How well does the software perform on industry narratives?
To identify and categorize SOIC errors, unique industry and occupation narratives were reviewed separately. Using Bureau of the Census industry divisions as a guide, SOIC and manually assigned codes were compared. Of the 48,067 total cases, 16,096 cases contained unique industry narratives (all duplicates were removed, e.g., the file contained the term “construction roofing” in the industry narrative field only once compared to the numerous times it may have appeared in the original file of 48,067 cases). More than half (9,262) were coded correctly. The software incorrectly coded 3,296 narratives and could not code 3,538 narratives. Problems that were identified are characterized below.
