I'd add a fourth - politicians/elites have an inaccurate picture of public sentiment on immigration.

People will write whole articles about how we Must Do X to See Off Farage - and then when you ask what evidence there is that that would achieve that, just silence.

You can cut migration, sure - but do you have a plan for how you'll fund universities when there are less international students? A plan for staffing the NHS when it takes 3 years to train a nurse, 7 a doctor? A plan for how you'll have a large enough tax base to pay for pensions?

Thank you, and thank you again for the comments on it!

That's a v useful comment, thank you!

Hah - I'm not sure I have but I thought it would be a fun exercise to try!

In other words: lots of work exists causally identifying pathways from demographics/ideology to vote choice.

I'm attempting to take those pathways as given based on previous work, and compare how predictive of vote choice they actually are

I think Dowding says something v similar, in that comparison of predictive performance (as I've done) is only interesting when you can assume scientific prediction for both models (i.e. causal identification).

& I think that's what I've tried to say, but less eloquently/precisely

Fair critique on combining being conditioning on the instrument, I hadn't thought of that.

I do only reach the conclusion on I being a mediator but D not a confounder post-hoc, to explore the possible confounding. Not sure what to add to address your point other than acknowledging the problem

Ah yeah that's probably my bad for being unclear. The figure is intended to visualise possible pathways to motivate the discussion on non-inclusion of mediators.

However, a critique of this would be (following Dowding, etc) that arbitrarily including any variable in the model would not produce theoretically interesting comparisons. Hence the need to make assumptions about causal structure for thus criteria to work

The lack of causal estimand is because I'm not trying to estimate a causal quantity - I'm trying to take previous research as given and establish prediction as a criteria for which established pathways offer a better explanation in the abductive framework following Spirling & Stewart

I mean that it would be a mistake to disregard theories such as the role of parental socialisation as an explanation for ideology/vote choice even if we no longer treat them as part of a wider theory linking vote choice to demographics.

If X -> Z -> Y holds but X -> Y is 0 (or is negligible in size), then Z mediates X but X does not confound Y (or effectively does not if the effect is small)

I think my overall interpretation is close to yours, from the limitations/robustness section and the conclusion, respectively:

Yes - though not very much as the demographics models don't get much better accuracy than assuming all voters support the largest party

One of the things I got out of using vote choice (beyond capturing knowing how someone votes given some characteristic) is a straightforward relationship between the predictability of ideology/demographic models + largest 2/1 party vote share 3/3

I do explore including constituency variables in the paper, at least for the UK portion, to capture the issue of tactical voting etc.

TLDR is even these models helped w/ LD, PC, SNP, but weren't sufficient for smaller third parties. Partly because main party voters denser in the same space 2/3

Agree w/ most of this (especially high dimensional ideology > specific issues). And good critique on PTV scores - I should include predicting them as a robustness check in a future version. 1/3

Thank you! The main paper I've previously seen on this and which was a major inspiration for mine is:

doi.org/10.1007/s111...

I'm not _aware_ of other papers showing demographics are not _predictive_ of vote choice, but releasing the preprint will hopefully help me find any that do already exist