I came up with this code and tested it successfully on stat1005:

import json
import statsd
from kafka import KafkaConsumer

consumer = KafkaConsumer('eventlogging_QuickSurveysResponses', group_id='perf-survey-kafka-to-statsd', bootstrap_servers=['kafka-jumbo1001.eqiad.wmnet:9092'])
statsd_client = statsd.StatsClient('statsd.eqiad.wmnet', 8125)

for msg in consumer:
    object = json.loads(msg.value)
    wiki = object['wiki']
    # Example: ext-quicksurveys-example-internal-survey-answer-neutral
    response = object['event']['surveyResponseValue'][48:]
    key = 'performance.survey.{}.{}'.format(wiki, response)
    print key
    statsd_client.incr(key)

It's a small daemon that forwards the response values per wiki to statsd from the EventLogging kafka. This will allow us to easily graph this in Grafana.

It's probably best to run this code on one of the performance team's production servers. I'll find a home for it in Puppet.

Turns out, it can piggy-back on our navtiming python daemon!

Change 437731 had a related patch set uploaded (by Gilles; owner: Gilles):
[performance/navtiming@master] Collect performance survey response metrics

https://gerrit.wikimedia.org/r/437731

Change 437731 merged by jenkins-bot:
[performance/navtiming@master] Collect performance survey response metrics

https://gerrit.wikimedia.org/r/437731

Here it is: https://grafana.wikimedia.org/dashboard/db/performance-perception-survey?orgId=1

Not enough data for even cawiki to render properly, though. When there's more data I'll switch to aggregating by 24-hour period and looking at a 7-day period.