Implicit bias test

[Gorgon]

https://implicit.harvard.edu/implicit/uk/takeatest.html

 

I took this test for giggles. Apparently I have a slight bias for associating the male gender with the sciences and the female gender with the humanities.

 

So proof that the patriarchy is real ?

 

I suspect the learning curve is a big factor though. They test reaction time, and the whole time I felt that the test was trying to tell me that there was a connection. More precisely, if I know for a fact that one valid outcome, one that does not result in an error or taking too long which invalidates the result, is playing the test like a misogynist. Well, I'm a power gamer, I do whatever works best.

 

The only frame I got perfect was the last one and I played that one trying to anticipate the answers. 

[Bartimaeus]

"If you are unprepared to encounter interpretations that you might find objectionable, please do not proceed further."

 

I did the fat-thin one, and came up with a strong automatic preference for thin people compared to fat people. The test is biased in of itself, though, because they make you do it one way the first time (where you learn to associate "good" things with thin people), and then reverse it for the second part of the test...by which time you've already had it ingrained in you to do it that way. I wish they'd done it in reverse, or more accurately, went back to the original way, because doing it in that manner introduces bias in of itself. Since it's difficult to avoid that original mental programming, perhaps it would be best to test 50% of people going one way, and then 50% the other way, to see how much that truly affects the results. That would be more about empiric results than individual, though.

Edited February 20, 2018 by Bartimaeus

[injurai]

I really question all this interest in implicit bias. Looking at the mind from the science of cognition, we aptly recognize that we are pattern building machines that build internal microcosmic models of externalities in order to operate over those models. We do this for all things. So when reality reflects certain distributions, the ability to acknowledge the raw distributions does not seem like it is of the sort of thing that would be called implicit bias. It is instead memory building, database building. Because of this necessary aspect our functioning, you will never get a pure enough reading on why someone's associations between things is as they are, and a lot of interest in this topic revolves around narrative building.

 

Now implicit bias does exist, clearly, but many of the proponents of it as a mainstream language to talk about the social landscape seem to themselves miss their own bias (implicit or not, who cares!) with regards to the entire theory itself. Namely that recognition, which as an associative function, is not the same sort of thing as the bias that they so much want to talk about. The bias that is defined by the underlying models, which comes out of statistics and machine learning, is actually the error of the model and the priors. Which is very different than what we humans talk about as bias.

 

For a questionnaire, it's clear certain leading questions will force people to pull the trigger one way or the other. If anything, that shows a bias towards the more familiar language.

 

I speak generally, and not specifically about the fine researchers running that particular questionnaire. They at least give a disclaimer on the fact that they will be interpreting results and any interpretation by not be valid with respect to the subject. Well being anonymous they are admitting to interpreting soiled results and warning the user to be prepared that you are feeding them data that they can choose to interpret as they wish, (which will be based on other results?) Kicking the can down the road, when they could solicit deeper inquiries directly with subjects. Better yet, do this all as a double blind.

 

But don't let me get in the way of their grants, I'm sure they just want to feed their families and retire with tenure.

[Gorgon]

"If you are unprepared to encounter interpretations that you might find objectionable, please do not proceed further."

 

I did the fat-thin one, and came up with a strong automatic preference for thin people compared to fat people. The test is biased in of itself, though, because they make you do it one way the first time (where you learn to associate "good" things with thin people), and then reverse it for the second part of the test...by which time you've already had it ingrained in you to do it that way. I wish they'd done it in reverse, or more accurately, went back to the original way, because doing it in that manner introduces bias in of itself. Since it's difficult to avoid that original mental programming, perhaps it would be best to test 50% of people going one way, and then 50% the other way, to see how much that truly affects the results. That would be more about empiric results than individual, though.

mmm yeah, they show you a bunch of pictures of people who are thin and some who are neither particularly thin nor fat but somewhere in between and teach you that they are all 'thin'. That's a false dichotomy.  Thin isn't the opposite of fat - or rather, those people aren't all 'thin' some of them are regular size, you know, where you want to be if you aren't obsessive. 

 

Indeed being extra thin can have a bunch of negative associations as well.

 

What about healthy and not healthy. Most regular humans fall somewhere in between. 

[Valsuelm]

I hope this test is a bad joke.

 

If it isn't it's an excellent example of just how truly unintelligent and uninspired some of those 'teaching' at Harvard really are.

