« Forecasting 2012: How much does ideology matter? | Main | The attack on election forecasting straw men »

November 14, 2011


If a callback sample size of 102 is "tiny", what adjective applies to a sample size of 17 used for Presidential election models? Lillipution? Negligible? Teensy-weensy? :)

Seriously, in both cases, a proper analysis using a small sample is a lot better than just winging it.

The difference is that the n=102 means they have no power to detect changes from the previous survey due to the margin of error in survey sampling (plus there are concerns about whether the callback subset is representative). By contrast, there's no sampling for presidential elections. N=17 is obviously tiny though - no getting around that.

It's true in a sense that there's no sampling for presidential elections. One uses all the recent electiions.

However, the typical casualty actuary's view would be that the 17 past elections represent a sample from some unknown underlying distribution. The point of a model is to estimate as best one can what this underlying distribution looks like. So, our formulas would treat the last 17 elections just like a sample.

The Boston Globe seems not to understand the difference between real ideas and empty slogans. "Hope and Change" is an empty slogan. It has no specific meaning. People can interpret that phrase however they like.

OTOH a proposal to shut down the Departments of Commerce, Education and Energy is a specific policy idea. The fact that Perry uses this proposal as a repetitive talking point doesn't deprive it of meaning.

That Skeptical Science (SS) post uses a rhetorical trick. Instead of dealing with what Dr. Pielke and Dr. Curry actually said, it deals with some hypothetical extensions which they didn't say.

SS is correct that a trend line fitted to 1973 - 2011 has an upward slope. OTOH a trend line fitted to 1998 - 2011 has a slight downward slope. SS points out that one could pick other periods since 1973 that had a downward slope. However, there are three important differences between the earlier periods and the period 1998 - 2011:

1. The earlier periods SS picked out are shorter than 1998 - 2011.

2. The earlier periods SS chose were cherry-picked because they had a downward slope. The period 1998 - 2011 is the the most current data. (More precisely, both the beginning and the end of the earlier periods were cherry picked. Only the beginning of 1998 - 2011 was cherry-picked.)

3. Neither Dr. Curry nor Dr. Pielke made any claim about those earlier brief peiods.

As I read SS, they seem to more-or-less imply that Drs. Pielke and Curry are disputing the long-term upward trend since 1973. That's not the case. What Drs. Pielke and Curry say is that the existing temperature models are flawed or incomplete. They don't explain why temperature leveled out over the last 13 years.

P.S. Despite its title, Skeptical Science is not a climate skeptics' site. It's pretty much a warmists' site.

P.P.S. Drs. Curry and Pielke. are eminent senior scientists, experts in climate modeling, with emormous numbers of relevant publications. In fact, Dr. Curry isn't even a denier. I believe she accepts the theory of anthropogenic global warming.

These two scientists deserve a heck of a lot more credibility than some anonymous blogger called dana1981. Brendan presumably relied on dana1981 because he like what dana1981 said. In other words, Brendan is guilty of motivated reasoning.

Perhaps another reason Brendan fell for that Skeptical Science post is that it was simple and pictorial. The real science is hard.

E.g., here's one of the posts from Curry's blog on pretty much the same topic. In order to make sense of it, one must know a lot of statistics and be familiar with a lot of climate research papers and data sources.

The comments to this entry are closed.