I passed this along to a family member with a PhD in the field.
She also voted for Biden, so take that with a grain of salt.
Thoughts?
I don’t see how what he’s suggesting makes sense. There are a number of issues with this.
First, you can’t say that because X% of people were straight party Republican voters, that means that X% of voters in that precinct are Republican. Instead, that means that X% of the straight party voters were Republican. But what proportion of voters voted straight party? Without knowing what the denominator is, this is meaningless. E.g., if only 10% of all voters in that precinct were straight party voters, then this means that X% of that 10% were Republican. More specifically, if 20% of straight party voters were Republican, but only 10% of all voters voted straight party, this means that 2% of voters were straight party Republicans. This highlights how without providing the promotion of straight party vs. individual votes, he doesn’t give you enough detail to know what the political leanings are for that precinct. You just don’t have enough data.
Second, the y-axis values he uses are the differences between straight party and individual vote %s for Trump. But why does that matter? He argues that there’s a negative slope, such that as the % of straight party Republican voters increases, the % of individual voters decreases. This is entirely plausible. For instance, if people find a political candidate particularly unlikeable, you might see more willingness to use individual voting than straight party so that a person can deviate from their party in casting a vote for a particular candidate. E.g., one might generally vote Republican, but really despise Trump, so they cast their vote for Biden using the individual voting method. But again, without knowing what proportion of people were straight party vs individual voters, this is all pretty meaningless.
Third, his claim that you rarely see linear effects when plotting data is also nonsense. As a sample size increases, it becomes more likely that you’ll see more systematic effects emerge due to increased statistical power. Basically, as the number of cases increases, random error (or strange cases) will be less noticeable because the more typical effects will be much more numerous.
To be honest, I stopped watching at about the 35 minute point because the logic here was nonsense. And when he threw out an ungrounded statement about votes being given from one candidate to another without any evidence to support the claim, he completely lost me. So then I did a little research on who this person is, and now I’m even more confident that this presentation is nonsense.
I don't know enough about stats.
I passed this along to a family member with a PhD in the field.
She also voted for Biden, so take that with a grain of salt.
Thoughts?
I don’t see how what he’s suggesting makes sense. There are a number of issues with this.
First, you can’t say that because X% of people were straight party Republican voters, that means that X% of voters in that precinct are Republican. Instead, that means that X% of the straight party voters were Republican. But what proportion of voters voted straight party? Without knowing what the denominator is, this is meaningless. E.g., if only 10% of all voters in that precinct were straight party voters, then this means that X% of that 10% were Republican. More specifically, if 20% of straight party voters were Republican, but only 10% of all voters voted straight party, this means that 2% of voters were straight party Republicans. This highlights how without providing the promotion of straight party vs. individual votes, he doesn’t give you enough detail to know what the political leanings are for that precinct. You just don’t have enough data.
Second, the y-axis values he uses are the differences between straight party and individual vote %s for Trump. But why does that matter? He argues that there’s a negative slope, such that as the % of straight party Republican voters increases, the % of individual voters decreases. This is entirely plausible. For instance, if people find a political candidate particularly unlikeable, you might see more willingness to use individual voting than straight party so that a person can deviate from their party in casting a vote for a particular candidate. E.g., one might generally vote Republican, but really despise Trump, so they cast their vote for Biden using the individual voting method. But again, without knowing what proportion of people were straight party vs individual voters, this is all pretty meaningless.
Third, his claim that you rarely see linear effects when plotting data is also nonsense. As a sample size increases, it becomes more likely that you’ll see more systematic effects emerge due to increased statistical power. Basically, as the number of cases increases, random error (or strange cases) will be less noticeable because the more typical effects will be much more numerous.
To be honest, I stopped watching at about the 35 minute point because the logic here was nonsense. And when he threw out an ungrounded statement about votes being given from one candidate to another without any evidence to support the claim, he completely lost me. So then I did a little research on who this person is, and now I’m even more confident that this presentation is nonsense.