Is statistics poorly taught and explained in traditional learning institutions?
Statistics is poorly taught at all levels, for one main reason: emphasis on techniques and procedures rather than on concepts.At the undergraduate level, this means students learn in a ritualistic way how to carry out statistical tests, compute p-values and form confidence intervals without thinking too hard about what they are doing.It always amazes me that at the same time students are learning the big difference between P(A|B) and P(B|A) in their probability classes, they are learning to gloss over the difference between P(Hypothesis | Data) and P(Data | Hypothesis) in their statistics classes. The fallacious analogy of hypothesis testing to reductio ad absurdum is a commonplace explanation and rarely questioned. Students learn to treat a 95% confidence interval of (30, 40) as a "plausible" range of values for an unknown parameter but not why we are dancing around the word probability by saying "confidence" and "plausible."Early graduate training is possibly even worse, as it conveys that statistics is mainly about doing hard math (or these days, maybe computing). Instead of focusing on technical intricacies, this would be a good time to ask broad questions like:Under what circumstances does it make sense to follow the convention of fixing probability of Type I error and maximizing power subject to that constraint?Is it better to draw weak conclusions with weak assumptions (non-parametric methods) or strong conclusions with strong assumptions (Bayesian methods)?When are asymptotic results useful, given that real data is always finite?The philosophical (aka "thinking about what you are doing") aspects of statistics are sorely neglected everywhere in the curriculum I'm familiar with. As a result, many students of statistics either learn rituals (which can be worse than learning nothing) or techniques (mathematical or otherwise) which aren't squarely aimed at making the best use of data.
Do we believe scientists on global warming? Are they too all over the place to be believed?
"Isaac Newton had something to say about all this: In his seminal “Principia Mathematica,” he noted that if separate data sets are best explained by one theory or idea, that explanation is most likely the true explanation." READ:http://www.livescience.com/environment/0...
Why do climatologists need at least 30 years of data to describe climate?
Climate data is statistical and to be able to draw any conclusions from the data you need a period of at least 30 years. 30 years is the minimum value so that any trends can be seen as statistically significant when comparing year to year trends. *if I didn't explain that very well just say and I will try and rewrite it (long day at work)* @pegminer: it also has a lot to do with the palaeoclimate proxy data that is used to look at past climate and comparing it to todays climate. You can often get yearly resolution in sediment cores but anything below that is a big dodgy so 30 data points is what is required to get significant conclusions when it relates to climate. Anything less than that and your stats will hit a lot of snags. I have worked on cores with resolutions of 1 year per cm3 to hundreds and thousands of years per cm3 of sediment. You also have to consider the uncertainty in the age model as no age model is ever perfect. @pegminer mk 2: Yes they can look at shorter time frames but for you to be able to make predictions about global climate and have robust enough data to do so you require the data to be over 30 years. Now you can talk about climate variation in a less than 30 year period but the stats to compare year on year trends would be less clear.
Do you think that the majority of statistics data are unreliable?
I’m not sure what you mean exactly by “statistics data”. There’s always a risk of error when you make statistical inferences.Now, I think that surveys and polls which are designed with a reasonable methodology are overall reliable. They are usually explicit about their methodology, and explain how reliable they are (e.g. they give their margins of error).Polls and surveys that are not carefully designed are not reliable.I don’t know if there are more well-designed polls than badly-designed polls out there. If I had to guess, there are probably more bad ones, as it’s easier and way cheaper to conduct a poll without using any method.You can definitely see a lot of badly-designed online polls in some newspapers and other websites. They are not reliable at all; anyway, in this case, they are not designed to provide information, but entertainment.However, if you’re interested in the truth, you can’t dismiss all polls and all surveys just because a part of them is not reliable. You need to check the methodology they use.