Thumbs inflation
category: general [glöplog]
For years I had a strong feeling that we're not thumbing down things enough anymore. Now there's a clear proof.
Methodology:
Latest 20260826 prods dump is taken (why is it gz, not xz?).
Unfortunately, dumps only contain aggregate number of votes, so it is impossible to get exact vote date. So an assumption is made: prods get most of their votes not so long after they're released, so we can proxy-attribute voting behavior to prod release date.
A total of all thumbs-up, pigs, thumbs-down is calculated from all the prods released in a given month.
These values are normalized by total per-month count of votes. An month-average vote is computed from all average prods vote. Percentiles are computed too to shape the overall average-vote distribution.
Chart:
There are clearly 5 different sections:
A conclusion is not made. As a person of eastern-european culture provenance, I am not happy.
Disclosure: analysis has been vibe-coded running DeepSeek-V4-Flash-0731 locally fully on solar power.
P.S.: I wish we had individual votes data in dumps. If those existed, we could also potentially have cool recommendation systems like "give me the first top-N prod in a given category that I haven't commented yet", and "give me a prod that I haven't commented, but which had been thumbed-up by people why mostly agree with my thumbs", etc.
Methodology:
Latest 20260826 prods dump is taken (why is it gz, not xz?).
Unfortunately, dumps only contain aggregate number of votes, so it is impossible to get exact vote date. So an assumption is made: prods get most of their votes not so long after they're released, so we can proxy-attribute voting behavior to prod release date.
A total of all thumbs-up, pigs, thumbs-down is calculated from all the prods released in a given month.
These values are normalized by total per-month count of votes. An month-average vote is computed from all average prods vote. Percentiles are computed too to shape the overall average-vote distribution.
Chart:
There are clearly 5 different sections:
- Pre-1986. There are not enough prods/data, so it's just noise.
- 1986-2000. 1000 votes per prod per month. Slight downward trend on thumbs-up, and upward on thumbs-down, but overall things are pretty stable regardless, average vote around 0.4.
- 2000-2010. Total number of votes is all time high, arount 3000/prod-month. And we're mostly honest, do not hesitate to call things that are shit shit, etc. Average vote stays the same, around 0.4.
- 2010-2020. A strong trend emerges: more thumbs-up, fewer thumbs-down and pigs. Average vote rises 2x: from 0.4 to 0.8. Total votes goes down to below 1500/prod-month.
- 2020+. Barely any thumb-downs at all. Avg vote stabilizes around 0.8. Total votes around 1000-1500/prod-month.
A conclusion is not made. As a person of eastern-european culture provenance, I am not happy.
Disclosure: analysis has been vibe-coded running DeepSeek-V4-Flash-0731 locally fully on solar power.
P.S.: I wish we had individual votes data in dumps. If those existed, we could also potentially have cool recommendation systems like "give me the first top-N prod in a given category that I haven't commented yet", and "give me a prod that I haven't commented, but which had been thumbed-up by people why mostly agree with my thumbs", etc.
Fascinating.
As a demoscene person I don’t thumb down anymore (or really occasionally).
When seeing prod I don’t like I just ignore it.
Maybe this is what happens when you get older?
Maybe a regression model between the age and the amount of 👎 would show us something, but age of Poueters is not easy to infer.
As a demoscene person I don’t thumb down anymore (or really occasionally).
When seeing prod I don’t like I just ignore it.
Maybe this is what happens when you get older?
Maybe a regression model between the age and the amount of 👎 would show us something, but age of Poueters is not easy to infer.
Thumbing down feels extremely discouraging; I want more demoscene, so it seems counterproductive to do.
Thumbing down is also for me something, that i very rarely do. This might only happen when absolute crap was ranked high on a demoparty. I prefer more a piggy and type then positive points and not so positive to keep people motivated. Tumbing up is that, what i do mostly as it feels simple better, to value peoples work as they invest often much time, love and a passion for our little demoscene world.
