Research:MoodBar/Response time

MoodBar	Pilot: Early data (July 2011 - September 2011)	Stage 1: Usage and UX (September 2011 - May 2012)	Stage 2: Impact on editor engagement (May 2012 - September 2012)
MoodBar Feedback Dashboard Mark as Helpful	First month of activity	Time to first feedback Response time Effects of UI changes Email confirmation	Observational study Controlled experiment design Controlled experiment results Response team call to action

This page in a nutshell: This page is a work-in-progress of a report of research on MoodBar, an experimental feature of the English Wikipedia, part of the New Editor Engagement initiative.

This text is a placeholder

Random information I need to note down somewhere

The number of feedback that the dashboard shows is 20. Every time an editor scrolls down, it loads 20 more.
The dashboard already allows to filter by: mood, unresponded to.
The dashboard does not show to the responders whether the feedback was posted while editing or not.
The dashboard shows the top responders (of the day?) but not how many responders are currently active.

Testing weekly and circadian effects

Figure 1:
Response time (in log(days)) versus hour of day of posting mood.

Figure 2:
Histogram of response time (in log(days)) versus hour of day of posting mood.

Figure 3:
Response time (in log(days)) versus day of week of posting mood.

Figure 4:
Histogram of response time (in log(days)) versus day of of week of posting mood.

This stuff is here only temporarily.

> # test weekly differences
> kruskal.test(log(response) ~ wday + mood, data = moodbar[moodbar$status == 1,])

	Kruskal-Wallis rank sum test

data:  log(response) by wday by mood 
Kruskal-Wallis chi-squared = 53.3929, df = 6, p-value = 9.778e-10


> # test hourly differences
> kruskal.test(log(response) ~ hour + mood, data = moodbar[moodbar$status == 1,])

	Kruskal-Wallis rank sum test

data:  log(response) by hour by mood 
Kruskal-Wallis chi-squared = 126.4277, df = 23, p-value = 2.863e-16

Survival analysis

Survivor function plots:

Figure 1:
Survivor Function (while editing)

Figure 2:
Survivor function (while not editing)

Still to do: hazard function plots

This stuff is here only temporarily. The table should be wikified:

> print(km.fit)
Call: survfit(formula = moodbar.formula, data = moodbar)

                                               records n.max n.start events median 0.95LCL 0.95UCL
mood=confused, strata(is_editing)=is_editing=0    3159  3159    3159   1877  0.324   0.261   0.402
mood=confused, strata(is_editing)=is_editing=1     547   547     547    330  0.300   0.188   0.556
mood=happy, strata(is_editing)=is_editing=0       7259  7259    7259   1459     NA      NA      NA
mood=happy, strata(is_editing)=is_editing=1       1050  1050    1050    211     NA      NA      NA
mood=sad, strata(is_editing)=is_editing=0         1683  1683    1683    953  0.676   0.466   1.022
mood=sad, strata(is_editing)=is_editing=1          187   187     187    110  0.358   0.145   1.230

> # Peto test for difference of survival
> survdiff(moodbar.formula, data = moodbar, rho = 1)
Call:
survdiff(formula = moodbar.formula, data = moodbar, rho = 1)

                 N Observed Expected (O-E)^2/E (O-E)^2/V
mood=confused 3706     1805      936       808      1252
mood=happy    8309     1398     2634       581      1980
mood=sad      1870      858      491       274       371

 Chisq= 1992  on 2 degrees of freedom, p= 0 

> m1 <- moodbar[moodbar$mood != "happy",]
> m1$wday <- factor(as.POSIXlt(m1$mood_time, tz = "UTC")$wday)
> levels(m1$wday) <- c("Mon", "Tue", "Wed", "Thu", "Fri", "Sat", "Sun")
> survdiff(Surv(ttresponse, status) ~ mood + strata(is_editing, wday), data = m1)
Call:
survdiff(formula = Surv(ttresponse, status) ~ mood + strata(is_editing, 
    wday), data = m1)

                 N Observed Expected (O-E)^2/E (O-E)^2/V
mood=confused 3706     2207     2142      1.95       5.7
mood=sad      1870     1063     1128      3.70       5.7

 Chisq= 5.7  on 1 degrees of freedom, p= 0.0169