Add percentiles and newer literature

docs/02-geocoding.Rmd

@@ -416,4 +416,15 @@ all_distances = all_distances[which(!is.na(all_geocodes$truth.lat)),]
 save(all_distances, file = '../results/all_distances.rda')
 ```
 
+### Compute Supplementary Statistics for
+Upon review, it was recognized that RMSE alone is an insufficient measure of accuracy. So in addition, we compute percentiles [median (50), 75, 90, max (100)] for distances of each geocoder.
+
+```{r}
+load('../results/all_distances.rda')
+dist_ntiles <- apply(all_distances[,2:15], 2, function(result){
+  quantile(result, na.rm = TRUE, probs = c(.5, .75, .9, 1)) %>% round(1)
+})
+print(dist_ntiles)
+write.csv(dist_ntiles, file = '../results/ntile_distances.csv')
+```
 

results/ntile_distances.csv

@@ -0,0 +1,5 @@
+"","BingSingle","EsriSingle","Google","HereSingle","OpenCage","TomTomSingle","CensusSingle","GeocodioSingle","TomTomBatch","BingBatch","CensusBatch","GeocodioBatch","EsriBatch","HereBatch"
+"50%",11.7,24.9,14.3,11.4,34.3,15.1,40.6,11.7,15.1,11.7,41,11.7,24.9,11.4
+"75%",13.3,33.5,19.8,12.3,61.1,26,58.2,12.4,26,13.2,59.8,12.4,33.5,12.3
+"90%",30.5,40,36.1,15.3,198.8,105.7,90.9,13.8,105.7,29.8,102.5,13.8,40,15.3
+"100%",1252263.9,13462,15882.2,14267.9,13478.5,47013.3,14608.1,19972.5,47013.3,2812794.9,21240.3,19972.5,13462,14267.9