Recent Comments:
9/11 Silverado featured in Olympics commercial {Autoblog}
Aug 13th 2008 10:35AM I thought this commercial was about tough, long-lasting trucks, not some debate on foreign oil and the Iraq war. This reminds me of something I heard regularly in Costa Rica. If you want to go off-road or through mud and water, nothing can beat an American truck. They can take much more of a beating over thousands of miles than many of the non-American vehicles. Keep in mind, Costa Rica is a country where people NEED SUVs and trucks and drive on terrible dirt roads on a daily basis. They also have Land Rovers and Toyota Land Cruisers to compare to American trucks. Land Rovers are better in one area--the aluminum body doesn't rust (which is important in a rain forest.)
Will housing prices drop 50% from their peak? {BloggingStocks}
Apr 1st 2008 2:17PM I think that the idea of a 50% decline is suggested to be a bit of an outlier by the pricing models. Most of the models discussed in the article suggest a significantly smaller decline. One point the article does make is that error terms (variation of the actual data from the model) are not normally distributed. There is evidence that there are more large errors than would be predicted by a normal distribution.
Ford Edge headed to Brazil {Autoblog}
Jan 17th 2008 11:11AM To the marais,
All Brasilian cars that run on gas have to be able to run on ethanol. They're far ahead of the US on ethanol distribution and production.
Kia asks for the rock, buys its way into NBA {Autoblog}
Jan 11th 2008 9:32AM This cracks me up. If a celebrity endorses cars, you need to believe that they'd actually drive the car. When Tiger first endorsed Buick (pre-Enclave), did anyone think for a second that he really drove a Buick? NBA players in Kias is an even bigger joke. Well I guess a few players might be like Shawn Kemp. To quote Kanye West "won the Super Bowl and drove off in a Hyundai." Anyone can be broke if you pay enough in child support.
Consumer Reports rates Toyota Highlander No. 1, won't recommend it {Autoblog}
Dec 7th 2007 4:08PM Capt Vroom
Thank you for finding that page, somehow I'd missed it. They seem to punt on the issue of suppressing data other than saying that they don't publish with a sample size less than 100.
My basic problem with their data is that when the average, as mentioned below in a quote from their website, for a problem is quite small, such as 2 or 3 percent, the statistical difference between an above average model and an average model is not statistically significant. It's a statistical tie. Yes, of course, as they mention, their data is better for popular models with a larger sample size. However, even with these larger sample sizes on popular model, finding a statistically significant difference can be very difficult for infrequent problems (such as engine trouble.) In many ways, it's much harder to accurately measure rare things in statistics that common things. To quote from their website:
"An exception to this rule occurs when the mean problem rate is quite small (less than 2.5 percent), as is often the case in newer model years. In order to avoid making distinctions that are too fine to be meaningful, we do not assign a unless the problem rate is at least 3 percent, or a unless the problem rate is at least 4 percent."
Think about the difference, with a small sample size, with rounding, a single respondent can push a model from average to above average. 2 respondents could push it to far above average. In general, if they posted the confidence intervals, sample sizes, along with the actual percentages, consumers could make a much more educated decision. Presidential election polls always give the margin of error, so I think that most smart people could at least have a basic understanding of the results.
Consumer Reports rates Toyota Highlander No. 1, won't recommend it {Autoblog}
Dec 7th 2007 2:13PM Bob-obomb
You want a lesson in stats, here goes. Given the sample sizes that they're working with and the small number of problems that occur with a single part, the variation between the worst and the best vehicles is quite small. Confindence Intervals that you can achieve with even a million vehicle survey for individual parts of individual models would be so large that they would overlap. This means that the predicted reliability of single parts from the best model vs. the worst model could be completely wrong. With overlapping confindence intervals, it is quite possible that in the population (meaning every car on the road) the worst and the best models could be the opposite of the survey. Now, their data has more validity for the overall number of problems for vehicles. The overall data would be better because they're not splitting the sample multiple times.
In my experience running data from much larger surveys than their survey, if you split it by the number of models on the road and then look for individual types of problems within those models, you would have very wide confidence intervals.
A real statistical publication should at least publish their methodology for when the error gets too large, and they suppress the data because it lacks validity. Despite looking, I've never been able to find their methodolgy.
Consumer Reports rates Toyota Highlander No. 1, won't recommend it {Autoblog}
Dec 7th 2007 1:06PM The other day on the CR blog, they discussed small CUVs. The staffers who drove a Toyota Rav4 (3 of them) all thought that it was still the best vehicle in its class. The staffer who drove a Forrester thought his vehicle was still the best in class. Have they ever considered that their own bias of proving they made the right decision would lead them to be biased? Not one of the three staffers who owned a Rav4, thought that anything had surpaseed it...Hmm, confirmation bias?
In addition, their statistics for reliability of individual parts of a car are terrible. Anyone who has worked in stats would look at their data and question their methodology. They deliberately hide their statistical methods and their methods for scoring a car? Why? Would someone who understood statistics just rip them to shreds?
Video: Pinks, Romanian style {Autoblog}
Nov 29th 2007 10:38AM I love the outfits the girls are wearing. Those really aren't uncommon on the streets of Bucharest (during summer). It is so weird in downtown Bucharest, you'll see cars like this all the time. At the same time, I did see a Cayenne Turbo racing an Audi S4 down a narrow street. I remember when a mobster got arrested a couple of years ago. He had a Dacia 1310 and an AMG Mercedes in his garage. He also had a pet lion. You gotta love Romania.
Verizon fesses up, officially adds Discovery HD to FiOS {Engadget HD}
Nov 9th 2007 3:23PM Fios also just added Comcast Sportsnet MidAtlantic HD in Arlington, VA. YEAH! Finally a new HD channel that I'm excited about--I could care less about Food channel or wealth TV in HD.
Toyota hit hard by new Consumer Reports predicted reliability ratings {Autoblog}
Oct 16th 2007 3:46PM As someone who has experience working with survey data with similar or larger sample sizes to CR's data, I can pretty confidently say that much of their results must be crap. While the overall results for individual models might have some statistical validity, establishing reliablity for a specific part, such as the transmission, when the average percent of problems is around 2%, is simply bad statistics. They never say what their confidence intervals (CI) are. 90%, 95%, 50%, what is it? Because, based upon the data, I'd bet that the 90% CI of their data would range on a specific part from below to above average. How do you publish results where the CI on transmissions could give you any possible result?
