Posts by BenWilson
Last ←Newer Page 1 2 3 4 5 Older→ First
-
Polity: Meet the middle, in reply to Steve Barnes,
I probably won't be using it. That was just because someone who already had a bunch of SAS code helped me out. I can use SAS, but I still prefer R. I can share the code with you guys and you don't have to buy SAS, which is really quite expensive.
Not to mention that it's patriotic to use R, as well as fully righteous. Join the revolution! you have nothing to lose but your IP chains.
-
Polity: Meet the middle, in reply to Rich of Observationz,
There's tons of interesting stuff to dig out of those surveys. The parental voting and when people made up their minds is certainly covered.
Studying the voting across time is done to some extent, both in questions about how people voted last time, and also in actual comparisons for individuals that span multiple surveys. That's longitudinal, but I don't know how far it goes back.
I'm quite interested also in questions about how people behave when they change their vote, either to another party or to not-voting. So many questions, so little time.
You've got the skills, surely, Rich. Have a dig, see what you find.
-
Polity: Meet the middle, in reply to Steve Barnes,
Will we be going 3D?
I've been looking at it like that on my workstation, but I don't know how to give that out on the web. It would need some kind of java applet or something...
I could show a 3D model in 2D if I could rotate it to find something useful to look at. But essentially that's what PCA is doing anyway - you're looking at this data rotated for maximal interest and then projected onto 2D. In this case, 95% of what is interesting is actually there in those 2 dimensions anyway - a third dimension might add like 2% to it. This is the whole point of PCA. It's a way of reducing dimensions so that we can look at them. The data underlying the graph has as many dimensions as there are variables shown there. The analysis just turned the whole thing around in every dimension until the data looked nice and spread out, so you can look at it (it also scales and centers it)
It would be fun to rotate it around, but ultimately not particularly illuminating. This doesn't alway happen though. I'll be sure to show you anything I do find that's cool in 3D. I expect when I add in questions that are more differentiating for the small parties that less of the information would be captured in the first dimension here.
That's really where I'm going with this. It's kind of cool that you can pretty much gauge left-vs-right on the subset of questions here, so long as you go with Rob's view that left and right are defined relative to the two major parties. I think it's also somewhat illuminating, that the right here are mostly the beneficiary and union bashing questions rather than broader questions of economic organization or management. That's possibly an actual insight, at least for Labour, and for all of us trying to comprehend what moving to the right actually means. It means, quite literally, moving toward those variables which (confusingly) are shown on the left, clustered around National and ACT.
I'm not convinced, looking at this, that the pickings in that direction are really that rich for Labour. But this graph is, once again, very much a subset of political opinion questions. I want to get to a situation where most of the variables are clustered around most of the votes. This graph is not like that, and if one didn't know that National is dominant, one might think the majority of NZers were voting left.
What that shows is that the questions here are not adequate to explain their voting behaviour. Many have said that valence issues are bigger than policy and positioning ones...I'd like to be able to confirm that, when I get the time. Gut feeling is that it's true.
-
Polity: Meet the middle, in reply to Rich of Observationz,
That’s a higher level of literacy than most people can manage, so it’d quite possible that the anomalous results are caused by a failure to understand the question.
Could be. Or maybe it isn't really that much of a distinction between the main parties' voters.. Can dig deeper here. Won't be until the end of next week, though.
-
Polity: Meet the middle, in reply to BenWilson,
The strange location of taxsolve I can’t explain
Well, OK, I can try. It's something that isn't important in the Labour/National distinction, but is important in the second dimension. In other words, it's not something that could distinguish Labour voters from National ones, but could help you tell NZF ones apart from Green ones, since they're very close in the first dimension. That and views on immigration.
Obviously there aren't any questions about environmentalism here otherwise there would be some that were near the Greens. This is not deliberate in that I don't think it's important - I just didn't use the B1-5 questions in the survey yet because dealing with the "don't knows" is more problematic. I'll get around to it at some point.
So I guess the point I made above about explaining 90% of the variation should be made with this in mind. It tells you as much about the questions as about general positioning. Had all the questions been about immigration then NZF and Maori Party would have ended up on the main axis at one end, UF and Greens on the other, and National and Labour close together in the middle, with Labour probably closer to the NZF end.
-
Polity: Meet the middle, in reply to Rich of Observationz,
Yes, principal components. But you knew that, I'm sure, you pendantic sew and sew.
-
Polity: Meet the middle, in reply to Steve Barnes,
Interesting work Ben. That National and Labour seem to be on the same plain.
Thank Steve, although it's far from being as complete as I'd like. I'll be fleshing it out a lot more over the next month.
As for the same plane thing, well the analysis itself looks for the plane (well, hyperplane, technically) that most separates the data out. Since most people voted for either National or Labour, I'd expect them to have ended up along the first principle axis somewhere. Essentially, it was looking for the plane that would go between them, rather than luckily finding one. I did not tell the analysis anything about the left-rightness of the questions, nor how the people voted. It just looked for the best way to split them into two groups based on all the questions. Then it does it again for the second dimension, and so on. There's more dimensions, but I only plotted the first 2.
There are some interesting things on there. taxsolve is quite a long way from ACT and National. This is "Reducing taxes in general would help solve our economic problems". immig is a long way from NZ First and actually from everywhere, but this is because it was worded in the reverse direction "Do you think the number of immigrants allowed into New Zealand nowadays should be" with increased score meaning "increased". ie you'd expect NZF supporters to disagree with this. So the second axis (up-down) is most strongly associated with views on immigration, it seems. The strange location of taxsolve I can't explain.
-
Polity: Meet the middle, in reply to Steve Barnes,
Something I’ve had to put on hold while getting other things done. Using opinion on subjects rather than about left-right positioning, to look for principle components. I had someone do this in SAS for me, I’ll attempt to do the same thing in R when I get the time.
The party points are the center of mass for all the people that voted for them. The other labels are the variable loading positions for the entire population on the questions that they correspond to. You’ll have to ask me which actual questions they were if you're interested in one particular question.
But the thing that leaps out at me from this graph is that while of course we know that politics is multidimensional (or at least anyone who knows what that means probably thinks that), in this graph Labour and National are sitting almost squarely on the principle axis, and 90% of the variation in the data is explained by this axis.
So if we did actually like Rob’s definition of left and right as being the direction between Labour and National, then just on opinions about politics, as asked in 2011 in this survey, it does look like we’re almost in a uni-dimensional situation. I didn’t go off every question, there’s a lot more tidying to do on this…but it’s bloody interesting. These were the general questions you’d ask someone to find out whether they’re left or right wing without just asking them if they’re left or right wing.
This was built on that helpful someone I don’t want to name without their consent giving the suggestion to binarize the choices, because unfortunately the data allowed “Don’t know” on the same continuum as the answers and I lose almost all the data if I keep only complete cases. It seemed like a good alternative to imputing the don’t knows.
-
Legal Beagle: Into the River/Interim Restrictions, in reply to Rich Lock,
I remember seeing a DVD copy sitting next to ‘I spit on your grave’ in the local video hire shop when we moved to Devonport in 2006.
BAN THIS SICK FILTH!
When faced as a young teen with the choice and some lads who wanted to watch something not-allowed, I Spit on your Grave was eliminated because only I hadn't already seen it. When we picked Caligula instead, and the question is "It it art, or obsenity?" was being pondered, the consensus was that we hoped it wasn't art. We were all a bit disappointed.
-
"Banned in NZ". It's like being damned with faint praise.
Last ←Newer Page 1 … 131 132 133 134 135 … 1066 Older→ First