 

Never minding the ridiculous methodology and a complete lack of a neutral appraisement before we get to the nitty gritty, the test reveals itself as being bias before one even takes it.

 

Edited February 20, 2018 by Valsuelm

[Tigranes]

Yeah, the fact that you're trained in one combo then made to do a reversed combo is a massive red flag. 

 

I tried the Skin one, then quickly peeked at the Countries one - the test design is far less plausible for 'appraising nationalism', and it does raise questions that they felt this one should also be included. 

 

I could potentially see the 'spoilers' Valsuelm mentions as being intended, i.e. activating many users' existing intention to prove that they are not biased towards males/thins/whites, but that would still be a questionable method & needless risk. 

 

Is it 'accurate' for my own implicit biases? I certainly can't tell you, otherwise it wouldn't be implicit. But the test certainly doesn't persuade me of its own validity or competence.

[majestic]

Little to no automatic preference between the United Kingdom and the United States.

 

Gee, really. One's the appendix of the other anyway, so why would there be.

 

Oh, wait. Uhm...

 

Moderate automatic preference for Thin People compared to Fat People.

 

In this test I was first given the categories Fat and Good as a combination, so they seem to mix it up. And the moderate automatic preference had nothing to do with my personal bias and everything to do with the fact that during the "thin good, fat bad" portion of the test I got like 90% short positive words and thin faces.

 

Kept hitting the same button and had to correct mistakes on the fat side. What a load of crap.

[Malcador]

An online test being sketchy, odd

[HoonDing]

guess im a bigot

[Tigranes]

guess im a bigot

 

I reverse my judgment, test must be true

[LadyCrimson]

I got as far as "insert your zip code" and went "nah."

[majestic]

I got as far as "insert your zip code" and went "nah."

 

You can just click on "OK" for any of the survey parts. It won't complain if you don't enter any data. Just hit the buttons until the testing starts.

[ShadySands]

I always use 90211 or 10271 for things that ask for my ZIP

[Gorgon]

Let the white suits try and come get me. I'm dogsitting a rottweiler. 

[teknoman2]

i tried the countries. it showed a slight preference towards UK over US. it was due to a difference of a few ms when making some choices.

it interpreted that delay as an association bias while it was actually just a small lapse in concentration.

Edited February 20, 2018 by teknoman2

[majestic]

Let the white suits try and come get me. I'm dogsitting a rottweiler. 

 

What's got four legs and an arm?

 

 

A rottweiler at the playground.

 

 

Yeah I know, old jokes and all...

[LadyCrimson]

 

I got as far as "insert your zip code" and went "nah."

 

You can just click on "OK" for any of the survey parts. It won't complain if you don't enter any data. Just hit the buttons until the testing starts.

 

Oh, ok. My first thought was some kind of mild profiling by zip code and 2nd thought was my own over-protective anal-retentive privacy concerns. :D

[Orogun01]

I think that testing site has an implicit bias since it doesn't have Hispanic as an option but apparently White Irish is a thing. Plus is a bunch of dumb questions like, do you like thing X?

[Bartimaeus]

I think the "what do you like/prefer" questions are there for their research purposes as well as to illustrate what you think you think vs. your results (as flawed as they may be).

Edited February 21, 2018 by Bartimaeus

[Orogun01]

I think the "what do you like/prefer" questions are there for their research purposes as well as to illustrate what you think you think vs. your results (as flawed as they may be).

They are inaccurate as they poorly reflect intent and depth and as such are not suited for sciences.

[Zoraptor]

I think that testing site has an implicit bias since it doesn't have Hispanic as an option but apparently White Irish is a thing. Plus is a bunch of dumb questions like, do you like thing X?

 

Odd questions are often put in surveys to weed out bots (for internet surveys) and people who are doing a click through without actually reading. Classic example is something like 'do you like sushi?' (or whatever) then later asking 'do you hate sushi?'; if the respondent says 'yes' to both then their answers are obviously not consistent.

 

I also wouldn't discount the site having a deliberate implicit bias either- a decade or so ago there was a bias test where they would serve different introduction texts designed to prime people to answer the test in particular ways so as to test that facet of implicit bias as well. To use the example Vals quoted on the previous page, one version would say that people tended to be biased towards lighter skin tones while an alternative introduction would make no comment on expected results etc.