This summarises the use of my shiny apps (at https://shiny.psyctc.org/). The page is regenerated to reflect the latest data usually a bit after 03.00 UTC (previously GMT).
Info | Value |
|---|---|
First date in data | 2024-02-07 |
Last date in data | 2026-09-11 |
This analysis time/date | 03:13 on 11/09/2026 |
Number of days spanned | 947 |
Total number of sessions | 12571 |
Mean sessions per day | 13.27 |
I am not using any way to separate different users and session is per app, so if someone used multiple apps during one visit to the server, each app used is counts as a separate session.
I can identify one early spike after the apps were publicised through the Systemic Research Centre Email list (5.iii.24) and a smaller one after a posting to the IDANET list (9.iii.24). There are later bursts that I can’t directly ascribe to any publicity.
More sensibly, here is the plot by week, actually plotting the sessions per day and counting from the launch from the launch on 7.ii.24. Where the last week is still an incomplete week that has been taken into account in the calculations, i.e. the plot shows the mean for the number of days so far in that most recent week. 95% CIs are Poisson model estimates.
Breaking that down by app gives me this.
And facetting by app gives this.
This is using a bar chart representation and fixed y axis so the use of the different apps can be compared.
This is using line and point plotting and I’ve freed the y axis so this helps understand changes over time within each app but completely loses usage differences between apps.
Again, this is mapping mean number of sessions per day, but broken down by month not week.
The first month was incomplete and the last month will usually be incomplete, that is taken into consideration in computing these session per day rates.
Here is an interactive table with the usage stats for each app. The names of the apps are clickable links to the apps. You can search the apps and perhaps most usefully, you can change the column sorting using the headers of the columns and you can export the data should you have such an appetite!
… and I am trimming it!
The dotted reference line marks the overall usage as a proportion of days available across all the apps.
Here’s a map of usage per app against dates. The sizes of the points show how many times the app was used on that day. The y axis sorts by first date used and then by descending total number of times used.
That shows that many of the apps were first used on the same day (7.ii.2024) which was the day I set up this logging. I tested all the then existing apps that day so all appear on that day.
Weekday | n | percent |
|---|---|---|
Mon | 17,398 | 17% |
Tue | 16,815 | 17% |
Wed | 17,625 | 17% |
Thu | 14,340 | 14% |
Fri | 14,888 | 15% |
Sat | 9,472 | 9% |
Sun | 10,438 | 10% |
Same sorted!
Weekday | n | percent |
|---|---|---|
Wed | 17,625 | 17% |
Mon | 17,398 | 17% |
Tue | 16,815 | 17% |
Fri | 14,888 | 15% |
Thu | 14,340 | 14% |
Sun | 10,438 | 10% |
Sat | 9,472 | 9% |
I’ve broken this down by hour. The server is to some extent protected behind a proxy at my ISP which is good for forcing https access but it does mean that I don’t know where people come from so this is all UMT (i.e. old “GMT”: internet time). I think it also suggests, assuming that most accesses are during working hours, that most visitors/users are coming to the site from Europe or the Americas.
Hour | n | percent |
|---|---|---|
0 | 211 | 2% |
1 | 224 | 2% |
2 | 212 | 2% |
3 | 207 | 2% |
4 | 327 | 3% |
5 | 438 | 3% |
6 | 617 | 5% |
7 | 666 | 5% |
8 | 688 | 5% |
9 | 682 | 5% |
10 | 749 | 6% |
11 | 687 | 5% |
12 | 686 | 5% |
13 | 717 | 6% |
14 | 723 | 6% |
15 | 912 | 7% |
16 | 640 | 5% |
17 | 624 | 5% |
18 | 503 | 4% |
19 | 600 | 5% |
20 | 518 | 4% |
21 | 393 | 3% |
22 | 316 | 3% |
23 | 231 | 2% |
Same sorted.
