Are Ludwicks more common in the UK?!
Well, much like the administrators of the Affordable Health Care Act , I’ve learned the hard way how difficult it can be to anticipate and manage an excited tidal wave of interest surging through the internet toward one’s web portal.
Yes, “tomorrow” has arrived, but because I’ve been inundated with so many 10^3’s of serious entries for the latest MAPKIA, I’ve been unable to process them all, even with the help of my CCP state-of-the-art “big data” MAPKIA automated processor [cut & paste: http://www.palantir.net/2001/tma1/wav/foolprf.wav]
So taking a page from the President’s playbook, I’m extending the deadline of “tomorrow” to “tomorrow,” which is when I’ll post the “results” of the “Where is Ludwick” MAPKIA. In the meantime, entries will continue to be accepted.
But while we wait, how about some related info relevant to an issue that came up in discussion of the ongoing MAPKIA?
In response to my observation that Ludwick’s are “rare”—less than 3% of the U.S. population--@PaulMathews stated that “Ludwicks are not a rare species” in the UK but rather
are quite common. For example, two of our most prominent climate campaigners, Mark Lynas and George Monbiot, are pro-nuclear and pro-GMO.
Well it so happens that I have data that enables an estimation of the population frequency of Ludwicks—that is, individuals who are simultaneously (a) concerned about climate change risk but not much concerned about the risks of (b) nuclear power and (c) GM foods—in England.
Not the UK, certainly, but I think better evidence of what the true frequency is in the UK than reference to a list of commentators (indeed, compiling lists of “how many of x” one can think of is clearly an invalid way to estimate such things, given the obvious sampling bias involved, not to mention the abundant number of even people with very rare combinations of whatever in countries with populations in the tens or hundreds of millions).
It turns out that Ludwicks are even rarer in England than in the U.S. Consider:
Again, a scatterplot of survey respondents (1300 individuals from a nationally representative sample of individuals recruited to participate in CCP “cross-cultural cultural cognition” studies—including the one in our forthcoming paper “Geoengineering and Climate Change Polarization”) arrayed in relationship to their perceptions of nuclear power and climate change risks.
I’ve defined a Ludwick as an individual whose scores on a 0-10 industrial strength risk perception measure (ISRPM10) are ≥ 9 for global warming, ≤ 2 for nuclear power, and ≤ 2 for GM foods.
Those numbers are pretty close equivalents for the scores I used to compute U.S. Ludwicks on the 0-7 industrial strength risk perception measure (≥ 6, ≤ 2, & ≤ 2, respectively) in the data set I used for the MAPKIA (I determined equivalence by comparing the z-scores on the respective ISRPM7 and ISRPM10 scales).
As I said, less than 3% of the US population holds the Ludwick combination of risk perceptions.
But in England, less than 2% do!
But @PaulMathews shouldn’t feel bad—it’s just not easy to gauge these things by personal observation! I trust my own intuitions, and those of any socially competent and informed observe (@Paulmathews certainly is) but verify with empirical measurement to compensate for the inevitably partial perspective any individual is constrained to have.
There are some other cool things that can be gleaned from this cross-cultural comparison—ones, in fact, that definitely surprised me but might well have informed @Paulmathews’ conjecture.
One is that there’s not nearly as much of an affinity between climate change risk perceptions and nuclear ones in the England (r = 0.26, p < 0.01) as there is in the U.S. (r = 0.47, p < 0.01).
The reason that this surprised me is that in our study of “cross-cultural cultural cognition,” we definitely found that climate change risk perceptions in England fit the cultural-polarization profile (“hierarch individualists, skeptical” vs. “egalitarian communitarians, concerned”) that is familiar here.
Another thing: while the population frequency of Ludwicks is lower than in England than in the U.S., the probability of being a Ludwick conditional on holding the nonconformist pairing of high concern for climate and low for nuclear risks is higher in England.
In the scatterplot of English respondents, I’m defining the “Monbiot region” as the space occupied by survey respondents whose ISRPM10 scores for global warming and nuclear were ≥ 9 for global warming, ≤ 2, respectively.
The analogous neighborhood in the U.S. is the “Ropeik region” (global warming ISRPM7 ≥ 6 and nuclear power ISRPM7 ≤ 2).
Whereas about 33% of the residents of the U.S. Ropeik region are Ludwicks, over 60% of the residents of Monbiot are Ludwicks.
Huh!
What does this signify?
No doubt something interesting, but I’m not sure what!
Do others have views? People who have a better grasp of English cultural meanings & who would be more likely than I to venture sensible interpretations (ones, obviously, that would still need to be empirically verified, of course)?
Could this information be of any use in constructing a successful Ludwick profile in the US (or in England for that matter)?
