BIO 330 Flashcards Preview

Biology > BIO 330 > Flashcards

Flashcards in BIO 330 Deck (379)
Loading flashcards...

file-drawer problem

studies that are not published- grad thesis, government research


look for effects of study quality, Meta-analysis

do differences in n or methodology matter
-correlation btw n and ES?
-difference in observ. and exp. studies?
-base meta-analysis on higher quality studies


pros of Meta-analysis

tells overall strength & variability of effect
can increase statistical power, reduce Type II error
can reveal publication bias
can reveal associations btw study type and study outcome


cons/challenges of meta-analysis

assumes studies are directly comparable and unbiased samples
limited to accessible studies including necessary summary data
may have higher Type I error if publication bias is present


what do we get out of the statistical process

a probability statement
this process is called Frequentist statistics, most commonly used


What does frequentist statistics do

-answer probability statements if/given the null is true
-infer properties of a population using samples
-doesn't tell if null is true, not proof of anything
-useful, but must understand so not overinterpreted


frequentists statistics developed

Cohen, 1994; Null Hypothesis Sifnificance Testing


why use frequentist statistics

appears to be objective and exact
readily available and easily used
everyone else uses it
scientists are taught to use it
supervisors & journals require it


limits of frequentist statistics

–provides binary info only: significant or not
–does not provide means for assessing relative strength of support for alternate hypotheses
–failing to reject Ho does not mean Ho is true
–does not answer real question


does not provide means for assessing relative strength of support for alternate hypotheses

ex. conclude the slope of the line is not 0, how strong is the evidence that the slope is 0.4 vs 0.5


real question

whether scientific hypothesis is true or false
-treatment has an effect (however small)
-if so, then Ho of no effect is false, but we are unable to show that Ho is false (or true)
-we can only show the probability of getting the data, if Ho is true


question we CAN answer

about the data, not the hypothesis- given the data, how likely is Ho to be true


more limitations for frequentist stats

whether a result is significant depends on n, ES, alpha
significant does not always mean important


larger n, ES, alpha

increase likelihood of rejecting Ho- getting significant result


significant does not necessarily mean important

effects can be tiny and still statistically significant


focus on p-values and Ho rejection

distracts from the real goal- deciding whether data support scientific hypotheses and are practically/biologically important


mostly we should be interested in

size/strength/direction of an effect


Bayesian statistics

incorporate beliefs or knowledge of parameter values into analyses to contain population estimate


frequentists vs. bayesian example

100 coin flips all give 95 heads, what is the probability that the next flip will be a head?
freq.- 50%
bay.- 95%