Hour | n | percent |
|---|---|---|
15 | 912 | 7% |
10 | 749 | 6% |
14 | 723 | 6% |
13 | 717 | 6% |
8 | 688 | 5% |
11 | 687 | 5% |
12 | 686 | 5% |
9 | 682 | 5% |
7 | 666 | 5% |
16 | 640 | 5% |
17 | 624 | 5% |
6 | 617 | 5% |
19 | 600 | 5% |
20 | 518 | 4% |
18 | 503 | 4% |
5 | 438 | 3% |
21 | 393 | 3% |
4 | 327 | 3% |
22 | 316 | 3% |
23 | 231 | 2% |
1 | 224 | 2% |
2 | 212 | 2% |
0 | 211 | 2% |
3 | 207 | 2% |
For what little it’s worth, here are the browser IDs picked up by shiny (in descending order of frequency).
The value of “ahrefs.com/robot/” is my translation of accesses that identify their browser as: “Netscape 5.0 (compatible; AhrefsBot/7.0; +http://ahrefs.com/robot/) -?”.
For reasons I don’t understand, my open source shiny does not seem to detect Microsoft Edge. I have used the apps with Edge (ugh) and it didn’t show up here. If you know why, or even how to detect Edge, do tell me (https://www.psyctc.org/psyctc/contact-me/)!
Browser | n |
|---|---|
Chrome | 8,370 |
Safari | 2,151 |
Firefox | 1,399 |
Other | 213 |
Opera | 63 |
The “Other” there refers to visits from browsers not identifying as one of Chrome, Firefox, Opera or Safari. These are usually or always crawlers, the breakdown of them was as follows.
A bit more interesting is the durations of the sessions.
Some sessions don’t have a recorded termination time, currently that’s
true for 2828, i.e. 22.5% of the sessions. This could include occasional
session still active at the time at which the copy of the database was
pulled. However, I think most will be where someone leaves the session
open. I have capped the sessions at one hour in the analyses below.
Here are the descriptive statistics.
name | nNA | nOK | min | lqrt | mean | uqrt | max |
|---|---|---|---|---|---|---|---|
durMinsAll | 2,828 | 9,742 | 0.0 | 1.0 | 39.0 | 18.0 | 9,564.0 |
durMinsCapped | 2,828 | 9,742 | 0.0 | 1.0 | 14.1 | 18.0 | 60.0 |
durMinsCensored | 4,093 | 8,477 | 0.0 | 1.0 | 7.2 | 7.0 | 60.0 |
durMinsAll includes all the sessions so far,
durMinsCapped treats all sessions recorded as lasting 60
minutes as such, more realistically, durMinsCensored
ignores those sessions assuming that they were abandoned sessions. (This
shows a maximum duration of 60 minutes as session durations were
measured to a fraction of a second so any duration of over 59’30” and
less than 60’0” is rounded up to 60 minutes and counted as a genuine 60
minutes!).
Most of the sessions, as you would expect given the nature of the apps, are sessions lasting only a few minutes. If I use the censoring and ignore all the sessions that lasted more than an hour on the plausible assumption that they were abandoned sessions rather than someone continuing to try different parameters for any app for more than an hour then there have been 8477 such sessions so far. Of these 1033 lasted under a minute. I guess it’s possible to launch an app and get useful output if only wanting the default parameters in under a minute but I think it would be rare so I think we can regard these as “just looking” sessions and they represent 12.2% of the 8477 uncensored sessions.
The number of sessions lasting a minute (rounding to the nearest minute) was 4061, i.e. 47.9% of the uncensored sessions. I think these probably represent very quick but perhaps genuine uses of an app.
That leaves 3383 sessions lasting longer than a minute but less than an hour i.e. 39.9% of the uncensored sessions, I think these can be regarded as sessions in which someone entered parameters and perhaps played around with different parameters and perhaps noted or pulled down outputs.
For now (August 2024) I see those as pretty sensible breakdown proportions. I guess that as time goes by it may be interesting to break things down by months and by apps but for now the numbers don’t really merit that and the effects of different apps being added at different times mean that the two variables of app and month are structurally entwined.