I’ve actually been thinking about this since something similar was suggested over on Meta Stack Overflow but just hadn’t gotten around to pulling some numbers. This was the kick in the pants I needed to pull something together.
I looked at the past 90 days of activity, where a user posted a question and deleted it. I’ve also limited the results to the sites that had more than 10 deleted questions with answers.
First, I looked at the total number of questions deleted and then the total number of these questions had answers. I also included how many were closed, and flagged as either a typo or a duplicate question, and how many of each of those had answers.
# Qs # Closed # Qs w/ Ans # Closed w/ Ans # Typo Flagged # Typo Flagged w/ Ans # Dup Flagged # Dup Flagged w/ Ans Site Name
----- -------- ----------- --------------- -------------- --------------------- ------------- -------------------- ---------
62681 8604 6626 797 939 197 3465 413 Stack Overflow
4024 451 348 38 0 0 150 12 Mathematics
1106 120 61 8 0 0 40 3 Super User
926 114 57 10 9 0 69 7 Ubuntu
739 233 52 19 0 0 44 3 Physics
852 267 45 7 0 0 32 1 Server Fault
344 72 33 10 0 0 12 1 Electronics and Robotics
473 67 34 4 0 0 15 1 Unix and Linux
373 45 24 1 22 2 4 1 WordPress
310 43 24 4 0 0 7 1 Magento
250 50 24 5 0 0 11 0 Drupal Answers
262 64 20 4 0 0 21 1 English Language and Usage
546 177 20 7 0 0 4 0 MathOverflow
780 288 19 12 137 4 16 0 Programmers
211 35 18 1 0 0 7 1 Salesforce
651 164 15 2 0 0 24 2 Statistical Analysis
247 52 15 5 0 0 7 1 Database Administrators
493 220 14 4 0 0 0 0 Code Review
239 55 14 1 0 0 7 0 Game Developers
180 26 13 0 0 0 3 0 Apple
204 77 12 2 27 2 26 3 Gaming
257 60 10 1 0 0 9 1 Computer Science
Besides telling us how many times a user self-deletes a question that has an answer, this doesn’t tell us a lot.
Next, I looked at how many edits were on each answer to see if blocking deletes based on edit totals would result in a large number of questions being saved.
Max # Edits On Answer # Qs Site Name
--------------------- ---- ---------
0 5532 Stack Overflow
1 838 Stack Overflow
2 171 Stack Overflow
3 58 Stack Overflow
4 14 Stack Overflow
5 3 Stack Overflow
6 5 Stack Overflow
7 2 Stack Overflow
8 2 Stack Overflow
9 1 Stack Overflow
0 292 Mathematics
1 41 Mathematics
2 11 Mathematics
3 4 Mathematics
0 55 Super User
1 6 Super User
2 1 Super User
0 47 Ubuntu
1 9 Ubuntu
2 1 Ubuntu
0 39 Physics
1 12 Physics
5 1 Physics
0 41 Server Fault
1 4 Server Fault
0 31 Electronics and Robotics
1 2 Electronics and Robotics
2 1 Electronics and Robotics
0 28 Unix and Linux
1 4 Unix and Linux
2 1 Unix and Linux
3 1 Unix and Linux
0 19 WordPress
1 4 WordPress
2 1 WordPress
0 20 Magento
1 4 Magento
0 19 Drupal Answers
1 4 Drupal Answers
3 1 Drupal Answers
0 18 English Language and Usage
1 1 English Language and Usage
3 1 English Language and Usage
0 18 MathOverflow
1 2 MathOverflow
0 16 Programmers
1 2 Programmers
2 1 Programmers
0 16 Salesforce
1 2 Salesforce
0 14 Statistical Analysis
1 1 Statistical Analysis
0 10 Database Administrators
1 5 Database Administrators
0 11 Code Review
1 1 Code Review
2 1 Code Review
3 1 Code Review
0 12 Game Developers
1 2 Game Developers
0 13 Apple
0 12 Gaming
0 10 Computer Science
Just looking at the number of edits, we're not preventing a lot of stuff from getting deleted. On Stack Overflow, if we blocked deletes because an answer had >= 3 edits, we'd only prevent 85 questions from being deleted.
