Exploratory Data Analysis

[ian-whitestone]

• [ ] How many high distress conversations did we miss?
• [ ] Do highly distressed kids drop off more? Or wait longer?
• [ ] How often do kids try and chat before the conversation has started?
• [ ] How does wait-time influence drop-off rate?
• [ ] Can you quantify repeat callers?
  • By nickname? By survey information entered?
• [ ] How do factors like chat time, message length, response time, wait time etc. affect NPS
• [ ] How long do counsellors wait before ending an inactive conversation?
• [ ] Any differences between convos that are terminated by chats vs counsellors?
  • In length? NPS? Handle time? Word count? Responsiveness?
• [ ] How do chats end, and how does it relate to NPS?
  • Connection error? Terminated by user? Terminated by counsellor?

[ian-whitestone]

Some observations while trying to join icarol + ICE data:

• some of the end_time's are just wrong, only saw this for ~5

  • contact_ids: 231490, 231497
  • end_time is taken from icescape extraction pipeline

• end times reported by icarol and ice can be off by a few minutes (also causing handle times to vary)

• expect records to be missing from icarol (5-10% expected according to Rob)

• some times are just systematically off by 3 hours

  • 51439, 52864

• 19241. This call was actually a hang up, counsellor took another one right after19242, then entered the icarol details for that under 19241

• Matching on call start time and mapped name (need to map agent_id to ICE name to icarol name) results in 60% match rate (excluding records where handle time < 3 min)

[ian-whitestone]

Questions for RH

1) Why is Call Information - Emergency Referral - was an Emergency Referral Provided? question blank for bulk of icarol records: Answer: it's a new survey field..

2) Name mismatches between ICE ID and phoneworkername...