150 likes | 255 Vues
Standard Normal Distribution. μ =0 and σ 2 =1. Confidence Intervals. Scientists often use a sample standard deviation to construct a confidence interval around the mean. For a normally distributed random variable:
E N D
Standard Normal Distribution • μ=0 and σ2=1
Confidence Intervals • Scientists often use a sample standard deviation to construct a confidence interval around the mean. • For a normally distributed random variable: approximately 67 % of the observations occur within + 1 standard deviation of the mean approximately 96 % of the observations occur within + 2 standard deviations of the mean
What does it mean? Because our sample mean and sample standard error of the mean are derived from a single sample, this confidence interval WILL change if we sample again. Thus, this expression asserts that the “true” population mean μ will fall within a single calculated confidence in 95% of the iterations
By extension: • If we were to repeatedly sample the population (keeping sample size and all conditions equal), 5% of the time we would expect that the true population mean μ would fall outside of this confidence interval
Bad news • This is not satisfying!!!! • It is not exactly what you like to assert when you construct a confidence interval!! • You would like to say how confident you are that the confidence interval contains the population mean • A frequentist statistician, however, can’t assert that !!!!
Good news • A Bayesian approach turns this around. Because the confidence interval is fixed (by your sample data), a Bayesian statistician can calculate the probability that the population mean (itself a random variable) occurs within the confidence interval. • Bayesians refer to this as: Credibility intervals
More news • Bayesian credibility intervals and frequentist confidence intervals are usually numerically similar if the Bayesian prior probability distribution is uninformative. • Note that: • When the intervals are identical, the choice does not matter. • When the intervals are different, only the Bayesian approach provides logical results.
T-distribution Generalized Confidence Intervals
Some t-distributions: 8 df 4 df 20 df
t- distribution http://www.statsoft.com/textbook/sttable.html#t
Calculating the tail probability: Student’s t table http://www.statsoft.com/textbook/sttable.html#t
Why use R? • Search of Google Scholar from 2002-2008 with the search phrase “R Development” and Publication Name of “Ecology* or Evolution*”
Why learn programming? • “One of the most important things you can do is to take the time to learn a real programming language… • Unfortunately, learning to program is like learning to speak a foreign language—it takes time and practice, and there is no immediate payoff…But if you can overcome the steep learning curve, the scientific payoff is tremendous [emphasis added].” • Excerpted from footnote on p. 116 of Gotelli & Ellison (2004)
Why learn programming? (cont.) • “In offering advice to graduate students in almost any branch of ecology, one of the most important recommendations is to acquire at least some programming skills.” Excerpted from p. 320 of Fortin, M.-J. & M. Dale. 2005. Spatial Analysis: A Guide for Ecologists. Cambridge University Press. NOTE: Gotelli, Ellison, Fortin, and Dale are all field ecologists, not “just theoretical modelers”!