In generating some data to respond to an interesting observation/query from @Niv, I discovered that I hadn't adjusted the "color coding" of the observations to reflect the difference between the 11-point English industrial strength risk perception measure & the 8-point US one for GM foods. As a result, the "GM food risk believers" (red) were underepresented in the scatterplot, which also had the effect of visually concealing the strength of the correlation (r = 0.26) between nuclear risk and climate change risk perceptions in the (English) sample.
@NIV's observation was that the effect doesn't look very impressive. I'm guessing he is likely to think that it still doesn't -- and that's because it is in fact quite modest.
@NIV also wondered whether the effect reflected in the correlation was being driven by values at one or both extremes, obscuring that the effect is even closer to nil across most of the range. This is a good question -- and it illustrates how important it is for analysts to allow critical readers to observe the raw data rather than just report summary statistics that might in fact hide relationships of consequence (particuarly nonlinear ones) in the data.
Likely he & others can see more clearly now whether the positive correlation obtains in the "middle" part of the plot. But just to enhance everyone's visual acuity, I've superimposed a lowess line rather than a fitted regression one in the version below:
A lowess plot reflects a "locally weighted" regession. In the family of regression "smoothers," it basically breaks the data into a series of tiny slices along the x-axis, fits a regression to each slice, and then connects the resulting series of plotted slopes. Obviously, it is "overfitting" in that sense. But one of the main values of lowess and related "smoothing" techniques is to make the "shape" of the distribution of the raw data even more apparent, thereby faciliting judgment about whether that shape is close enough to the one that a particular statistical models superimpose on the data to make that model a reasonable one for representing the relationships between variables of interest.
I think the lowess line here suggests that the "linear" model inherent in describing the relationship between nuclear and global warming risk perceptions as "r = 0.26" is defensible -- i.e., less wrong than the statistical characteriation that would be be generated by any alternative, nonlinear model.
But the point is you should be able to see the data & make judgments like this for yourself!
For record, btw, the values I selected for risk "GMO risk neutral" and "GMO risk high" were 3-8 & ≥ 9 on the 0-10 ISRPM
Reader Comments (4)
When you say 2% is "rare", what does that mean?
We have a 10x10x10 cube, and the 'Ludwick' region is a 1x2x2 box in the corner. If people are uniformly spread, we would surely expect 4/1000 = 0.4% in the box.
If we consider just nuclear and climate change risks, the box is 1x2 out of a 10x10 arena, so we ought to expect 2% by chance to be nuclear sceptical/climate sensitive, with all opinions on GM.
My impression from looking at that chart is that there is a cluster around NR = 5 to 8 and GWR = 5 to 10, and that for the rest there is a slight slant towards high GWR but otherwise they're fairly uniform. It might just be the choice of colours, but I get the distinct impression of blue to the left of the chart and red/orange from the centre to the right. NR and GMR are correlated (we have a long history of anti-nuclear protesters who are just the sort to also go in for anti-GM) and the weak correlation with GWR seems to be the result of that cluster being positioned above and slightly to the right of centre.
Incidentally, I know you've put a 0.01 p-value on that chart, but the correlation looks pretty weak to me. I'm guessing the p-value is calculated assuming normality and independence, or something like that. A handy way to check that out is to show the lines for (y regressed on x) and (x regressed on y) separately. If they line up in more or less the same direction, it shows a strong relationship. If they cross at a wide angle, it indicates the relationship is weak. Just a suggestion.
Incidentally, so far as I know Monbiot isn't pro-GMO but very much anti-, although he objects to them on economic/political grounds rather than safety. He has said in the past that there's no scientific evidence that they are a health risk - so by all means include him here - but I suspect he'd be upset to have been described as 'pro-GMO'.
@NIV
good points.
1. In creating a new graphic w/ a lowess line to address the one about the "flat" relationship between nuke & gw risk perceptions, I noticed a (color-)coding error in my scatterplot that resulted in the under-representation (in the scatterplot only) of a certain proportion of the subjects most concerned about GM foods. In addition to skewing the visual impression one would get of their number, that error muted the visual impress of the GMO-Nuclear relationship, since a decent proportion of the GMO risk believers are also very concerned about both nuclear and climate. But I agree the relationship is unimpressive -- but so is any r = 0.26 relationship.
2. On "rarity" of Ludwicks-- I was thinking in sort of simple population frequency terms: 1 person in 30 or 1 person in 50 seems "rare" to me. But I agree it is informative to figure out what we'd expect to see if risks were "randomly" distributed
You actually ask how many we should expect to see if risk perceptions are uniformly distributed, which is one version of random, although not a particularly likely one.
I think, though, that we'd expect to see about the number we do in fact see, in that case.
My math is a little different from yours. See if you think I'm doing it right.
The size of the "risk grid" is 121 (0-10 x 0-10 or 11 x 11) units (the ISRPMs are actually 11 point!)
