Fuck Cars
A place to discuss problems of car centric infrastructure or how it hurts us all. Let's explore the bad world of Cars!
Rules
1. Be Civil
You may not agree on ideas, but please do not be needlessly rude or insulting to other people in this community.
2. No hate speech
Don't discriminate or disparage people on the basis of sex, gender, race, ethnicity, nationality, religion, or sexuality.
3. Don't harass people
Don't follow people you disagree with into multiple threads or into PMs to insult, disparage, or otherwise attack them. And certainly don't doxx any non-public figures.
4. Stay on topic
This community is about cars, their externalities in society, car-dependency, and solutions to these.
5. No reposts
Do not repost content that has already been posted in this community.
Moderator discretion will be used to judge reports with regard to the above rules.
Posting Guidelines
In the absence of a flair system on lemmy yet, let’s try to make it easier to scan through posts by type in here by using tags:
- [meta] for discussions/suggestions about this community itself
- [article] for news articles
- [blog] for any blog-style content
- [video] for video resources
- [academic] for academic studies and sources
- [discussion] for text post questions, rants, and/or discussions
- [meme] for memes
- [image] for any non-meme images
- [misc] for anything that doesn’t fall cleanly into any of the other categories
Recommended communities:
view the rest of the comments
What are you talking about? A correlation coefficient of .5 is in the ballpark of or bigger than the correlation between human height and weight. I wouldn't be surprised if the bottleneck isn't in the reliability of the measurement.
Unmodeled interactions here also would only be able to suppress the explained variance - adding them in could only increase the R-squared!
"They produced a regression model and deduced that because the F-test had a low p value that the dark tetrad scores predicted the car score. The F-test, for clarity, determines if a model predicts the response variable better than a model with no explanatory variables. "
Yes, when you wanna know if a variable predicts another, one thing you can do is that you compare how well a model with the predictor included fares compared to a model without the predictor. One way of doing that is by using an F-test.
In case your 101 course hasn't covered that yet: F-tests are also commonly used when performing an analysis of variance.
"As is it's impossible to say if the model they found is actually very good."
You say that after quoting explained variance, which is much more useful (could use confidence intervals.. but significance substitutes here a little) in this context for judging how good a model is in absolute terms than some model comparison would be (which could give relative goodness).
Your criticism amounts to "maybe they are understating the evidence".
Do you think the paper drew sensible conclusions, or do you just not like my arguments?
This is fair enough, my background is not in social research so to me 0.5 is a moderate correlation. Not sure what you mean by the 'bottleneck' here, are you suggesting that the correlations could be higher with a different survey?
Given that the explanatory variables are in some cases more strongly correlated with each other than the response, do you think the model without interactions is likely to be an appropriate way to analyse the relationship between the response and the explanatory variables? It doesn't at all make sense to me to do one single regression model and say "The F test says this is a good model, so the explanatory variables explain the response", especially with a relatively low R^2, and given the fact that there is evidence of multicollinearity presented alongside!
The paper presents the fact that they have done a regression model with a few good significances without any real analysis of if that model is good. We don't see if the relationships are linear, we don't see if the model assumptions are met. Just doing a regression is not enough, in my opinion.
There's no need to be rude. It's perfectly acceptable to disagree with me, but you could do it politely.
Yes, I'm well aware, although I'm not sure what your point is. They haven't done any analysis of variance.
My point is that they haven't made any effort to find a model that best fits the data, they have just taken all the available variables, smacked them into python or R or whatever, and written down the statistics it spits out. There's no consideration in the paper given to interpreting the statistics, or to confirming their validity.
From the study:
Not only was p-value for age clearly not significant, the confidence interval for the coefficient was [–.21, .17]..... This includes 0 ffs! There's no evidence here that there is greater endorsement of the car items in younger respondents. Why was age even included in the model in the first place, given that the correlation was near 0?
Like I said - there is some evidence here of an interaction, I'll concede that in context the correlation isn't bad for 2 of the dark tetrad items, Wild and Crafty, but the analysis they have used to present this information is not well thought out or presented. Personally I don't think that a linear regression model is even the right way to analyse the data they have, I especially don't think this regression model is a good way to analyse the data.