocoded cases will tend to be out in the farmland.

PLoS ONE Alerts: New Articles Frank C. Curriero et al. 2010

Alternatives to Strategies 2 and 3 then might be to somehow include the outcome variable which like age and race is likely known for the nongeocoded records or other auxiliary variables (risk factors) that correlate with the spatial distribution of the outcome under investigation.

PLoS ONE Alerts: New Articles Frank C. Curriero et al. 2010

Alternatively, imputation performance could be evaluated on the individual case level, identifying for example, whether each experimental nongeocoded case was imputed to its correct Census unit.

PLoS ONE Alerts: New Articles Frank C. Curriero et al. 2010

In such scenarios distributing the nongeocoded cases along population at risk patterns may overlook underlying unusual distribution of cases and hence potentially lead to an under-detection of such patterns.

PLoS ONE Alerts: New Articles Frank C. Curriero et al. 2010

As was restated here it is well known that nongeocoded data are often not geographically uniform and commonly a function of population density with rural areas being more susceptible.

PLoS ONE Alerts: New Articles Frank C. Curriero et al. 2010