The Monbiot region is 6 (0-2 x 9-10 or 3 x 2) units. (I have no idea what the real Monbiot's GM food risk perceptions are; @Paulmathews tells me Monbiot is GM food risk skeptical, but his region reflects only nuclear-risk skeptical & climate-change risk concerned)
So if nuclear and and GM food risk perceptions were distributed randomly across the space, we'd expect to see 6/121 or about 5% of the observations in the Monbiot region.
In fact, we see about 2.4% in this sample.
conclusion: Monbiot's are underrepresented by that measure.
The GM food ISRPM is 0-11, also, and Ludwicks are ISRPM ≤ 2. We'd thus expect Ludwicks, if GM food risk percpetions were also uniformly distributed across the grid, to make up 3/11 of the population of the Monbiot population.
If the Monbiot region had the expected 5% of the observations, we'd anticipate that Ludwicks make up about 1.4% (3/11 * .05) of the observations.
In fact, they make up 1.8%. So I guess they are a tiny bit overrepresented -- by that measure.
We know in fact that they are about 60% rather than 3/11 (27%) of the Monbiot region in this sample.
Of course, as I'm sure you realize, it's not very satisfying to represent something like risk perceptions as "uniformly" distributed when we are trying to figure out how different what we are observing is from what we'd see if observations were "randomly" distributed.
Probably we'd assume a "normal" distribution. In that case, of course, there'd be fewer observations at the extreme ends of each risk perception continuum than if the observations were uniformly distributed.
How should this effect our thinking about the "expected" number of Ludwicks??
If I tell you the means & SD's of the 3 risk perceptions in this sample, you should be able to calculate what fraction of the observations we'd expect to see by chance in the Monbiot region & what fraction of those we'd expect to be Ludwicks -- and the product of those woudl be our answer.
So:
global warming: M = 6.5, SD = 2.5
nuclear: M = 5.8, SD = 2.8
GM food: M = 5.5, SD = 2.8
Looking forward to the answer -- and your interpretation of the significance thereof!
Ah. I see. The plot gave me the subconscious impression that the values were continuous. Of course, you've just randomly jittered the points around each location so they don't overlay one another.
The background distribution is an interesting question. If you suppose it's approximating a continuous value from 0 to 10, and being rounded to the nearest whole number, then you would expect half as many on the two extremes. (i.e. 1/20 for 0 and 10, with 1/10 for all the others.) On the other hand, read as a set of discrete categories, uniform on the eleven values might make more sense. Sounds like something that ought to be tested empirically, rather than us guessing.
As for assumptions about distributions, I ***definitely*** wouldn't assume they were Normal! In fact, that's one of my 'hot button' topics, that can normally trigger an extended rant from me should I ever come across it. I shall spare you on this occasion, but don't make a habit of it!
To be honest, I picked uniform as the background distribution because that's what I'd probably assume, if everyone gave a simple opinion and there was no differentiation across the population.
Depending on how the questionnaire was constructed, I might possibly assume a binomial distribution. You could argue for it by saying that if the score is the total for 11 binary questions, and they pick one or the other randomly and independently because they have no opinion to distinguish them, then you would get B(11,0.5). I'm guessing that might be what you was thinking of.
There is also the point that if people don't know, they'll often pick '5' to express that. Strictly speaking, assessment of risk and confidence in the assessment are two separate variables, and '5' doesn't distinguish between 'absolutely certain the risk is half way' and ' might be 0, might be 10, I don't know'. If people are estimating it to be uniform over a range and then taking the midpoint of the range, and you pick intervals uniformly, then I think you get a symmetric triangular distribution.
Another point is that people often judge their belief on some floating scale, normalised around what they think other people's opinions are. If everyone they know think x is in the range 50 to 60, and they think it's 51, they'll judge their value low. If everyone they know assesses it 40 to 50, the same person will assess their own view to be high. So we're measuring not only the opinion but its social context.
Also, somebody might want to 'turn the dial up to 11' on expressing a strongly held opinion, but be limited by the scale, so we might expect some truncation effect. The extremes 0 and 10 ought to get more than their neighbours from those people trying to express the strength of their opinions beyond the ends of your scale. Conversely, people will sometimes moderate strongly held opinions in order to give themselves more credibility, to indicate that they are more open minded and sceptical. Also, there are people who are well aware of the political uses to which such surveys are put, and having political views themselves, may be inclined to exaggerate or moderate their views to encourage a particular outcome.
The net effect of all these influences can only be guessed at. Or you could maybe design your surveys with a 'control' question on a made-up topic that nobody could possibly have any knowledge of, to see what distribution *does* arise. I'm not sure if that will help, though.
Ideally, this stuff needs to be empirically determined. I'm not sure how it could be, though. What alternative hypothesis would we be considering?
Interesting question!