The 2006 data file has many cleanliness issues.
Doing a quick frequency count shows that instead of detailCM, the 2006 data has two identical columns -- detail1_ and details_ -- that seem to share the same kind of values as the detailCM columns in all the other years.
csvcut -c \
detail1_ data/fetched/nypd-stop-and-frisk/csv/2006.csv \
| sort | uniq -c | sort -rn | head -n 10No other year has a wepfound column. So we omit it from the compiled file.