PDA

View Full Version : Issue with distrowatch's poll: When it comes to recommending a Linux distro



trpted
November 18th, 2020, 10:18 PM
The current distrowatch weekly at https://distrowatch.com/weekly.php?issue=20201116 has a poll that ask.

When it comes to recommending a Linux distro:

a) I recommend the distro I run:
b) I recommend another distro in the same family:
c) I recommend a distro I do not run:
d) I do not make distro recommendations:

#1 Their poll assumes that users use only one distro on the computer(s) that that they use.

#2 It does not address different needs of the intended user.

#3 Is there any other issues with their poll, besides what I said?

Please and thank you

TheFu
November 18th, 2020, 11:04 PM
Their poll assumes that the same recommendation applies to everyone. It doesn't.

Sometimes I recommend the distro I run.
Sometimes I recommend the distro I run, with a different DE or no DE.
Sometimes I recommend a distro in the same family.
Sometimes I recommend the distro that I don't run.
Sometimes I recommend that the person not bother with Linux and use BSD, Android, or Windows instead.


But they aren't going to change a poll that has been around for decades. That would break their historical data.

In statistics 101, we are taught that a 100% random sample is required for any questions that don't include the entire population. All web surveys are flawed in the same way - they are never random and never 100% of the desired population.

mastablasta
November 19th, 2020, 07:10 AM
these polls are more meant for fun. they are not scientific measurement. also the group being polled is only those that actually visit the site and decided to leave a vote. the error margin must be quite big.

but they might provide some indicator of trends.

QIII
November 19th, 2020, 07:23 AM
DistroWatch's "polls" are not scientific. Somewhat like "referral bias" in medicine. A foot doctor might get the impression that everyone in the world has corns because everyone they see has corns. But people without corns don't go to the doctor to complain about not having corns. That poll depends on who shows up and decides to take it, so it is skewed. That is the biggest problem with it.

And the "popularity" values on the right column on the page are meaningless and based on page hits. They go like this:

Go to the laundry detergent aisle at the grocery store. Pick up and look at Brand A because you heard about it on a really cute commercial. Hit. Pick up and look at Brand B. Hit. Pick up Brand A again. Hit. Notice Brand C and pick it up. Hit. Meh. Pick up Brand A again. Hit. Meh. Pick up Brand X because that is your favorite. Hit. Take Brand X to the cashier and check out.

Brand A got three hits, but you left with Brand X. Mulitply that by 10 people. Brand A gets thirty hits, Brand X only ten. Which is the most popular brand? Brand A? No.

The numbers at DistroWatch are really meaningless.

Bean count here is meaningless, too. If JoeSnuffy has 10 beans, but every post is spot on and helps someone, is he less prolific than SallySilly who posts 100 times a day and says nothing of value -- or even leads people astray?

I'm a Mathematician by education and most of my work in software revolves around modeling of systems of some sort. Ask me what two plus two is and after a smirk I'll ask you "What do you want it to be?" Numbers can mislead, if not outright lie.