What is meta? ×
Meta Stack Exchange is where users like you discuss bugs, features, and support issues that affect the software powering all 131 Stack Exchange communities.

It seems to me that SO questions go in some sort of a cycle, with waves of newbie questions occurring for a week or two, then somewhat less naive questions for a few weeks. Don't know if it would be related to school schedules or what, and don't have a clue as to how to analyze it, but it's a potentially interesting phenomenon (if my impressions are true).

share|improve this question
4  
There are certain periods during the school schedule that produce waves of such questions. –  Robert Harvey Jun 6 '12 at 3:29
8  
The calendar is stuck on September –  random Jun 6 '12 at 3:30
1  
We could start a study - drag and drop questions into bins, and sort those into dates, then count. Have fun implementing it! :) –  minitech Jun 6 '12 at 3:30
2  
We need to factor in the lunar phase as well. See if there is a full moon when all the crazy questions come out. –  slugster Jun 6 '12 at 4:00
2  
What about Friday, eh? It's always Friday in Iceland! –  Manishearth Jun 6 '12 at 4:05
    
@random Eternal Friday? –  Michael Mrozek Jun 6 '12 at 4:19
1  
I can certainly vouch that somewhere in the vicinity of 10 PM-midnight U.S. time every Sunday, the quality of iOS-related questions (and answers) takes a nosedive. –  Josh Caswell Jun 6 '12 at 5:14
    
@Josh it's morning in India then - that's pretty certainly a factor (no offense to our plenty of valued contributors from India, of course!) –  Pëkka Jun 6 '12 at 7:25
    
@Pekka: Yup, the start of the workweek on the other side of the world. –  Josh Caswell Jun 6 '12 at 17:32

2 Answers 2

You can look at score and close count percentage to see if there's a change month over month. I'll leave it to you to see if you can find a pattern month over month.

Avg_Score CloseCountPercentage month year 
--------- -------------------- ----- ---- 
161.2     0                    7     2008 
15.833602 0.0594607052316202   8     2008 
11.089009 0.0412796697626419   9     2008 
8.554738  0.0279286764235956   10    2008 
6.202574  0.0240381037433951   11    2008 
5.782043  0.0256190624513316   12    2008 
5.48562   0.0234699940582294   1     2009 
4.672511  0.0182930129350003   2     2009 
3.935539  0.0164080446298814   3     2009 
3.556143  0.0145408850192981   4     2009 
3.344831  0.0137977326968974   5     2009 
3.136148  0.0147454452579602   6     2009 
2.805524  0.0135439511587932   7     2009 
2.521266  0.0112299938102396   8     2009 
2.264269  0.00830870951698312  9     2009 
2.118176  0.00930170575692964  10    2009 
1.932727  0.00780955729101273  11    2009 
1.801787  0.00942167751819226  12    2009 
2.339628  0.00969013113977476  1     2010 
2.42432   0.0108960666278292   2     2010 
2.167111  0.0076658149094545   3     2010 
2.051308  0.00723337769804988  4     2010 
1.936797  0.0103513523540979   5     2010 
1.752515  0.0194672307348364   6     2010 
1.60956   0.019977574580313    7     2010 
1.610515  0.0192521901436012   8     2010 
1.552725  0.0191118693943995   9     2010 
1.437876  0.0208977167031714   10    2010 
1.441277  0.0206715505037316   11    2010 
1.345706  0.019308147598841    12    2010 
1.374392  0.0188202314493136   1     2011 
1.282048  0.0219543891527607   2     2011 
1.173771  0.0252765209834796   3     2011 
1.122367  0.0286872218690401   4     2011 
1.217903  0.032668700515726    5     2011 
1.130133  0.03313007023466     6     2011 
0.976285  0.0348148082527927   7     2011 
0.970282  0.0357422232578267   8     2011 
0.87364   0.0389819783115611   9     2011 
0.862309  0.0326890203813281   10    2011 
0.846628  0.037586483259458    11    2011 
0.804665  0.0388254737581243   12    2011 
0.764135  0.0366887325780297   1     2012 
0.65271   0.030412795818063    2     2012 
0.569333  0.0292261815429388   3     2012 
0.500984  0.0444985682783135   4     2012 
0.38183   0.0415258490881713   5     2012 
share|improve this answer
    
Well, an autocorrelation or an FFT would tease out any rhythm. –  The Dark Avenger Jun 6 '12 at 10:47
    
Using a very crude autocorrelation, there appears to be a 4-month cycle in the CloseCountPercentage. Can't analyze Avg_Score with this code since there's too much tilt. –  The Dark Avenger Jun 7 '12 at 0:05
    
@TheDarkAvenger I've marked this answer as Community Wiki so feel free to add any analysis –  Some Helpful Commenter Jun 7 '12 at 15:14

A naive non-Bayesian approach (no NoSQL)

Your application for running this analysis will be the StackExchange Data Explorer.

Inquiry
Is there correlation* between one or more cyclical time series and the frequency of "basic questions" submitted to StackOverflow?

Preliminary requirements
What is a "basic question"? Establish criteria for defining a "basic question". Here are some to consider:

Question is asked by user s.t. (reputation points < x) where 1 <= x <= 50
    OR [Question is asked by user s.t. (100 < reputation points < 102)  
       AND (User has other SE accounts) ]

    OR [ (Question is asked by user s.t. reputation < 500)   
       AND (time as SO user < 3 months)  
       AND (question is first question asked by that user for any given tag) ]

    AND [Duplicate prior open or closed questions > 2]

Identify cyclical event
What are some empirically observed tendencies?

  1. Sunday night prior to homework due on Monday
  2. End of semester projects postponed until the last minute by university (or post-graduate level...?) students
  3. Any scheduled, repetitive event where there are high levels of participation by SO users, yet is not a proctored examination, but rather, a project type of activity.

Consider the global distribution of SO user base and prevalence of user location in deciding which events are plausible candidates for analysis.

  • Correlation is not causation!
    Yes, I know. But it is a first step in gaining a preliminary understanding of the data and whether it is even feasible to answer the question. One needs to get acquainted with the data before analyzing it. We are subject matter experts, to some extent, as SO users, and even more so, as Meta SO users, which should help! But trying to formulate and run a preliminary analysis is needed for data profiling.

Also, consider the library of existing StackExchange Data Explorer queries. Browse through them, and maybe build off them for your initial data stratification.

Analytic method
If you get this far, and are able to identify one or more plausible time-series and criteria for selecting data, start with basic statistical methods such as ANOVA to look for correlations and trends. Watch out for colinearity.

Cross-post at Meta CrossValidated SE for further advice on statistical methodology.

share|improve this answer

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .