Collapses locatr's several trust signals into one calibrated
match_confidence on a 0-1 scale plus a short confidence_reason string, so
a reviewer can sort or threshold on a single column instead of reading
match_status, validation_status, nm_status, and review_status
together. Higher is more trustworthy.
Arguments
- data
A data frame from the batch pipeline or from
geocode_address().
Value
data with two added columns: match_confidence (0-1, rounded to
three decimals) and confidence_reason.
Details
The right scoring model is chosen from the columns present:
Pipeline output (from
geocode_records()/ the crosswalk): scored from the tier that placed the row (geocode_pass), the match and validation status, and the name-tier confidence. Rejected or unplaced rows score near zero; reference-verified and manual rows score highest.Candidate output (from
geocode_address()): scored from the ArcGIS match score, discounted by how coarse the address type is, and capped when the point falls outside a suppliedbbox.
The score is a transparent, rule-based prior - deliberately explainable
rather than a black-box model - so every value can be traced to its
confidence_reason.
Examples
add_match_confidence(data.frame(
geocode_pass = c("pass_1_census_structured", "pass_4_name_lookup"),
match_status = c("matched", "matched_low_confidence"),
validation_status = c("coordinate_ok", "coordinate_ok"),
latitude = c(40.2, 40.3), longitude = c(-74.7, -74.8)
))
#> geocode_pass match_status validation_status latitude
#> 1 pass_1_census_structured matched coordinate_ok 40.2
#> 2 pass_4_name_lookup matched_low_confidence coordinate_ok 40.3
#> longitude match_confidence confidence_reason
#> 1 -74.7 0.90 census structured match
#> 2 -74.8 0.35 low-confidence name match