1
$\begingroup$

I have a dataset that has when a user clicked over a period of 31 days in an app. Its possible that all users won't have 31 days of clicks as they might not be using the app on a particular day.

The questions I am trying to answer "Would you conclude Rafa and Roger's average numbers of click per day to be different?". Hence, I was considering to do a T test

But I am not sure if sample sizes should be 23 and 26 respectively as Rafa and Roger have only 26 and 23 days of data in the month of Jan'19 or it should be 31 for both of them?

Here is the dataset:

            Rafa    Roger
1/1/2019    264     1
1/2/2019    352     98
1/3/2019    243     396
1/4/2019    243     4
1/5/2019    3       631
1/8/2019    637     438
1/9/2019    1247    229
1/10/2019   634     7
1/11/2019   4       4
1/12/2019   711     3
1/13/2019           2
1/15/2019   58      963
1/16/2019   1005    729
1/17/2019   782     108
1/18/2019   613     237
1/19/2019   565     838
1/20/2019           291
1/21/2019   2       11
1/22/2019   786     466
1/23/2019   968     417
1/24/2019   638     1940
1/26/2019           2
1/27/2019   4       20
1/29/2019   676     546
1/30/2019   530     149
1/31/2019   573     483
$\endgroup$
7
  • $\begingroup$ if you are running a paired t.test, then you can only use complete observations right? it depends on the test you want to use $\endgroup$
    – StupidWolf
    Apr 24 '20 at 23:40
  • $\begingroup$ You should use all data available. A two sample t-test does not restrict both samples to have the same number of observations. $\endgroup$ Apr 24 '20 at 23:55
  • $\begingroup$ The question that i have is: Would you conclude their average numbers of clicks per day to be different? What would you think would be correct? $\endgroup$
    – user136542
    Apr 24 '20 at 23:55
  • $\begingroup$ @DemetriPananos: are u saying I should consider n=23 and 26 for Rafa and Roger respectively, instead of n=31 for both and assuming 0 clicks for days when there is no data? $\endgroup$
    – user136542
    Apr 24 '20 at 23:58
  • 2
    $\begingroup$ For a time series like this one it should be much better to use the paired t-test despite missing a few data, because the data are likely to be subject to common fluctuations related to time (such as weekends vs. weekdays) and the paired test controls for that. $\endgroup$
    – whuber
    Apr 25 '20 at 0:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.