Regression, Anovas, Generilized Linear Models Flashcards
(14 cards)
Is ecological data normally normally distrubuted?
No
What summerized stats assume normal distribution
CI’S, SD, mean, varience, SE
Why do ecologist still use mean, standard deviation, standard error, and confidence interval, even though they may not have normally distributed data?
- All of these metrics are just ways of summarizing sample, not formal statistical test
- statistics use individual data points like raw data not summarize samples
What is statistical robustness?
That the results of a given statistical test are still reliable, even though the underlying assumptions of the test are not fully met by the data
What does a Levine test use for?
In an anova, it can be helpful to detect equal variances, but this can be overly sensitive because they often respond to non-normality
Which is more important equal appearances or normality in parametric test?
Equal valances is more important than normality
How can you improve equality of appearances?
They are often linked to non-normality such as data with positive skewness can have various increases with mean, we can use log transformation or for transformations to improve data and normality and equal variences
How can we detect outliers?
In the ANOVA we can do box plots in exploratory analysis or in regression analysis we can use cooks D test
How do we deal with outliers?
Always double check data entry. ask yourself if there are errors in the lab work and remove, but you need prior reasoning
- if outline is a combined with non-normality, then transforming the data can normally help
-run analysis with and without them and compare the conclusions
-
Compare box plots, and histogram and what do they checking for?
Histograms allow checking for normality while box plots detect for non-normality, unequal variances, and outliers
What is type i vs a type ii error? Which one is least common?
I) effect detected; none exists (False positive?)
II) effect not detected; but exists - (false negative?)
What is statistical power?
The probability of detecting a given effect in a statisitcal population with our statistical sample (or samples) if it occurs in this population.
Simple version: the probability of detecting a ‘true’ effect when it exists in our data.
Power= 1- B (beta, probability of making a type ii error)
Should be >0.80 (B<0.20)
What does stats power depend on? What increases power?
1) effect size - larger
2) sample size - larger
3) variance - less variation
4) significance level, a (alpha) - typically a= 0.05, power increases when power is set at eg. 0.10. These are not permanet values