Posts by BenWilson
Last ←Newer Page 1 2 3 4 5 Older→ First
-
While I'm grabbing that, note that David gave a breakdown of the don't knows by how they voted http://publicaddress.net/system/cafe/speaker-meet-the-middle/?p=346845#post346845above.
I'll give overall totals.
-
Polity: Meet the middle, in reply to Mr Mark,
Am I right in assuming that those respondents who didn’t know where particular Parties (or where they themselves) were located on the Left-Right spectrum – were excluded from the data ?
I excluded them from my positioning plots. I'm treating the left-right evaluations where they occur as numbers, rather than factors. This makes sense, otherwise talking about the median doesn't make sense. Since they're numbers the Don't Knows can't be plotted. Gimme a sec and I'll tell you the proportions.
-
Polity: Meet the middle, in reply to David Hood,
My main reason for not wanting to calculate to much detail is not so much the granularity, it is the number of people that answered both that they didn’t vote and answered which party they did vote for, which suggests an error rate that means not reading too much into the data.
There's error. But the means and medians didn't seem very sensitive to whether I included or excluded the non-voters. Which suggests that the differences between the groups is not great. Perhaps I should do an ANOVA...
-
Polity: Meet the middle, in reply to David Hood,
But it could just as much be Labour 4.4 round to 4, centre 4.5 round to 5, and National 8.4 round to 8, and National is actually 39 units further away.
It could be...but in this actual case I think we got something along the lines of what I'm suggesting. The median self reported position is 5, but the revised median along the lines I'm suggesting is 5.41, which pushes the center 0.41 closer to National, and the same amount further away from Labour.
I haven't done the same calculations for the positions of Labour and National under the breakdowns that you go through above yet. But you seem to be assuming people had an some kind of judgment about where the center is. I don't think it's totally obvious that they automatically judged that the center was 5.
I don't think it's at all crazy to think that people in general might think that the political center in NZ is somewhat to the right. It's not a contradiction in terms, it's a simple observation that the National government was elected, something that most people are aware of.
Again, it comes back to what does the center mean? Is it where people are, by their own judgment? Is it where people in general think everyone else is? Or is it 5, by definition, because that's how the survey was designed? Or is it relative to the Labour and National Party, as Rob defined it?
-
Polity: Meet the middle, in reply to David Hood,
Using mean gets around the granularity. But you were talking about the median voter theorem, so the mean won't do for arguments about that. Perhaps a revised median? It would be nice when a median lies more than halfway along all the data points it is equal to that it could linearly scale up toward the value it would take if it straddled the jump to the next value. So if the data is integer and the median is 2, then the next possible value it could take is 2.5 if the split is between scores of 2 and 3. So if the split point was actually halfway between the center of the 2s and the edge of the 3s then the revised median would take the value 2.25. If it was nearly at the split, it would nearly take the value 2.5. Is such a measure sometime used when the values in data are few in count, but the population is large?
I know this is not a true median and won't have the same mathematical properties. It's a compromise between median and mean without having the property the mean does of being dragged around by extreme positions. I think it captures the idea of the middle-most data point.
Then again, the mean is easily calculated and because the range is limited, it could serve much the same purpose.
-
Polity: Meet the middle, in reply to David Hood,
people moving from Labour to National were doing so while thinking National was a greater distance from the centre
I think we need to exercise care on how we judge perception of position. There's lots of skew for the party positions and the mode will always be an integer and the median is highly likely to be as well (it could only not be if there were an even number of individuals, and the exact center happened to be right on the boundary between two integer values). So the judgments are highly granular.
What I'm saying here is that although the median is 5, I don't think that perfectly represents the center. The actual split point between two values could very close to either boundary, so it could be 5-but-almost-4, or 5-but-almost-6. Either case shows up as 5. Similarly for Labour and National's position. It's quite possible that Labour could show up as 4, National as 7, the center as 5, so that makes National look twice as far from the center. But if they're 7-but-almost-6, the center is 5-but-almost-6, and Labour is 4-but-almost-3, that in reality Labour is nearly 2 times further away from the center. If I use fractional values to represent "but almost", what I'm saying is that
5-4 = 1, 7-5=2. So 7 is twice as far from 5 as 4
But 5.99 - 4.01 = 1.98, 7.01 - 5.99 = 1.02. So 4.01 is nearly twice as far from 5.99 as 7.01. -
Polity: Meet the middle, in reply to David Hood,
Nice. Very nice. Neatly illustrates the incoherence of the single dimensional view.
Is grey everyone else beside the top 4 parties? Looks like it from the code :-)
Can you plot the variable coefficient positions too? All 9 of them?
-
Polity: Meet the middle, in reply to Brent Jackson,
like you, I having trouble drawing any conclusions about it
Well I kind of do have one conclusion. Self-evaluated centrists themselves see less political distinction between left and right. Tallies with the political science wisdom that they’re more influenced by valence issues than policy ones. Perceptions of honesty, competence, etc, are what distinguishes the parties for them. That’s a hypothesis at this point anyway. Not a new one. But one that perhaps the coplot I gave might enable the geek minded to actually see. Literally see, that is. With their eyes.
-
Polity: Meet the middle, in reply to Brent Jackson,
The rest are confused (or uncategorisable at least) ?
Or maybe it's we who are confused? In this question, the entire thread hangs. Do we listen to them and find what they mean by left and right? Or decide what left and right means for them, and divide them into the confused and not confused, then left and right? Possibly throwing away half of our population in the process.
Other possibilities: Data capture/entry error. Deliberately confused answers put in to frustrate the statistical process on purpose. Genuine belief that Labour is right because they're "feminazis" and National are left because they're "neoLIBERALs".
I think we can only really progress by trying it as many ways as possible and seeing how it fits. I've also got another idea, but I'd like to surprise people with it, and it will take some time to get it out. Gut feeling is that good stats will mean that the small size of those "confused corner" clusters would mean they don't really affect the models.
-
Polity: Meet the middle, in reply to Brent Jackson,
Can we categorise people in some other way ? For example :
There's a very large number* of possibilities. Rob Salmond mentioned upthread what his method was. It's not crazy, but it does use National and Labour as the reference point and is thus, as Sacha comments, FPP flavoured. He mentions his method for centrists, and I presume left and right are the leftovers in each direction, although he doesn't actually say so.
We've also got the respondent's own self rating of which party they are closest to. I think that might be useful for finding "Labourites" and "Nationalites", possibly more useful than what you suggest. This self rating is in two separate questions, just for more multi-dimensional confusion :-). One is "which are you close to". The other is "which are you closest to". I wonder how much trouble respondants who were not highly competent in English might have had seeing the difference here.
Last ←Newer Page 1 … 135 136 137 138 139 … 1066 Older→ First