Being stimulated by good initiatives of @coinlocket$, today I would like to start my new topic, namely:
In the topic, I will show you interesting facts related to the distributions of merit over time on weekly data.
3) Time series.
For now, I don't have time to actually compose a totally new contents for the topic, so let you all discover some points which I posted on
I simply created the topic, marked my ideas for further works later.
Rising from 3580 to 9684, then dropping from 9684 to 4510.
This is what I expected to see after watching your previous thread (last week). There are many demoted Junior members who ranked up back to this rank with only one new merit earned after the day newest ranking system launched weeks ago.
As I suggested in other threads, it might be better if you, or someone else, can collect specific statistics on the total demoted Junior Members who ranked up back and till now has had only one merit (enough figure for ranking up back).
Not sure, but I guess those cases and total merits received by them accounted for dominant proportions of those abnormal sudden rises of merit distribution last two week.
I really hope that you or someone else can do it someday.
I spent a couple of minutes to run simple analysis and grap box plots from those merit data. Here you go:
(1) Basic statistics:Notes:Dropped data means I dropped the data of the week from 17/9/2018 to 23/9/2018 out of the analysis because this one is a extremely outlier.
As you all can see that the means of those two datasets (4876, and 47111 for Full, and Dropped datasets, respectively) are much different than the Medians (4440 and 4431, for Full and Dropped datasets, respectively).
This is one of magical meanings of the Median statistic. In statistics, Medians are called as 'true' means of variables.
Besides the Medians, with the current datasets, statistics outside 3903 and 5043 (for Full dataset) or 4820 (for Dropped dataset) can be called as potential outliers because they are outside the Q3 + 1.5 IQR or Q1 - 1.5 IQR, with Q1 ~ 25th quartiel; Q3 ~ 75th quartile; IQR ~ Interquatile range.
For example:
You can look back at the table above, for the full merit dataset, we have:
- 25th quartile is
- 75th quartile is
so, the IQR = 5043 - 3903 = 1140
Q3 + 1.5 IQR = 5043 +1.5*1140 = 6753
Q1 - 1.5 IQR = 3903 - 1710 = 2193.
For the case, weeks which has total merit distributed
over 6.7k or below 2.2k should be taken into deeply consideration to find out where are the reasons behinds those un-normal merit distributions for those weeks. And, please remember that those ones are only
'Potential Outliers'.From the current dataset, you can easily see two important things:
(1) There are four potential weekly outliers [the first three weeks on the top, and the second from the bottom of the table given above], and all of them have values above 6.7k. I don't remember what actually happened in March, but I guess there were some significant changes on those weeks, maybe more new merit sources added. Highly appreciated information to explain these Potential Outliers.
(2) More interestingly, none of them has value below 2.2k. Thanks merit sources, at least they have been actively worked and kept distributing allocated merits.
Over time, when we have more weekly data, the whole picture will be clearer, and more reliable.
By now, the dataset which I used from @coinlocket$ contains only 30 data point (30 weeks), it's not large enough.
So what does it mean in real life in the BTT forum? It means when we see the total merits distributed per week above 4.8k, we can start thingking of some potential internal changes in the forum, such as significant new active merit sources, or new rules/ systems implemented recently. Interesting, right?
These potential outliers presented as red circle in the below box plots.
(2) Box plots and histograms:Box plots presents both medians, interquartile ranges, potential outliers.
Historgrams present that whether data (merit distribution in this case) is normal distribution by following the bell curve, or un-normal distribution (by not following the bell curve).
2.1. Full data:
2.2. Dropped data:
One more time, as you all can see that both box plots, and histograms obviously present that these dataset on merit distributions are un-nonrmal distributions.
Notes:- I will do a time-series analysis with given datasets from @LoyceV later.
- Highly appreciated anyone help with the sort of data in Excel format (I don't need excell file, if someone can get it, simply give me a snapshot of your sheet. It is enough for me, I will input those given figures manually into my sheet). Something like this one:
- I will dedicate my spare time to do it in my coming own topic. Stay tuned, please.
1) Collecting more data (older weeks in January and February; and merit distributions on forum boards over time), maybe I will get them from LoyceV datasets.
2) Making a time series analysis on merit distribution.
3) Adjusting graphs.
