Worked example
Average Text Reply Time: A Worked Example
Most articles answer this with a survey number: "the average person replies in 90 minutes". That number is useless for your thread, because reply time depends on the person, the hour, the relationship and the channel.
The better approach is to compute your own baseline once. Here is a worked example with real arithmetic, so you can repeat it on your own export.
The setup
A 16-message thread across three sessions:
| Time | Sender |
|---|---|
| Mon 09:12 | Maya |
| Mon 09:14 | Sam |
| Mon 09:15 | Sam |
| Mon 09:31 | Maya |
| Mon 09:33 | Sam |
| Mon 20:02 | Maya |
| Mon 20:40 | Sam |
| Tue 08:55 | Sam |
| Tue 09:05 | Maya |
| Tue 13:20 | Maya |
| Tue 18:47 | Sam |
| Wed 10:01 | Sam |
| Wed 10:03 | Maya |
| Wed 10:04 | Maya |
| Wed 10:30 | Sam |
| Wed 21:15 | Maya |
Step 1: pick the rule before you compute
A reply is a message from one person that follows a message from the other. Two decisions have to be made first, because they change the answer:
- Do you count sessions? A reply 14 hours later is not a reply, it is a new conversation. This is why the gap should be capped — the analyzer uses six hours.
- Do you use the mean or the median? The mean is dragged by one slow week. The median is the middle value and describes the typical reply. Use the median.
Step 2: compute Maya's replies
Maya replies after Sam at four points: 09:14→09:31 (17 min), Mon 20:40→Tue 09:05 (12h 25m, over the cap, dropped), Wed 10:01→10:03 (2 min), Wed 10:30→21:15 (10h 45m, dropped).
Two replies inside the cap: 17 min and 2 min. Median: 9.5 minutes.
Step 3: compute Sam's replies
Sam replies after Maya at: Mon 09:12→09:14 (2 min), 09:31→09:33 (2 min), Tue 09:05→13:20 (4h 15m), Tue 13:20→18:47 (5h 27m), Wed 10:03→10:04 (1 min, but that is Maya again — see below), Wed 10:30 after Maya's 10:04 double text...
This is where hand counting gets fiddly: consecutive messages from the same person mean several gaps, but only the last one is a real reply. Sam has six measurable replies inside the cap; the median lands at 2 minutes.
Step 4: read the result properly
Maya's median is 9.5 minutes, Sam's is 2. The useful observation is not "Sam likes Maya more". It is that Sam's replies cluster at the start of sessions and Maya's are spread out. One person answers immediately, the other answers when they pick up the phone.
That is a typing style. What would make it a signal is a change: if Maya's median was 3 minutes last month, the shift from 3 to 9.5 across two months is worth noticing.
The three mistakes this example exposes
- Uncapped gaps. Include a 14-hour overnight gap and every median becomes meaningless.
- Counting double texts as replies. If Maya sends two messages into silence, that is not two replies from anyone.
- Comparing people instead of periods. Absolute speed describes a person's phone habits. Change over time describes the relationship.
Do it once, then stop
Compute your own median for two months, write both numbers down, and put the export away. You now know what "normal" looks like for these two people, which is the only baseline worth having — survey averages are about nobody in particular.