#sampling bias

[curiositysavesthecat]

*This poll was submitted to us and we simply posted it so people could vote and discuss their opinions on the matter. If you’d like for us to ask the internet a question for you, feel free to drop the poll of your choice in our inbox and we’ll post them anonymously (for more info, please check our pinned post).

[sharknado-three]

My pet peeve is people who tag their polls in a way that would bias their sample. Like, for example, if you make a "how much do you like anime" poll, and tag it as "anime", then it's going to be shown to disproportionally more people who strongly like anime, since those are typically the people who browse this tag. Sometimes it doesn't matter, because you may only care about the opinion of people who browse certain tags, or the poll isn't supposed to be serious. Or maybe you just want qualitative data. But if you want to get a sample that is representative of a larger population (of Tumblr users who do polls for fun), or if you want to interpret the results as if that was the case, then you have to keep tag-induced sampling bias in mind.

[bloodonmybarbieshirt]

I really wonder about the difference between online and offline leftist voting discussion. Online, I see a TON of "no votes for genocide kamala!" posts, but I've just... never met anyone IRL who is like that?

It's literally always been "I hate her and wish 'no genocide' was an option, but she's a lot better than trump so I'm voting for her" [paraphrased].

Maybe this is a swing state thing? Maybe it's because almost all my IRLs are queer and know the stakes? I have a decent sample size but my sampling bias is high.

[shade-of-violet]

From my experience, the average tumblr user has a distinct understanding of sampling bias

[casualnearenjoyer]

New favorite gif

[blizardstar]

I feel like sampling bias tumblr would like to know about how I saw my third “sampling bias” based post today and my immediate thought was “damn the new thing is sampling bias” followed by “what kind of (reactions to) polls have these people been seeing that they’re so up in arms about tumblr needing to understand sampling bias so bad” and then “haha. Experience bias, I need to post this.”

Edit: I saw yet another post about it and like. I always knew it was about the ‘is Tumblr gay’ polls, but I figured there HAD to be something more going on cause like. I swear I’ve seen more of the sampling bias posts than I have straight up “are you queer” polls. Maybe that’s recency bias tho. All sorts of bias.

[imkeepinit]

[manticore239]

Everyone I asked said their worst accidents happened while wearing a seatbelt.

[raffaellopalandri]

The Epistemology of Algorithmic Bias Detection: A Multidisciplinary Exploration at the Intersection of Linguistics, Philosophy, and Artificial Intelligence

We live in an increasingly data-driven world, where algorithms permeate nearly every facet of our existence, from the mundane suggestions of online retailers and products to the critical decisions impacting healthcare and justice systems. Photo by Tara Winstead on Pexels.com These algorithms, while often presented as objective and impartial, are inherently products of human design and the data…

[hugintheraven]

This is a bad idea but:

[strangesmallbard]

please reblog to increase sample size!

EDIT: i definitely should have included some sort of “multiple of these” option, so if that applies, you can click “yes and [something else]”!

EDIT 2: feel free to also include your location!

[9ofspades]

Personally I just interpret polls in absolutes rather than as a representative sample for this very reason. Like, I’ve no idea if it’s actually 31.6% of people on tumblr who were born before 1990 (probably not), but I know at least 41,000 people clicked that option on the poll. Assuming all of them clicked correctly (admittedly a pretty big assumption), then there are at least 41,000 people aged 33 or older who used their tumblr account in the last week.

That said, of course lies, bots, and mistakes can play a role. I guess the only actual factual piece of information that’s safe to conclude from this poll is that because those things can’t account for 41k+ votes (since they usually occur in small numbers), there are a nonzero number of people born before 1990 who used tumblr in the last week.

Also a nonzero number of people born after 2000. Everything else could’ve been glitches. I could be a glitch.

I’m curious about how an actually representative sample could be achieved, though. I don’t think there’s any way to do it that wouldn’t be horribly detrimental to tumblr as a company or the user experience, but I’m thinking the way to do it would be to do the Blaze thing, where a poll: goes out to a random 2000 users, cannot be reblogged, and must be answered before the user can continue scrolling their dashboard.

This would still exclude

Users who stopped using tumblr but left their blogs active

users who are so stubborn that they would create a new blog and nuke their old one rather than answer a poll

users who are so stubborn that they would give up tumblr entirely and move to a forest to become lumberjacks rather than answer a poll

This might still include

Accidental false clicks

Deliberate lies by users

Sending a poll to two accounts owned by the same person with multiple email addresses

Bots with accounts who are advanced enough to answer polls

The only way to avoid all of these would be to Looney Tunes IRS agent the 2000 random users by going to their house based on IP address, strapping them in an impossibly accurate lie detector, and holding them hostage until they answered the question. Also generating a replacement user for each user that turned out to be a bot, a repeat, or a deceased user.

For now, I think it’s just barely safe to draw conclusions from tumblr polls along the lines of “the number of users born before 1990 is somewhere between 0 and 41,739, non-inclusive”.

Let's consider the sampling bias of a classic polling method: the telephone survey.

In many jurisdictions, robo-calling cell phone numbers is illegal, so right off the jump, our sample is limited to people with landlines.

Second, our survey's calling centre probably doesn't operate 24/7, and you can only answer a home landline when you're at home, so we're also selecting for people who tend to be at home during our calling centre's office hours.

Third, most people who have landlines probably also have answering services and caller ID, so we're additionally selecting for people who answer unknown numbers rather than letting them go to the machine.

Fourth, our recipient needs to be able to participate in the survey, so we're also selecting for people who speak the language(s) in which the survey is being administered.

Finally, after all this, most people will just hang up once they figure out they're being polled, so in sum, we're selecting for people who:

have landlines;

are usually at home during our calling centre's office hours;

customarily answer unknown numbers;

speak the language(s) in which the survey is administered; and

are actually interested in responding to surveys.

Any one of these factors is likely to introduce very serious bias into our results; all of them taken together are going to render our data practically meaningless for most purposes.

Now, understand that this still represents less selection bias than trying to do demographic surveys by reblogging Tumblr polls.

[selfindulgentcompetition]

TRYING AGAIN WITH CLEARER WORDING. PLS READ BEFORE VOTING

*Meaning: When did you stop wearing a mask to a majority of your public activities? Wearing a mask when you feel sick or very rarely for specific events/reasons counts as “stopping”

[More Questions Here]