Makes sense.
“We” (I) should try to get back on topic!
1 Like
I was about to get my data science hat on and complain about the same thing. If you were trying to build a model to predict these outcomes, then your model would suffer from multicolinearity:
And that’s the underlying variable that you’d put in the model, with the others only coming in if they really represented improvements over and above the base variable.
Of course, the interesting thing about American culture is that so many of these things do go together. On the face of it, owning a gun and a liking for a certain brand of peanut butter shouldn’t correlate at all, but the fact that they do (and that these persist over time) tells us something meaningful about the way that American society has become polarised.
Of course, I wonder what the correlations would be outside of the USA. Do other countries have similar cultural touchstones that would divide people quite as sharply (probably yes)?
7 Likes
How about the fact that all the percentages are around 55%-65%? Does that make the correlations adequate or not? I don’t know enough about statistics.
3 Likes
The paper is pay-walled, so I can’t see exactly what their percentages relate to, but I’m guessing that these are correlation coefficients
In that case, a 55-65% figure is excellent for social science data (That’s not a dig, data about physical processes is inherently more predictable than fuzzy, emotional free-willed humans).
(Note- the above is generally true, but for a binary outcome like this you’d more likely use a logistic fit, and the statistical tests that go along with that, but that’s way too much of a rabbit hole to go down right now).
5 Likes
Yeah I am really skeptical of the math behind this. The factors should be as independent as possible. For example, I scored 3 out of 10 on the social indicators of whiteness. Scoring all of the top ten indicators as independent variables, this works out to a probability of whiteness of 93.32% for me. However, they are clearly not independent variables. Two out of those three were “homosexual sex isn’t wrong at all” and “premarital sex isn’t wrong at all”. I don’t have access to their data, but I’m fairly confident that those two are highly correlated. Also, I’m sure it’s no accident that I believe these two things and am not confident in the Executive branch (namely, Trump and the current state of the police force).
Also, assuming that the “not confident in the Executive branch” refers to Obama specifically, then I change my answer… and still score an 83.76%.
That makes sense, because that’s what it usually means, but the variables are binary here. You’re absolutely right, this is a logistic regression problem, not a linear regression problem, so I can’t see why they would use a Pearson r. My guess is that it’s some metric of predicting the outcome using the dependent variable as a predictor. But what type of metric? Precision, recall, accuracy, etc? I assume it’s correct classification rate, but I have so many questions.
4 Likes
I think it might just be the journalists making mistakes, but I can’t tell without more details.
Because of the way these are presented, I’d say that the listed variables are the most highly correlated ones in each category- these all being variants of the same underlying variable would make sense in that case. I think the actual model will be different. If you read the article closely, it doesn’t actually say that these are the characteristics that make up the best model, just that these are the most predictive individual characteristics.
Yeah, at this point, we’re just stumped without details of exactly what they have used. As they’re using a decision tree based approach, I guess it’s some form of correct classification or information gain.
5 Likes
Thanks for the explanation!
4 Likes
Owns a flashlight.
OK, this one genuinely mystifies me. Who doesn’t own a flashlight? I mean that literally. A flashlight is a common household object, isn’t it? I mean, everyone has one somewhere around their place, right?
I find the implication that there are enough people w/out flashlights to make this question predictive of anything oddly disturbing.
11 Likes
Owning a flashlight would be odd, to me. Everyone I know has multiple. Regardless of home ownership status (which so much of the other stuff is tied to).
9 Likes
Doubly so in the great PNW… power outages are common enough that they are a necessity. I have one in the desk of my basement office.
9/10 on the first… cause Don’t need an AC where I live. But seriously that is a do you live in a house, then yes you will have just about all that. Maybe 8/10 depending what qualifies for ‘sporting equipment’
4/10? on the second…
On the last sadly I have reasonably recently had Arby’s (do not judge me). Peanut butter brands vary by what is on sale so who knows. I have Wrangler jeans mostly cause they are what Costco (where we never buy enough to make the membership really worth it) rebrands and they fit me well and they are cheap at Target. There used to be a Sonic near me but it is now a pizza place but the building gives it away.
5 Likes
I think the other issue here is a lot of these are less “If you are white, you…” and more “If you do/believe this, you are most likely quite privileged (i.e. white).”
I mean, you’re going to be dealing with a rarity when you find a NASCAR-watching black woman who thinks the cops are justified in striking people. Not hard to find a white guy, though how respresentative he is, is questionable. It’s like saying that if you believe in Transubstantiation, there’s a really good chance that you are Catholic.
It’s very easy to predict something when you know the answer before you ask the question.
11 Likes
Exactly! I could locate two right this very minute, and I’m sure I have more.
I was w/out power for about 36 hours earlier this summer, and I dug out the candles for that. In hindsight, I’m a bit surprised that I still have candles (the box they were in goes back a decade, at least), but when I realized the power was going to be out for more than a few hours, I was like ‘time to bust out the candles, and I know exactly where they are’. And I’ve probably still got a kerosene lamp squirreled away somewhere. OK, maybe I’m an outlier on the non-electrical lighting options spectrum… 
As for Sonic, I refuse to eat at the one near me because they consistently suck, but I do have a soft spot for a chili-cheese cony (w/ mustard & onions) and some tots. And when I was a little kid, getting a slushie from Sonic was definitely a treat.
6 Likes
I think Stephen Colbert turned me off Arby’s for the rest of my life. As for peanut butter… does anyone else buy it in bulk? It’s a really slow process, but at least you know what goes into it.
5 Likes
Question for those counting numbers like 8/10 etc. - did I miss a link to take the quiz and get the results?
Aside from that, am I reading it wrong or did they color code things so that the whiter you are the darker brown your responses would be and vice-versa?
I guess if I rent then “owns a smoke/fire detector” and “owns a hot water heater” don’t apply?
I imagine it might profile me fairly well, but it’s confusing and I think I’m missing things.
5 Likes
Just look at each criterion and give yourself a point if it applies to you.
I never figured out the colour scheme either.
4 Likes
Colour scheme is so you can spot the same item across multiple years in the article.
4 Likes
Then why didn’t they use reds, greens, and blues, instead of whites and browns? I wondered if there was extra data on race I would have gathered if I’d read the article carefully (which I didn’t because the study seemed so questionable).
3 Likes
¯_(ツ)_/¯
ETA: Although they did use blues.
It looks like browns show items that have moved up the ranking (darker is more movement), and blues are items that have moved down the ranking (ditto).
2 Likes
Makes sense.
Using colors to indicate race would be pretty rude, now that I think about it.
3 Likes