Ah! Thanks!
Individually, fairly weak at predicting ideology, none of them being over 57%. But in the aggregate the paper cites 70% accuracy in predicting ideology based of purchases.
To my mind that means that the claims in the OP about fishing rods or Arby’s being significant on their own are over inflated.
As an aside, the source paper isn’t really about predicting political belief based on purchases, that is just small part of the paper which seeks to measure whether the extent of cultural divides have changed over the years or not - and they conclude “not”, but the paper was published in 2018, based on data from before that. In the current political climate the paper might need an update.
Anyway, here’s what the paper had to say about purchases and political ideology:
As indicated in Figure 1, our ability to correctly predict political ideology based on the basket of goods and brands consumed hovers around the low 70 percent range throughout the sample period. Figure 12 shows this pattern is similar when we restrict the consumer behavior information to either products or brands, with the exception of a noticeable increase in ideological distance based on brands at the end of the 1990s.37
…The list of products individually most distinctive of ideology is interesting (cf: Table A.12). In all years, liberals distinguish themselves from conservatives by drinking alcohol. Conservatives, on the other hand, are much more likely to engage in fishing. Ideology-specific brands are mostly food, primarily with brands indicative of conservatives who disproportionately buy Jell-O gelatin desserts and eat at Arby’s.53
Note that “much more likely to engage in fishing” is actually speculative. The data actually says liberals are mildly less likely to own a range of fishing equipment, which may or may not mean they actually fish less or conservatives fish more. A 57% chance of guessing correctly whether a person is liberal or conservative based on owning a fishing rod does not blow me away in terms of predictive power.
I’d never thought of them as fancy. I just like it better than toast for somethings like breakfast sandwiches I’m not even sure “muffin” is right they’re less fancy than anything else I’ve seen called a muffin.
I’d rather be flayed.
But it has ENGLISH right there in the name… of course they’re fancy!
I’m gonna have to see some citizenship papers, buster. You don’t sound properly white American to me…
When I look at that list I see that it’s mostly a toss up. Allowing for statistical error, it’s practically 50/50 on everything.
most of what I own is books…(and tools)
Re: Conservatives and Fishing Poles
What if you put ranch dressing on an English muffin? Is that like crossing the streams? Would every molecule in your body explode at the speed of light?
There’s some overlap. Many college professors are weird dads. I should know, my dad is a retired professor who builds and restores bamboo flyrods.
I think the popularity of flyfishing in a given area is proportional to the ease of access to good trout streams and the ability of the angler to not give a fuck about what the neighbors think about their practice casting on the lawn.
@ejeffrey “calling “statistical analysis” “AI” in order to get more attention”
I agree. AI, machine learning, blah, blah, blah. It is almost certainly just statistics. Was the system doing this ‘learning’ or iteratively changing to try and make the system better or more efficient?
Also agree. Does anything in what they find prove that one thing drives the other? Can you change one variable and get different results or outcomes?
And look at the percentages - most of the ‘indicators’ are in the low to mid 50s. Not sure how that would really show liberal vs. conservative. For example, I’m pretty liberal and own three fishing poles (or whatever you want to call them). Jiff Extra Crunchy is my favorite peanut butter.
Finally, my 83 year old father, was raised in rural Wisconsin, served in the Navy, worked most of his life as a mechanic, active church member, doesn’t really buy any of the ‘liberal’ things in the list and he is about as liberal as you can get. Voted for Jerry Brown when he ran for president. BTW - I may be prejudiced as I think he is one of the best people I have ever met in my life. Go Dad.
What I’ve noticed is it’s getting more popular. Like I said there is a sudden, and quite new popularity for saltwater fly fishing in the US. Along with newly available salt water specific fly fishing equipment and a lot of influence from fly fishing on standard salt water equipment. Seems to go hand in hand with increasing interest in surfcasting.
If I had to guess it’s because millennials can not afford boats. These are the two most down right geeky styles of fishing from the shore. So if you’re going to get serious about fishing, you go down that path these days instead of tricking out your boat and moving into deeper and deeper water.
Fly equipment is comparatively expensive vs almost anything else. But it’s more dual use than most other things. And it’s not boat expensive.
Sorta like how camping and RVing has exploded in popularity. It’s just a more affordable, flexible approach to travel than a condo in Florida or a Perillo Tour.
Here’s what the study authors have to say… If that means anything to you (I don’t understand half of it), enjoy.
3.3 Machine learning
We use a machine-learning ensemble method to determine how predictable group membership is
from the variables in each dataset (i.e., time use, social attitudes, media consumption, and consumer
behavior) in each year. The ensemble method consists of running separate prediction algorithms
(we employ elastic net, regression tree, and random forest) and then combining the predictions of
these algorithms with weights chosen by OLS (Mullainathan and Spiess 2017). For each dataset,
year, and group division (e.g., time use data by gender in 2010), we first split the dataset into a
training sample (70% of the data) and a hold-out sample (30% of the data). We empirically tune
each algorithm on the training sample by cross-validation. In particular, we partition the training
data into five folds. For a given fold, we fit the algorithm on the other folds for every value of
the tuning parameter. Through this process, we obtain a prediction (e.g., probability that the
respondent is a woman) for every observation in the training sample for every value of the tuning
parameter. We then average the squared-error loss function for each tuning parameter over the
full training sample and choose the tuning parameter that minimizes the loss. This gives us a
prediction for every observation in the training sample for each of the three algorithms. We regress
(using simple OLS) group membership on the three predictions (from the three algorithms) in the
full training sample. We use the coefficients from this regression to combine the three algorithms
into the ensemble prediction in the next step.
We then turn to our hold-out sample. For each observation in the hold-out sample, we derive the prediction of each algorithm using the model estimated in the training sample under the
optimal tuning parameter. We then compute the ensemble prediction for that observation using
the aforementioned OLS coefficients. We then guess a respondent’s group affiliation based on the
ensemble prediction: if the probability that a respondent is in a group is above 1
/2, we guess that
she is in that group; otherwise, we guess that she is in the other group. We define cultural distance
(for each dataset, year, and demographic category) as the predictability of the group membership,
i.e., the share of the guesses in the hold-out sample that are correct.
Oh yeah, it’s definitely gotten more popular.
This topic was automatically closed after 5 days. New replies are no longer allowed.
