Posts by BenWilson
Last ←Newer Page 1 2 3 4 5 Older→ First
-
Looking more closely at the subsetting there, it does seem to divide Labour into 2 clusters. The dense cloud on the left near where they think National is, and the big sparse cloud on the right (the dimensions are reversed on that graph, it seems), in which their variable loading falls. Not sure what to make of that.
-
Polity: Meet the middle, in reply to David Hood,
I could also have gone to decimals.
Ah, that explains a lot.
I think the graph is fundamentally misleading in that it is based solely on the Labour National dimension, and winning the median of that means a party is pretty closed to being able to govern by themselves- it completely ignores coalitions.
Yes, it’s an approximation with a lot of bias. There are so many ways in which the basic assumptions of the median voter theorem are violated. The multidimensionality is the biggest first point, especially in NZ under MMP. But also the theorem assumes unimodality and I think it’s pretty clear from the way voters placed themselves that unimodality is highly questionable. The theorem also assumes all voters vote (which we know to be false), and that they vote for their preferences (rather than tactically). I think the subsetted PCA here shows that that is very much questionable in the case of NZF. It looks like they actually place themselves closer to Labour. This shows that the left-right placement is not the dominant factor in their political choice, despite it being pretty clear that NZF voters are close to the LR median.
In other words, it looks like LR matters most to people who have more extreme LR. Voters in the center don’t care about it so much, so LR moves won’t be what captures them.
-
Polity: Meet the middle, in reply to David Hood,
As I understand it the left right metaphor is one of strategic positioning (hence the move to to middle arguments) rather than a multidimensional strength of signal metaphor.
OK, but I'm not seeing that last graphic undermining the strategic positioning argument. It looks like there's a big chunk of people between the parties that National contested better. If I can see anything it that graph it's almost confirming Rob's point. The median is closer to National, so that's where the median voters went.
That said, I don't really get how you made that graph, why you scaled up massively. There's some math about scaling between 2 fixed points that you haven't explained.
-
Polity: Meet the middle, in reply to David Hood,
Even doing everything to make it look like a right left case, I still read it as the influence of National was extending further than the influence of Labour, there are more people left of Labour voting National than there are right of National voting Labour.
Why does that seem so silly? National got more votes period by a substantial margin, so I'd expect their reach to be further.
-
Polity: Meet the middle, in reply to David Hood,
Clearly most people go “Labour could be more left wing, National could be more right wing, there is about a 4 point difference” but some people go “Labour is 0 and National is 10”
Is 4 the median there? I'd have thought it was 5, despite not being the mode. I say this because 4 looks like it's got a lot more to the right of it than the left.
-
Polity: Meet the middle, in reply to David Hood,
More to come, building on this, focusing specifically on the left-right axis Rob was posting about (which I don’t like, I say again).
Keep it up, I'm reading it. I'm going to leave off these specific variables now*, since you're going hard on them. I'm going to take a different tack - modelling the respondents positions based on their opinions on issues that don't pertain to left-right-ness. In other words, instead of looking at where they place themselves in this contested model, it's just where they place themselves on the general opinion space. This will take quite some time because there are a LOT of variables.
*Well I will probably answer specific questions if anyone has them, and you don't get there first.
-
Polity: Meet the middle, in reply to David Hood,
I would much rather do something, and be right eventually, than try to be right from the very start and never achieve anything
Yup, at the very least, it's interesting to ourselves.
if you do botch things early on, you just go back to the code of the early stages, change it, and rerun, and in a second or two you have inorporated the corrections into the analysis
Totally. I was able to very quickly check if my labels were messed up the way you warned of. I'm using the "foreign" library in R so I don't think my SPSS read function was the same as yours. It failed on some things, but they didn't seem significant - so far as I could tell, R just can't read some of the features of SPSS files using that library, but they seem to be SPSS icing, and not compromising of the data frame that I constructed. But if they had been messed up, I could have produced all my graphs again after fixing what was wrong, in a few minutes.
-
Polity: Meet the middle, in reply to Sacha,
Interesting work, thanks, though I get lost in some of the stats detail.
Me too. Part of the point of doing this on here is to get an idea what a "lay" audience can understand. By "lay" I'm meaning a clever and interested audience, who are not themselves statistics specialists, but could well have higher than average ability to follow it. But also people who aren't even any good at stats - it's interesting to hear which visuals they can get anything out of.
Also, the masters can help me debug my own misunderstandings. I've got to be a bit bulletproof to the possibility of being shamed out massively by getting something completely wrong. I tend to think that when that happens, it's actually a really good teachable moment for everyone participating. Not being someone with their reputation on the line, I think I can take being the useful idiot on the chin.
You have to risk being wrong to get better at being right (or at least less wrong). As a group, you only need some people to do that.
-
Polity: Meet the middle, in reply to Sacha,
How about modelling potential MMP coalitions as single entities?
There's a heck of a lot of them. The number of potential coalitions is 2^(number of parties), roughly (every party is either in or out of the coalition). For the top 8 parties, that's 256 potentials. But of the probable ones the list is much shorter. Can you say what groupings you're most interested in? Maybe narrow down the "swinging" parties. If that list is only {UF, NZF, Maori} then we've got it down to 8. Do you consider the Greens to be potential swingers? If so, it doubles the number of options. Could be interesting, though.
What kind of analysis are you interested in? For the positioning, I could add the totals for each score for each party in the coalition to make a coalition left-right distribution. So adding those who gave National a 7 to those who gave ACT a 7 and those who gave UF a 7 and those who gave Maori a 7 - do that for each number and get a master distribution for the right-coalition? Try that with a few combos? That sort of thing?
-
Polity: Meet the middle, in reply to BenWilson,
I think this suggests that how people rated themselves has little bearing on their choice between Labour and National, more so for the smaller parties, particularly NZF.
I change the wording of that entirely. How people rated* themselves has little bearing on how they rated* Labour and National, but does have some bearing on how they rated* NZF and the other smaller parties.
*rated on the left to right scale.
Last ←Newer Page 1 … 133 134 135 136 137 … 1066 Older→ First