Monday, May 19, 2008

Nakagawa sees an end to the Fukuda slide

Break out the champagne! The prime minister's slide has stopped!

So declares Nakagawa Hidenao, in response to a Hodo 2001 poll that showed Mr. Fukuda's approval rating rise four points to 24.2% and his disapproval rating fall two points to 70%.

A net six-point shift in a single poll (which is limited to 500 people in the Tokyo area) and Mr. Nakagawa sees an end to the "headwind"? When you do the math, a shift from 72.2% to 70% is a shift of eleven people; a shift from 20.2% to 24.2% is a shift of twenty people. So the prime minister's fortunes now rest on the opinions of some thirty citizens in the Tokyo area?

Perhaps it's a function of having recently read Nassim Nicholas Taleb's Fooled by Randomness, but a poll like this seems to be fairly insignificant. Mr. Nakagawa provides a number of explanations for this minuscule shift — discontent with the DPJ's calls to revert to the old over-75 health care system, confidence that Mr. Fukuda will be a reformer thanks to his securing a cabinet decision on his road construction reform plan — but this strikes me as excessive for a shift that may in fact just be random, the product of having found by chance a slighter larger handful of citizens in the Tokyo area who support the prime minister than in previous weeks.

Admittedly my knowledge of statistics is wanting (something that graduate school will correct sooner rather than later), but this strikes me as a thin basis for arguing that the prime minister is poised to reverse his fortunes.

Come back to me when the newspapers release their latest polls — or, better yet, when a media outlet conducts a poll that's actually representative of the public at large. The opinion polls conducted by the newspapers and other media outlets seem useful only in providing a big-picture look at public sentiment, a sense of the trends and the general public mood (supportive, opposed, skeptical, resigned, etc.); the precise numbers don't seem particularly important, given the dubious (or unknown) methodology of these polls.

UPDATE: Asahi's latest poll recorded a one-point fall in Mr. Fukuda's approval rating to 19% and a six-point increase in his disapproval rating to 65%.

UPDATE TWO: In Yomiuri's latest poll, Mr. Fukuda's approval rating reached its lowest point yet (26.1%) in the polls conducted by Yomiuri. His disapproval rating also rose 6.3 points to 64.7%.

So hold off on the champagne.


Chris ( said...

A back-of-the-envelope calculation suggests that the change in the disapproval result is not statistically significant at the 95% confidence level, while the change in the approval result is borderline.

In other words, if you did the same survey 20 times with different groups of 500 people selected from the same population of Japan, and assuming that the "true" figure was an unchanged 72.2%, you would get a range of results for Fukuda's disapproval ranging from around 68% to 76%.

Statistical literacy is one of the greatest (and often most profitable) gifts you can give yourself in these data-driven times. Don't wait till grad school, and more importantly don't trust grad school to teach you properly - unless you are specializing in stats, they will often give you a poor and cursory overview.

Find a good introductory text on Amazon that uses real-world examples (and shows real-world abuses). You already seem to be thinking critically about the relevance of survey size, so you have the right mindset. And remember that 93.74% of all statistics are made up on the spot.

Japan Observer said...


Appreciate your back-of-the-envelope contribution.

I can probably wait until September, when I'll be starting graduate school, but I appreciate the advice.