60 likes | 195 Views
Statistics as a Distillation of Everyday Experience Gerald van Belle, University of Washington, Seattle, WA. Comments by Victor van Daal Universitetet i Stavanger, Norge. Why it is good…. The link of methodology and statistics with everyday experience
E N D
Statistics as a Distillation of Everyday ExperienceGerald van Belle, University of Washington,Seattle, WA Comments byVictor van Daal Universitetet i Stavanger, Norge
Why it is good… • The link of methodology and statistics with everyday experience • We look for consistency in a noisy system: we look for ways to categorise things, we look for causes • Variation makes it harder to find out whether a thing belongs to a certain category or not • Establishing a link between everyday-life reasoning and doing research is important because we can use it to train our students • How to tackle issues of variation and causation:the methodology • How to chose proper tools to examine variation and causation: the statistics www.lesesenteret.no
…and it’s good because… • I recognise a lot in it • I’m biased… • Personally: • The example from a study on Alzheimer’s • Professionally: • British Cohort Studies • Very large data set (17,000 births in week 20 in 1970) with a follow-up at ages 6, 10, 16, 21, and 30 • Look for predictors and consequences of developmental disorders • PIRLS (2001) and (2006) • Comparison of 43 countries on reading comprehension performance in 10-year-old children, including measures of home, school and teacher characteristics
…but… • I don’t think it’s just variation and causation, but also measurement, especially because a lot of attention is given to IRT (Item Response Theory) • Or could it be that measurement is in fact classification? • ‘We sort into genres’ could be extended that we use more than one dimension to sort into (with again lots of individual variation) • How do we know that we have a random sample? • It must be random to avoid bias • Small samples tend to be biased • Therefore big samples?
More variations on variation… • There is no end to controlling variation • Just by trying to match on very many variables, you may unintentionally create differences on an unknown variable • My question would therefore be ‘Why do you wish to control variation?’ • May be it’s better to find the sources of variation • What’s there so slippery about representativeness? • Missing values • When to impute and how to impute? • When not to impute?
For the discussion… • Everyday experience may lead us astray in many ways • There are more ingredients to a lethal mix • How to assess that you’re on the same track? • The great divide • Anticipate possible problems • In hindsight: how to fix the problems afterwards? • TheoryExample: developmental disorders • Progressive modularisation, genetic constructivism, levels involved: genetics/biology, cognitive, behavioural • Research strategy • Problem solving, responsibilities, …