For counters statsd sends two metrics to graphite, a "count" and a
"rate". Because the counter starts at 0 after each flush, "count" is
essentially the rate per flushInterval, whereas "rate" is the rate per
second. graphite's scaleToSeconds function works correctly only with the
rate per flushInterval metric.
You could keep the "rate" metric and aggregate it with average, but
graphite can do the calculations itself. No need to keep redundant data.