I'm not sure looking at the number of edits offers a high enough signal to determine if the question should be kept around.
So, I decided to look a bit closer at the total number of edits and the length of the answer. Assuming that a longer answer might mean it's a higher quality post.
Total Questions with Answers of
A Specific Length
Max # Edits On Answer < 200 200 - 500 500 - 1000 1000 - 2500 > 2500 Site Name
--------------------- ----- --------- ---------- ----------- ------ ---------
0 1205 2320 1384 544 78 Stack Overflow
1 73 232 300 202 30 Stack Overflow
2 3 41 60 60 7 Stack Overflow
3 1 9 16 26 6 Stack Overflow
4 1 0 4 9 0 Stack Overflow
5 0 0 2 1 0 Stack Overflow
6 0 0 1 2 2 Stack Overflow
7 0 0 0 0 2 Stack Overflow
8 0 1 0 1 0 Stack Overflow
9 0 0 0 0 1 Stack Overflow
0 89 119 64 19 1 Mathematics
1 6 20 8 7 0 Mathematics
2 0 3 2 6 0 Mathematics
3 1 1 1 1 0 Mathematics
0 11 22 15 7 0 Super User
1 1 3 0 2 0 Super User
2 0 0 0 1 0 Super User
0 11 18 15 3 0 Ubuntu
1 1 5 2 1 0 Ubuntu
2 0 0 0 1 0 Ubuntu
0 8 10 12 8 1 Physics
1 0 3 5 4 0 Physics
5 0 0 1 0 0 Physics
0 9 15 17 0 0 Server Fault
1 0 1 1 2 0 Server Fault
0 5 8 13 5 0 Electronics and Robotics
1 0 1 0 1 0 Electronics and Robotics
2 0 0 1 0 0 Electronics and Robotics
0 7 14 6 1 0 Unix and Linux
1 0 2 0 2 0 Unix and Linux
2 0 0 0 1 0 Unix and Linux
3 0 1 0 0 0 Unix and Linux
0 7 5 4 2 1 WordPress
1 2 1 1 0 0 WordPress
2 0 0 0 1 0 WordPress
0 5 9 4 2 0 Magento
1 1 1 2 0 0 Magento
0 5 5 7 3 0 Drupal Answers
1 0 0 2 2 0 Drupal Answers
3 0 0 0 1 0 Drupal Answers
0 6 8 4 0 0 English Language and Usage
1 0 1 0 0 0 English Language and Usage
3 0 0 1 0 0 English Language and Usage
0 4 6 5 3 0 MathOverflow
1 0 1 1 0 0 MathOverflow
0 4 6 5 1 0 Programmers
1 0 0 1 1 0 Programmers
2 0 0 0 1 0 Programmers
0 4 7 4 1 0 Salesforce
1 0 1 0 1 0 Salesforce
0 4 1 4 4 1 Statistical Analysis
1 0 0 1 0 0 Statistical Analysis
0 2 4 2 1 1 Database Administrators
1 1 2 0 2 0 Database Administrators
0 0 4 5 2 0 Code Review
1 0 0 0 1 0 Code Review
2 0 0 0 0 1 Code Review
3 0 0 0 1 0 Code Review
0 1 7 2 2 0 Game Developers
1 0 0 2 0 0 Game Developers
0 5 5 2 1 0 Apple
0 2 6 1 1 2 Gaming
0 1 6 1 2 0 Computer Science
If we checked the length of an answer - let's say 500+ characters we'd be preventing a lot of more questions from being deleted. At least on Stack Overflow, if we blocked deletion if it contained an answer with a length greater than 500, we'd stop 2.7k+ questions from being deleted versus the 85 by solely looking at edit totals.
Maybe the signal we should look for to prevent self-deletes is answer length instead of edit totals?